Solar Now

Deliveries will begin in 2027 for three utility-scale projects in North Carolina and Texas, US.
Inox Solar Americas has signed an agreement to supply 767MW of photovoltaic (PV) modules to a US renewable energy developer and independent power producer (IPP).
The customer, which was not named, develops, finances, owns and operates utility-scale solar and energy storage projects.
The gold standard of business intelligence.
Find out more
Combine business intelligence and editorial excellence to reach engaged professionals across 36 leading media platforms.
Find out more
The modules will be used for three utility-scale solar developments in North Carolina and Texas. The individual capacities are approximately 71MW, 102MW and 594MW, with deliveries due to start in 2027.
Inox Solar Americas, which manufactures solar PV modules and cells in the US, said the projects would use its Vega Series bifacial modules. These are offered in single-glass and dual-glass configurations and use Galaxion N-Type PV cells.
The company said the modules are designed for utility-scale use and support high levels of US domestic content.
Inox Solar Americas president and CEO Ashok Nair said: “Our customers are looking beyond module performance to domestic content, supply-chain transparency, regulatory compliance, product reliability and long-term bankability.
“This agreement demonstrates our ability to meet these priorities with reliable, high-performance PV modules manufactured in the US.”
Inox Solar’s US manufacturing and sourcing plans are being built to comply with Prohibited Foreign Entity and Foreign Entity of Concern rules, along with domestic content, supply chain traceability and applicable US trade and energy policy requirements.
Measures cited by the manufacturer include tighter supplier qualification, more domestic sourcing, and greater visibility over components and their origin. It said the aim is to help projects meet compliance obligations and to lower supply-chain risk.
Inox Solar Americas stated that the award reflected the technical, commercial, supply chain and risk factors that developers, IPPs, investors and lenders weigh when choosing module suppliers for long-term solar assets.
The contract will serve large project pipelines through domestic manufacturing and long-term customer support.
The gold standard of business intelligence.
Find out more
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Find out more
Give your business an edge with our leading industry insights.
Give your business an edge with our leading industry insights.



Visit our Privacy Policy for more information about our services, how we may use, process and share your personal data, including information on your rights in respect of your personal data and how you can unsubscribe from future marketing communications. Our services are intended for corporate subscribers and you warrant that the email address submitted is your corporate email address.
View all newsletters from across the GlobalData Media network.
Power industry news, data and in-depth articles on the global trends driving power generation, renewables and innovation
Powered by
© GlobalData Plc 2026

source

Controversial plans to build one of the UK's largest solar farms have been delayed further by the government, which says it needs additional time to "consider further information".
Developer Photovolt Development Partners (PVDP) wants to build the Botley West Solar Farm across more than 2,000 acres of land to the north and west of Oxford.
A decision on the scheme had been due earlier this year, before being pushed back to September 2026 by then-Energy Secretary Ed Miliband.
In a statement on Wednesday, the government confirmed that its decision on the proposals would now be further postponed until 10 November at the latest.
On the latest setback, Minister for Local Energy and Jobs Martin McCluskey said it would "enable my department and other interested parties to consider further information received from the applicant".
"The decision to set the new deadline for this application is without prejudice to the decision on whether to grant or refuse development consent," he added.
The final decision on the proposals will fall to Miatta Fahnbulleh, who succeeded Miliband as the secretary of state for energy security and net zero in July.
PVDP previously said the project was crucial to meet the UK's climate and energy security goals, and could power the equivalent of 330,000 homes.
If approved, the £800m solar farm would cover about 1,000 hectares (2,471 acres) across three areas – north of Woodstock, west of Kidlington and west of Botley.
Reacting to the latest delay, the developers said it was a "common part" of the process for a project "of this scale and significance".
"The additional time will allow the department to review the further information submitted and ensure all relevant considerations are properly assessed before a decision is reached," the company said.
"We remain confident in the strength of our application, which has been shaped by extensive consultation and detailed environmental and technical assessment."
But the proposals have proved particularly controversial in the local area, with campaigners previously saying the development would harm an 11km (7-mile) rural corridor.
On Wednesday's further delay, the group Stop Botley West said the government was allowing PVDP "even more time to do its homework and answer longstanding challenges".
This, it said, was an "abuse of what should be a strict and fair planning process".
"We know that hundreds of Oxfordshire residents are angered and frustrated by the further delay and feel unfairly let down by a planning system that we've been asked to respect," the group said.
Bicester and Woodstock MP Calum Miller, who has been a vocal opponent of the plans, said the latest postponement "crystallises what was already clear: this Botley West application is full of holes and should be rejected outright".
"The new secretary of state should reject this proposal and back renewable energy schemes that deliver genuine community benefit and respect the places where they are built," he added.
The scheme aims to reduce energy costs at Fieldhead Hospital in Wakefield.
The use of solar panels and biofuel is on the rise across the island.
Officials announce the huge illegal waste dump near Kidlington has been fully cleared.
People living near the planned solar farm site in South Yorkshire will get longer to have their say.
The UK government claimes panels could help homes "significantly cut" energy bills.
Copyright 2026 BBC. All rights reserved. The BBC is not responsible for the content of external sites. Read about our approach to external linking.
 

source

© 2025 THE COOL DOWN COMPANY. All Rights Reserved. Do not sell or share my personal information. Reach us at hello@thecooldown.com.
The measure also prohibits a range of deceptive conduct, including pretending to be affiliated with a utility or government agency.
Photo Credit: iStock
A broad set of new Texas laws takes effect on September 1, and three of them could be especially noticeable for consumers: stricter oversight of residential solar sellers, expanded repair rights for device owners, and legal-tender status for certain gold and silver bullion.
Together, the measures could shape how Texans buy energy equipment, repair broken electronics, and consider alternative forms of payment.
Starting September 1, companies that sell solar panels in Texas — along with the sales representatives who work for them — must register with the state, according to Lone Star 92.3. Senate Bill 1036 also requires those businesses to carry liability insurance and actively oversee their sales teams.
Some consumer safeguards in that bill are in force. Texans who buy or lease solar equipment must be given 5 days to cancel without penalty, and any financing associated with the transaction must be canceled as well.
Texas will also become the ninth state with a right-to-repair law under House Bill 2963. Makers of certain digital electronic devices must offer owners and independent repair shops the same specialty tools, parts, and manuals that authorized repair providers have access to.
House Bill 1056 also gives qualifying gold and silver bullion legal-tender status in Texas, although private businesses and individuals still are not required to accept bullion as payment.
💡These best-sellers from Quince deliver affordable, sustainable luxury for all
Starting at $50
Starting at $99
Starting at $60
Starting at $80
The solar rules target a fast-growing industry that has also drawn complaints about aggressive sales tactics and confusing contracts.
When devices can be repaired rather than replaced, consumers can save money and keep usable electronics out of landfills. That can also reduce demand for the raw materials and energy needed to manufacture brand-new products.
Not every product is covered by the repair law. Exemptions include gaming consoles, home appliances, motor vehicles, farm equipment, medical devices, and industrial or aerospace technology.
The law also does not require manufacturers to disclose trade secrets or source code, nor do they have to provide tools that would weaken security protections.
For bullion to qualify, it must plainly list its weight and purity, and it cannot include markings that resemble federal minting.
Beginning May 1, 2027, the Texas comptroller may create or approve electronic payment systems for vendors and account holders using currency backed by precious metals stored in the state vault.
Other Texas laws taking effect on September 1 include limits on local bans on manufactured homes, authority for departments to compensate volunteer firefighters, and a broader list of counties officially designated as border counties.
Overall, the September 1 changes mean stronger protections for solar deals, more opportunities to repair electronics rather than replace them, and a legal shift in how certain gold and silver bullion can be used in Texas.
These new consumer laws land amid bigger debates over home energy and how Texas steers its power market. The articles here add context on grid reliability and the state’s growing renewable buildout.
• Across ERCOT, batteries and solar help Texas through punishing summer demand spikes.
• After completing a half-billion-dollar project, Ørsted makes massive donation to support Texas grasslands.
They show how consumer protections fit into a much larger Texas energy story. That wider view can help readers judge what the new solar rules may mean in practice.
Get TCD’s free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.
© 2025 THE COOL DOWN COMPANY. All Rights Reserved. Do not sell or share my personal information. Reach us at hello@thecooldown.com.

source

Located in the municipality of Eurajoki in the Satakunta region on Finland’s west coast, the solar farm is built, owned and operated by Alight and adds new renewable electricity generation to the grid under a long-term virtual power purchase agreement (PPA) with Autoliv. For Autoliv, the agreement supports its transition to renewable electricity in the EMEA region, strengthens long-term energy resilience, and contributes to the company’s ambition to achieve carbon-neutral own operations by 2030.
The project is the first commissioned in Alight’s rapidly growing Finnish portfolio. Alight’s solar and storage pipeline in the country already exceeds 1 GW, and the Eurajoki park alone adds 12 percent to Finland’s entire utility-scale solar capacity, which stood at 842 MWp across 45 parks at the end of June 2026, according to Renewables Finland.
The facility is expected to generate around 100 GWh of renewable electricity annually, equivalent to the consumption of approximately 20,000 households. It is connected to the Finnish electricity system through Caruna’s network via a shared substation.
“Today’s inauguration is an important milestone for Alight in Finland and a strong example of what long-term partnerships can achieve” said Warren Campbell, CEO at Alight. “Together with Autoliv, we have brought one of Finland’s largest solar parks into operation, adding new renewable electricity to the grid and helping demonstrate the role solar can play in the Finnish energy system.”
Long-term corporate PPAs such as this one are becoming an increasingly common route for businesses to secure predictable energy costs while helping bring new renewable capacity to the grid. For Autoliv, the agreement is part of a broader, global climate strategy across its operations.
“At Autoliv, sustainability is integrated into everything we do” added Kaisa Tarna-Mani, Vice President Sustainability at Autoliv. “We are committed to reducing the environmental impact of our operations while continuing to deliver world-class safety products to our customers. By supporting the development of new renewable electricity generation through this PPA, we are taking another important step toward achieving carbon neutrality in our own operations by 2030, and at the same time strengthening the resilience a of our energy supply.”
The Eurajoki Solar Park is made up of two distinct sites, brought together as a single energy asset and linked to a shared substation. The two-site design has allowed the project’s environmental protection measures to be tailored to local conditions at each location. One site’s biodiversity plan includes habitat restoration and improved wetland edges, while a new pond has been created at the other site to support the surrounding ecosystem.
The inauguration brought together representatives from Alight, Autoliv, politicians, local stakeholders, grid partners, financiers and project partners to mark the park’s entry into operation. The project was made possible through 46 million euros of senior debt from banks ABN AMRO and SEB.
For additional information:
Alight
Autoliv

source

Subscribe
Today’s print edition
Home Delivery
For many Europeans, this summer’s heat was a powerful reminder of the damage climate change has wrought — from illness and devastated crops to high energy bills. For Andrew Austin, a 34-year-old homeowner in Shropshire, England, the stubbornly high temperatures were a chance to make some money.
Power demand was rising as London crossed 34 degrees Celsius on June 24. Energy prices spiked as the sun set and solar panels across the U.K. stopped generating electricity. That’s when Austin knew he could take advantage of his 200 kilowatt-hour home battery, which he had charged up using cheap power from the grid during the daytime. He sold the power back to the grid for £60 ($82) from 4 p.m. to 9 p.m. In just a few hours, Austin had earned about as much as an apartment dweller would typically pay for electricity every month.

All this happened without Austin having to press a single button. He has set up a computer to conduct this energy trading while he goes about his life as normal. The computer downloads electricity prices for every 30-minute period from his utility Octopus Energy, which publishes them at 4 p.m. each day. That allows him to know when electricity will be cheap to charge his batteries, and when power prices on the grid are high, the computer sells it back to the grid.
In a time of both misinformation and too much information,
quality journalism is more crucial than ever.
By subscribing, you can help us get the story right.
With your current subscription plan you can comment on stories. However, before writing your first comment, please create a display name in the Profile section of your subscriber account page.
Your subscription plan doesn’t allow commenting. To learn more see our FAQ
Sponsored contents planned and edited by JT Media Enterprise Division.
広告出稿に関するおといあわせはこちらまで
Read more

source

GREW Solar secures INR 430 crore repeat order from leading IPP for high efficiency G12R TOPCon modules across India.
September 04, 2026. By EI News Network

Naveen Gupta, Trev Mobility, Says Premium EV Mobility Must Compete on Reliability and Service

Advanced Coated Steel Can Improve Solar Asset Life, Reduce Maintenance, Says Ranjan Dhar

India’s CNG, CBG and Hydrogen Push Makes it Strategic Market for KonveGas: Alexander Enulescu

GreenLine Mobility CEO Madhur Taneja Explains the Shift Towards Integrated Green Freight

Bondada Group Targets Major BESS Expansion Amid Rising Demand for LDES: Dr. Raghavendra Rao

source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 17, Article number: 6109 (2026) Cite this article
4033 Accesses
Metrics details
Organic photovoltaics (OPVs), with their intrinsic lightweight nature, flexibility, and low energy payback time, are promising power sources for Ambient Internet of Things (A-IoT) nodes. Yet, the large variation in shunt resistance compromises OPV reproducibility, especially for OPV modules operating under low-light environments, which are the typical working conditions of A-IoT nodes. This study reveals that random presence of mesoscale non-fullerene acceptor agglomeration is the primary contributor to leakage current in high-performance OPVs and demonstrates an effective shunt passivation method by applying a large, continuous reverse bias (RB) on as-fabricated devices. OPVs exhibit excellent stability during RB treatment, with leakage current flowing preferentially through shunted regions to generate spatially confined Joule heat, thereby promoting local molecular diffusion to selectively cure mesoscale shunt pathways. The RB-treated module, with an effective area of only 0.24 cm², enables the continuous operation of our self-designed A-IoT temperature sensor under a minimal illuminance of 200 lux, representing the smallest self-powered A-IoT node operating under extremely low-light conditions. Our work presents a universally applicable method to overcome the key practical limitation in OPV module reliability, paving the way towards miniaturised, self-powered A-IoT nodes.
Ambient Internet of Things (A-IoT) refers to an interconnected ecosystem of devices seamlessly integrated into environments that continuously provide data for analysis and decision-making to enhance daily life. Each individual device (referred to as an A-IoT node) is sustainably powered by ambient energy-harvesting technologies, among which photovoltaic (PV) is a widely adopted, technologically mature approach. Given that ambient lighting is often low in intensity and subject to temporal fluctuations (within the 200–1000 lux range), PVs must deliver superior performance in terms of efficiency, reliability, and deployability to meet stringent Quality of Service requirements^1,2,3. In this context, organic photovoltaic (OPV) emerges as a promising third-generation PV technology, utilising non-toxic organic molecules as light absorbers and offering intrinsic lightweight properties, flexibility, and a low energy payback time^1,3,4,5. While power conversion efficiencies (PCEs) of single-junction OPVs have surpassed 20% under outdoor light conditions and 30% under indoor light conditions, they still suffer from poor reproducibility (wide performance statistics) due to significant variations in shunt resistance (R_sh) in as-fabricated devices^6,7. A low R_sh compromises the open-circuit voltage (V_OC) and fill factor (FF) of an OPV device, particularly under low-light indoor conditions where the magnitude of photocurrent becomes comparable to the leakage current⁸. For commercial modules with an increasing number of serially connected subcells, the random presence of shunted subcells further degrades the module’s performance under varying light intensities⁹. Consequently, large-area OPV modules are often employed to ensure sufficient power output, which in turn impedes the downsizing of A-IoT nodes for better deployability^3,10,11,12.
The shunting of OPV is a two-step process involving the thermally activated injection of charge carriers from electrodes, followed by subsequent charge transport through the active layer¹³. Interfacial-induced shunting has been extensively investigated in early-generation ITO-free OPV devices incorporating heavily-doped PEDOT:PSS layers as electrodes. Due to the lack of charge carrier selectivity at the active layer:PEDOT:PSS (anode) interface, those devices exhibit initial ohmic shunting. This can be cured by applying a short reverse-bias pulse (10 ms) to induce electron accumulation, which de-dopes the PEDOT:PSS at the interface and forms a uniform electron-blocking layer, thereby improving the rectifying characteristics of the devices¹⁴. In high-performance OPVs, such interfacial-induced shunting has been eliminated through the employment of electron- and hole-transport layers (ETLs and HTLs) with excellent uniformity and charge carrier selectivity^{13,15,16,17,18}. Therefore, recent works studying indoor OPVs primarily focused on passivating shunt pathways within the bulk of the active layer. This is usually achieved by optimising nanoscale phase-separated structures of the active layer through modifications to molecular structures, additive incorporation, and adjustments to post-annealing conditions^{19,20,21,22,23}. However, since nanoscale morphology is also highly relevant to the generation and extraction of photo-generated charge carriers, those studies often result in case-specific conclusions that lack universal applicability²⁴. On the other hand, the presence of mesoscale shunt pathways, which have been extensively studied for inorganic PVs, remains overlooked for OPVs^25,26.
In this work, we revealed for the first time the presence of mesoscale shunt pathways in high-performance OPV systems and developed a universally applicable method for shunt passivation. Impedance spectroscopy, along with thickness- and composition-dependent dark J–V measurements, confirmed that the magnitude of leakage current in the prototypical PM6:Y6-based OPVs is determined by bulk-limited, filamentary-type charge transport through Y6-rich phases. Upon applying a large and continuous reverse bias (referred to as the RB treatment), we observed significant improvements in R_sh and device reproducibility. Conductive-atomic force microscopy (c-AFM) mappings revealed mesoscale Y6-rich agglomerates exceeding 1 μm in size within shunted devices. During the RB treatment, leakage current preferentially flows through these shunted regions, inducing spatially confined Joule heat that facilitates the local diffusion of Y6 molecules, thereby restoring mesoscale homogeneity. In the meantime, nanoscale phase separation and molecular orientation remain unmodified. The general applicability of our RB treatment is further confirmed across a variety of OPV systems utilising different active and charge-transport layer materials, in both normal and inverted structures. Encouragingly, the RB treatment is particularly effective at curing shunted subcells within OPV modules. With an effective area of only 0.24 cm², our RB-treated OPV module powers our self-designed A-IoT temperature sensor under a minimal illuminance of 200 lux, representing the smallest self-powered A-IoT node operating under extremely low-light conditions. Our work significantly enhances the reliability of OPV devices, particularly in improving module performance under low-light conditions, thereby showcasing their potential as power sources in miniaturised, self-powered A-IoT nodes.
We began with a prototypical OPV system employing a 100 nm blend film of polymer donor PM6 and small-molecule non-fullerene acceptor (NFA) Y6 as the active layer sandwiched within a conventional device structure of ITO/PEDOT:PSS/active layer/PNDIT-F3N/Ag (details of device fabrication are provided in the Methods). As illustrated in Fig. 1a, the reverse bias (RB) treatment involves applying a −10 V bias to the ITO anode (relative to the Ag cathode) in as-fabricated devices under dark conditions for 15 seconds, during which the dark current density undergoes a fast initial drop followed by saturation. Despite the thin active layer, OPVs appear robust during RB treatment with negligible degradation in device performance after prolonging the duration to 1 hour (Fig. S1). This is in stark contrast to perovskite PVs, which show semi-irreversible degradation even under mild reverse bias^27,28. The magnitude of voltage used in the RB treatment (−10 V) has been optimised to maximise the curing effect while reducing the risk of device breakdown, as discussed in detail in Figs. S2–4. In contrast, Figure S5 demonstrates that shunt passivation cannot be achieved under forward bias (FB) treatment; instead, large FB degrades the device performance.
a The real-time dark current density extracted during the RB treatment for 15 s. The setup of the RB treatment is shown in the inset, with −10 V bias applied to the ITO electrode (anode). b Dark J-V curves, c indoor (dashed lines) and outdoor (solid lines) light J-V curves of the same device before (blue lines) and after the RB treatment (orange lines). Statistical data of devices from the same batch: d R_sh and J_leak, e PCE, f V_OC, g FF, and h J_SC under outdoor and indoor light conditions. Light intensity-dependent performance of the same device before and after the RB treatment: i PCE, j V_OC, k FF, and l J_SC. The power-exponent α is shown in the inset of l.
As shown in Figs. 1b, c, S6, dark and light J-V curves were measured for the same batch of 20 devices under outdoor (AM1.5 G) and indoor (2600 K LED with an illuminance of 1000 lux) light conditions before and after the RB treatment, labelled as ‘initial’ and ‘RB’, respectively. The dark current density (J_d) in an OPV device is composed of dark saturation current density (J₀) and leakage current density (J_leak)¹³. By fitting the exponential part of the dark J–V curve using the Shockley equation (Fig. 1b), we obtained the J₀ on the order of 10⁻¹⁰mA cm⁻² and confirmed that J_leak is indeed the main contributor to J_d, typical for OPVs with thin active layers. Therefore, we used J_d measured at −1 V to estimate J_leak and used the inverse differential of the dark J–V curve at 0 V to calculate R_sh. As shown in the statistical data (Fig. 1d–h, Table S1), initial devices suffer from a large scatter in the magnitudes of R_sh and J_leak, which compromises device reproducibility primarily by influencing FF and V_OC. Encouragingly, the J_d of the RB-treated device is reduced by two orders of magnitude (Fig. 1b), while light J-V curves become much more square-like with enhanced FF and V_OC (Figs. 1c, S6). Benefitting from the higher R_sh and suppressed J_leak, RB-treated devices exhibit better average performance with a much narrower distribution than initial devices (Fig. 1e). Compared to outdoor light conditions, the effect of the RB treatment is more pronounced under indoor light conditions (Fig. 1c), where J_leak has a greater effect on device performance due to the much lower photocurrent.
To further understand the roles of shunt pathways on key device metrics, we performed light-intensity (I)-dependent and temperature-dependent (T)-dependent J–V measurements on the same device before and after the RB treatment, as shown in Figs. 1i–l, S7. FF and V_OC drop significantly with decreasing light intensity in the initial device, as previously explained using a simple equivalent circuit model incorporating a finite shunt resistance⁸. In contrast, the V_OC of the RB-treated device strictly follows the ideal kT/qln(I) dependence throughout the measured intensity range, while FF remains almost constant due to effective shunt passivation. On the other hand, the intensity-dependent J_SC curve shows no discernible difference before and after the RB treatment, consistent with the device statistics (Fig. 1h) and results of external quantum efficiency (EQE) measurements (Fig. S8). This is also consistent with our equivalent circuit modelling results, which indicate that J_SC is barely influenced by R_sh when it is larger than 100 Ω⋅cm² (Fig. S9). Recent work has also suggested that high leakage current influences temperature-dependent V_OC measurements, causing an anomalous turnover at low temperatures²⁹. Encouragingly, we found that the RB treatment can also cure such behaviour, restoring the expected near-linear V_OC increase with decreasing temperature (Fig. S7). Overall, the alignment between device statistics, light-intensity-dependent, and temperature-dependent measurements clearly demonstrates the effectiveness of our RB treatment in enhancing the performance and reproducibility of OPV devices, particularly under low-light conditions.
Next, we elucidated the conduction mechanism of J_leak in our devices via thickness-dependent dark J–V measurements. By plotting dark current density against the electric field (Fig. 2a), we observed a gradual reduction in J_leak with increasing active layer thickness. This excludes the possibility of injection-limited J_leak, which is expected to be independent of active layer thickness under the same electric field³⁰. To understand the nature of this bulk-limited conduction mechanism, impedance spectroscopy measurements were conducted at a DC bias of –1 V under dark conditions. As shown in Fig. 2b, at frequencies above 10⁴Hz, where the impedance is dominated by capacitive response, the Bode (phase) plots of both the initial and RB-treated devices merge. This is consistent with the capacitance spectra of initial and RB-treated devices, which show identical magnitudes and slopes within the measured frequency range (Fig. 2c), ruling out the potential contribution of bulk traps to J_leak^31,32. At frequencies below 10⁴Hz, where the impedance is dominated by R_sh, the phase angle of the RB-treated device remains at around −90° while that of the initial device shows significant deviation due to insufficient R_sh. Consistently, the initial device shows a much smaller semicircle radius compared to the RB-treated device in Nyquist plots (Fig. 2d). Those results point out that J_leak in PM6:Y6 devices is dominated by bulk-limited charge transport at local shunted regions of the active layer, also known as filamentary-type conduction³¹. To determine the composition of conductive filaments, we systematically varied the D:A ratios of the active layer while maintaining a constant thickness of 100 nm to monitor the change in J_leak³³. Fig. 2e, f demonstrate that the increase of Y6 content led to a reduction of R_sh and an increase of J_leak for both initial and RB-treated devices. This indicates that the Y6-rich phase functions as local conductive filaments for J_leak.
a J_d of RB-treated devices with different active layer thicknesses plotted against the electric field. b Bode (phase) plots, c capacitance spectra, and d Nyquist plots of initial and RB-treated devices at a DC bias of −1 V. e Statistics of R_sh for initial and RB-treated devices with different D/A ratios. f Dark J-V curves of typical devices with different D/A ratios.
To investigate the curing effect of the RB treatment on Y6-rich conductive filaments, c-AFM measurements were performed on initial and RB-treated devices after peeling off the top electrode and electron transport layer (ETL) using tape, thereby exposing the top surface of the active layer, as shown in Fig. 3a. The PM6:Y6 device processed with chlorobenzene (CB) was studied first (dark J–V curves shown in Fig. 3b and photovoltaic performance shown in Fig. S10, Table S2), as this system exhibits stronger phase segregation, allowing more confident assignments of PM6- and Y6-rich phases³⁴. By applying a positive bias to the PEDOT:PSS/ITO substrate for hole injection, and using a grounded c-AFM tip to collect holes that reach the top surface, our c-AFM measurement probes the difference in the 3-D hole transport network within the bulk active layer³⁵. Considering interfacial energetic alignment (Fig. S11), holes can be effectively injected from the ITO/PEDOT:PSS substrate to PM6-rich phases due to its shallower highest occupied molecular orbital (HOMO), while the injection current falls significantly in the Y6-rich phases with a much deeper HOMO, as confirmed by c-AFM mappings of pure PM6 and Y6 films (Fig. S12). Therefore, the increased magnitude of (negative) hole current upon the RB treatment, as shown in Fig. 3g, indicates the formation of more interconnected PM6-rich hole-transport pathways. To better visualise the morphology change, we re-rendered the c-AFM mappings (Fig. 3c, d) using a three-colour scale (Fig. 3e, f). As shown in Fig. 3c, d, c-AFM mappings of both initial and RB-treated devices exhibit bright spherical regions with local currents between 0 and −17 pA. Those regions, rendered white, are assigned to Y6 crystalline domains with a size of around 50 nm (Fig. S13), consistent with our previous works³⁴. In the initial device, there are large, interconnected low-current regions with sizes exceeding 1μm surrounding Y6 crystalline domains. Those regions, with local current between −17 pA and −34 pA (the intersection point between the current histograms of the initial and the RB-treated devices), are assigned to the mesoscale Y6-rich agglomerates and rendered grey. At last, regions with the magnitudes of local current exceeding 34 pA, which appear predominantly in the RB-treated device, are rendered black. Those regions are assigned to homogenised mesoscale phases with more interconnected PM6-rich hole-transport pathways, thereby suppressing Y6-rich shunt pathways. Based on the re-rendered c-AFM mappings (Fig. 3e, f), it becomes clear that the main impact of the RB treatment is to annihilate mesoscale Y6-rich agglomerations that function as local shunt pathways. Consistent results were also obtained when the mapping area was increased from 2 × 2 µm² to 5 × 5 µm² (Fig. S14).
a Setup of c-AFM measurement. The positive bias was applied to the substrate for hole injection, while the grounded probe was placed on the top surface of the active layer after peeling off the top electrode and ETL. b Dark J-V curves of PM6:Y6 (CB) devices without and with the RB treatment for c-AFM measurements. The current mappings (c) without and (d) with the RB treatment. e and f The re-colorized c-AFM mappings corresponding to c and d. The orange box in (e) with a size of 1 µm×1 µm is used to highlight the presence of mesoscale Y6-rich agglomerates. g The current histograms of two c-AFM mappings. The corresponding micro-PL spectra are shown in (h).
To distinguish our RB treatment from the well-known electrical annealing method, which affects nanoscale morphology and the molecular orientation, we measured micro-photoluminescence (PL) spectra on initial and RB-treated devices, which completely overlap with each other, as shown in Fig. 3h^{36,37,38,39,40,41}. This is consistent with the results of J_SC statistics (Fig. S10d), light-intensity-dependent J_SC (Fig. S10i), and EQE measurements (Fig. S15), confirming that the nanoscale phase separation, which governs the yield of exciton dissociation and charge generation, is unmodified by RB treatment²⁴. Additionally, the topography and surface potential (SP) mappings obtained by tapping-mode AFM and Kevlin Probe Force Microscopy (KPFM, Fig. S16) also show no discernible difference with and without the RB treatment. Due to the large molecular quadrupole moment associated with Y-series NFAs, any change in molecular orientation is expected to result in a notable change in the SP of the film⁴². Therefore, the identical SP indicates that the molecular orientation remains unchanged after the RB treatment. Finally, optical microscopy images of the initial and RB device appear identical and smooth (Fig. S17), which excludes the potential impact of macroscopic defects, such as pinholes, on R_sh. Consistent results were also observed in high-performance CF-processed PM6:Y6 devices, as shown in Fig. S18–S21.
Based on the above results, we propose the fundamental mechanism of RB treatment, as illustrated in Fig. 4. The random presence of mesoscale Y6-rich agglomerates is the key contributor to the large variation of R_sh in initial devices. This likely arises from the much worse rectifying characteristic of Y6 than PM6 (Fig. 2e, f), due to the smaller bandgap of Y6 as well as its stronger tendency to aggregate (via face-on π-π stackings)⁴³. During the RB treatment, large leakage current preferentially flows through those Y6-rich agglomerates, which generates spatially confined Joule heat to induce local diffusion of Y6 molecules, selectively annihilating those mesoscale shunted pathways. In the meantime, the nanoscale morphology and molecular orientation are unaffected by the RB treatment, maintaining efficient photocharge generation and extraction. In contrast, the joule heat generated under FB treatment is uniformly distributed across the entire active layer, which, as proposed by Maria et al., is equivalent to thermally annealing the device at the same temperature⁴¹. In fact, the excessive joule heat generated under the FB treatment, which is over five orders of magnitude higher than that under the RB treatment (as determined by comparing the magnitudes of current flow during the RB treatment at −5 V shown in Fig. S3b and the FB treatment at 5 V shown in Fig. S5a), can degrade the device performance. Therefore, the unique advantage of the RB treatment is that it leverages the rectifying characteristics of the diode structure, allowing leakage current to selectively anneal and cure the local shunted region.
a The initial D/A network. b The molecule diffusion due to the thermal gradient during the RB treatment. c The D/A network after the RB treatment.
To assess the general applicability of our RB treatment, we applied it to a wide range of NFA-OPV systems, with different active layer materials and charge transport layers, in both normal and inverted structures, as summarised in Fig. S22 and Table S3. Before the RB treatment, all devices showed scattered R_sh values, which resulted in large variations in indoor photovoltaic performance. For BTRCl:Y6, small shunt resistance readily harms outdoor performance, so the corresponding indoor performance was not measured. Promisingly, after RB treatment, all devices showed improved photovoltaic performances, along with much better statistics. The general applicability of the RB treatment on NFA-OPV systems with vastly different optoelectronic and morphological properties suggests that the presence of mesoscale inhomogeneity within the bulk active layer is a universal characteristic of solution-processed NFA-OPVs. Therefore, the RB treatment can be employed as a standard post-treatment method to improve the reproducibility of high-performance NFA-OPVs, especially for those designed for indoor applications.
To evaluate the effectiveness of our RB treatment at the module level, we implemented it in our custom-designed OPV module comprising four 0.06 cm² subcells connected in series with a device structure of ITO/2PACZ/PM6:L8BO (100 nm)/PDINN/Ag, as shown in Fig. 5a, b. Under outdoor light conditions, the initial module shows compromised FF due to an ‘early turn-on’ in current below the built-in voltage, as shown in Fig. 5c (blue solid line) and Fig. S23a. Under indoor light conditions (Fig. 5c, blue dashed line), the initial module shows a more significant reduction in PCE due to an additional V_OC loss. To examine the origin of inferior module performance, we performed separate dark J–V measurements on each of the four subcells. As shown in Fig. 5e and Table S4, four subcells exhibit vastly different degrees of shunting with R_sh ranging from 4 Ω⋅cm² to 5.4 × 10⁵ Ω⋅cm². To understand the underlying mechanism, we simulated the module J–V curves under various light intensities using a four-diode model as shown in Fig. S24a. We set the R_sh of three diodes to 1 × 10⁶ Ω⋅cm², which is close to the R_sh measured in RB-treated devices, and varied the R_sh of the fourth diode (R_sh4) from 1 × 10⁶ to 10 Ω⋅cm². As R_sh4 decreases, the leakage current first influences the diode current near V_OC, resulting in the “early turn-on” observed in the outdoor light J–V curve that compromises FF (Figs. 5d, S24b–e). Upon further decreasing light intensities and R_sh4, the magnitude of photocurrent becomes comparable to the leakage current in the fourth diode, so it behaves like a resistor, giving rise to the additional V_OC loss observed under indoor light conditions. Therefore, our simulation suggested that the large variation of R_sh among subcells is the main cause of the compromised OPV module performance under outdoor and indoor light conditions. By applying the RB treatment to each subcell, their average R_sh was significantly increased to over 10⁶ Ω⋅cm², and dark current density was decreased by several orders of magnitude (Fig. 5e, Table S4), leading to a largely suppressed dark current within the entire module, as shown in Fig. 5f. As a result, light J–V curves of the module become square-like under both outdoor and indoor conditions, with significant improvements in all photovoltaic parameters (Figs. 5g, S23, S25, Table S5), consistent with the simulation results. We further measured the performance of the RB-treated OPV module within a wide illuminance range from 4000 to 200 lux (Fig. S26, Table S5). Remarkably, the maximum power output (P_max) of the RB-treated module maintains a nearly linear proportionality against the illuminance (Fig. 5h), showcasing its excellent reliability.
a, b The photo and schematic of the OPV module. c Indoor (dashed lines) and outdoor (solid lines) light J-V curves before (blue lines) and after the RB treatment (orange lines). d The simulated J-V curves under light intensity of 1 mW⋅cm⁻² using a four-diode model. Dark J-V curves of e each subcell and f the whole module before (blue lines) and after (orange lines) the RB treatment. g Statistical P_max of 8 independent modules before and after the RB treatment. h P_max of an RB-treated module plotted against the illuminance. i System-level block diagrams of the A-IoT temperature sensor integrated with an OPV module. j Real-time temperature monitoring under different illuminances. k The real-time transmission of temperature and location data to a smartphone. l A comparison of illuminance and the area of PV module used as power supply for A-IoT nodes in the last five years, with the demonstration in this work. Detailed parameters are included in Table S7. For fair comparison, we include both the total area of our module (1.08 cm²) and the active area (0.24 cm²). The solid line represents a constant illuminance–module size product, serving as a visual guideline.
To demonstrate real-world applicability, we employed our RB-treated OPV module as the power source in a miniaturised, self-powered A-IoT node. The power consumption of typical A-IoT nodes is on the mW-scale, often requiring panel-scale PV modules with an effective area of tens of cm² (Table S7), which constrains downsizing and deployment^10,12,44. Herein, we designed a BLE (Bluetooth Low Energy)-equipped temperature sensor with an ultra-low power consumption of only 4 µW. This is achieved by avoiding the use of high-consumption components, such as DC-DC converters or low-dropout (LDO) regulators, in our circuit design. Instead, we utilise the simplest dual Under-Voltage Lockout (UVLO) circuit to manage energy flow (Fig. S27), which fundamentally eliminates unnecessary circuit complexity and switching losses. The temperature sensor and an RB-treated OPV module with an active area of only 0.24 cm² are integrated at a micro-patch scale (Fig. 5i), achieving self-powered operation within a wide illuminance range from 4000 to 200 lux (Fig. 5j, S28). During operation, data collected from the device is delivered via the BLE beacon to a smartphone and synchronised to the cloud (Fig. 5j, k). The minimal execution interval of around 200 ms can be achieved for illuminance above 250 lux, while the average execution interval decreases continuously with increasing illuminance, reaching 566 ms at 4000 lux (Fig. S29). Remarkably, under an extremely low illuminance of 200 lux, at which the P_max of our OPV module (4.62 µW, see Table S5) is close to the startup power of our A-IoT node (4 µW), the node still operates continuously with a decent average execution interval of 14 s and a minimal execution interval of 12 s, as shown in Table S6. By benchmarking our results with previous works (Fig. 5l), it becomes apparent that our prototype represents the smallest self-powered A-IoT node that can operate under extremely low-light conditions. The downsizing of A-IoT nodes, achieved through the joint efforts of our RB treatment and innovative circuit design, will enable the transition from panel-scale deployment to sticker-scale deployment, representing a crucial step towards further expanding the application scenarios of A-IoTs.
Overall, our work identifies mesoscale NFA agglomerations as the primary contributors to leakage currents in high‑performance OPVs. Such mesoscale inhomogeneity randomly presents within solution‑processed active layers, leading to scattered shunt resistance and compromised device reproducibility. This effect is particularly detrimental under low-light indoor environments, which are the typical working conditions for A-IoT nodes. To address this challenge, we developed a simple yet effective shunt-passivation method, known as the RB treatment. Despite the thin active layer, OPVs remain robust during RB treatment, as leakage current preferentially flows through shunted NFA filaments to generate spatially confined Joule heat, which induces local molecular diffusion to annihilate shunt pathways. Unlike conventional post‑treatments, RB treatment selectively cures mesoscale shunted regions without disturbing nanoscale morphology and molecular orientation, making it broadly compatible with diverse fabrication protocols. Since light-intensity-dependent and temperature-dependent J–V measurements that are widely applied to study the recombination kinetics and energetics of OPVs can be influenced by shunt pathways, the RB treatment is suggested to be performed before those measurements to eliminate the random shunting effect and thereby better reveal the intrinsic properties of the materials systems under investigation. Finally, the practical utility of RB treatment is further demonstrated by our OPV module with an effective area of only 0.24 cm², which powers our self-designed BLE-equipped temperature sensor under an extremely low illuminance of 200 lux. By ensuring predictable power delivery at sub‑cm² scale while sustaining standard‑protocol connectivity, our approach advances passive labels into active endpoints and transitions OPV-IoT from pilot‑scale demonstrations toward pervasive, sticker‑scale infrastructure.
Chloroform (CF), chlorobenzene (CB), 1-chloronapthalene (CN), Zn(CH₃COO)₂·2H₂O, 2-Methoxyethanol, ethanolamine, and MoO₃ were purchased from Sigma-Aldrich. 1,4-diiodobenzene (DIB) and 2PACz were purchased from Meryer Co., LTD. PM6 and Y6 were purchased from Solarmer Inc. (Beijing). BTRCl, L8BO, Y6-1O, PDINN, and PDINO were purchased from Derthon Optoelectronic Materials Science Technology Co., Ltd. PNDIT-F3N was purchased from eFlexPV Ltd. PEDOT:PSS (Al 4083) was purchased from Heraeus Ltd. All chemicals were used as received without further purification.
For conventional devices and homemade modules, the organic photovoltaics (OPVs) were fabricated with the structure of ITO/hole transport layer (PEDOT:PSS or 2PACz)/active layer/electron transport layer (PNDIT-F3N, PDINN, or PDINO)/Ag (100 nm). The ITO substrates were sequentially ultrasonicated for 20 min using detergent, deionized water, acetone, and isopropanol, and then treated with UV-ozone for 20 min. PEDOT:PSS was spin-coated onto ITO at 4000 rpm for 30 s and dried at 120 ˚C for 20 min in ambient. 2PACz (0.3 mg/ml) in ethanol was spin-coated onto ITO at 3000 rpm for 30 s after 15 s resting and heated at 100 ˚ C for 10 min in the N₂-filled glovebox. The material for the active layer was dissolved in different solvents. For PM6:Y6 and PM6:Y6-1O-based devices, materials were dissolved in CF with 0.5 vol% CN as an additive. For PM6:BTP-eC9, PM6:L8BO, and BTRCl:Y6-based devices, materials were dissolved in CF with 10 mg/ml DIB as an additive. For PM6:L8BO-based indoor devices, materials were dissolved in CF without an additive. For PM6:Y6 (in CB)-based devices, materials were dissolved in CB with 0.5 vol% CN as an additive. The D/A ratio is 1:1.2, except the BTRCl:Y6 (1.7:1). The active layer was spin-coated at 3000 rpm for 30 s and annealed at 100 ˚ C for 5 mins. The active layer, composed of PM6:L8BO without an additive, was annealed at 80 ˚ C for 5 minutes. PNDIT-F3N (0.5 mg/ml with 0.5 vol% ethanoic acid), PDINN (1 mg/ml), or PDINO (0.5 mg/ml) in methanol was deposited at 2000 rpm for 30 s. Then, the Ag electrode with a thickness of around 100 nm was thermally evaporated at 10^-4 Pa.
For an inverted device, OPVs were fabricated with the structure of ITO/ZnO/active layer/MoO₃(2.6 nm)/Ag (100 nm). The ZnO precursor was prepared by dissolving Zn(CH₃COO)₂·2H₂O (100 mg) in 2-methoxyethanol (973 µL) with ethanolamine (28.29 µL). After fully mixing, the precursor was stirred at 60 ˚C for 10 min, followed by stirring at room temperature overnight. The ZnO layer was spin-coated onto ITO substrates at 4000 rpm for 30 s in air and annealed at 200 ˚ C for 30 min. The active layer was deposited as the conventional device. Then, the MoO₃ layer and the Ag electrode were thermally evaporated at 10⁻⁴ Pa.
The current density-voltage (J–V) curves of devices were measured by a Keithley 2400 Source Metre in a N₂-filled glove box under various light sources. For the outdoor condition, a solar simulator (SS-F5-3A, Enlitech) with AM 1.5G (100 mW⋅cm⁻²) spectrum was used, and the intensity was calibrated by a reference silicon solar cell (SRC2020, Enlitech). For the indoor condition, a white LED (iwata M1 Pro RGB Mini, iwata Tech) with adjustable colour temperature and intensity was used. The illuminance was calibrated by a light metre (TES-1334N, TES Electrical Corp.). The external quantum efficiency (EQE) was measured by a QE/IPCE system (Enli Technology Co. Ltd., China) in a wavelength range of 300−1000 nm. The thickness of the active layer was measured by a profilometer (Bruker Dektak XT).
Current density–voltage characteristics were measured using an automatic photovoltaic efficiency measurement system equipped with a commercial solar simulator (LIV-1220, LightSky Technology Co., Ltd.). For temperature-dependent measurements, the devices were mounted on a Linkam HFS600E-PB4 heating/freezing stage integrated into the optical path of the measurement system. The stage provides a temperature range from −196 to 600 °C with a temperature stability of 0.1 °C. During the measurements, the sample chamber was continuously purged with dry nitrogen to minimise moisture accumulation and ice condensation at low temperatures. At each target temperature, the stage was set to the desired value and held until thermal equilibrium was reached, as confirmed by a temperature fluctuation within ±0.1 °C, after which the J–V curve was recorded. All measurements were performed under a fixed one sun intensity condition.
The impedance spectroscopy was measured by ZAHNER ZENNIUM Electrochemical Workstation. All devices were encapsulated and measured in air. During the measurement, a 10 meV AC bias was applied to the device with frequency scanning from 4 MHz to 100 Hz at the reverse bias (−1.0 V) to obtain the complex impedance of the device under dark conditions. The capacitance spectrum was derived from the complex impedance using Eq. 1³²:
where Z’ and Z” are the real and imaginary parts of the complex impedance, ω is the angular frequency, R_s is the series resistance of the devices derived from the dark J-V curve, and L is the parasitic inductance.
All AFM images were conducted using the JPK NanoWizard NanoOptics from Bruker. All samples were fabricated by peeling off the top electrode and electron transport layer using tape and measured in air. Topography images and surface potential mapping were taken through Klevin force probe microscopy on TappingMode^tm technology using the conductive ElectriMulti75-G probe (Pt overall coating, Budget Sensors) in the dark condition. c-AFM images were taken under the contact mode via the conductive ElectriCont-G probe (Pt overall coating, Budget Sensors). The 2 V bias was applied to ITO to inject holes into the active layer in the dark conditions.
The Renishaw inVia Qontor Micro Raman was used to measure PL spectroscopy with a 785 nm excitation laser. The sample was fabricated by peeling off the electron transport layer and Ag electrode to expose the surface of the active layer for measurement.
The Olympus BX60 with AxioCam MRc 5 (ZEISS) and 10× objective lens (Olympus) was used to measure optical microscopy. The sample was fabricated by peeling off the electron transport layer and Ag electrode to expose the surface of the active layer for measurement.
The four-diode model was simulated module J–V curves using MATLAB Simulink. The solar cell block was parameterized by s/c current and o/c voltage, 5 parameters. A PS constant block was used to simulate the light source. A resistor block placed in parallel with the solar cell block was used to simulate the shunt resistance. A resistor block placed in series with the solar cell block was used to simulate the series resistance. A Piecewise Linear Voltage Source block was used to generate a voltage for the circuit from −5V to 5 V. A current sensor, a voltage sensor, and a scope were used to measure the voltage and current in the circuit.
The A-IoT sensor circuit consists of three main components: the Energy Harvester Unit (EHU), the Energy Management Unit (EMU), and the Energy Utilization Unit (EUU). The EMU includes two capacitors for energy storage: a (1,mu {{rm{F}}}) capacitor, which ensures the normal operation of the energy management circuit, and a (47,mu {{rm{F}}}) capacitor, denoted as ({C}_{{{rm{sto}}}}), which serves as the primary energy reservoir for the EUU. The EMU also integrates a hysteresis comparator, a window comparator, and three switches that control the energy flow. The hysteresis comparator features high and low threshold voltages (({V}_{{{rm{DET}}}})) of 1.55 V and 1.45 V, respectively. When the voltage at the EHU input to the EMU reaches the upper threshold of the hysteresis comparator (1.55 V), switch SW2 (Fig. S27) is activated, establishing a direct connection between the EHU and the (47,mu {{rm{F}}}) storage capacitor. At this point, the (1,mu {{rm{F}}}) capacitor maintains the EHU voltage and continues to power the energy management circuit. Conversely, when the EHU voltage falls below the lower threshold of the hysteresis comparator (1.45 V), the internal control circuit deactivates SW2, preventing reverse discharge from the storage capacitor to the EHU, and ensuring readiness for the next power cycle. The window comparator embedded in the circuit has a high threshold voltage (({V}_{{{rm{thH}}}})) of 2.13 V and a low threshold voltage (({V}_{{{rm{thL}}}})) of 1.29 V. When the storage capacitor voltage (({V}_{{{rm{storage}}}})) exceeds 2.13 V, the internal control circuit engages SW3 while simultaneously disengaging SW2, thereby ensuring that power is drawn exclusively from the storage capacitor to supply the EUU. As the storage capacitor voltage decreases to an intermediate value within the high-low threshold range, SW2 is reactivated, allowing both the EHU and storage capacitor to supply power to the EUU concurrently. SW3 remains engaged until the storage capacitor voltage falls below 1.29 V, at which point it is deactivated, effectively disconnecting the EUU from the power supply. The total energy delivered by the EMU to the EUU can be approximated as follows, yielding a value of approximately 67.88 μJ:
By measuring the capacitor voltage under varying input power conditions, it was determined that a 47 μF capacitor can be charged to 2.13 V within approximately 28.9 seconds, completing a full cold start of the entire circuit. Based on the following Eq. 2, the minimum average input power requirement ({P}_{{average}}) is calculated to be 3.71 μW. Taking into account potential power losses at the EHU and EMU interfaces, the estimated minimum average input power required is approximately 4 μW.
In the EUU, the control circuit acquires temperature data from an onboard sensor. During the continuous operation of the EUU, it cyclically executes tasks such as temperature data acquisition and transmission via the Bluetooth protocol. The execution interval is configurable, with a default setting of 100 ms, and persists until the EUU is powered down. The collected data is broadcast to a mobile device. The mobile device synchronizes the temperature data with local positioning information, uploads it to the cloud, and displays it on an app. This process provides users with real-time temperature readings and location-based insights.
This mechanism ensures precise control of the energy delivered to the EUU during each cycle. Notably, the EMU does not incorporate conventional switching circuits. This design choice not only reduces the overall circuit board footprint but also enhances the integration of Bluetooth and other RF functionalities within compact devices.
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
The data supporting the findings of this study are available within the main text and the Supplementary Information. Additional data are available from the corresponding authors upon request.
Müller, D. et al. Indoor photovoltaics for the Internet-of-Things – a comparison of state-of-the-art devices from different photovoltaic technologies. ACS Appl. Energy Mater. 6, 10404–10414 (2023).
Article  ADS  Google Scholar 
Lübke, D., Hartnagel, P., Hülsbeck, M. & Kirchartz, T. Understanding the thickness and light-intensity dependent performance of green-solvent processed organic solar cells. ACS Mater. Au 3, 215–230 (2023).
Article  PubMed  PubMed Central  Google Scholar 
Grandhi, G. K. et al. Promises and challenges of indoor photovoltaics. Nat. Rev. Clean. Technol. 1, 132–147 (2025).
Article  Google Scholar 
Pecunia, V., Occhipinti, L. G. & Hoye, R. L. Z. Emerging indoor photovoltaic technologies for sustainable Internet of Things. Adv. Energy Mater. 11, 2100698 (2021).
Article  CAS  Google Scholar 
Wang, Z. et al. Mechanically robust and stretchable organic solar cells plasticized by small-molecule acceptors. Science 387, 381–387 (2025).
Article  ADS  CAS  PubMed  Google Scholar 
Zhang, R. et al. Equally high efficiencies of organic solar cells processed from different solvents reveal key factors for morphology control. Nat. Energy 10, 124–134 (2025).
Article  ADS  CAS  Google Scholar 
Zhou, X. et al. Over 31% efficient indoor organic photovoltaics enabled by simultaneously reduced trap-assisted recombination and non-radiative recombination voltage loss. Mater. Horiz. 10, 566–575 (2023).
Article  CAS  PubMed  Google Scholar 
Proctor, C. M. & Nguyen, T.-Q. Effect of leakage current and shunt resistance on the light intensity dependence of organic solar cells. Appl. Phys. Lett. 106 https://doi.org/10.1063/1.4913589 (2015).
Kumar, R. & Gupta, R. Shunts in crystalline silicon PV modules: A comprehensive review of investigation, characterization, and mitigation. Sol. Energy Mater. Sol. Cells 277, 113121 (2024).
Article  CAS  Google Scholar 
Burwell, G. et al. Scaling considerations for organic photovoltaics for indoor applications. Sol. RRL 6, 2200315 (2022).
Article  CAS  Google Scholar 
Cortese, A. J. et al. Microscopic sensors using optical wireless integrated circuits. Proc. Natl. Acad. Sci. 117, 9173–9179 (2020).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Jahandar, M., Kim, S., Kim, Y. H. & Lim, D. C. Large-area wide bandgap indoor organic photovoltaics for self-sustainable IoT applications. Adv. Energy Sustain. Res. 4, 2200117 (2023).
Article  CAS  Google Scholar 
Dongaonkar, S. et al. Universality of non-Ohmic shunt leakage in thin-film solar cells. J. Appl. Phys. 108. https://doi.org/10.1063/1.3518509.
Larsen-Olsen, T. T., Søndergaard, R. R., Norrman, K., Jørgensen, M. & Krebs, F. C. All printed transparent electrodes through an electrical switching mechanism: A convincing alternative to indium-tin-oxide, silver and vacuum. Energy Environ. Sci. 5, 9467–9471 (2012).
Article  CAS  Google Scholar 
Ma, L.-K. et al. High-Efficiency Indoor Organic Photovoltaics with a Band-Aligned Interlayer. Joule 4, 1486–1500 (2020).
Article  CAS  Google Scholar 
Phuong, L. Q. et al. Quantifying quasi-fermi level splitting and open-circuit voltage losses in highly efficient nonfullerene organic solar cells. Sol. RRL 5, 2000649 (2021).
Article  CAS  Google Scholar 
Wang, J. et al. Intrinsically stretchable organic photovoltaics by redistributing strain to PEDOT:PSS with enhanced stretchability and interfacial adhesion. Nat. Commun. 15, 4902 (2024).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Jayaraman, E. et al. Flexible ITO-free organic solar modules using fully roll-to-roll processable top illumination design. Adv. Energy Mater, e04465 https://doi.org/10.1002/aenm.202504465.
Zhou, X. et al. Different morphology dependence for efficient indoor organic photovoltaics: the role of the leakage current and recombination losses. ACS Appl. Mater. Interfaces 13, 44604–44614 (2021).
Article  CAS  PubMed  Google Scholar 
Go, E. et al. Unraveling the origin of dark current in organic bulk heterojunction photodiodes for achieving high near-infrared detectivity. ACS Photonics 9, 2056–2065 (2022).
Article  CAS  Google Scholar 
Park, S. et al. High-performance and stable nonfullerene acceptor-based organic solar cells for indoor to outdoor light. ACS Energy Lett. 5, 170–179 (2020).
Article  CAS  Google Scholar 
Zhang, T. et al. A medium-bandgap nonfullerene acceptor enabling organic photovoltaic cells with 30% efficiency under indoor artificial light. Adv. Mater. 34, 2207009 (2022).
Article  CAS  Google Scholar 
Li, M., Li, Z., Wang, M., Tang, Z. & Ma, Z. Suppressing shunt and trap-assisted recombination in organic photovoltaic devices for improved indoor light harvesting efficiency. ACS Appl. Energy Mater. 7, 11900–11909 (2024).
Article  CAS  Google Scholar 
Snaith, H. J., Arias, A. C., Morteani, A. C., Silva, C. & Friend, R. H. Charge generation kinetics and transport mechanisms in blended polyfluorene photovoltaic devices. Nano Lett. 2, 1353–1357 (2002).
Article  ADS  CAS  Google Scholar 
Li, W. et al. Sparkling hot spots in perovskite solar cells under reverse bias. ChemPhysMater 1, 71–76 (2022).
Article  CAS  Google Scholar 
Simon, M. & Meyer, E. L. Detection and analysis of hot-spot formation in solar cells. Sol. Energy Mater. Sol. Cells 94, 106–113 (2010).
Article  CAS  Google Scholar 
Najafi, L. et al. Reverse-bias and temperature behaviors of perovskite solar cells at extended voltage range. ACS Appl. Energy Mater. 5, 1378–1384 (2022).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Jiang, F. et al. Improved reverse bias stability in p–i–n perovskite solar cells with optimized hole transport materials and less reactive electrodes. Nat. Energy 9, 1275–1284 (2024).
Article  ADS  CAS  Google Scholar 
Tang, Y. et al. Origin of open-circuit voltage turnover in organic solar cells at low temperature. Sol. RRL 4, 2000375 (2020).
Article  CAS  Google Scholar 
Chiu, F.-C. A review on conduction mechanisms in dielectric films. Adv. Mater. Sci. Eng. 2014, 578168 (2014).
Article  Google Scholar 
Nau, S., Sax, S. & List-Kratochvil, E. J. W. Unravelling the nature of unipolar resistance switching in organic devices by utilizing the photovoltaic effect. Adv. Mater. 26, 2508–2513 (2014).
Article  CAS  PubMed  Google Scholar 
Brus, V. V. On impedance spectroscopy analysis of nonideal heterojunctions. Semicond. Sci. Technol. 27, 035024 (2012).
Article  ADS  Google Scholar 
Keivanidis, P. E., Ho, P. K. H., Friend, R. H. & Greenham, N. C. The dependence of device dark current on the active-layer morphology of solution-processed organic photodetectors. Adv. Funct. Mater. 20, 3895–3903 (2010).
Article  CAS  Google Scholar 
Fu, Y. et al. Enhancing inter-domain connectivity by reducing fractal dimensions: the key to passivating deep traps in organic photovoltaics. Energy Environ. Sci. https://doi.org/10.1039/D4EE02961E.
Fernando, P. S. & Mativetsky, J. M. Unambiguous measurement of local hole current in organic semiconductors using conductive atomic force microscopy. J. Phys. Chem. C. 127, 9903–9910 (2023).
Article  CAS  Google Scholar 
Sentein, C., Fiorini, C., Lorin, A. & Nunzi, J.-M. Molecular rectification in oriented polymer structures. Adv. Mater. 9, 809–811 (1997).
Article  CAS  Google Scholar 
Sentein, C. et al. Poling induced improvement of organic polymer device efficiency. Synth. Met. 102, 989–990 (1999).
Article  CAS  Google Scholar 
Padinger, F., Rittberger, R. S. & Sariciftci, N. S. Effects of postproduction treatment on plastic solar cells. Adv. Funct. Mater. 13, 85–88 (2003).
Article  CAS  Google Scholar 
Li, Y. et al. Thermal treatment under reverse bias: Effective tool for polymer/fullerene bulk heterojunction solar cells. Synth. Met. 158, 190–193 (2008).
Article  CAS  Google Scholar 
Devi, B. P., Thiyagu, S. & Pei, Z. Electrical annealing” effect in bulk heterojunction polymer solar cells. Thin Solid Films 529, 54–57 (2013).
Article  ADS  CAS  Google Scholar 
Méndez, M., Fernández, D., Viterisi, A., Martínez-Ferrero, E. & Palomares, E. Joule-heating annealing to increase organic solar cells performance: a comparative study. Appl. Sci. 12, 2552 (2022).
Article  Google Scholar 
Fu, Y. et al. Molecular orientation-dependent energetic shifts in solution-processed non-fullerene acceptors and their impact on organic photovoltaic performance. Nat. Commun. 14, 1870 (2023).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Liao, X. et al. Regulating crystallinity mismatch between donor and acceptor to improve exciton/charge transport in efficient organic solar cells. Angew. Chem. Int. Ed. 63, e202318595 (2024).
Article  CAS  Google Scholar 
Wu, Q. et al. High-performance organic photovoltaic modules using eco-friendly solvents for various indoor application scenarios. Joule 6, 2138–2151 (2022).
Article  CAS  Google Scholar 
Download references
Funding The authors disclose support for the research of this work from the RGC Research Fellow Scheme (RFS) [grant number RFS2425-4S05]. X. Li and M. Xiao disclose support for research of this work from Guangdong Basic and Applied Basic Research Foundation [Grant number 2025A1515011342]. W. Liao discloses support for the research of this work from The Chinese University of Hong Kong [grant number 3134164].
These authors contributed equally: Luhang Xu, Yuang Fu.
Department of Physics, The Chinese University of Hong Kong, Shatin, Hong Kong, China
Luhang Xu, Yuang Fu & Xinhui Lu
Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong, China
Mianxin Xiao & Wei-Hsin Liao
Department of Chemistry and Hong Kong Branch of Chinese National Engineering Research Centre for Tissue Restoration and Reconstruction, The Hong Kong University of Science and Technology, Kowloon, Hong Kong, China
Ho Ming Ng & He Yan
Guangdong Basic Research Centre of Excellence for Aggregate Science, School of Science and Engineering, The Chinese University of Hong Kong (Shenzhen), Shenzhen, Guangdong, China
Wenzhi Ma & Jun Yan
School of Civil Engineering, Harbin Institute of Technology, Harbin, Heilongjiang, China
Xin Li
Institute of Intelligent Design and Manufacturing, The Chinese University of Hong Kong, Shatin, Hong Kong, China
Wei-Hsin Liao
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
L.Xu, Y.Fu, and X.Lu conceived the idea and designed the experiments. X.Li, W.Liao, and X.Lu supervised the project. L.Xu. and Y.Fu fabricated and characterised devices and modules. Y.Fu conducted the EIS measurements. L.Xu. performed KPFM and c-AFM measurements. H.Ng and H.Yan help set up the photovoltaic testing equipment under indoor light conditions. M.Xiao, X.Li, and W.Liao designed the A-IoT temperature sensor and developed ViPSN (the software on cell phones that receives signals from the sensor). W.Ma and J.Yan performed temperature-dependent J–V measurements. L.Xu, Y.Fu, and X.Lu analysed the results and wrote the manuscript. All authors provided revisions.
Correspondence to Xin Li, Wei-Hsin Liao or Xinhui Lu.
The authors declare no competing interests.
Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Xu, L., Fu, Y., Xiao, M. et al. Reverse-bias enabled mesoscale shunt passivation for organic photovoltaic modules to power miniaturised Ambient IoTs under low-light conditions. Nat Commun 17, 6109 (2026). https://doi.org/10.1038/s41467-026-72623-1
Download citation
Received: 04 February 2026
Accepted: 20 April 2026
Published: 06 May 2026
Version of record: 13 July 2026
DOI: https://doi.org/10.1038/s41467-026-72623-1
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Collection
Advertisement
Nature Communications (Nat Commun)
ISSN 2041-1723 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Texas-based OCI Energy and Israeli utility-scale developer Arava Power have broken ground at the site of the SunRoper Solar project, a 347 MWdc solar installation located in Wharton County Texas, about 60 miles southwest of Houston.
Backed by a 20-year PPA with an undisclosed Fortune 100 company, the installation is expected to begin commercial operations in December 2027. 
WHC Energy Services, a Louisiana-based EPC company with a significant presence in the Texas market, will serve as the contractor on the project.
“WHC is proud to serve as EPC contractor on the SunRoper Solar project, bringing our construction expertise to bear on a facility that will deliver meaningful power to the Houston region. This groundbreaking reflects months of careful planning and coordination with OCI Energy, Arava Power and our project partners, and we look forward to executing a safe, high-quality build through to completion in 2027,” said Randel Badeaux, WHC’s president of power for North America.
The installation is the third collaboration between the two developers in Texas, adding to a list of projects that includes the 270 MW Sunray Solar farm in Uvalde county and the 670 MWdc La Salle Solar facility located about 60 miles northwest of Lardeo. 
“SunRoper demonstrates how strategic partnerships can help meet Texas’ growing demand for electricity through investments in critical energy infrastructure,” said OCI Energy president Sabah Bayatli. “Today’s groundbreaking also marks the beginning of an important new chapter for the partnership between OCI Energy and Arava Power and reflects the strength of collaboration across development, financing, construction and energy procurement.”
The 1.3 GW under development by the companies represents a significant portion of the projected capacity of installations in Texas. According to the Solar Energy Industries of America, the state is expected to see 39 GW in new installations over the next 5 years.  
Financing for the SunRoper project closed in February, with ING Capital acting as the sole coordinating lead arranger, bookrunner, and green loan coordinator for the transaction, with additional support from BHI and Bank of Hapoalim. The financing package included a construction-to-term loan, a tax equity bridge loan, and various letters of credit.
Arava Power now lists a portfolio of more than 2 GW of generation capacity either operating or under development across Israel and North America. OCI Energy said it is targeting a portfolio of up to 10 GW in projects under development and management by 2028.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.

You have no items in your basket.

source

Florida seniors call solar purchase their ‘biggest’ financial mistake after questionable installation  Yahoo Finance
source

You are logged in
Loading…
You don’t have any Active Subscription.
Subscribed with another email? Logout and Login with that one.
Your active subscription(s)
Account subscription benefits alongside Premium Stories, Editorials, Opinions and more. Unlock these with Subscription
Products you’ve access to
Additional Subscription Benefits
Account Settings
Need help with your subscription?
September 4, 2026e-Paper
The View From India Looking at World Affairs from the Indian perspective.
First Day First Show News and reviews from the world of cinema and streaming.
Today's Cache Your download of the top 5 technology stories of the day.
Science For All The weekly newsletter from science writers takes the jargon out of science and puts the fun in!
Data Point Decoding the headlines with facts, figures, and numbers
THEdge At the cutting edge of education and careers
Health Matters Ramya Kannan writes to you on getting to good health, and staying there
Gender Agenda Stories from beyond the binary.
The Hindu On Books Books of the week, reviews, excerpts, new titles and features.
September 4, 2026e-Paper
Published – September 04, 2026 12:40 pm IST – Bengaluru
WeWork is a flexible workspace developer promoted by real estate developer Embassy Group. The solar plant is expected to generate approximately 15–16 million units of clean electricity annually, and is targeted for commissioning in FY27.  | Photo Credit: Shannon Stapleton
WeWork India Management Limited, a flexible workspace developer promoted by real estate developer Embassy Group, plans to develop a 10 MWp (DC) ground-mounted solar power plant in Karnataka with an investment of ₹35 crore. The location is yet to be identified.
Targeted for commissioning in FY27, the plant is expected to generate approximately 15–16 million units of clean electricity annually. Once operational, it is expected to increase the share of renewable electricity across WeWork India’s portfolio from close to 40% currently to approximately 50%, bringing the company closer to its 100% renewable electricity goal by 2028.
Nearly 80% of the plant’s output would support WeWork India’s operations, including 10 centres in Bengaluru. The remaining 20% of the plant’s output will be used for their future growth.
Karan Virwani, Managing Director & CEO, WeWork India, said, “Bengaluru is our largest market, making it the natural starting point for an investment of this scale. As we grow, we want a greater share of that growth to be powered by clean energy, but we also want sustainability to make strong business sense.’’
Building the company’s own renewable generation capacity allows it to both reduce the carbon footprint of its operations while creating greater certainty over energy costs for the next 25 years. “This is the kind of long-term investment we believe can make sustainability an integral part of how we scale, rather than an initiative that sits alongside the business,” Mr. Virwani added.
According to the company, Karnataka is a strategic location for the investment, with Bengaluru representing WeWork India’s largest market at 30 operational centres. The State accounts for approximately 30% of the company’s total electricity consumption across its portfolio, making it a natural market to drive renewable energy adoption at scale.
The solar plant is expected to enable WeWork India to meet a meaningful share of this demand through renewable power, while reducing its reliance on conventional grid electricity.
Published – September 04, 2026 12:40 pm IST
Copyright© 2026, THG PUBLISHING PVT LTD. or its affiliated companies. All rights reserved.
BACK TO TOP
Terms & conditions  |  Institutional Subscriber
Comments have to be in English, and in full sentences. They cannot be abusive or personal. Please abide by our community guidelines for posting your comments.
We have migrated to a new commenting platform. If you are already a registered user of The Hindu and logged in, you may continue to engage with our articles. If you do not have an account please register and login to post comments. Users can access their older comments by logging into their accounts on Vuukle.

source

Stephen Barnfield, a Regulatory and Insurance Partner at Forbes Solicitors, looks at whether a maturing risk profile is required for solar panels in schools.

Progressive solar power
The education sector is embracing solar technology in a move to manage both costs and carbon emissions. In recent weeks, the Department for Education (DfE) announced that a new wave of schools and colleges across England are set to save £220 million on energy bills over the lifetime of solar panels. The aim is to free up money that would have otherwise be spent on energy, so that it can be reinvested in children’s education.
According to the DfE’s July 2026 update, 245 schools and colleges already have government-funded solar panels, and 100 more will now join the Great British Solar Partnership. A further 150 schools and colleges will also be embracing Photovoltaic (PV) technology through private sector support, with a pilot programme seeing installation and maintenance of panels at no upfront cost.
It’s clear there’s progressive adoption of PV technology throughout education, but as panels become an increasingly familiar sight on school roofs, is there a danger they are being treated as a ‘fit and forget’ technology? Recent reports suggest this is a real risk, and one that could impact insurance cover for buildings, contents and business / operational interruption, as well as employer, governors’ and trustees’, and public liability.
Fire safety risks
In June this year, the BBC reported that Suffolk County Council had taken the decision to switch off solar panels at about 80 schools across the county. This came after a fire at Sidegate School in Ipswich, which the local fire service confirmed as being caused by solar panels on the roof. Thankfully, there were no reports of anyone being injured by the fire.
It’s also believed that fires at two other Suffolk schools over the last year were linked to PV panels. And in 2025, Northumberland County Council switched off solar panels on 141 of its buildings following a fire at a school, as well as a community centre. Solar panel safety issues aren’t isolated to public sector owned and managed buildings, with PV panel fires also reported across domestic dwellings and commercial properties, including a rooftop fire at a supermarket’s regional distribution centre in Peterborough in 2024.
Solar panel fire risks seem to be more commonly associated with the age and maintenance of PV technology, which should prompt wider consideration about ongoing management. Local authorities, academy trusts, schools and colleges, and insurers must treat solar panels as critical infrastructure, supported by risk management and fire safety frameworks that account for the changing risk profile of PV systems over time.
Moving away from ‘fit and forget’
Solar technology is often perceived as a relatively straightforward system, with no moving parts and low upkeep. These solutions are often positioned as long-term investments that require little maintenance to generate affordable, clean energy. Internet searches even reveal benefits such as natural rainwater cleaning panels by washing away dust and dirt, or a panel installation only needing a quick check or wash a couple of times per year.
Once installed, solar panels can quietly generate electricity for years with little visible intervention, creating a risk that they are viewed as a ‘fit and forget’ technology. Strategies and programmes must avoid this approach.
PV systems are made up of a range of electrical components, including panels, wiring, inverters and connectors, and in some cases, battery storage systems. Each part represents a potential point of failure and like many products, are subject to the ‘bathtub curve’ of reliability. Failure rates can be typically higher during installation and commissioning, before falling (the downward part of the curve) to a relatively low level during normal operation, and then rising again (upward curve) as systems age and components begin to deteriorate.
Based on reported school fires, attention seems to be increasingly focusing on ageing solar installations. The PV panels at Sidegate Primary School were installed in 2012 and Suffolk County Council’s decision to switch off panels at about 80 schools applied to systems installed between 2011 and 2016. Components can become more susceptible to wear, deterioration and potential failure as they mature.
However, risks aren’t only age-related. Solar panel problems can also be caused by faults and damage and may not always provide visible warning signs until an issue becomes more significant.

Managing the risks
To effectively manage the risks and potential liabilities of PV systems, organisations should implement a structured risk management framework based on hazard identification, risk likelihood and impact evaluation, and monitoring and inspections. The framework should cover the entire lifecycle of the installation, from design, installation and commissioning, through to operation, maintenance and component replacement. By aligning inspections and performance reviews with the expected lifespan of key assets, organisations can identify any emerging issues early to ensure systems remain safe and operating as intended.
System inspections though, should not be limited solely to component lifespans. Risk management frameworks need to also account for external factors that could affect the condition and performance of PV systems over time. These may include weather exposure, storms and high winds, as well as accidental damage. For example, school and college settings may mean that solar panels are subjected to repeated impacts from footballs or other sports equipment. There is also a possibility of overhanging trees, falling branches and vegetation growth causing damage.
Wildlife should also be considered as part of risk assessments. Birds’ nesting activity could cause a buildup of twigs, leaves and features beneath panels and close to components, potentially restricting ventilation, concealing defects and increasing fire risks when overheating occurs. Risk management should be questioning: how frequently are systems inspected, and are procedures in place to assess installations following events such as storms or impacts from falling objects?
Detailed, formalised records should support risk management frameworks, with documentation covering asset registers, risk assessments, maintenance schedules, inspections and performance monitoring, incident reporting, and end-of-life planning. This information will provide an audit trail, demonstrating accountability and helping organisations more clearly determine fact, responsibilities and possible liabilities if a fire occurs. The management of formalised records also helps to keep PV systems front of mind, helping avoid the ‘fit and forget’ approach.
Consideration needs also to be given to how the inspections are to be carried out and who is going to conduct them. Will it be in-house or managed by contractors? In either case, the activity itself needs to be managed safely, as it will inevitably involve work at height and proximity to electrical systems.
As the education sector increasingly embraces PV technology, it’s important that systems are not treated as one-off capital investments. Mindsets must shift towards managing solar panels as critical infrastructure that requires ongoing oversight.
About Forbes Solicitors:
Forbes Solicitors is an award-winning law firm, with nine offices across England that look after the interests of clients nationwide.
The firm has 64 partners and an overall headcount of nearly 400, advising on a wide range of commercial and personal matters. Offering services to public sector organisations, commercial entities and consumers, Forbes specialises in providing legal expertise in practice areas including litigation, commercial law, intellectual property, corporate legal services, employment law, business immigration, HR consultancy, insurance, property litigation, commercial property, insolvency and debt recovery, and individual services.
Forbes holds the ISO9001 Quality Certification and in its recent assessment it was described as “exceptional”. The firm is ranked as a Legal 500 Top Tier Firm and a Chambers and Partners Leading Firm, receiving 90 nominations in the latest editions. Furthermore, a number of its professionals are included in the elite “Hall of Fame” list, “Leading Partners” list, the “Next Generation Lawyers” list, the ‘Leading Associates’ list, and 36 of Forbes’ solicitors are listed as recommended lawyers. Forbes is also a member of LawPact® – the international association of independent business law firms – which supports the expansion of its national and global reach.

Descartes Underwriting and Generali Extend Partnership
ESG Book, a global leader in sustainability data and technology, today released new analysis showing that global ESG regulations have increased by 155 percent over the past decade, as the rapid growth of sustainability-based policy […]
Swiss Re Institute is already looking ahead to next year, with a raft of predictions on their Sigma site. Here’s the word; After a resilient 2023 powered by strong US economic growth, the world economy […]
The latest risk insights report from Beazley; Businesses are navigating a risk environment more layered and unpredictable than ever. Beazley’s 2025 Geopolitical and Economic Risk & Resilience report surveyed 3,500 senior business leaders, placing a […]
This site uses Akismet to reduce spam. Learn how your comment data is processed.
Enter your email address to subscribe to Insurance Edge and receive notifications of new posts by email.
Email Address
Follow Insurance Edge
We are interested in the latest news, new products, partnerships and much more, so email us at; editorial@insurance-edge.net. READ MORE….
Copyright © 2026 | MH Magazine WordPress Theme by MH Themes
Subscribe now to keep reading and get access to the full archive.
Type your email…
Subscribe
Continue reading

source

Solar Power World
By Kelly Pickerel | September 3, 2026
Cypress Creek Energy’s Ostrea Solar project in Yakima County, Washington, has reached commercial operation.
“Washington’s economy is growing, and that growth requires more electricity,” said Kevin Smith, CEO of Cypress Creek Energy. “Ostrea shows how we can responsibly bring new power online quickly and affordably, while creating lasting benefits for the communities where we operate, and caring for the land for generations to come.”
Ostrea was supported by $180 million in private funds, and nearby headquartered Microsoft is the long-term offtaker of energy and environmental products generated by Ostrea Solar. The project is interconnected to the Bonneville Power Administration transmission system, adding 104 MW_DC of new generation to the region’s power system.
Cypress Creek will own and operate Ostrea for up to 40 years, and the company made thoughtful land management and environmental stewardship an important part of the project throughout its operating life. Cypress Creek worked closely with the Washington State Energy Facility Site Evaluation Council (EFSEC) and the Washington Department of Fish and Wildlife (WDFW), including modifying the project design to conserve approximately 275 acres of critical shrubsteppe habitat.
Shrubsteppe is one of Washington’s most diverse ecosystems, providing habitat for species found nowhere else in the state. It is also increasingly scarce, with an estimated 80% of Washington’s historic shrubsteppe lost or degraded to development and agriculture. Cypress Creek also contributed funding to WDFW to support additional conservation work elsewhere in the region.
Ostrea was also designed with wildlife corridors to maintain migratory pathways through the project area, including for Rocky Mountain elk that move through the region.
Cypress Creek’s stewardship efforts extend to the land beneath and around the solar panels. Portions of the project site include degraded former cropland that has not been actively farmed for more than 25 years and where invasive and non-native plant species had become established. During construction, invasive plants that can threaten native habitats were removed, and disturbed areas were revegetated.
The project was then seeded with carefully selected mixes of grasses and pollinator-friendly flowering plants. Beneath and around the solar arrays, lower-growing species were selected to establish healthy vegetation without interfering with the panels. Other restored areas include taller grasses and flowering species that provide additional habitat and forage for pollinators.
 
PCL Solar Constructors served as the project’s engineering, procurement, and construction contractor. Construction created approximately 300 jobs, all paid at prevailing wage rates, and included an apprenticeship program that provided opportunities for workers to gain valuable experience and develop skills for long-term careers in the construction trades. Approximately 75 apprentices contributed 39,000 hours to the project during construction.
News item from Cypress Creek
Kelly Pickerel has more than 15 years of experience reporting on the U.S. solar industry and is currently editor in chief of Solar Power World. Email Kelly.

Comment *
Name *
Email *
Website


Δ
Copyright © 2026 Arrowfly LLC. All Rights Reserved. The material on this site may not be reproduced, distributed, transmitted, cached or otherwise used, except with the prior written permission of Arrowfly LLC
Privacy Policy | RSS

source

Nordic solar developer Alight and automotive safety supplier Autoliv have inaugurated the 101 MW Eurajoki solar park.
Located on Finland’s west coast, the park is built, owned and operated by Alight. It will generate around 100 GWh annually, equivalent to the consumption of approximately 20,000 households, making it one of the largest in the country.
The electricity generated by the plant will be delivered to Autoliv under a long-term virtual power purchase agreement (PPA) first signed in April 2025, billed as the largest in Finland’s history at the time.
A statement published by Alight explains that long-term corporate PPAs are becoming an increasingly common route for businesses to secure predictable energy costs, with this agreement part of a broader, global climate strategy across Autoliv’s operations.
The solar park is made up of two distinct sites brought together by a shared substation. Alight says the two-site design allowed the project’s environmental protection measures to be tailored to local conditions at each location, with one site’s biodiversity plan including habitat restoration and improved wetland edges, and the other featuring a new pond.
The project was backed by €46 million ($53.5 million) of senior debt from banks ABN AMRO and SEB.
Alexander Rudberg, Alight’s Head of Development, told pv magazine the company is planning to add a co-located battery storage to the Eurajoki solar park.
“The current plan is for a 30 MW/60 MWh battery system to be in place in 2027,” he confirmed.
Rudberg added that in the Finnish region of Satakunta, where the Eurajoki project is located, Alight has a further four projects under development totaling 360 MW of solar and 150 MW of battery energy storage.
“They are in various stages of development, for example one has a grid connection agreement signed and two have received approved permits,” he shared.
Additional figures from Alight put its total solar and storage pipeline across Finland at in excess of 1 GW.
Finland installed 478 MW of utility-scale solar during the first half of 2026, taking cumulative utility-scale capacity to 842 MW. The commissioning of the Eurajoki project brings the country closer to 1 GW of large-scale solar.
Finland’s largest operational solar project is the 204 MW Kalanti solar park, which came online earlier this year.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Renewable energy company Alight has inaugurated the 101 MW Eurajoki Solar Park in western Finland. The project is one of the country’s largest solar facilities, according to Alight. Built, owned and operated by Alight, the park will generate around 100 GWh of renewable electricity annually, enough to match the consumption of about 20,000 households. Swedish automotive safety company Autoliv will access the project’s output through a long-term virtual power purchase agreement (VPPA). The project is Alight’s first commissioned solar facility in Finland. It is expected to add about 12% to the country’s utility-scale solar capacity.
UK renewable energy developer Anesco has secured project financing from Lombard for three projects in the UK. The financing covers the already operational 21 MW Woodwalton Solar Farm in Cambridgeshire; the 50 MW/100 MWh Rothienorman battery energy storage system (BESS) in Scotland, which is nearing commissioning; and the 48.5 MW Coven Solar Farm in Staffordshire, which has reached financial close. Anesco CEO Hildagarde McCarville said the financing would help move the projects forward and contribute to the UK’s energy transition and security.
METLEN, an energy and metals company with operations in renewable energy, has signed a 10-year power purchase agreement (PPA) with Coca-Cola Tria Epsilon. The latter is the Coca-Cola bottling company in Greece. The agreement covers electricity from an approximately 12 MW solar plant in Mikro Perivolaki, near Velestino in Greece. Coca-Cola Tria Epsilon will purchase 100% of the plant’s output, estimated at about 16.5 GWh annually. The company claims to already source all electricity used at its production facilities from renewable sources. The agreement is also linked to parent company Coca-Cola HBC’s target of reaching net-zero carbon emissions by 2040.
Vattenfall, a European energy company active in renewable power generation, has officially opened its 46 MW Nauen Solar Park near Berlin, Germany. The project became operational in June 2026 and was connected to the regional distribution grid. The solar park has around 80,000 panels across 40 hectares and is expected to generate about 47 GWh of electricity annually. It was built without government subsidies. The project’s economics are supported by a 10-year power purchase agreement (PPA) with Wieland Group, a copper and copper-alloy semi-finished products manufacturer. Wieland will purchase the plant’s entire electricity output, which is expected to cover around 15% of the electricity it buys annually for its German sites.
TEAL, a renewable energy company focused on developing clean energy projects, has launched TEAL Renovables, a new development company targeting the Spanish market. The new entity will focus on the early-stage development of renewable energy assets in Spain. TEAL said the country was selected because of its policy framework, growing demand for clean energy and plans to expand renewable power deployment. TEAL Renovables will work with local partners and stakeholders to develop renewable energy projects supporting Spain’s decarbonization and electrification goals.
Orrön Energy, a publicly listed renewable energy and infrastructure company within the Lundin Group of Sweden, has completed its strategic transaction with Cloudberry Clean Energy. It has now become the largest shareholder of Cloudberry with 27.01% stake. This deal, including two board positions, has created a Nordic independent power producer (IPP) with about 2.1 TWh of annual proportionate power generation, says Orrön. The latter retains the 86 MW Karskruv Wind Farm in Sweden and a European development portfolio of about 12 GW spanning solar, battery storage and data center projects.
Catalonia’s regional government, led by President Salvador Illa, plans to triple the region’s renewable energy capacity by 2030, and by 12 times by 2050 with solar power playing a major role in the expansion. The government said it aims to significantly increase the number of solar parks while accelerating renewable energy deployment across Catalonia. The target is part of its broader plan to increase renewable electricity generation and reduce dependence on fossil fuels. Catalonia has over 2 GW of operational solar energy capacity, including self-consumption. Illa was speaking during the inauguration of the 300 MW Alcarràs Solar Park. It is to be expanded by 100 MW along with the addition of batteries with up to 4 hours duration.
TaiyangNews 2024

source

Vertical solar specialist Over Easy Solar has installed its first vertical PV system in Ireland.
Located in Dublin, the 5.37 kW system features 21 of Over Easy Solar’s VPV Units comprised of the company’s third-generation XM-3 QUATTRO-256S, a preassembled, lightweight vertical bifacial photovoltaic unit.
The installation was installed directly on a green roof through Over Easy Solar’s ongoing partnership with Sempergreen, a Dutch company specializing in sustainable urban nature solutions.
Keelin Currivan, International Customer Solutions Advisor at Over Easy Solar, told pv magazine such installations prove that customers do not have to choose between having a green roof and a rooftop solar installation.
“The panels manage rainwater and provide thermal regulation for the roof, while the vegetation boosts panel output via the albedo effect without shading the plants themselves,” Currivan explained. “It’s a genuine case for combining biodiversity and clean energy generation on commercial rooftops, rather than treating them as competing uses of the same space.”
The installation was carried out by Irish renewable energy and electrical contractor company Solar Precision. In a statement to pv magazine, the company explained it chose to utilize Sempergreen’s green roof system with the Over Easy Solar’s vertical PV system “because it offered an innovative way to maximize renewable energy generation while preserving the environmental and biodiversity benefits of a green roof.”
“The system aligns with our commitment to delivering sustainable solar solutions and gave us the opportunity to be the first solar PV company in Ireland to install this technology as a case study,” the company’s statement continues.
Solar Precision added that since completion, the project has demonstrated that vertical solar panels can be successfully integrated with a green roof without compromising performance or maintenance. 
“As an early adopter of this system in Ireland, we’ve gained valuable knowledge and confidence in the technology, and we’re excited about the opportunities it creates for future commercial and residential green roof projects,” the company said.
Over Easy Solar’s latest expansion follows its first installation in the US earlier this year with a 100 kW vertical PV system in New York. Since then, the company also installed its first system in Vancouver, Canada.
In April, Over Easy Solar shared that its vertical bifacial PV system had outperformed a conventionally-tilted monofacial rooftop PV system in the UK across all seasons during a year-long study.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

With Lombard’s funding, Anesco will be able to support its UK projects through their development phases.
September 3, 2026
UK renewable energy developer Anesco has secured funding from financier Lombard to develop and operate three renewable sites in the UK.
Lombard’s financing will cover a portfolio of Anesco’s projects, which are all at various stages of delivery:
Woodwalton solar farm in Cambridgeshire (21MWp) is currently operational and generating renewable power.
Rothienorman BESS in Aberdeenshire (50MW/100MWh) is approaching the final stages of construction before being commissioned.
Coven in Staffordshire (48.5MWp) has achieved financial close and will begin the next phase of development.
For Lombard, the financing agreement with Anesco is part of the company’s drive to fund renewable projects which are reinforcing the UK’s energy security.
Tom Bosson, head of renewable energy lending for Lombard, commented: “By working with experienced developers such as Anesco, we can help bring forward investment in renewable energy and storage technologies that are expected to play an important role in supporting a more resilient energy system and the UK's pathway to net-zero.”
Related:EY: UK is 2nd most attractive European energy investment market, with strong renewables outlook
“The Woodwalton, Rothienorman and Coven projects demonstrate the scale and variety of infrastructure supporting the UK's energy transition.”
Hildagarde McCarville, Anesco’s CEO, added that the completion of the projects’ financing—at different stages of development—was a testament to the company’s pipeline, delivery and partnerships.
“We are pleased to be working with Lombard to move forward with these projects, which will contribute to the UK’s energy transition and energy security, while creating long-term value for our stakeholders,” McCarville concluded.
The news comes as Anesco continues to expand its foothold across northern Europe. As part of Ara Partners and Nature Infrastructure Capital, Anesco has delivered 1.3GW of solar and storage assets and additionally manages a portfolio for investors totalling over 1.3GW.
Read more about:
Catie Owen
Contributing writer
Since 2019, Catie has been writing news, interviews, client content and editing magazines. In recent years, her interest in sustainability has led her to pursue renewable energy as her primary beat. Having written primarily about solar energy and storage, Catie also enjoys covering the positive human impact of renewable technology.
Google Preferred Source
Google Preferred Source
Copyright © 2026 Informa PLC. Informa PLC is registered in England and Wales with company number 8860726 whose registered and head office is 5 Howick Place, London, SW1P 1WG.

source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 16, Article number: 16141 (2026) Cite this article
1696 Accesses
1 Citations
30 Altmetric
Metrics details
This study experimentally investigates the performance enhancement of photovoltaic (PV) panels using different cooling techniques under real outdoor operating conditions. Three cooling approaches are evaluated, including water-spray cooling, serpentine water circulation, and a fame-glass configuration. The objective is to assess their impact on PV surface temperature and electrical performance. Experimental measurements are conducted under similar environmental conditions, and the current–voltage (I–V) characteristics of the PV modules were recorded for each cooling configuration. The results demonstrate that active cooling techniques significantly reduced the operating temperature of the PV module, which consequently improved its electrical performance. Among the tested methods, the water-spray cooling technique achieved the most effective temperature reduction and the highest improvement in power output. In contrast, the fame-glass configuration showed a reduction in electrical power despite lowering the panel temperature, mainly due to optical losses and partial shading caused by the glass frame structure. The findings highlight the importance of selecting appropriate cooling strategies that balance thermal management and optical performance. Overall, the results confirm that effective cooling techniques can enhance PV efficiency and contribute to improved energy production in hot climate regions.
Solar photovoltaic (PV) systems are widely recognized as a key renewable energy technology for sustainable power generation, particularly in regions with high solar irradiance^1,2. However, the electrical performance of PV modules is strongly influenced by their operating temperature, which becomes a critical limiting factor in hot and arid climates^3,4. PV module performance is strongly influenced by operating temperature, with 25 °C defined as the reference temperature under standard test conditions (STCs). A decrease in cell temperature below this value generally enhances electrical performance. Although increased solar irradiance leads to higher power output, it may also cause a rise in cell temperature, which can result in a reduction in conversion efficiency. In such environments, a significant portion of the incident solar energy is converted into heat rather than electricity, leading to elevated module temperatures and subsequent efficiency degradation⁵. It has been reported that the efficiency of silicon-based PV modules decreases by approximately 0.4–0.5% for every 1 °C increase in cell temperature above STCs (25 °C)^6,7. Consequently, in regions such as Upper Egypt, where ambient temperatures can exceed 45–50 °C during summer months, effective thermal management is essential to maintain PV performance, ensure operational reliability, and mitigate long-term thermal degradation^3,8.
Various approaches have been proposed to address PV temperature rise, including optical concentration, tracking systems, and structural modifications aimed at increasing the incident solar irradiance^7,8. While these methods can enhance power output, they often intensify thermal stress on PV modules, which may further reduce efficiency and accelerate performance degradation if adequate cooling is not provided^3,4. Several studies have investigated different PV cooling techniques to enhance electrical efficiency and reduce module temperature. For example, previous works have examined water-spray cooling^9,10, serpentine water circulation systems^11,12, and other passive or hybrid cooling approaches^13,14,15,16.
The effectiveness, cost, and suitability of these cooling techniques remain highly dependent on local climatic conditions, particularly in high-temperature regions³.
The average efficiency of commercial PV panels is between 15 and 20% under ideal condition, monocrystalline PV can gain efficiency up to 22%, while polycrystalline can gain 18% efficiency⁵.
Under strong solar irradiance conditions, particularly in hot climates, PV modules tend to operate at elevated temperatures, which negatively affect their electrical performance. Therefore, effective cooling is required to limit temperature rise while maintaining high irradiance utilization.
As a result, keeping the units’ temperature below thermal degradation is crucial. It has been shown that, the efficiency of a silicon-based PV panel decreases by 0.5% for each degree Celsius of temperature increase⁶. When the PV panel efficiency reaches its maximum under the given conditions, the surface temperature is around 25 °C⁷. The solar radiation factor states that when more sunlight reaches a PV module’s surface, the module’s efficiency will rise. This is often achieved by using investigation if there are specific research goals or hypotheses, Trackers, lenses, and/or condensers. By lowering the angle of incidence, these methods seek to raise the solar energy density on the PV panel^7,8. The issue with these techniques is that they cause the temperature of the solar cells to increase over the safe operating range, which can lead to reduced cell efficiency and, in severe cases, even cell damage. As a result, PV modules that are cooled down will perform better, particularly in warm weather³.
Efficiency is not over 20% even in the best-case scenario under STCs. Because it depends on several factors, maintaining this efficiency at its best level is the biggest problem. The PV modules must be cooled to maintain the lowest temperature with the maximum radiation.
Many researchers have proposed cooling as a means of increasing PV performance and mitigating the impact of cell temperature on performance. A few of the research that have been done to find out more about this subject will be highlighted in this section. Solar panel cooling methods that are most often used are shown in Fig. 1. Cooling systems are categorized into passive and active. Passive cooling solutions don’t require operating energy; however, active cooling solutions require additional energy to cool by extracting heat using pumps, fans, etc. PV modules are rated under standard test conditions, which correspond to a solar irradiance of 1000 W/m² and a cell temperature of 25 °C.
The most popular methods for cooling solar PV panels.
As illustrated in Fig. 1, PV cooling techniques can be broadly categorized into passive and active methods. Passive cooling approaches, including heat sinks and radiative cooling, reduce the operating temperature of PV modules without consuming auxiliary energy by relying on natural heat dissipation mechanisms. Conversely, active cooling techniques, such as spray cooling and hybrid photovoltaic/thermal (PV/T) systems, utilize externally driven fluid flow or coupled thermal–electrical systems to extract heat more effectively, particularly under high irradiance and elevated ambient temperature conditions. Accordingly, Fig. 1 has been updated to reflect a complete and consistent classification of PV cooling techniques discussed in this study.
Passive cooling methods are often chosen because they are simple to use and inexpensive. Natural convection serves as the primary cooling mechanism in passive cooling solutions. In addition to natural convection, other uses, including PCMs and thermosiphon systems, could contribute to the cooling¹⁶.
There are many subcategories of passive cooling techniques are covered in separate discussions. The following are these categories: a) Passive cooling methods based on gas (air). b) Passive cooling methods based on liquid. c) Passive cooling methods based on radiative. d) Passive cooling methods based on phase change material (PCM).
The leading coolant in the passive cooling method, which uses natural convection to decrease the temperature of PV panel, is air. But it does not cool as much as other methods. Gas (air)-cooling methods have so far been the subject of much theoretical and experimental investigation by several researchers.
Theoretically, polycrystalline panels in¹⁷ and¹⁸ can be cooled by passive air cooling using devices like heat sinks, which can reduce temperature by 17.61 °C and 16 °C, respectively. For mono crystal panels in¹⁹, air ducts (flat and curved fins) attached to the back of panel can reduce temperature by 18.09 °C. Additional experimental studies, In Dhahran, Saudi Arabia, aluminium fins with a nano coating placed on the back of polycrystalline panels lowered temperature by 8 °C and increased efficiency by 2.88²⁰. In Tiruchirappalli, India, fin-attached heat sinks lowered temperature by 9.45 °C and increased efficiency by 1.072²¹. In Selangor, Malaysia, polycrystalline panels cooled by both longitudinal and lapped fin heat sinks, which decreased temperature by 22.7 and 14.28 °C, respectively, and increased efficiency by 12.33 and 8.86, respectively²².
Water and other fluids, including chemical fluids like glycol, as well as nanofluids, are the primary coolants employed in these passive cooling methods to reduce the PV panel’s operating temperature. These methods use gravity, thermosiphon effects, heat pipes, floating, and other natural fluid circulation to cool PV panels without requiring any additional electricity. Compared to gas (air)-based passive cooling approaches, these methods have a superior cooling effect. In theoretical study²³, polycrystalline panels are cooling using both Heat sink based on passive air, and Thermosiphon loop based on passive water, the temperature decreased by 25 °C, 36.5 °C, and the efficiency increased by 30%, and 41.5%, respectively.
Theoretically, a solar-driven rainwater cooling system was studied for cooling poly crystal panels; the temperature decreased by 19 °C and the efficiency increased by 8.3%²⁴, in other study the porous channel cooling method increased the efficiency by 4.17%²⁵.
Experimentally in Karnataka (India) applied Evaporative cooling, and Gravity assisted flow cooling techniques for cooling poly crystal, The results showed decreases in the temperature by 15.9 °C, and 8.7 °C and increases in efficiency by 45%, and 33%, respectively²⁶. After using floating cooling with fin assistance, the efficiency of polycrystalline panels at Port Said, Egypt, increase by 22.24%²⁷. In Baghdad, Iraq, monocrystalline solar panels’ efficiency increased by 9.7% following the use of evaporative cooling, which dropped the temperature by 18.3 °C²⁸. At Thuwal (Saudi Arabia), following the application of a solar-driven PV cooling technology that lowered the temperature by 8 °C, their efficiency increased by 47%²⁹.
Through the phase transition from solid to liquid in temperature latent, PCMs store thermal energy. By switching phases, PV panels may be cooled without using any energy. More cooling occurs than with passive air-based cooling. Theoretically, Authors have used PCMs as coolant materials, in³⁰ using RT27 increased the efficiency By 12%, in³¹ using RT42 paraffin wax decreased temperature by 32 °C, and in³² using RT35HC paraffin wax decreased temperature by 24.9 °C and increased efficiency by 11.03%.
On the other side, many research works have applied experimentally PCM as a cooling technique, OM47 was employed as PCM cooling in Karnataka, India, to cool polycrystalline PV panels, this method reduced temperature by 8.1 °C and raised efficiency by 13%²⁶. In Jahrom, Iran, the poly-crystalline panels were cooled with PEG1500 passive PCM cooling in addition to a heat sink. This resulted in a 28 °C drop in temperature and a 3% gain in efficiency³³. Paraffin is used for cooling poly-crystalline panels in Dezful (Iran) that leads to decrease temperature by 6.8 °C and increase efficiency by 12.5%³⁴. When using Glauber salt to cool monocrystalline panels in Tamil Nadu, India, the temperature decreased by 31 °C and the efficiency increased by 11.5%³⁵.
Energy is used throughout the cooling process in active cooling systems. Forced convection is the primary cooling mechanism in active cooling systems. Three distinct categories are used to categorize active cooling methods. These consist of a) Active Gas (air) cooling techniques b) Active Liquid cooling techniques c) Active Thermoelectric cooling techniques.
In active cooling approaches, air is employed as the primary coolant to drive convection and lower the temperature of the PV panel. But compared to gas (air) cooling methods, there is more cooling.
Studies like³⁶, which used both active fan cooling and passive nature cooling in an experimental to cool mono-crystalline panels, compare passive and active air cooling. The results showed a 33.33% reduction in the panel’s performance assessment cost.
A 50 kW poly-crystalline PV experimental system in Mathura, India was cooled utilizing both passive natural cooling and active artificial cooling which led to an increase in the performance of the cell to 95.9% and 98.2%³⁷.
The forced air cooling was applied on the front side of the polycrystalline panels in Ismailia, Egypt, the temperature decreased by 7 °C and the efficiency increased by 2.29%³⁸. Whereas active air cooling with a channel under the monocrystalline panels is used experimentally in Tehran, Iran, the temperature drops by 5 °C and the efficiency increases by 2.6%³⁹.
In Benha, Egypt, researchers conducted an experimental comparison of fans and blowers for forced convection cooling on the rear side of monocrystalline PV panels. The results showed that efficiency increased by 3.9% and 7%, respectively, and temperatures decreased by 5.4 °C and 9 °C, respectively⁴⁰.
Water and other fluids, including chemical and nanofluids, are the most common coolants utilized in this active cooling method to lower panel temperatures. This approach cools the panels by forcing the fluids to circulate using energy. This method provides a greater quantity of cooling than passive liquid cooling methods.
Techniques for active liquid cooling are researched. Theoretically⁴¹, addressed the effects of jet-impingement the cooling with SiC-water nanofluid on panels, resulting in a temperature reduction of 32.1 °C. Similarly⁴², examined the effects of water spraying at the back side of the panels as a cooling method, resulting in a temperature reduction of 10◦C and an increase in efficiency of 7.3%.
However, they are used experimentally in many studies, including the following: at Alcala de Henares (Spain), where the temperature lowered by 14 °C and efficiency improved by 9.8% after using heat exchanger at the back side of the poly-crystalline panels⁴³; in⁴⁴ at Ankara (Turkey) where the temperature lowered by 1.88 °C and efficiency increased by 13.69%, using water film that flowing on the upper side of the panel; and in⁴⁵ using spraying of water on the upper side of the monocrystalline panel that lead to increase efficiency by 15.73%.
Nanofluid was used to improve the cooling systems as follows: in⁴⁶ at Xi’an (China), nanofluid including 2% Al₂O₃ combined with heat pipe groundwater and spiral pipe was used to improve the cooling systems for mono-crystalline panels; and in⁴⁷ at Baghdad (Iraq), nanofluid with Zn-H₂O combined with a heat exchanger on the back of the panel was used to cool mono-crystalline panels, resulting in a decrease in temperature of 18 °C and an increase in efficiency of 7.8%.
This active cooling technique lowers the temperature of PV panels via the thermoelectric effect. Electric energy is used to cool using active thermoelectric cooling technology. With this method, electrical current is used to cool the panels. The junction of two different wires cooled when a little current was sent through it. The Peltier effect is a phenomenon that is fundamental to thermoelectric refrigeration. Heat is taken in from the medium to be cooled and rejected in a warmer environment in this refrigeration circuit. The net electrical input that is still available is the difference between these heat levels. Compared to other active cooling strategies, this method provides less cooling⁴⁸.
A thermoelectric module installed on the back of a photovoltaic panel is used to theoretically analyse the cooling effects. The cooling improved efficiency by 18%⁴⁸. However, an experimental study on how a thermoelectric module mounted to the back of the panel for cooling the panel. The corresponding cooling strategy reduced the temperature by 35 °C and increased efficiency by 18%⁴⁹.
This research work tested experimentally in the Renewable Energy Laboratory at Assiut University, Egypt. The experiment was done under high temperatures in the summer months which reached 50 °C. So that three different cooling systems are chosen to deal with high temperature and selected the most effective. These techniques are selected between passive and active categories. They are spray water, serpentine with water, and fame glass.
The main goal of the paper is to search for effective cooling systems that deal with PV panels in hot places.
Search for the benefits and drawbacks of using spray water for cooling PV systems. Investigate the advantages and disadvantages of employing a serpentine water-cooling system for PV panels. Research the properties and applications of “fame glass” in the context of PV cooling, considering if it might be a specific type of glass or “fame glass”.
The fame-glass technique represents a passive cooling approach in which a conventional glass sheet is coated with a thin semi-shading layer that allows partial transmission of solar radiation. Unlike reflective coatings or advanced spectrally selective films reported in previous studies, the fame-glass used in this work relies on commercially available treated glass similar to automotive tinted glass. This approach provides partial irradiance reduction and thermal mitigation without complex fabrication processes, offering a simple and practical passive cooling alternative for photovoltaic applications.
Previous studies have reported significant performance improvements using active water-based cooling techniques for PV modules. Spray cooling methods have demonstrated efficiency enhancements ranging from approximately 7% to 24%, with corresponding surface temperature reductions of 10–22 °C under high irradiance conditions^42,45. Similarly, serpentine or channel-based water cooling systems have achieved temperature reductions of 15–25 °C and efficiency improvements between 9 and 25%, depending on flow configuration and operating conditions^43,44,45.
In contrast, passive cooling approaches based on glass covers or shading elements have been shown to reduce PV surface temperature by 8–12 °C; however, several studies report substantial power losses, in some cases exceeding 30–60%, due to irradiance attenuation caused by reflection, refraction, and partial shading effects^24,26. These findings indicate that while passive glass-based cooling can lower operating temperature, it often introduces a significant trade-off between thermal mitigation and electrical output.
Accordingly, the present study experimentally compares spray cooling, serpentine water cooling, and glass-frame cooling under identical outdoor conditions to quantitatively assess their thermal and electrical performance trade-offs in hot arid climates.
The motivation of this work is to analyze the factors influencing the effectiveness of different PV cooling techniques and to evaluate their suitability for hot climatic regions:
Analyse the characteristics of the climate in Asyut Governate, Egypt, and how these three cooling techniques might be particularly suitable for this environment.
Compare and contrast the expected cooling performance, cost-effectiveness, and complexity of implementation for each of the three techniques: spray water, serpentine water cooling, and “fame glass”.
Look for studies that compare active cooling methods like spray water and serpentine water cooling with passive methods potentially represented by “fame glass” to understand the reasons for including both categories.
Investigate if there are specific research goals or hypotheses in the mentioned research that necessitated the selection of these particular three cooling methods.
Explore research publications or studies from the Renewable Energy Laboratory at Assiut University in Egypt that discuss the rationale behind selecting spray water, serpentine water cooling, and “fame glass” for their PV cooling experiments.
The rest of the paper is organized as follows: Section “Methodology” presents the proposed methods for cooling PV panels, then Section “Measurements” describes tools which used in measurement, and the steps of measuring power, the results and their discussion displayed in Section “Results“, and finally, the conclusion of the proposed method in Section “Limitations of the study“.
Asyut Governate, located in Upper Egypt, is characterized by an arid climate with prolonged periods of high solar irradiance exceeding 1000 W/m² and ambient temperatures that frequently surpass 45–50 °C during summer months. Under such conditions, PV modules experience severe thermal stress, leading to pronounced efficiency degradation and accelerated aging. Therefore, cooling techniques selected for this study were required to be effective under extreme temperatures, technically simple, and adaptable to real outdoor installations.
Spray water cooling was selected due to its high heat removal capability through combined convective and evaporative mechanisms, which makes it particularly effective during peak temperature and irradiance periods. This method is well suited to regions where water is locally available and rapid temperature reduction is required. In contrast, the serpentine water-cooling technique was chosen to represent a more controlled and water-efficient active cooling approach, capable of providing stable thermal regulation and improved average power output throughout the operating day.
The glass-frame configuration was included as a passive cooling technique to evaluate a low-cost, energy-free thermal mitigation strategy that relies on partial shading and reduced radiative heat gain. Although such passive approaches may reduce module temperature, their suitability in high-irradiance arid climates remains uncertain due to potential optical losses. Accordingly, the inclusion of the glass-frame system enables a direct experimental assessment of the trade-off between temperature reduction and irradiance attenuation under the specific climatic conditions of Assiut.
Despite the extensive body of research on photovoltaic cooling techniques, a clear gap remains in experimentally comparing active and passive cooling strategies under identical real outdoor conditions in extremely hot arid climates. Many previous studies focus on individual cooling methods or rely on controlled laboratory environments, which limits the direct comparability of their results and their applicability to real operating conditions.
Motivated by the harsh climatic conditions of Upper Egypt, this study aims to provide a side-by-side experimental evaluation of three representative cooling techniques—spray water cooling, serpentine water cooling, and glass-frame cooling—using identical photovoltaic modules operating simultaneously under the same environmental conditions. The novelty of this work lies in its systematic assessment of the thermal–electrical trade-offs between highly effective active cooling methods and low-cost passive cooling approaches, with particular emphasis on water consumption, temperature reduction, and electrical performance degradation.
The findings of this study offer practical insights for selecting suitable PV cooling strategies in hot arid regions, supporting the deployment of more reliable and efficient photovoltaic systems under extreme environmental conditions.
This section of the study introduces three techniques for cooling photovoltaic panels. PV panels have been established on the laboratory rooftop, facing south at a tilt angle of 30 degrees, in line with the site’s latitude. Before the readings, every PV panel was maintained clean. The datasheet data is listed in .
Table 1.
The first technique, known as the Water-spray technique. The spray cooling system was designed as an open cycle system and had a pump with 45 watts, and maximum flow rate 1000 Liter / Hour, but the real flow is 2 Liter / minute, and provided with a 50-L storage tank as shown in Fig. 2. Dimensions of spray system consist of a 1-cm diameter of PVC tube which is fixed in the top of PV module with holes separated 2.5-cm apart. This PVC tube connected with a tank of water with a pump and control unit. The spray system works to spray water over the panel for 5 min at a time, pausing for 10 min and repeating all day or time of work, from 9:00 am to 3:00 pm, as shown in Fig. 3.
The schematic form of the spray water technique.
The water-spray system for cooling PV panels.
For the spray cooling technique, water was supplied at a flow rate of approximately 2 L/min, and the system operated intermittently for 5 min followed by a 10-min pause. Accordingly, each spraying session consumed approximately 10 L of water. This operating strategy was selected to balance effective evaporative cooling with water consumption efficiency, particularly under arid climatic conditions.
The spray system consisted of a PVC pipe with a diameter of 1 cm, fixed along the upper edge of the PV module frame. The pipe was perforated with evenly spaced holes at 2.5 cm intervals, allowing water to be uniformly distributed across the module surface. This configuration ensured homogeneous spray coverage and effective heat removal over the entire panel area, Fig. 4.
The schematic form of the serpentine cooling technique.
To mitigate any potential shading effects, the PVC pipe was positioned outside the active photovoltaic cell area and aligned with the module frame. Owing to its small diameter and peripheral placement, the shading impact on the incident solar irradiance was negligible and did not measurably affect the electrical output of the panel during operation.
The second cooling approach, referred to as the serpentine water-cooling technique, employs a closed-loop liquid circulation system installed on the backside of the photovoltaic module. Water was circulated through the serpentine at a constant flow rate of approximately 2 L/min. This flow rate was selected as a compromise between achieving sufficient convective heat removal and minimizing pumping power consumption, based on preliminary trials and the pump operating characteristics.
The cooling system was driven by a 45 W pump with a maximum rated capacity of 1000 L/h and connected to a 50-L storage tank. To enhance the cooling potential during high-temperature conditions, the water in the tank was pre-cooled using ice, maintaining the inlet water temperature below 10 °C. Water circulated continuously through the serpentine from 9:00 am to 3:00 pm, corresponding to the full operational period of the PV modules.
The serpentine heat exchanger was fabricated from a copper tube with a diameter of 0.5 cm and a total length of approximately 5 m, covering nearly 80–85% of the PV module backside area. The serpentine was firmly fixed directly onto the rear surface of the PV panel using mechanical clamps to ensure intimate thermal contact. This direct attachment minimized thermal contact resistance and enabled efficient heat transfer from the PV module to the circulating water, Fig. 5.
Iron-serpentine with water technique for cooling PV panels.
The fame-glass was implemented by placing a transparent glass sheet above the photovoltaic module with a uniform air gap of approximately 2 cm. This separation distance was selected to allow natural air circulation between the glass and the PV surface while reducing direct solar heat gain through partial shading and radiative attenuation, Fig. 6. The term “fame-glass” refers to a conventional glass sheet coated with a thin semi-shading layer that partially transmits solar radiation. This type of treated glass is similar to automotive tinted glass, which is commonly used to reduce solar heat gain while allowing partial sunlight transmission. In this study, the fame-glass layer is used as a passive cooling approach to limit excessive irradiance and thermal loading on the photovoltaic module.
The schematic form of the fame glass technique.
The glass used in this configuration had an average solar transmittance of approximately 85–90% in the visible spectrum. Despite its relatively high transmittance, the presence of the glass inevitably reduced the effective irradiance reaching the PV surface due to reflection and refraction losses.
The glass frame was mechanically supported using longitudinal mounting clamps, as shown in Fig. 4. These clamps introduced localized shading on the PV surface, which contributed to additional power losses beyond those associated with optical attenuation alone. While this shading effect affects direct comparability with the actively cooled configurations, it realistically represents practical installation constraints often encountered in low-cost passive shading systems. Consequently, the glass-frame technique was intentionally included to experimentally assess the thermal–electrical trade-off and the limitations of passive shading-based cooling approaches under real outdoor conditions, Fig. 7.
Fame-glass technique for cooling PV panels.
In this section, tools these used in measurements are presented, then steps for measuring output power are described in specific subsection.
Panel voltage (V), current (I), temperature (T^o C), and solar irradiance (w/m^2) were measured every hour from 9:00 am to 3:00 pm for six weeks. The measurement instruments are displayed in Fig. 8: Fig. 8-A shows both the voltmeter and clamp meter for reading voltage and current, respectively; Fig. 8-B shows the digital thermometer for temperature measurement; Fig. 8-C shows the light meter (lux-meter) for solar radiation measurement and; Fig. 8-D shows the variable resistance.
Measurement instruments for PV panel voltage, current, temperature, and radiation; Renewable Energy Laboratory at Assiut University, Egypt.
The current–voltage (I–V) characteristics of the photovoltaic modules were obtained using a variable resistive load. During each measurement cycle, the resistance value was gradually increased from near-zero (short-circuit condition) to a maximum value of 25 Ω (open-circuit condition). The resistance adjustment was performed in 19 discrete steps, ensuring sufficient resolution to accurately capture the full I–V curve.
At each resistance step, the corresponding current and voltage values were recorded after allowing a short stabilization period to ensure steady-state operation. This stepwise loading method enabled consistent I–V curve acquisition at each hourly measurement interval and ensured comparability between different cooling configurations under identical environmental conditions.
Table 2 listed the accuracy of the measurement instruments used in the experimental work.
First measurements are recorded of both the ambient temperature and sun irradiation readings, using digital thermometers and lux meters were linked.
Second, about panel:
For determining the short circuit current, immediately connect the panel’s two outer ends to form a short circuit, and then use the clamp for recording the current’s value.
The voltmeter is connected to both terminals of the cell, for reading the open circuit voltage and recording its value.
The clamp meter is wrapped around the wire to determine the current value, and the voltmeter is connected in parallel with the PV cell’s terminal to measure the output voltage value. The variable resistance is set to a low value and connected to the two outer ends of the PV panel.
Gradually raising the resistance value to create the I-V curve for PV panel.
Third, proceed these steps with all Panels by repeating the earlier processes.
Fourth, compared the cooling approaches using the measurement reading.
Finally, the ratio of the measured maximum output power (produced) of PV panels with its input power (amount of solar irradiance striking the array) is used to determine the panel’s efficiency⁵⁰. The power gain % (({eta})) obtained due to the application of the cooling technique relative to the reference PV panel (({P}_{max,Panel})) is calculated as follows:
where, ({P}_{max,Ref}) is the maximum output power measured with Reference Panel A.
This modification clarifies that the parameter represents the relative improvement in power output rather than the intrinsic PV efficiency.
To assess the reliability and repeatability of the experimental measurements, an uncertainty analysis based on standard deviation and error propagation was conducted. The standard deviation (σ) of a measured quantity was defined as follows:
where, ({x}_{i})​ represents the individual measured values, (overline{x }) is the mean value, and N is the number of measurements.
However, due to the outdoor nature of the experiment and the continuously varying environmental conditions (solar irradiance, wind speed, and ambient temperature), repeated measurements under identical conditions were not always feasible. Therefore, the uncertainty of the electrical power output was quantified using the propagation of uncertainty method, based on the measured voltage and current and their respective instrument accuracies.
The combined uncertainty in power output (ΔP) was calculated using:
where, P = VI and ΔV and ΔI represent the uncertainties in voltage and current measurements, respectively. This approach provides a reliable quantitative estimate of measurement uncertainty and ensures the comparability and credibility of the reported experimental results.
The outdoor experimental measurements were conducted using a manual I–V curve acquisition approach, which requires sufficient time to ensure stable operating conditions and accurate electrical characterization of each PV panel. Consequently, an hourly measurement interval was adopted to allow system stabilization and complete I–V data acquisition for all tested configurations under identical environmental conditions.
Unlike short-interval point measurements, each recorded data point in this study represents a full electrical characterization cycle rather than an instantaneous snapshot. Transient environmental fluctuations, such as short-term wind gusts or brief cloud passages, were inherently averaged during the measurement process. Furthermore, data reliability and consistency were ensured through uncertainty propagation analysis based on the specified accuracies of the measurement instruments, allowing anomalous deviations to be quantitatively bounded.
Accordingly, the selected measurement interval provides a reliable representation of the thermal and electrical performance trends of the PV panels under real outdoor conditions, while maintaining experimental feasibility and repeatability.
The electrical performance of the PV module was monitored using a calibrated digital multimeter to measure voltage and current, while the maximum output power was determined from the measured I–V characteristics. In addition, a temperature sensor was used to record the PV surface temperature, and a solar power meter was used to measure the incident solar irradiance during the experimental tests. All measurements were conducted under similar environmental conditions to ensure consistency and comparability between the different cooling configurations.
The study was done at Assiut University in Egypt in the Renewable Energy Laboratory. To guarantee maximum energy output year-round. The six-week period from 1^st September to mid-October 2022 saw the recording of results, which were checked hourly between 9:00 am and 3:00 pm. practically, to research the best PV cell cooling methods.
The thermal impact of several PV panel cooling methods is displayed. A few comparisons are shown, including the experimental I-V curves of each cooling method used during the day, the experimental output power of solar PV panels using various cooling techniques during the day, and a comparison of the output power of solar PV panels at various temperatures during the day using various cooling techniques.
In this study, four identical photovoltaic panels are used, each with a different cooling technique. As seen in Fig. 9, as follows:
Panel A: Reference Panel, Fig. 9-A, which was used to examine the impact of cooling methods on other PV panels but did not have any cooling techniques.
Panel B: As seen in Fig. 9-B, this PV panel cooling method uses water spraying.
Panel C: Water-based iron-serpentine process, as illustrated in Fig. 9-C.
Panel D: The fame-glass method, presented in Fig. 9-D, employs a different fame-glass.
PV panels with different cooling techniques, (A) Reference panel; (B) Water-spray technique; (C) Iron-serpentine with water technique; and (D) Fame-glass technique.
The experimental results were measured over 3 days on 1^st September, 15^th September, and 1^st October. Results of each day have a table that includes columns as follows (Time) has an hour of readings, (Temperature) has the reading of ambient temperature in Celsius degrees. (Irradiance) has the irradiance of solar in watts per meter square, and the maximum power per each panel in watts as (Panel (A)) (Panel (B)) (Panel (C)) (Panel (D)), respectively. This table also has the maximum efficiency (Max η), the maximum value (Max) and average (Mean) values of each column.
The following figures are displayed the I-V and P–V curves for the four panels included in the study. The black curves correspond to the reference panel (Panel A), which did not have any cooling techniques; the red curves correspond to the panel that applied the spray water technique (Panel B); the blue curves correspond to the panel that applied the iron-serpentine with water technique (Panel C); and the purple curves correspond to the panel that applied the fame-glass technique (Panel D).
The operating temperature of PV panels is a key factor influencing their electrical performance and overall efficiency, particularly under hot climatic conditions. In this study, the surface temperature of each PV panel was measured hourly at each data collection time step (from 9:00 am to 3:00 pm) using a calibrated digital thermometer, simultaneously with the electrical measurements.
Table 3 illustrates the temporal variation of PV panel temperature for the reference panel (Panel A) and the panels equipped with different cooling techniques (Panels B, C, and D). As expected, the reference panel exhibited the highest operating temperature throughout the day, reaching peak values during midday hours due to high solar irradiance and elevated ambient temperatures, which in some cases exceeded 50 °C.
The application of active cooling techniques resulted in a significant reduction in PV surface temperature. The water-spray cooling method (Panel B) demonstrated the most pronounced temperature reduction during peak irradiance periods, particularly around noon, due to the combined effects of convective heat transfer and evaporative cooling. This thermal reduction directly corresponded to the observed enhancement in output power and conversion efficiency during high-temperature intervals.
Similarly, the iron-serpentine water-cooling technique (Panel C) provided a consistent and stable temperature reduction over the operating period. Although its instantaneous cooling effect was slightly lower than that of the spray system at peak temperatures, it maintained lower average panel temperatures during morning and early afternoon hours, which explains its superior average power output observed in several measurement intervals.
In contrast, the fame-glass technique (Panel D) reduced the PV panel temperature by partially blocking direct solar radiation; however, this temperature reduction was accompanied by a substantial decrease in incident irradiance on the PV surface. Consequently, despite the lower operating temperature, the electrical output and efficiency were consistently inferior compared to both the reference and actively cooled panels.
Overall, the temperature variation analysis confirms that active cooling techniques are more effective than passive shading-based approaches in high-temperature environments. The results highlight a strong correlation between PV temperature reduction and electrical performance enhancement, validating the necessity of thermal management for PV systems operating in arid and hot climates such as Upper Egypt.
Tables 3, 4, 5 include the experimentally measured PV surface temperature values corresponding to the four analyzed cooling scenarios, which are used to interpret the observed variations in electrical power output.
The first day is the 1^st of September in 2022; the reading is during the daytime from 9 am to 1 pm. Its reading is listed in Table 3 which shows the comparison between the maximum output powers of each type of panels during this day. The Panel B records the highest maximum output power, while the Panel C records the best average output power, and maximum efficiency. However, the Panel D records the worst power output.
The results of the experiment of the I-V and P–V curves using various cooling methods of the same PV panels at 9 am, are displayed in Fig. 10. Panel B’s efficiency improves by approximately 5%, Panel C’s efficiency increases by approximately 18%, and Panel D’s efficiency declines by approximately 54%. Then, the experimental results of the I-V and P–V of PV panels at 10 am, with a maximum irradiation of 998 w/m² and maximum temperature of 50 °C, are displayed in Fig. 11. Panel B’s efficiency improves by approximately 21%, Panel C’s efficiency increases by approximately 9%, and Panel D’s efficiency decreases by approximately 41%. While Fig. 12 shows the experimental results of the I-V and P–V curves of PV panels at 11 am. Panel B’s efficiency improves by approximately 16%, Panel C’s efficiency increases by approximately 11%, and Panel D’s efficiency decreased by approximately 42%. Subsequently, at 12 pm the experimental results of PV panels’ I-V and P–V curves are displayed in Fig. 13. Panel B’s efficiency improves by approximately 16%, Panel C’s efficiency improves by approximately 25%, and Panel D’s efficiency decreased by approximately 36%. Finally, Fig. 14 shows the experimental results of the I-V and P–V curves of PV panels at 1 pm.
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 1^st, 2022; 9 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 1^st, 2022; 10 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 1^st, 2022; 11 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 1^st, 2022; 12 pm).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 1^st, 2022; 1 pm).
According to the results, on September 1^st, the Panel B cooling panels with spray water method recorded a maximum output power of 133 w and an enhanced efficiency of 21%; on the other hand, the Panel C cooling panels with iron-serpentine with water technique recorded the best average power 117.2 w, and best efficiency of 25% at the morning. Panel B outperforms Panel C between 11 am and 12 pm., whereas Panel C outperforms Panel B at other reading times.
The second day is the 15^th of September in 2022; the readings are during the daytime from 9 am and extend to 2 pm. In this case, the maximum temperature reading is 50 °C at 12:55 pm. Its reading is listed in Table 4 which shows a comparison between the maximum output power of each type of panel during this day. The Panel B records the highest maximum output power, while the Panel C records the best average output power, and the best efficiency. However, the Panel D records the worst power output.
The experimental results of the I-V and P–V curves utilizing various cooling methods of the same PV panels on September 15, 2022, at 9 am, are displayed in Fig. 15. Panel B’s efficiency gains approximately 1% over Panel A’s, Panel C’s efficiency gains approximately 2% over Panel A, and Panel D’s efficiency decreases approximately 70% less than Panel A’s reference panel. Then Fig. 16 displayed curves at 10 am. Panel B’s efficiency increases by approximately 1% compared to Panel A’s reference panel; Panel C’s efficiency increases by approximately 19% compared to Panel A; and Panel D’s efficiency decreases by approximately 40% compared to Panel A’s reference panel. Furthermore, Fig. 17 shows the experimental results of I-V and P–V curves using different cooling techniques of the same PV panels at 11 am with maximum irradiance = 1132 w/m². Panel B’s efficiency increases by approximately 18%, Panel C’s efficiency increases by approximately 14%, and Panel D’s efficiency decreases by approximately 42%. While Fig. 18 shows the experimental results of I-V and P–V curves using different cooling techniques of the same PV panels at 12 pm with maximum temperature = 50.7° C. Panel B’s efficiency is increased by approximately 23%, Panel C’s efficiency is increased by approximately 21%, and Panel D’s efficiency is decreased by approximately 32%, in this reading, panels with cooling techniques records the maximum produce of power in the day. Whereas in Fig. 19 that display experimental results at 1 pm records the maximum output of daily records for reference panel A, and Panel B’s efficiency is increased by approximately 11%, Panel C’s efficiency is increased by approximately 3%, and Panel D’s efficiency is decreased by approximately 50%. Finally, Fig. 20 shows the experimental results at 2 PM. Panel B’s efficiency is increased by approximately 21%, Panel C’s efficiency is increased by approximately 24%, and Panel D’s efficiency is decreased by approximately 37%.
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 9 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 10 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 11 am).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 12 pm).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 1 pm).
Experimental results of using different cooling techniques of same PV panels; (A) I-V curve, and (B) P–V curve; (September 15^th, 2022; 2 pm).
As shown in Fig. 19, Panel D exhibits a lower operating temperature compared to the reference single PV module. However, this temperature reduction does not translate into higher electrical power output. This behavior is attributed to the glass-frame configuration, which introduces partial shading and optical attenuation that significantly reduce the effective solar irradiance reaching the PV surface. Consequently, the irradiance loss dominates over the thermal gain, resulting in a net reduction in power output. This outcome highlights the inherent trade-off associated with passive glass-based cooling techniques in high-irradiance environments.
From these experimental results, on the 15^th of September, the maximum output power is produced from Panel B equal to 136 w with a maximum efficiency of Panel B at 23%. This recording in the maximum temperature record in reading, while Panel C records the best average power in the day equal 120.5 w, and the best efficiency 24% at after noon. Panel C did not record better efficiency in the mid-day time. On the other hand, Panel D records the worst solution to produce power.
The thermal impact of various PV panel cooling systems is shown in Fig. 21 at 12:55 PM with an irradiation of 1070 w/m² and an ambient temperature of 50 – 60 °C. The reference panel (Panel A) has high temperatures that reach 55–65 °C. Various cooling procedures are used to lower the temperatures, including water spraying (Panel B), iron-serpentine (Panel C), and fame-glass (Panel D). Under the high-temperature case, the effective solutions to reduce the temperature of PV panels are water spraying and fame-glass. However, fame-glass technique shaded the sunlight on the PV, which led to reduced production power.
Thermal effect resulting of different cooling techniques of PV Panels; (A) normal picture and (B) thermal picture. (September 15^th, 2022; 12:55 pm).
The third day is the 1^st of October in 2022, and the reading during the daytime is extended to 4 pm. To study the effect of different cooling systems before sunset, which shows the comparison between the maximum output power of each type of panel during this day. The Panel B records the highest maximum output power, while the Panel C records the best average output power. However, the panel D records the worst power output.
Utilizing various cooling strategies on the same PV panels. Figure 22 displays the experimental findings of the I-V and P–V curves at 11 am, In comparison to Panel A, the efficiency of Panel B using the spray-water technique increases by approximately 15%. In Panel C, the efficiency of the panel increases by approximately 3% over the reference panel (Panel A). In Panel D, the efficiency of the panel decreases by approximately 46% less than in Panel A when using the serpentine with water technique. Figure 23 displays the experimental findings of the I-V and P–V curves at 12 pm. Panel B’s efficiency increases by approximately 24% compared to Panel A’s reference panel; Panel C’s efficiency increases by approximately 6% compared to Panel A; and Panel D’s efficiency decreases by approximately 31% compared to Panel A’s reference panel. At 4 pm, Fig. 24 displays the experimental findings of the I-V and P–V curves. Compared to the reference panel (Panel A), the efficiency of panel B rises by around 0.1%; on the other hand, the efficiency of Panel C rises by approximately 2%; and Panel D, with an approximate decrease in efficiency of 68%.
Experimental results of using different cooling techniques of same PV Panels; (A) I-V curve, and (B) P–V curve; (October 1^st, 2022; 11 am).
Experimental results of using different cooling techniques of same PV Panels; (A) I-V curve, and (B) P–V curve; (October 1^st, 2022; 12 pm).
Experimental results of using different cooling techniques of same PV Panels; (A) I-V curve, and (B) P–V curve; (October 1^st, 2022; 4 pm).
As proven by the results, it is preferable to use the Serpentine with water approach (Panel C) rather than alternative methods for cooling solar PV panels at low temperatures. Using the spray water method (Panel B) is a better way to cool during high temperatures than the serpentine with water approach (Panel C). The Fame glass approach (Panel D) not recommended to use since its low power production in all measurements.
It should be noted that the observed performance enhancement under the different cooling configurations is mainly reflected in increased electrical power output due to reduced operating temperature. While power gains are significant, the corresponding conversion efficiency improvements are reported separately and remain comparatively modest, as summarized by the maximum efficiency values presented in Tables 3 and 4.
Although the fame-glass configuration (Panel D) resulted in a noticeable reduction in PV surface temperature, the experimental results presented in Tables 3, 4, 5 reveal a substantial decrease in output power that exceeds what can be explained by temperature effects alone. This behavior indicates the presence of additional loss mechanisms beyond thermal influence.
One major contributing factor is partial shading caused by the mechanical clamps used to secure the glass frame, as clearly visible in the thermal image shown in Fig. 6. These clamps obstruct a portion of the incident solar radiation, leading to localized shading and non-uniform irradiance distribution across the PV surface. Such shading effects are known to significantly reduce PV output power, even when the shaded area is relatively small.
Furthermore, the fame-glass layer introduces inherent optical losses due to reflection, refraction, and attenuation of incoming solar radiation, which reduce the effective irradiance reaching the PV cells. Consequently, despite operating at a lower surface temperature, Panel D consistently exhibited the lowest electrical performance among all tested configurations.
These combined effects explain the disproportionately large power reduction observed in the fame-glass scenario and confirm that this technique does not provide a fair thermal–electrical trade-off when compared to active cooling methods. Therefore, the fame-glass approach is not recommended for practical PV cooling applications, especially in high-irradiance and hot climatic conditions.
The noticeably lower power output recorded for Panel D is primarily attributed to optical losses caused by partial shading from the glass-frame mounting clamps rather than thermal effects. Although the glass-frame configuration achieved a measurable reduction in PV surface temperature, the associated irradiance attenuation and localized shading significantly limited the electrical output. Consequently, Panel D is considered as a reference passive shading case, and its results are discussed qualitatively to highlight the limitations of glass-based cooling approaches under practical outdoor installation conditions, rather than as a directly comparable alternative to the active cooling techniques.
The hourly variation of PV surface temperature for the investigated cooling configurations is presented to illustrate the dynamic thermal response of the modules under outdoor conditions. These temperature trends directly influence the corresponding electrical power output and efficiency behavior throughout the day. While efficiency variations are not plotted separately for each hour, their impact is inherently captured through the measured power output and summarized by the maximum efficiency values reported for each configuration.
The fame-glass technique demonstrates that temperature reduction alone is insufficient to guarantee performance enhancement when accompanied by significant irradiance attenuation.
The findings of this study provide preliminary insights into the applicability of PV cooling techniques, while long-term field validation remains a necessary step for large-scale deployment.
In this section, the cost analysis is presented for proposed cooling techniques. This analysis determined components with their capital and operation costs.
The capital cost is listed in .
Table 6, the item prices are from amazon expected the serpentine and fame glass sheet from local market. This table shows the spray water cooling system is the most effective system, and fame glass is the simplest. Serpentine has the high capital cost in its system.
Table 7 has tabulated the operation cost for various operation cost, fame glass as a passive system hasn’t needed to energy or water. While the cost in the active system, Spray water is effective in reduced energy consumption. However, it needs to water continuously. In the other side the serpentine cooling system most effective in saving water since it’s a closed cycle. Serpentine uses more energy because pump always works.
The main key to choose the suitable cooling system is the environment, is the water is available and with low cost, or not, then the ambient temperature degrees.
Despite the promising results obtained in this study, several limitations should be acknowledged. First, the experiments were conducted on a single PV module under specific outdoor environmental conditions, which may limit the generalization of the results to larger PV systems. Second, the study focused on a limited number of cooling techniques and a specific experimental configuration. Additionally, the experiments were carried out during a defined testing period, and seasonal variations were not fully considered. Future work may extend this study by investigating long-term performance under different climatic conditions and applying the proposed cooling techniques to larger PV installations.
Photovoltaic (PV) panels in normal operation without cooling as Panel (A), the highest efficiency of the panel is at noon, it is in the middle at 9 and 10 am and the lowest efficiency is at two in the afternoon. Since the effect of high ambient temperature. The fame glass technique has reduced the temperature of PV, whereas it reduced the efficiency all the time as shown in results in Panel (D), since this technique reduced the irradiance also.
The water spray technique has the highest efficiency of the panel at noon. But the efficiency is often low in the early morning and late afternoon. The iron serpentine technique has the highest efficiency most of the time except in the early morning and late afternoon as proven by results in Panels (B) and (C).
The results demonstrate that active cooling techniques, particularly spray and serpentine cooling, lead to noticeable improvements in PV electrical performance due to effective reduction of operating temperature. In contrast, the glass-frame technique, although successful in lowering the module temperature, results in a reduction in electrical output because of decreased incident irradiance. These findings confirm that temperature reduction alone does not guarantee efficiency enhancement, and that the balance between thermal cooling and irradiance availability is critical for achieving net performance gains.
This study has been recommended using the Serpentine with water technique, at low temperature. Whereas the water spray cooling technique is recommended at the high temperature. Finally, this study hasn’t been recommended the Fame-glass technique.
Other cooling techniques can be studied in future works such as Active air-cooling, PCM, Continuous water spraying cooling, and Back-water spraying cooling.
Testing the spray system with various flowrate and operation time to determine other settings.
Testing system of serpentine with different (on / off) times, and cooling the water with other method than icing to provide the more effective method and economic cost.
Using advanced devices to measure VI and PV curves.
Using individual temperature sensors to measure each panel temperature.
Using wind sensor to measure the effect of wind speed.
The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.
Abdelsattar, M. et al. Optimal integration of photovoltaic and shunt compensator considering irradiance and load changes. Comput. Electr. Eng. 97(2022), 1–20 (2022).
Google Scholar 
Sayed, N. G. E. et al. Electrical grid linked to PV/wind system based fuzzy controller and PID controller tuned by PSO for improving LVRT. Sci. Rep. 15, 7698 (2025).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Oda, E. S. et al. Optimal allocation of a hybrid photovoltaic-based DG and DSTATCOM under the load and irradiance variability. Int. Trans. Electr. Energy Syst. 31(11), e13131 (2021).
Article  Google Scholar 
Bahaidarah, H., Subhan, A., Gandhidasan, P. & Rehman, S. Performance evaluation of a PV (photovoltaic) module by back surface water cooling for hot climatic conditions. Energy 59, 445–453 (2013).
Article  Google Scholar 
Chandrasekar, M., Suresh, S., Senthilkumar, T. & Karthikeyan, M. G. Passive cooling of standalone flat PV module with cotton wick structures. Energy Convers. Manag. 71, 43–50 (2013).
Article  ADS  CAS  Google Scholar 
Choubineh, N., Jannesari, H. & Kasaeian, A. Experimental study of the effect of using phase change materials on the performance of an air-cooled photovoltaic system. Renew. Sustain. Energy Rev. 101, 103–111 (2019).
Article  Google Scholar 
Nethra, M. & Kalidasan, B. Earth tube heat exchanger design for efficiency enhancement of PV panel. Mater. Today Proc. 45, 587–591 (2021).
Article  Google Scholar 
Abdelsattar, M., Azim, M. A., AbdelMoety, A. & Emad-Eldeen, A. Comparative analysis of deep learning architectures in solar power prediction. Sci. Rep. 15, 1–20 (2025) (31729).
Article  Google Scholar 
Singer, S., Rozenshtein, B. & Surazi, S. Characterization of PV array output using a small number of measured parameters. Sol. Energy 32(5), 603–607 (1984).
Article  ADS  Google Scholar 
Abdelsattar, M., Mohamed, H. A. & Abuelkhair, A. F. Negative effects and processing methods review of renewable energy sources on modern power system: A review. Int. J. Renew. Energy Res. 14(2), 385–394 (2024).
Google Scholar 
Petrone, G., Spagnuolo, G., Teodorescu, R., Veerachary, M. & Vitelli, M. Reliability issues in photovoltaic power processing systems. IEEE Trans. Ind. Electron. 55(7), 2569–2580 (2008).
Article  ADS  Google Scholar 
Benavides, N. D. & Chapman, P. L. Modeling the effect of voltage ripple on the power output of photovoltaic modules. IEEE Trans. Ind. Electron. 55(7), 2638–2643 (2008).
Article  ADS  Google Scholar 
King, D. L. Photovoltaic module and array performance characterization methods for all system operating conditions. AIP Conf. Proc. 394(1), 347–368 (1997).
Article  ADS  CAS  Google Scholar 
Rezk, H., Tyukhov, I., Al-Dhaifallah, M. & Tikhonov, A. Performance of data acquisition system for monitoring PV system parameters. Measurement 104, 204–211 (2017).
Article  ADS  Google Scholar 
Ulieru, V. D., Cepisca, C., Ivanovici, T., Pohoata, A., Husu, A., and Pascale, L. Measurement and analysis in PV systems,” in Proceedings of the international conference on circuits, ICC, pp. 22–24 (2010).
Sutanto, B., Indartono, Y. S., Wijayanta, A. T. & Iacovides, H. Enhancing the performance of floating photovoltaic system by using thermosiphon cooling method: Numerical and experimental analyses. Int. J. Therm. Sci. 180, 107727 (2022).
Article  Google Scholar 
Haque, M. A., Miah, M. A. K., Hossain, S. & Rahman, M. H. Passive cooling configurations for enhancing the photovoltaic efficiency in hot climatic conditions. J. Sol. Energy Eng. 144(1), 011009 (2022).
Article  CAS  Google Scholar 
Mankani, K., Chaudhry, H. N. & Calautit, J. K. Optimization of an air-cooled heat sink for cooling of a solar photovoltaic panel: A computational study. Energy Build. 270, 112274 (2022).
Article  Google Scholar 
Özcan, Z., Gülgün, M., Şen, E., Çam, N. Y. & Bilir, L. Cooling channel effect on photovoltaic panel energy generation. Sol. Energy 230, 943–953 (2021).
Article  ADS  Google Scholar 
Hamdan, M. A. Performance enhancement of a photovoltaic module by passive cooling using water-based aluminum oxide nano-fluid. J. Ecol. Eng. 23(4), 276–286 (2022).
Article  Google Scholar 
Marudaipillai, S. K., Ramaraj, B. K., Kottala, R. K. & Lakshmanan, M. Experimental study on thermal management and performance improvement of solar PV panel cooling using form stable phase change material. Energy Sources A. Recovery Util. Environ. Eff. 45(1), 160–177 (2023).
CAS  Google Scholar 
Elbreki, A. et al. Experimental and economic analysis of passive cooling PV module using fins and planar reflector. Case Stud. Therm. Eng. 23, 100801 (2021).
Article  Google Scholar 
Rajput, U. J. & Yang, J. Comparison of heat sink and water type PV/T collector for polycrystalline photovoltaic panel cooling. Renew. Energy 116, 479–491 (2018).
Article  CAS  Google Scholar 
Wu, S. & Xiong, Ch. T. Passive cooling technology for photovoltaic panels for domestic houses. Int. J. Low-Carbon Technol. 9(2), 118–126 (2014).
Article  Google Scholar 
Zhang, Y. et al. A novel porous channel to optimize the cooling performance of PV modules. Energy Built Environ. 3(2), 210–225 (2022).
Article  CAS  Google Scholar 
Malvika, A., Arunachala, U. & Varun, K. Sustainable passive cooling strategy for photovoltaic module using burlap fabric-gravity assisted flow: A comparative energy, exergy, economic, and enviroeconomic analysis. Appl. Energy 326, 120036 (2022).
Article  Google Scholar 
Elminshawy, N. A., El-Damhogi, D., Ibrahim, I., Elminshawy, A. & Osama, A. Assessment of floating photovoltaic productivity with fins-assisted passive cooling. Appl. Energy 325, 119810 (2022).
Article  Google Scholar 
Mahmood, D. M. & Aljubury, I. M. A. Experimental investigation of a hybrid photovoltaic evaporative cooling (PV/EC) system performance under arid conditions. Results Eng. 15, 100618 (2022).
Article  Google Scholar 
Wang, W. et al. Integrated solar-driven PV cooling and seawater desalination with zero liquid discharge. Joule 5(7), 1873–1887 (2021).
Article  Google Scholar 
Kumar, A., Singh, A. P. & Singh, O. P. Effect of new overhead phase change material enclosure designs on thermo-electric performance of a photovoltaic panel. J. Energy Storage 46, 103814 (2022).
Article  Google Scholar 
Bria, A. et al. Effect of PCM thickness on the performance of the finned PV/PCM system. Mater. Today Proc. 72, 3617–3625 (2023).
Article  CAS  Google Scholar 
Zhao, J., Li, Z. & Ma, T. Performance analysis of a photovoltaic panel integrated with phase change material. Energy Procedia 158, 1093–1098 (2019).
Article  Google Scholar 
Moein-Jahromi, M., Rahmanian-Koushkaki, H., Rahmanian, S. & Jahromi, S. Evaluation of nanostructured GNP and CuO compositions in PCM-based heat sinks for photovoltaic systems. J. Energy Storage 53, 105240 (2022).
Article  Google Scholar 
Lotfi, M., Shiravi, A. H. & Firoozzadeh, M. Experimental study on simultaneous use of phase change material and reflector to enhance the performance of photovoltaic modules. J. Energy Storage 54, 105342 (2022).
Article  Google Scholar 
Ganesan, K., Winston, D. P., Ravishankar, S. & Muthusamy, S. Investigational study on improving the yield from hybrid PV/T modified conventional solar still with enhanced evaporation and condensation technique-an experimental approach. Energy Sour. Part A: Recover. Utilization, Environ. Eff. 44(2), 5267–5286 (2022).
Article  CAS  Google Scholar 
Sultan, S. M., Tso, C. P., Ervina, E. & Abdullah, M. Z. Cost and time effective performance evaluation methods for photovoltaic module cooling techniques: Analytical and experimental study. Appl. Energy 326, 119940 (2022).
Article  Google Scholar 
Chandra, S. et al. Influence of artificial and natural cooling on performance parameters of a solar PV system: A case study. IEEE Access 9, 29449–29457 (2021).
Article  Google Scholar 
Nabil, T. & Mansour, T. M. Augmenting the performance of photovoltaic panel by decreasing its temperature using various cooling techniques. Results Eng. 15, 100564 (2022).
Article  CAS  Google Scholar 
Golzari, S., Kasaeian, A., Amidpour, M., Nasirivatan, S. & Mousavi, S. Experimental investigation of the effects of corona wind on the performance of an air-cooled PV/T. Renew. Energy 127, 284–297 (2018).
Article  Google Scholar 
Hussien, A., Eltayesh, A. & El-Batsh, H. Experimental and numerical investigation for PV cooling by forced convection. Alex. Eng. J. 64, 427–440 (2023).
Article  Google Scholar 
Mohammadpour, J., Salehi, F., Sheikholeslami, M. & Lee, A. J. R. E. A computational study on nanofluid impingement jets in thermal management of photovoltaic panel. Renew. Energy 189, 970–982 (2022).
Article  CAS  Google Scholar 
Zhao, Y., Gong, S., Zhang, C., Ge, M. & Xie, L. Performance analysis of a solar photovoltaic power generation system with spray cooling. Case Stud. Therm. Eng. 29, 101723 (2022).
Article  Google Scholar 
Valiente-Blanco, I. et al. Efficiency improvement of photovoltaic solar modules by cooling using an underground heat exchanger. J. Sol.Energy Eng. 144(6), 061015 (2022).
Article  Google Scholar 
Erdoğan, İ, Bilen, K. & Kıvrak, S. Experimental investigation of the efficiency of solar panel over which water film flows. J. Polytech. 27(2), 699–707 (2023).
Google Scholar 
Raju, M., Sarma, R. N., Suryan, A., Nair, P. P. & Nižetić, S. Investigation of optimal water utilization for water spray cooled photovoltaic panel: A three-dimensional computational study. Sustain. Energy Technol. Assess. 51, 101975 (2022).
Google Scholar 
Liu, Z., Yang, X., Ali, H. M., Liu, R. & Yan, J. Multi-objective optimizations and multi-criteria assessments for a nanofluid-aided geothermal PV hybrid system. Energy Rep. 9, 96–113 (2023).
Article  Google Scholar 
Hussein, H. A., Numan, A. H. & Abdulrahman, R. A. Improving the hybrid photovoltaic/thermal system performance using water-cooling technique and Zn-H₂O nanofluid. Int. J. Photoenergy https://doi.org/10.1155/2017/6919054 (2017).
Article  Google Scholar 
Kane, A., Verma, V., Singh, B. J. R. & Reviews, S. E. Optimization of thermoelectric cooling technology for an active cooling of photovoltaic panel. Renew. Sustain. Energy Rev. 75, 1295–1305 (2017).
Article  Google Scholar 
Fabbri, G. & Greppi, M. Numerical modeling of a new integrated PV-TE cooling system and support. Results Eng. 11, 100240 (2021).
Article  Google Scholar 
Alamri, H. R., Rezk, H., Abd-Elbary, H., Ziedan, H. A. & Elnozahy, A. J. C. Experimental investigation to improve the energy efficiency of solar PV panels using hydrophobic SiO₂ nanomaterial. Coatings 10(5), 503 (2020).
Article  CAS  Google Scholar 
Download references
Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).
Electrical Engineering Department, Faculty of Engineering, Qena University, Qena, 83523, Egypt
Montaser Abdelsattar
Electrical Engineering Department, Faculty of Engineering, Assiut University, Assiut, 71518, Egypt
Ola Mostafa A. Saleh, Alaa F. M. Ali & Hamdy A. Ziedan
Electrical Technology Department, Egyptian German College (EGC), Misr International Technological University (MITU), Assiut, Egypt
Ola Mostafa A. Saleh
Faculty of Industrial and Energy Technology, New Assiut Technological University, Assiut, Egypt
Ashraf N. Eldeen Mourad
Department of Electrical Engineering, College of Engineering, Jouf University, Sakaka, 72388, Saudi Arabia
Meshari D. Alanazi
Electrical Power Engineering Department, High Valley Institute for Engineering and Technology, Qalyubia, Egypt
Mohamed Abdelhamid
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
M. Abdelsattar, O. M. A. Saleh, A. F. M. Ali, A. N. Mourad, M. D. Alanazi, H.A. Ziedan and M. Abdelhamid contributed to the design and implementation of the research, the analysis of the results, the writing of the manuscript and preparing figures. All authors reviewed the manuscript.
Correspondence to Montaser Abdelsattar.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Abdelsattar, M., Saleh, O.M.A., Ali, A.F.M. et al. Enhancing photovoltaic efficiency in arid climates using cooling strategies. Sci Rep 16, 16141 (2026). https://doi.org/10.1038/s41598-026-50636-6
Download citation
Received: 22 May 2025
Accepted: 22 April 2026
Published: 25 May 2026
Version of record: 25 May 2026
DOI: https://doi.org/10.1038/s41598-026-50636-6
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Scientific Reports (Sci Rep)
ISSN 2045-2322 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Alberta bans solar panels from landfills  Airdrie City View
source

Certisolis, a French certification body specializing in PV products and systems, has updated its list of modules that have obtained a Simplified Carbon Assessment (Évaluation Carbone Simplifiée, ECS). These certificates demonstrate compliance with the carbon-footprint criteria required to qualify for certain French support mechanisms, notably the S21 feed-in tariff and tenders organized by the French Energy Regulatory Commission (CRE).
The updated list provides information on valid ECS certificates, including the relevant module references, manufacturers, technologies, maximum power ratings, and the carbon-footprint thresholds they meet.
To facilitate the transition between the previous and new certification systems, Certisolis has also launched an online calculator. The tool allows users to convert the carbon value of an existing PPE2 certificate, issued under France’s second Multiannual Energy Program (PPE), into a new value designated as “non-LCA,” referring to a calculation that does not use the previous life-cycle assessment (LCA) methodology.
This new method is specifically required for the new CRE PPE2 tenders for building-mounted PV projects. Rather than analyzing the carbon impact of each stage of the module manufacturing process, as under the previous methodology, it uses an overall score based on the countries in which production takes place.
Before production begins, manufacturers can also request a provisional, or temporary, ECS certificate. This allows them to market a “low-carbon” version of a module and bid for future projects before purchasing all the components required to manufacture it.
“Some module manufacturers wait until they have ‘firm’ orders for low-carbon modules—which qualify for the S21 tariff—before sourcing ‘low-carbon’ components, which are more expensive,” Certisolis notes in a publication on the subject.
Once a sale of low-carbon modules intended to qualify for the S21 tariff has been finalized, for example, the manufacturer must purchase the low-carbon components specified in the temporary certificate.
The temporary certificate, however, cannot be used for commissioning. The modules purchased and installed must correspond exactly to the part numbers and components specified in the final certificate. Their manufacturing date must also fall within the certificate’s validity period.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

The U.S. Department of Commerce has announced preliminary affirmative determinations in its anti-dumping (AD) duty investigations into solar imports from India, Indonesia, and Laos.
According to a fact sheet released by Commerce, the agency determined preliminary dumping margins of 123.04% for India, 35.17% for Indonesia, and 22.46% for Laos.
These investigations were initiated last August following a petition by the Alliance for American Solar Manufacturing and Trade, a coalition of domestic manufacturers including First Solar, Hanwha Qcells, and Mission Solar. The group argued that a surge of low-priced imports from these nations was undercutting the U.S. manufacturing sector at a critical period of domestic expansion.
These new AD duties are in addition to the preliminary countervailing duties (CVD) announced by the Commerce Department in February 2026, which targeted government subsidies. When combined, the total preliminary duty exposure for many exporters from these countries has risen sharply. For India, total duties now reach approximately 234% for most manufacturers. In Indonesia, combined rates range between 121% and 178%, while in Laos, the total preliminary rate stands at roughly 103%.
U.S. Customs and Border Protection (CBP) will now require importers to post cash deposits based on these preliminary rates. The targeted nations represent a massive portion of the U.S. supply chain; according to government data, India, Indonesia, and Laos accounted for $4.5 billion in solar imports in 2025—roughly two-thirds of the total volume entering the country.
While these rates take effect immediately as cash deposit requirements, they remain preliminary until final determinations are issued later this year. Final decisions for India and Indonesia are scheduled for July 13, 2026, while Laos is expected to receive a final determination on or around September 9, 2026.
The final step in the process rests with the U.S. International Trade Commission (ITC), which must determine if these imports have caused material injury to the domestic industry. The ITC’s final injury determination is currently scheduled for October 19, 2026. If the ITC issues a negative determination, the investigations will be terminated, and all collected deposits will be refunded. If affirmative, final duty orders will be issued on October 26, 2026.
The U.S. government ha also announced preliminary affirmative determinations in these countervailing duty (CVD) investigations in late February.
 
 
 
 
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Noise complaint exposes unreported 2019 sale of East Lyme solar farm  Hartford Business Journal
source

Access to solar is security. Europe and America should work together  euractiv.com
source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 16, Article number: 27041 (2026) Cite this article
1034 Accesses
7 Altmetric
Metrics details
Photovoltaic (PV) solar cells generate waste heat during field operations, which reduces their overall power output. One potential solution for future solar power technology is to integrate solar cells with thermoelectric generators (TEGs) to enable waste heat reclamation, thereby enhancing power output. However, the high TEG resistance (R_TEG) increases the series resistance of these devices, leading to significant power loss. Here, we demonstrated that PV operation at low current and high voltage sufficiently reduces the impact of R_TEG, facilitating field-scale PV–TEG coupling. Furthermore, to achieve low-current, high-voltage operation, a shingled PV module configuration proved effective. This module comprises narrow strip-shaped solar cells connected in series; hence, the current is divided, and the voltage output across the strips is increased. Consequently, lower current and higher voltage than those of an uncut cell of the same size are achieved. Particularly, for a 14-strip shingled module, a load-resilient shingled PV module was realized for a field-scale PV–TEG (170 cm²) that delivers 3.27 W with a P_loss of only 0.043%. This shingled configuration is versatile and can be applied to any solar cell type, including organic, perovskite, and state-of-the-art tandem solar cells. Our study provides potential solutions to the problem of high R_TEG and new directions for achieving load-resilient PV modules for reliable field-scale PV–TEG coupling.
The Sun is a fundamental and inexhaustible source of energy. To address global climate change, expanding the utilization of carbon-free solar energy is essential to power a sustainable future¹. In order to achieve multi-terawatt scale solar energy use, various studies have focused on effective material consumption², implementing cradle-to-cradle recycling³, and developing advanced solar energy harvesting devices^4,5. The use of photovoltaic (PV) solar cells to convert sunlight into electricity is a carbon-free solution that is more cost-effective than coal or nuclear energy⁶. In 2024, PVs accounted for 6.9% of electricity worldwide, and this proportion is expected to rise⁷. Crystalline silicon (c-Si) single-junction solar cells are the current mainstream solar cells of the industry. Recent progress has pushed the performance of silicon heterojunction solar cells toward fundamental efficiency limits⁸. To further enhance efficiency, specific strategies have been explored, such as developing transparent-conductive-oxide-free front contacts⁹, studies on defect control^10,11, carrier-selective contacts¹², and novel material growth techniques¹³. Finally, understanding and mitigating heat generation in these devices remains critical for reliable operation¹⁴.
However, the performance of these cells is approaching their maximum theoretical efficiency, as they absorb only photons with energies higher than their bandgap and transmit all other lower-energy photons in the solar spectrum⁸. Moreover, in PV cells, excess energy from high-energy photons is dissipated as heat^8,14. Tandem solar cells, in which two or more subcells with different bandgaps are stacked to capture a broader range of the solar spectrum, are expected to become the mainstream solar cell technology in the future^4,5,15. However, these tandem cells require complex fabrication and integration processes involving different materials and layer structures, thereby increasing manufacturing costs. This makes tandem solar cells more expensive than single-junction solar cells. Furthermore, to optimize the efficiency of tandem solar cells, the current output of each subcell must be balanced by splitting the solar spectrum, which is achieved based on a simulated solar spectrum corresponding to a solar zenith angle of 48.19° (AM1.5G). Consequently, tandem solar cell configurations, including three-terminal¹⁶ and two- or four-terminal devices¹⁷, face inherent operational challenges¹⁸. Specifically, these multi-junction systems are highly sensitive to variations in atmospheric parameters¹⁹ as well as variable spectra and ambient temperatures²⁰. This environmental sensitivity often causes spectral mismatch in monolithic tandem stacks²¹, which dictates the limiting efficiency for current-constrained two-terminal devices²² and fundamentally affects the overall limiting efficiencies in multigap systems²³. Additionally, waste heat generation is an inherent challenge in all solar cell applications. This waste heat, generated by infrared (IR) irradiation, recombination, and excess energy from high-energy photons during PV operation^8,14, decreases the overall power outputs of solar cells in both single-junction and tandem configurations.
Hybrid coupling, combining PV cells with thermoelectric generators (TEGs), provides a promising alternative to tandem solar cells²⁴. This hybridization strategy has been effectively applied to diverse photovoltaics, including dye-sensitized and organic polymer solar cells^25,26,27, as well as perovskite modules and tandem structures^28,29,30. Furthermore, the performance of various integrated architectures and monolithic power generators has been extensively investigated^31,32,33. Recent efforts have also focused on novel high-performance devices aiming for lossless hybridization^34,35, along with efficient integrations and numerical thermal analyses^36,37. Instead of relying on precise spectral splitting across the subcells, the TEG in a PV–TEG hybrid coupling (PV–TEG) reclaims the heat wasted during PV operation by converting the temperature gradient across the device into electricity via the Seebeck effect. This conversion provides an additional electromotive force that contributes to the power output (Fig. 1a). However, despite their excellent prospects, studies on PV–TEGs are currently confined to laboratory-scale testing. Transitioning to practical field-scale PV–TEGs remains challenging because, in a PV–TEG, the TEG resistance (R_TEG) exceeds the PV resistance (R_PV). The high series resistance (R_s, where R_PV is added to a very high R_TEG) decreases the fill factor (FF)^34,35. The FF is the largest rectangular area within the current–voltage (I–V) curve and determines the maximum power output capability of the device. This reduction in FF lowers the electric power output (Fig. 1b). Therefore, minimizing the effects of R_TEG is key to preventing power output losses and realizing field-scale PV–TEG. However, in the last decade, PV–TEG coupling has been characterized by small area and low power output^{25,26,27,28,29,30,31,32,33,34,35,36,37}, with FF values ranging from 40 to 60%^{25,26,27,28,29,30}. Under such conditions, the decline in the FF arising from R_TEG would appear insignificant. Alternatively, PV and TEG components can be operated individually in a four-terminal configuration^30,31,32. However, this approach is less preferable because it requires additional wiring and inverters, which increase back-end-of-the-line process and outdoor operation costs. Consequently, the effects of R_TEG on the power output have been consistently underestimated and/or overlooked.
PV–TEG characteristics. (a) Schematic illustration of the PV–TEG hybrid coupling. A PV component is connected in series with a TEG that converts the temperature gradient from the PV operation into electricity through the Seebeck effect, thereby providing an additional electromotive force that contributes to the power output. The additional power output from the TEG, combined with that from PV operation, renders the PV–TEG a promising next-generation solar power device. As shown in this work, the PV–TEG can be constructed from commercially available c-Si PV and a Bi₂Te₃ TEG connected in series. (b) Schematic comparison of I–V curves of conventional PV (black) and PV–TEG (red) fabricated with uncut c-Si cells. Owing to the high R_TEG, connecting the TEG and PV in series reduces the PV–TEG FF, thereby decreasing the electric power output. This effect is more pronounced in large-area devices, where high-current operation exacerbates R_TEG. Overcoming this challenge is crucial for minimizing the power output loss and transitioning the PV–TEG from the laboratory to the field. (c) Schematic comparison of I–V curves of PV (black) and PV–TEG (red). The impact of R_TEG can be minimized by operating the PV at low current and high voltage with a shingled PV module. This shingled module divides the current while increasing the voltage output across the strips, enabling the PV component to operate efficiently at low current and high voltage. V_PV electromotive force from PV, V_TEG thermoelectric electromotive force from TEG.
Here, we demonstrate the significant impact of R_TEG on the power output of a PV–TEG, and propose a general approach to minimize these effects, thus enabling the development of a load-resilient PV–TEG coupling. Our systematic studies on TEG behavior in the PV–TEG revealed that a high R_TEG reduces the power output. Moreover, Joule heating arising from current flow across the TEG during operation further increases R_TEG and exacerbates this loss in power output. However, we discovered that PV operation at low current and high voltage mitigates the impact of R_TEG (Fig. 1c). Shingled PV modules are ideal for this type of operation because they consist of narrow strips of c-Si solar cells connected in series, which divide the current while increasing the voltage output across each strip. This configuration enables the PV components to collectively deliver a lower current but a higher voltage than an uncut cell of the same size. By integrating a shingled module with a TEG, we realized a field-scale PV–TEG that effectively minimizes power loss under simulated temperature gradient conditions. In addition, we developed a numerical model that can accurately predict the power loss of PV–TEG across various PV parameters. We also studied the slow thermal response of the TEG in the PV–TEG during characterization and addressed such artifacts^38,39. Remarkably, because these general criteria—i.e., PV operation at low current and high voltage to achieve field-scale PV–TEG—do not depend on the material, this shingled design applies to any solar cell type, including organic, perovskite, and state-of-the-art tandem solar cells. Our study provides potential solutions to the problem of high R_TEG and new directions towards the realization of a load-resilient PV module for reliable field-scale PV–TEG coupling.
Commercial passivated emitter and rear contact (PERC) cells composed of c-Si, produced by Shinsung Engineering, were used as the basis for the fabricated shingled modules. These PERC cells were divided into narrow strips using a 1064-nm IR laser for laser scribing, followed by mechanical cleaving. For the shingled modules with three, five, or seven strips, the total designed area was 100 cm²; for the 14-strip shingled module, it was 170 cm². The strip dimensions for the shingled modules were as follows: 100 mm × 38.83 mm, 100 mm × 21.70 mm, 100 mm × 16.07 mm, and 85 mm × 16.07 mm for three, five, seven, and 14 strips, respectively. To establish electrical connections, the strips were connected in series using CA 3556HF electrically conductive adhesive, and subsequently hot-pressed and cured at 180 °C for 1 min. The PV tabbing ribbon was soldered at both ends of the shingled modules, which were then encapsulated with a front glass, ethylene-vinyl acetate (EVA) sheet, and polyethylene terephthalate (PET) back sheet. Details of the fabrication process of a shingled module from an uncut cell are provided in the Supplemental Information, along with the raw I–V measurements of the shingled PV modules (Supplementary Fig. S1 and S2, respectively).
Commercial thermoelectric (TE) elements were fabricated by Xinrong (China). The p- and n-type TE materials were Bi_0.5Sb_1.5Te₃ and Bi₂Te_2.7Se_0.3, respectively. The element dimensions were 2 × 2 × 5 mm³ (width × depth × height). The Seebeck coefficients for the p- and n-type elements were found to be 220–230 and 205–215 µV·K^− 1, respectively; their resistivities were 1000–1050 and 1000–1050 µΩ/cm, respectively; and their thermal conductivities were 1.4–1.6 and 1.3–1.4 mW·cm^− 1·K^− 1, respectively. The electrical parameters were measured using a Zem-3 measurement system (ULVAC, Japan) at 22 °C. The material figures of merit were calculated to be 0.777 and 0.748 for the p- and n-type elements, respectively. The device figures of merit were 0.711 and 0.828, respectively, as measured using the Harman method.
In this study, 100 cm² TEG arrays were fabricated without a ceramic substrate. The voids between the TE elements within the TEG array were filled with polymeric foam to ensure mechanical robustness. This design facilitated efficient heat transfer to the TEG from heat sources of arbitrary shapes. In most instances, a TEG array with a resistance of 0.88 Ω was employed in the PV–TEG. However, TEG arrays with lower (0.057 Ω) and higher (3.61 Ω) resistances were tested for comparative analysis.
The TEG array with 0.88-Ω resistance employed in this study comprised 77 TE elements arranged in series with a parallel combination of four serial connection lines (77 × 4). The low- (0.057 Ω) and high-resistance (3.61 Ω) TEG arrays comprised 14 × 22 and 154 × 2 TE elements, respectively. In all cases, the total number of TE elements remained consistent at 308, and the total device area was strictly maintained at 100 cm² to ensure a constant heat dissipation area and facilitate meaningful comparisons. The substrate-free TEG array was fabricated by temporarily bonding a patterned Cu film to an adhesive polyimide substrate. For practicality, the polyimide film was temporarily affixed to a glass substrate. Subsequently, a screen printer was used to apply solder paste to the Cu film according to a predetermined pattern. A surface-mounting tool was used to position the TE elements in areas where the solder paste had been deposited. After soldering with a reflow machine, the empty spaces between the TE elements were filled with a polymer to ensure mechanical stability. Finally, the polyimide films were removed to expose the Cu electrodes for electrical connections. Further details can be found in our previous study²⁷.
The electrical performance of the PV-TEG hybrid devices was evaluated in two different configurations : two-terminal (2T), where the PV and TEG components are connected in a direct series arrangement requiring only a single pair of external contacts ; and four-terminal (4T), where the PV and TEG are operated as separate electrical circuits with independent contacts to avoid series resistance penalties from the TEG. Throughout this study, the proposed PV–TEG was predominantly configured in the 2T setup, with the 4T configuration reserved specifically for a loss analysis. The flexible design of the TEG array employed in this work, which utilizes thin Cu electrodes and a substrate-free, polymer-filled structure, provided necessary structural adaptability. This adaptability allowed the Cu electrodes to conform securely to the target surface, thereby ensuring good thermal contact without needing a thermal interlayer between the PV and TEG. The 2T configuration involved a simple series connection between the PV and TEG components. In contrast, the 4T configuration incorporated separate electrical contacts for the current inputs and outputs for both the PV and TEG components.
For practical applications, the 2T configuration is preferable owing to its simpler wiring, which results in lower costs during later fabrication stages. In detail, the 4T configuration comprises a top PV and a bottom TEG, and the module is equipped with either a power optimizer or two junction boxes. All top- and bottom-junction boxes must be wired and linked to separate inverters, creating two distinct strings. Consequently, implementing the 4T configuration introduces additional expenses for wiring and inverters during the back-end-of-the-line process and outdoor operation.
The I–V characteristics of the PV–TEG were measured under two different temperature gradients (ΔT), at 0 and 25 °C. For ΔT = 0 °C, we aimed to investigate the behavior of the TEG within the PV–TEG and elucidate the operational mechanisms of this hybrid system. The TEG was wrapped with a thermal insulator to isolate it from external temperature effects during the I–V measurements. We measured I–V while exposing the PV to simulated AM1.5G and 1 sun illumination. To assess the thermal response of the TEG, the integration time (t_int) was varied to 0, 0.5, 1, and 30 s during the measurements. We recorded the temperature changes in the TEG during the I–V measurement using a Fortic 340 IR camera and a Fluke 52 II thermometer, such that the temperatures on the hot and cold sides of the TEG could be cross-verified.
Because the performance of the PV–TEG was significantly affected by the temperature gradient, we conducted I–V measurements under a constant, simulated ΔT = 25 °C. To precisely achieve this ΔT, the hot side (PV temperature) was maintained at a constant value of 75 °C using the transparent top heater, and the cold side was maintained at 50 °C using the bottom cooler. This setup was designed to simulate realistic operational conditions, as during prolonged exposure to 1 sun solar radiation, the PV module temperature gradually increases, reaching a saturation point of approximately 80 °C³⁹. This temperature contrasts with the standard test conditions (STC) for solar cells, which are conducted at room temperature (25 °C). We designed a custom measurement system equipped with a transparent Cu mesh heater on top and a cooler at the bottom to generate the external temperature gradient required for these experiments. This setup generates an external temperature gradient while simultaneously transmitting simulated solar radiation of AM1.5G and 1 sun through the transparent Cu mesh heater, ensuring the PV module receives the standard irradiance. Further details of this custom-built measurement system can be found in our previous work³⁹. Raw I–V measurement results for the individual PV, TEG, and PV–TEG are reported in Supplementary Fig. S2 and S3. Further details of the PV–TEG measurements are provided in Supplementary Note S1.
For electrical characterization, p- and n-type TE elements were separated from the TEG array. Hall-effect measurements were performed at room temperature using an HMS-5300 Hall effect measurement system (Ecopia, South Korea). Cu wires were connected to the four corners of the TE elements using conductive Ag paste, thereby forming contacts for the van der Pauw configuration. To accurately determine the electrical properties and reduce experimental uncertainty, the parameters extracted from 20 independent measurement results at different current levels were averaged.
To measure the change in resistance over time, electrical contacts were formed by applying the conductive Ag paste at both ends along the long axis. The TE elements were then connected to a DC power supply, and constant voltage biases of 0.1, 0.35, and 0.7 V were applied to introduce current flow levels of 0.06, 0.2, and 0.4 A, respectively. These voltage levels were chosen to simulate the representative current level for I–V measurements on the actual PV–TEG. Under each constant voltage bias, the current flowing across the TE elements was measured every 1 s.
For the numerical analysis, the PV component was evaluated using an explicit double-diode model (MDDM) solved via the Lambert W-function^40,41, and the TEG performance was formulated based on the standard governing equations for thermoelectric generators³⁸. The detailed formula transformations, comprehensive parameter sets, and step-by-step derivations are provided in the Supplementary Information (Supplementary Note S2). It should be explicitly clarified that the simulated I–V curves of the PV-TEG system presented in this study were achieved through mathematical fitting of the experimental data to extract actual operating parameters, such as R_TEG, rather than through pure forward calculation.
The I–V curve for this 2T PV–TEG can be expressed as follows:
where I is the PV–TEG current; V is the applied voltage; V_TEG is the TE electromotive force from the TEG; (:{I}_{text{P}text{V}}) is the photocurrent generated by the solar cell; (:{I}_{text{D}1}) and (:{I}_{text{D}2}) represent the diffusion and recombination currents, respectively; R_PV is the series resistance of the PV cell; R_TEG is the resistance of the TEG; and R_sh is the parallel resistance^35,40,41.
We can express I_D1 and I_D2 as follows:
where I₀₁ and I₀₂ are the recombination currents; N is the number of strips in the shingled module; n₁ and n₂ are ideality factors; and V_th is the thermal voltage defined by the Boltzmann constant k, operating temperature T, and electron charge q, and expressed as (:{V}_{text{t}text{h}}=frac{kT}{q}). For these two exponential functions to be converted to the Lambert W function, it was necessary to combine them into a single exponential function. Therefore, the following process was implemented:
The expression
can be further simplified because (:{I}_{01}) and (:{I}_{02}) are smaller than (:{I}_{D1}) and (:{I}_{D2}); therefore, (:{I}_{01}) and (:{I}_{02}) can be neglected. Equation (5) can be simplified to a linear equation and rewritten as follows:
Further, Eq. (1) can be split into Eqs. (7) and (8) by substituting Eq. (6) into the MDDM.
Furthermore, the equations comprising an exponential term can be expressed using the Lambert W function. Finally, our numerical model for the PV–TEG using MDDM based on the Lambert W function can be written as follows:
For the loss analysis, the maximum output power (P_max) of the 2T PV–TEG (P_max−2T) was calculated for various combinations of the current at maximum output power (I_mp) and the voltage at the maximum output power (V_mp). The calculated P_max−2T value was subsequently compared with the P_max of the 4T PV–TEG (P_max−4T). P_max−4T was derived by computing the PV-only output power using the MDDM models^40,41 and adding it to the measured output power of the TEG. Therefore, the power loss (P_loss) can be expressed as follows:
Here, the 4T configuration represents an ideal independent operation of the PV and TEG components without the series resistance penalty from R_TEG, whereas the 2T configuration represents the practical series-coupled operation. Therefore, this equation practically quantifies the percentage of generated power that is unavoidably dissipated due to the high R_TEG when transitioning to a realistic 2T PV-TEG system. Further details of our numerical model are provided in Supplementary Note S2.
To explore and identify the conditions that would reduce the impact of R_TEG on our PV–TEG, we used our previously reported custom-built I–V measurement setup (Fig. 2a)³⁹. This setup is equipped with a transparent Cu mesh heater on top and a cooler at the bottom, generating an external temperature gradient while simultaneously transmitting simulated solar radiation of AM1.5G and 1 sun. This configuration enabled accurate characterization of our PV–TEG under realistic operating conditions. During the measurements, ΔT was set to 25 °C, and the PV temperature (T_PV) was approximately 75 °C³⁹. Note that this elevated T_PV is closer to real-world outdoor operating conditions than the standard test condition of T_PV = 25 °C⁴².
Negligible P_loss in PV–TEG for PV operating at low current and high voltage to reduce R_TEG impact. (a) Three-dimensional (left) and cross-sectional (right) schematic illustrations of our previously reported custom-built setup for measuring the I–V characteristics of PV–TEG³⁹. This setup generates an external temperature gradient using a transparent Cu mesh heater while simultaneously transmitting a simulated solar spectrum of AM1.5G and 1 sun. (b) Equivalent circuit of our PV–TEG comprising PV and TEG connected in series. Both the thermoelectric electromotive force (V_TEG) from the TEG and the photovoltaic electromotive force (V_PV) from the PV contribute to the PV–TEG power output. However, the R_s of the device increases significantly because it includes a very high R_TEG. Therefore, minimizing R_TEG is key to optimizing PV–TEG performance. (c–f) I–V and (g–j) P–V curves comparing PV (black) and PV–TEG (red) couplings using either uncut cells (c,g) or shingled modules with three (d,h), five (e,i), and seven strips (f,j). Because shingled modules with more strips operate at lower current and higher voltage than those with fewer strips or uncut cells, they maintain a high FF and avoid P_loss. Thus, a PV–TEG with shingled modules is more resilient to the impact of R_TEG. The PV–TEG devices employing five- and seven-strip shingled modules demonstrated power gains. The external temperature gradient was 25 °C, and the PV temperature was approximately 75 °C. (k) Comparison of FF values between PV (black) and PV–TEG (red) devices using uncut cell and three-, five-, and seven-strip shingled modules. Owing to the high R_TEG, the PV–TEG devices displayed lower FF values than the PV-only devices. The decrease in the FF was mitigated when shingled modules were used, suggesting that shingled modules are a key design feature for loss-free PV–TEG.
Based on the equivalent circuit of our PV–TEG (Fig. 2b), the FF can be expressed as
where V_mp and I_mp are the voltage and current at the maximum power point, respectively; V_oc is the open-circuit voltage; I_sc is the short-circuit current; and R_Sh is the shunt resistance of the solar cell ⁴³. Because R_sh is extremely high, its effect can be neglected. Therefore, Eq. (11) can be simplified to
From Eq. (12), the R_s of the PV–TEG comprises R_PV and R_TEG; therefore, an R_TEG considerably larger than R_PV in the PV–TEG decreases the FF. The FF indicates how closely the I–V curve resembles a rectangle, and the area of the rectangle that fits within the I–V curve directly corresponds to the power output (P = IV). A high R_s causes the I–V curve to deviate from a rectangular shape. Consequently, the decrease in FF caused by a high R_TEG decreases the power output of the PV–TEG. Equation (11) also shows that reducing the I_mp-to-V_mp ratio by operating the solar cell under a low current and high voltage could mitigate the impact of R_TEG. As shingled modules operate at lower current and higher voltage than uncut cells of the same area, they are ideal PV components for PV–TEG.
Overall, to achieve loss-free PV–TEG for field-scale applications, the impact of R_TEG can be minimized by implementing one of two approaches: reducing R_TEG or incorporating a PV less susceptible to R_TEG. Although the R_TEG can be easily reduced in practice by decreasing the number of series connections and/or increasing the number of parallel connections in the TEG array, this approach is undesirable because it increases the operational current and heightens the risk of Joule heating proportional to I²R. Further, a TEG array with parallel connections provides a low thermoelectric electromotive force; thus, this configuration would not yield an efficient PV–TEG. Therefore, the best way to attain a high-performance PV–TEG is to use a PV component that is less affected by R_TEG.
Thus, we compared the I–V characteristics of PVs and PV–TEGs using uncut cells or shingled modules with three, five, or seven strips. The I–V curves (Fig. 2c–f) and the corresponding power–voltage (P–V) curves (Fig. 2g–j) show that increasing the number of strips in the shingled module increased FF and decreased P_loss. In addition, the PV–TEGs with five- and seven-strip shingled modules displayed power gains (Supplementary Tables S2 and S3). These results demonstrate that operating the PV component at low current and high voltage makes the PV–TEG less susceptible to the influence of R_TEG.
Consistent with previous studies^26,34,35, because of the high R_TEG, the PV–TEG in this study displayed a significantly lower FF than that of the PV-only devices (Fig. 2k; see Supplementary Fig. S2 and S3 for the raw I–V measurements for the PV, TEG, and PV–TEG). Importantly, when PVs were coupled with a TEG, a less pronounced decline in FF was observed for the shingled PV modules compared to the results for an uncut cell with the same area. This mitigated decrease in FF observed for shingled modules suggests a key design feature for a load-resilient PV for field-scale PV–TEG.
To understand how R_TEG contributes to power loss in the PV–TEG, we examined the I–V characteristics of the device with no external ΔT in the TEG, where the PV was illuminated with simulated solar radiation of AM1.5G and 1 sun (100 mW/cm²). As the current flowed, R_TEG increased, rendering characterization of the PV–TEG difficult. This is attributed to the TEG’s slow thermal response³⁸. With no external temperature gradient (ΔT = 0), infrared (IR) thermal imaging during I–V measurements revealed that the PV current flow generates a temperature gradient in the TEG via the Peltier effect, cooling the top of the TEG while heating the bottom (Fig. 3a–c). The temperatures of the top and bottom were recorded during I–V measurements under illumination, for the three different shingled module configurations (i.e., three, five, and seven strips). The temperature gradient forms rapidly within 5 s of the current flow being initiated (Fig. 3d). Meanwhile, current flow also introduces Joule heating (I²R) to the TEG, albeit considerably more slowly (i.e., over a period of 2 min after the start of current flow). Therefore, current flowing through the TEG introduces a rapid Peltier effect followed by considerably slower Joule heating. The differential thermal response in the TEG indicates that slow Joule heating is the primary reason for the gradual increase in R_TEG during I–V measurements. This slow thermal response significantly impacts the characterization of the PV–TEG. Particularly, the PV–TEG characterization results depend on the integration time (t_int), i.e., the duration during which the voltage at each measurement point is sustained. the I–V curves of the PV–TEG exhibited steeper slopes at lower t_int and more moderate slopes at higher t_int (Fig. 3e). This difference arose because, during a fast voltage sweep (t_int = 0 s), the TEG could not reach a thermal steady state owing to the slow Joule heating, and R_TEG increased during the I–V measurement. In other words, a faster voltage sweep can make the PV–TEG perform better, which is misleading. We also found that the slow thermal response introduced hysteresis into the I–V characteristics during a fast voltage sweep (Fig. 3f). The absence of hysteresis at a slow voltage sweep (t_int = 30 s) indicates that the TEG reached a thermal steady state under these conditions (Fig. 3g). Therefore, when measuring the I–V characteristics of the proposed PV–TEG, the thermal response of the TEG during operation must be carefully considered (for a detailed discussion of the TEG behavior during I–V measurement, see Supplementary Note S1, Supplementary Fig. S4–S6, and Supplementary Table S1).
High R_TEG deteriorates PV–TEG FF, and slow Joule heating in TEG further increases R_TEG. IR thermal images showing (a) top, (b) bottom, and (c) side views of TEG elements (assembled with p- and n-type Bi₂Te₃) in PV–TEG during I–V measurements. Current flow induces the Peltier effect, cooling the top (LT, RT) and heating the bottom (LB, RB). (d) Changes in TEG-element temperatures over time at the top (LT, RT), middle (LM, RM), and bottom (LB, RB) points are marked in (c). The Peltier effect occurred within 5 s of the onset of current flow, whereas Joule heating occurred more slowly over 2 min following the initiation of current flow, increasing the overall temperature of the TEG. (e) Effects of integration times (t_int) on PV-TEG I–V curves. Because of Joule heating, R_TEG increased slowly during I–V measurements, and the TEG did not reach a thermal steady state at low t_int. The I–V curves of the PV–TEG exhibited hysteresis at t_int = 0 s (f) but not at t_int = 30 s (g), because Joule heating is a slow process. Therefore, PV–TEG characterization at t_int = 30 s was more accurate because the TEG reached a thermal steady state. In the hysteresis test, first (black) and second (red) I–V sweeps were sequentially performed in the forward and reverse directions. Linear relationship between PV current and temperature gradient generated on TEG top and bottom for (h) three-, (i) five-, and (j) seven-strip shingled modules. Temperature gradients were generated on the top and bottom of the TEG as functions of the PV current, which was measured without external ΔT.
The temperature gradient at the top and bottom of the TEG exhibited a linear relationship with the PV current (Fig. 3h–j). The slopes of the linear functions for the three-, five-, and seven-strip shingled modules were found to be 10.7, 10.89, and 12.41, respectively. Note that the linear relationship between the generated temperature gradient |T_top – T_bottom| and the PV current comprises the definition of the Peltier coefficient. Considering the charge-carrier motion across the TEG during PV–TEG operation, the rate of heat generation (Q) at the contact can be expressed as a linear function of the current:
where (:varPi:) is the Peltier coefficient, which indicates the amount of heat energy released or absorbed at a junction. A higher Peltier coefficient suggests more efficient TE conversion⁴⁴. Furthermore, the Seebeck and Peltier coefficients are interdependent. Similar to the Seebeck coefficient, the Peltier coefficient (:varPi:) is defined as the rate of heat generation Q with respect to electrical current I, as given by
Here, S and ∆T are the Seebeck coefficient and external temperature gradient, respectively^45,46. Considering the slopes of the linear functions, the relationship between the Seebeck and Peltier coefficients suggests that the PV–TEG functions with considerably higher efficiency when a seven-strip shingled module is employed.
Based on these results, we propose a working mechanism for the TEG in the PV–TEG. Owing to the high R_TEG, connecting the TEG to the PV in series decreases FF, thus reducing the power output. Through Joule heating, the current flowing in the PV–TEG further increases R_TEG. Given that R_TEG reduces the power output of the PV–TEG, an increase in R_TEG via Joule heating would degrade the performance of the PV–TEG. Therefore, our results unequivocally show that the impact of R_TEG on the PV–TEG is significant and must be minimized.
Based on these experimental results, we developed a numerical model to predict the P_loss of the PV–TEG under various PV parameters (Supplementary Note S2, Supplementary Fig. S7 and S8). Our numerical model for the PV–TEG using MDDM employs the Lambert W function to calculate the I–V characteristics⁴⁰. Therefore, we simulated I–V curves and compared them with the measured I–V curves to obtain the maximum coefficient of determination. Loss analysis typically involves fitting the measured I–V curve to a numerical model that describes the modeled I–V curves of the PV–TEG. The fitting parameters of the numerical model were I₀₁, I₀₂, R_PV, R_TEG, and R_sh. By adjusting these parameters during the fitting process, the model was optimized to match the measured data. For example, the effective R_TEG of the PV–TEG can be deduced from the numerical model (Fig. 4a). Notably, by applying our numerical model to the measured I–V curves of the PV–TEG, the best fit for the effective R_TEG was obtained at 1.22 Ω, instead of the nominal value of 0.88 Ω. This modeling outcome implies an increase in R_TEG due to Joule heating. Once the model was well fitted to the experimental data, the PV–TEG parameters, such as the maximum output power (P_max), I_mp, V_mp, I_sc, V_oc, and FF, could be determined. Our numerical model predicted that the P_loss of the PV–TEG could be minimized to negligible levels when I_mp was as low as 0.5 A and V_mp as high as 7 V (Fig. 4b).
Accurate prediction of PV–TEG power loss by numerical model for various PV parameters. (a) Sample fitting results for PV–TEG R_TEG obtained using a numerical model and considering the maximum coefficient of determination. Applying our numerical model to the measured I–V curves of the PV–TEG, we obtained a best fit for R_TEG at 1.22 Ω instead of the nominal value of 0.88 Ω. This modeling outcome indicates that Joule heating increased R_TEG. (b) P_loss of PV–TEG calculated by our numerical model shown as a function of the PV I_mp and V_mp. Consistent with the experimental results, our numerical model also predicts that a low I_mp and high V_mp minimize the PV–TEG P_loss.
Encouraged by the experimental results and predictions of our numerical model, we examined a large-area (170 cm²) PV–TEG coupling using a 14-strip shingled module with a low I_mp of 0.5 A and a high V_mp of 7 V (Fig. 5a,b, Supplementary Table S4). Under a simulated condition of ΔT = 25 °C, our PV–TEG coupled device generated 3.27 W (Fig. 5c,d) and exhibited a P_loss of only 0.043%, as determined by the measured power outputs of each component in the PV–TEG: 3.096 W for the 14-strip shingled module and 0.1754 W for the TEG. Notably, the total power output of our PV–TEG far exceeds that of the best-performing devices reported in the literature, which currently stands at 1.15 W (Fig. 5e and Supplementary Table S5)³³. The shingled module, which divides the current while increasing the voltage across multiple strips, effectively suppressed the effect of R_TEG to achieve such a low P_loss and enabled our PV–TEG to maintain a low current and high voltage over a much larger area of 170 cm² than the largest area reported thus far in the literature, i.e., 68 cm² (Fig. 5e and Supplementary Table S5)³². These results indicate that the impact of R_TEG can be minimized to achieve field-scale PV–TEG coupling with negligible P_loss, and that our numerical model accurately predicts the P_loss of the PV–TEG. However, although we successfully demonstrated a load-resilient shingled PV module for a field-scale PV–TEG, this work was limited to a laboratory-based proof-of-concept demonstration, with no actual thermal couplings for realistic devices. Additionally, no outdoor operational tests were performed. Therefore, future research should focus on evaluating the operational reliability of PV–TEG through comprehensive outdoor testing under real-world conditions.
Load-resilient shingled PV module for TEG coupling with negligible P_loss. (a) Schematic illustration and (b) photograph of our field-scale (170 cm²) PV–TEG coupling constructed using a 14-strip shingled module and a TEG. (c) I–V and (d) P–V curves of 14-strip shingled modules with PV only (black) and PV–TEG (red), showing power outputs of 3.096 and 3.27 W, respectively. As the TEG power output was 0.1754 W, the PV–TEG P_loss was only 0.043%. As anticipated from our numerical model, the low-current, high-voltage operation of the 14-strip shingled module successfully suppressed the P_loss originating from R_TEG. The simulated, external temperature gradient was 25 °C, and the PV temperature was approximately 75 °C. (e) Compared to the current state-of-the-art PV–TEG devices reported in the literature^{25,26,27,28,29,30,31,32,33,34,35,36,37}(gray circles), our PV–TEG (red star) displays the highest power output (3.27 vs. 1.15 W) over the largest area (170 vs. 68 cm²). Inset: magnified view of a small-area, low-power region occupied by the best-performing PV–TEG devices reported previously.
Our systematic studies showed that R_TEG substantially affects the performance of a PV–TEG. Specifically, a high R_TEG decreases the FF of the PV–TEG, inducing a loss in the electric power output. We showed that current flowing through the TEG during operation introduces Joule heating, further elevating R_TEG. This heightened R_TEG contributes further to the loss in power output. Through numerical simulations and experiments, we demonstrated that PV operation at low current and high voltage mitigates the impact of R_TEG. Using a 14-strip shingled module, which divides the current while increasing the voltage across multiple strips, we realized a load-resilient shingled PV module for a field-scale PV–TEG (170 cm²) that delivers 3.27 W with a P_loss of only 0.043%. The scale and performance of our PV–TEG represent significant advances over the largest (68 cm²) and best-performing (1.15 W) devices reported thus far in the literature. Unlike tandem solar cells, which require complex monolithic integration and sophisticated spectral splitting, our PV–TEG involves only a straightforward connection of commercially available PV and TEG components, with no front-end-of-the-line fabrication being necessary.
In the future, our load-resilient PV–TEG coupled device should be tested for feasibility with optimized thermal couplings between the PV/TEG and TEG/heat sink interfaces, possibly in outdoor operational tests. Finally, because the general criteria of low-current and high-voltage operation reported here are independent of the material, the shingled PV module used in our PV–TEG can, in principle, be implemented with any type of solar cell, including (but not limited to) organic and perovskite solar cells, as well as state-of-the-art tandem solar cells. Materials with larger bandgaps tend to display lower current and higher voltage than c-Si. Consequently, their application as PV components in the shingled design employed herein could further improve the power output of the resultant PV–TEG.
The data that support the results of this study are available from the corresponding authors, Hee-eun Song and Ka-Hyun Kim, upon reasonable request.
Victoria, M. et al. Solar photovoltaics is ready to power a sustainable future. Joule 5, 1041–1056 (2021).
Article  CAS  Google Scholar 
Zhang, Y., Kim, M., Wang, L., Verlinden, P. & Hallam, B. Design considerations for multi-terawatt scale manufacturing of existing and future photovoltaic technologies: Challenges and opportunities related to silver, indium and bismuth consumption. Energy Environ. Sci. 14, 5587–5610 (2021).
Article  CAS  Google Scholar 
Peters, I. M., Hauch, J. & Brabec, C. Cradle-to-cradle recycling in terawatt photovoltaics: A vision of perpetual utility. Joule 8, 899–912 (2024).
Article  CAS  Google Scholar 
Wright, M. et al. Design considerations for the bottom cell in perovskite/silicon tandems: A terawatt scalability perspective. Energy Environ. Sci. 16, 4164–4190 (2023).
Article  CAS  Google Scholar 
Heydarian, M. et al. Recent progress in monolithic two-terminal perovskite-based triple-junction solar cells. Energy Environ. Sci. 17, 1781–1818 (2024).
Article  CAS  Google Scholar 
US Energy Information Administration. Cost and performance characteristics of new generating technologies. Annual energy outlook. (2022). https://www.eia.gov/outlooks/aeo/assumptions/pdf/table_8.2.pdf
Electricity production by source. Our World Data, (2024). https://ourworldindata.org/grapher/electricity-production-by-source
Sun, Z. et al. Toward efficiency limits of crystalline silicon solar cells: Recent progress in high-efficiency silicon heterojunction solar cells. Adv. Energy Mater. 12, 2200015 (2022).
Article  CAS  Google Scholar 
Li, S. et al. Transparent-conductive-oxide-free front contacts for high-efficiency silicon heterojunction solar cells. Joule 5, 1535–1547 (2021).
Article  Google Scholar 
Kim, D. H. et al. Unraveling mixed-defect transformations and passivation dynamics in silicon heterojunction solar cells. Adv. Funct. Mater. 35, e08814 (2025).
Article  CAS  Google Scholar 
An, J. H. et al. Silicon–hydrogen bonding configuration modified by layer stacking sequence in silicon heterojunction solar cells. ACS Appl. Energy Mater. 5, 15029–15037 (2022).
Article  CAS  Google Scholar 
Mo, S. I. et al. Design rule of electron- and hole-selective contacts for polycrystalline silicon-based passivating contact solar cells. ACS Appl. Mater. Interfaces. 15, 46849–46860 (2023).
Article  CAS  PubMed  Google Scholar 
Hong, J. E. et al. Fully bottom-up waste-free growth of ultrathin silicon wafer via self-releasing seed layer. Adv. Mater. 33, e2103708 (2021).
Article  PubMed  Google Scholar 
Xu, L. et al. Heat generation and mitigation in silicon solar cells and modules. Joule 5, 631–645 (2021).
Article  CAS  Google Scholar 
Hu, H. et al. Triple-junction perovskite–perovskite–silicon solar cells with power conversion efficiency of 24.4%. Energy Environ. Sci. 17, 2800–2814 (2024).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Kikelj, M. et al. Do all good things really come in threes? The true potential of 3-terminal perovskite–silicon tandem solar cell strings. Joule 8, 852–871 (2024).
Article  CAS  Google Scholar 
Raza, E. & Ahmad, Z. Review on two-terminal and four-terminal crystalline-silicon/perovskite tandem solar cells; progress, challenges, and future perspectives. Energy Rep. 8, 5820–5851 (2022).
Article  Google Scholar 
Baiju, A. & Yarema, M. Status and challenges of multi-junction solar cell technology. Front. Energy Res. 10, 971918 (2022).
Article  Google Scholar 
Theristis, M., Fernández, E. F., Stark, C. & O’Donovan, T. S. A theoretical analysis of the impact of atmospheric parameters on the spectral, electrical and thermal performance of a concentrating III–V triple-junction solar cell. Energy Convers. Manag. 117, 218–227 (2016).
Article  ADS  Google Scholar 
Theristis, M. & O’Donovan, T. S. Electrical-thermal analysis of III–V triple-junction solar cells under variable spectra and ambient temperatures. Sol Energy. 118, 533–546 (2015).
Article  ADS  Google Scholar 
Brown, A. S. & Green, M. A. Radiative coupling as a means to reduce spectral mismatch in monolithic tandem solar cell stacks theoretical considerations. Conf Rec IEEE Photovolt. Spec. Conf, 868–871 (2002).
Brown, A. S. & Green, M. A. Limiting efficiency for current-constrained two-terminal tandem cell stacks. Prog Photovolt. : Res. Appl. 10, 299–307 (2002).
Article  Google Scholar 
Martí, A. & Araújo, G. L. Limiting efficiencies for photovoltaic energy conversion in multigap systems. Sol Energy Mater. Sol Cells. 43, 203–222 (1996).
Article  Google Scholar 
Tang, J., Ni, H., Peng, R. L. & Wang, N. Zuo, L. A review on energy conversion using hybrid photovoltaic and thermoelectric systems. J. Power Sources. 562, 232785 (2023).
Article  CAS  Google Scholar 
Chen, T. et al. Thermoelectric Bi₂Te₃-improved charge collection for high-performance dye-sensitized solar cells. Energy Environ. Sci. 5, 6294–6298 (2012).
Article  CAS  Google Scholar 
Zhang, Y. et al. Integrated energy-harvesting system by combining the advantages of polymer solar cells and thermoelectric devices. J. Phys. Chem. C. 117, 24685–24691 (2013).
Article  ADS  CAS  Google Scholar 
Kim, Y. J., Park, S. E. & Cho, B. J. A wearable organic photovoltaic-thermoelectric (OPV-TE) hybrid generator to minimize the open-circuit voltage losses of OPV module. Nano Energy. 93, 106775 (2022).
Article  CAS  Google Scholar 
Liu, Z. et al. Novel integration of carbon counter electrode based perovskite solar cell with thermoelectric generator for efficient solar energy conversion. Nano Energy. 38, 457–466 (2017).
Article  CAS  Google Scholar 
Fu, P. et al. Integrating large-area perovskite solar module with thermoelectric generator for enhanced and stable power output. Nano Energy. 65, 104009 (2019).
Article  CAS  Google Scholar 
Guo, X. Z. et al. Hybrid tandem solar cell for concurrently converting light and heat energy with utilization of full solar spectrum. J. Power Sources. 195, 7684–7690 (2010).
Article  ADS  CAS  Google Scholar 
Cotfas, D. T., Cotfas, P. A., Machidon, O. M. & Ciobanu, D. Investigation of the photovoltaic cell/ thermoelectric element hybrid system performance. IOP Conf. Ser. : Mater. Sci. Eng. 133, 012037 (2016).
Article  Google Scholar 
Deng, Y., Zhu, W., Wang, Y. & Shi, Y. Enhanced performance of solar-driven photovoltaic–thermoelectric hybrid system in an integrated design. Sol Energy. 88, 182–191 (2013).
Article  ADS  Google Scholar 
Kim, Y. J. et al. High-performance monolithic photovoltaic–thermoelectric hybrid power generator using an exothermic reactive interlayer. ACS Appl. Energy Mater. 2, 2381–2386 (2019).
Article  CAS  Google Scholar 
Wang, N., Han, L., He, H., Park, N. H. & Koumoto, K. A novel high-performance photovoltaic–thermoelectric hybrid device. Energy Environ. Sci. 4, 3676–3679 (2011).
Article  CAS  Google Scholar 
Park, K. T. et al. Lossless hybridization between photovoltaic and thermoelectric devices. Sci. Rep. 3, 2123 (2013).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Xu, L. et al. Efficient perovskite photovoltaic-thermoelectric hybrid device. Adv. Energy Mater. 8, 1702937 (2018).
Article  Google Scholar 
Darkwa, J., Calautit, J., Du, D. & Kokogianakis, G. A numerical and experimental analysis of an integrated TEG-PCM power enhancement system for photovoltaic cells. Appl. Energy. 248, 688–701 (2019).
Article  ADS  Google Scholar 
Min, G. New formulation of the theory of thermoelectric generators operating under constant heat flux. Energy Environ. Sci. 15, 356–367 (2022).
Article  CAS  Google Scholar 
Kim, H. J. et al. Rapid evaluation system for photovoltaic cell and thermoelectric generator hybrid devices depending on temperature via transparent heater. ACS Appl. Energy Mater. 5, 2669–2674 (2022).
Article  CAS  Google Scholar 
Chen, Y., Sun, Y. & Meng, Z. An improved explicit double-diode model of solar cells: Fitness verification and parameter extraction. Energy Convers. Manag. 169, 345–358 (2018).
Article  ADS  Google Scholar 
Jain, A. & Kapoor, A. Exact analytical solutions of the parameters of real solar cells using Lambert W-function. Sol Energy Mater. Sol Cells. 81, 269–277 (2004).
Article  CAS  Google Scholar 
Kurnik, J., Jankovec, M., Brecl, K. & Topic, M. Outdoor testing of PV module temperature and performance under different mounting and operational conditions. Sol Energy Mater. Sol Cells. 95, 373–376 (2011).
Article  CAS  Google Scholar 
Green, M. A. & Cells, S. Operating principles, technology and system application 79–98 (Prentice Hall, Inc., 1982).
Kasap, S. O. Principles of electronic materials and devices 488–492 (McGraw-Hill, 2017).
Bhattacharya, P. & Comprehensive Semicond Sci. Technol. (Elsevier Sci., 326–358. (2011).
Letcher, T., Maidment, G. G. & Zi, Q. Comprehensive renewable energy second edn (Elsevier, 2022).
Download references
This work was supported by the Research and Development Program of the Korea Institute of Energy Research (KIER) [Grant Number C6-2403-14]; a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) [Grant Numbers RS-2024-00347775, 2017M1A2A2086911]; an Institute of Information & Communications Technology Planning & Evaluation (IITP)– Information Technology Research Center (ITRC) grant funded by the Korean government (Ministry of Science and ICT) [Grant Number IITP-RS-2024-00437284]; the Commercialization Promotion Agency for R&D Outcomes (COMPA) funded by the Ministry of Science and ICT (MSIT) (2710071341, Research Equipment Technician Training Program); and the Technology Innovation Program Development Program funded by the Ministry of Trade, Industry & Energy (MOTIE) (RS-2023-00265858). This work was conducted during the 2025 research year of Chungbuk National University.
Kyuhyeon Im and Sungeun Park contributed equally.
Photovoltaics Research Department, Korea Institute of Energy Research, Daejeon, 34129, South Korea
Kyuhyeon Im, Sungeun Park, Kwan Hong Min, Sang Hee Lee, Min Gu Kang, Kyung Taek Jeong & Hee-eun Song
Department of Physics, Chungbuk National University, Cheongju, 28644, South Korea
Kyuhyeon Im & Ka-Hyun Kim
Graduate School of Energy and Environment, Korea University, Seoul, 02841, South Korea
Kyuhyeon Im & Hae-Seok Lee
School of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, 34141, South Korea
Yong Jun Kim & Byung Jin Cho
Advanced Batteries Research Center, Korea Electronic Technology Institute, Seongnam, 13509, South Korea
Yonghwan Lee
AI Convergence R&D Group, R&D Innovation Division, Korea Institute of Ceramic Engineering and Technology, Jinju, 52851, South Korea
Soo Min Kim
Department of Materials Engineering and Convergence Technology, Gyeongsang National University, Jinju, 52828, South Korea
Tae Kyung Lee
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Kyuhyeon Im: Investigation (characterization) Sungeun Park: Investigation (numerical model) Yong Jun Kim: Investigation (sample fabrication) Yonghwan Lee: Investigation (characterization) Kwan Hong Min: Investigation (sample fabrication) Sang Hee Lee: Investigation (discussion) Soo Min Kim: Investigation (characterization) Min Gu Kang: Investigation (discussion) Kyung Taek Jeong: Investigation (discussion) Hae-Seok Lee: Investigation (discussion) Byung Jin Cho: Investigation (discussion) Hee-eun Song: Conceptualization, Funding acquisition, Project administration, Resources, Writing–review and editing, Investigation (discussion), Supervision Tae Kyung Lee: Conceptualization, Methodology, Investigation, Visualization, Writing–review and editing Ka-Hyun Kim: Conceptualization, Methodology, Investigation, Data curation, Visualization, Writing – original draft, Writing – review and editing, Supervision.
Correspondence to Hee-eun Song, Tae Kyung Lee or Ka-Hyun Kim.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Below is the link to the electronic supplementary material.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Im, K., Park, S., Kim, Y.J. et al. Load-resilient shingled photovoltaic module for field-scale thermoelectric coupling. Sci Rep 16, 27041 (2026). https://doi.org/10.1038/s41598-026-56895-7
Download citation
Received: 20 February 2026
Accepted: 03 June 2026
Published: 13 June 2026
Version of record: 28 August 2026
DOI: https://doi.org/10.1038/s41598-026-56895-7
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Scientific Reports (Sci Rep)
ISSN 2045-2322 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Perovskite Photovoltaics From Lab Efficiency to Mass Commercialization  IDTechEx
source

An engineer, representing Kimley-Horn and Soltage, outlined the solar project for about a dozen residents in attendance last Thursday.
PSC staff recommending approval of 18-megawatt project north of downtown
By Brian Shane
Staff Writer
(Sept. 3, 2026) A proposal to build what would become Worcester County’s largest solar farm has won the backing of state regulators as the state’s Public Service Commission moves closer to a final decision.
Public Service Commission staff and the Maryland departments of Natural Resources and the Environment are recommending approval of the 18-megawatt project, proposed for a site just north of downtown Berlin.
Their recommendations for a proposed solar array from New Jersey-based developer Soltage were presented Thursday during a public comment hearing at the Berlin branch of the Worcester County Library.
The project would cover about 113.7 acres of a 157-acre property at the northwest corner of Routes 50 and 113. The property is zoned for industrial use but is now being farmed.
Seven people spoke. Some residents questioned the loss of farmland, while a statewide environmental group supported the project as a source of cleaner and more affordable energy.
Berlin resident Andrew Cadogan said he was not opposed to solar energy but argued that panels should be installed atop existing structures instead of taking agricultural land out of production. As an example, he mentioned the solar-covered parking area at Wor-Wic Community College in Salisbury.
“We’re giving up a huge tract of land,” Cadogan said. “I don’t understand why we are not going the route, like they did with Wor-Wic, where we’re covering already developed areas, parking lots, buildings, and areas that are not farmland with these solar developments.”
Brian Gilliland, an Eastern Shore clean-energy organizer for the Maryland League of Conservation Voters, said community solar offers subscribers discounted electricity while helping Maryland reduce greenhouse-gas emissions.
“Marylanders are listing energy affordability among their top concerns, and community solar is a smart solution,” Gilliland said. “This project reflects the kind of thoughtful, forward-thinking energy development that Maryland currently needs.”
The Town of Berlin has not taken a position for or against the project.
District 1 Councilman Steve Green said the town still wants details about how the land would be restored once the solar farm reaches the end of its working life, a process known as decommissioning.
“Decades from now, we feel strongly that Berlin’s taxpayers and neighboring property owners should not inherit the responsibility for removing this infrastructure,” Green said at the hearing. “We are unaware at this point what the decommissioning plan is.”
Green’s district includes the project site. He requested that a copy of the developer’s decommissioning plan be sent to the town so it could be shared with residents. Green is also the editor of this newspaper.
Soltage’s paperwork with the state calls for the estimated cost of decommissioning, minus the equipment’s salvage value, to be recalculated every five years. The developer previously stated it would post a bond to cover the cost of decommissioning.
Project engineer Nick Leffner said the developer estimates the project could generate about $8.5 million in tax revenue for Berlin and $910,000 for Worcester County over its operating life, compared with approximately $130,000 from the land’s current agricultural use. The state application says the project could operate for 30 years or more.
Berlin resident Bob Mitchell, who is also Worcester County’s director of environmental programs, questioned those projections. He said comparing the solar project with the property’s current agricultural use did not account for its industrial zoning or the tax revenue and jobs that another type of development could generate.
Maryland generally limits individual community solar projects to 5 megawatts but allows multiple projects to share the same industrially zoned property – an arrangement known as co-location.
Soltage’s proposal consists of six separately connected arrays: two producing 5 megawatts, two producing 3 megawatts, and two producing 1 megawatt.
Plans call for landscaped buffers, agricultural-style fencing, and preservation of the property’s topsoil. Soltage previously said construction could begin next year after state approval and more local permitting.
Thursday’s meeting was the project’s second public comment hearing. The first was held June 22 at a community center in Willards.
The project still needs approval from the Public Service Commission. Before it makes that decision, a judge will hold a formal hearing at 10 a.m. Oct. 1, when the parties can present evidence and question witnesses.

source

Renewable energy company Alight has inaugurated the 101 MW Eurajoki Solar Park in western Finland. The project is one of the country’s largest solar facilities, according to Alight. Built, owned and operated by Alight, the park will generate around 100 GWh of renewable electricity annually, enough to match the consumption of about 20,000 households. Swedish automotive safety company Autoliv will access the project’s output through a long-term virtual power purchase agreement (VPPA). The project is Alight’s first commissioned solar facility in Finland. It is expected to add about 12% to the country’s utility-scale solar capacity.
UK renewable energy developer Anesco has secured project financing from Lombard for three projects in the UK. The financing covers the already operational 21 MW Woodwalton Solar Farm in Cambridgeshire; the 50 MW/100 MWh Rothienorman battery energy storage system (BESS) in Scotland, which is nearing commissioning; and the 48.5 MW Coven Solar Farm in Staffordshire, which has reached financial close. Anesco CEO Hildagarde McCarville said the financing would help move the projects forward and contribute to the UK’s energy transition and security.
METLEN, an energy and metals company with operations in renewable energy, has signed a 10-year power purchase agreement (PPA) with Coca-Cola Tria Epsilon. The latter is the Coca-Cola bottling company in Greece. The agreement covers electricity from an approximately 12 MW solar plant in Mikro Perivolaki, near Velestino in Greece. Coca-Cola Tria Epsilon will purchase 100% of the plant’s output, estimated at about 16.5 GWh annually. The company claims to already source all electricity used at its production facilities from renewable sources. The agreement is also linked to parent company Coca-Cola HBC’s target of reaching net-zero carbon emissions by 2040.
Vattenfall, a European energy company active in renewable power generation, has officially opened its 46 MW Nauen Solar Park near Berlin, Germany. The project became operational in June 2026 and was connected to the regional distribution grid. The solar park has around 80,000 panels across 40 hectares and is expected to generate about 47 GWh of electricity annually. It was built without government subsidies. The project’s economics are supported by a 10-year power purchase agreement (PPA) with Wieland Group, a copper and copper-alloy semi-finished products manufacturer. Wieland will purchase the plant’s entire electricity output, which is expected to cover around 15% of the electricity it buys annually for its German sites.
TEAL, a renewable energy company focused on developing clean energy projects, has launched TEAL Renovables, a new development company targeting the Spanish market. The new entity will focus on the early-stage development of renewable energy assets in Spain. TEAL said the country was selected because of its policy framework, growing demand for clean energy and plans to expand renewable power deployment. TEAL Renovables will work with local partners and stakeholders to develop renewable energy projects supporting Spain’s decarbonization and electrification goals.
Orrön Energy, a publicly listed renewable energy and infrastructure company within the Lundin Group of Sweden, has completed its strategic transaction with Cloudberry Clean Energy. It has now become the largest shareholder of Cloudberry with 27.01% stake. This deal, including two board positions, has created a Nordic independent power producer (IPP) with about 2.1 TWh of annual proportionate power generation, says Orrön. The latter retains the 86 MW Karskruv Wind Farm in Sweden and a European development portfolio of about 12 GW spanning solar, battery storage and data center projects.
Catalonia’s regional government, led by President Salvador Illa, plans to triple the region’s renewable energy capacity by 2030, and by 12 times by 2050 with solar power playing a major role in the expansion. The government said it aims to significantly increase the number of solar parks while accelerating renewable energy deployment across Catalonia. The target is part of its broader plan to increase renewable electricity generation and reduce dependence on fossil fuels. Catalonia has over 2 GW of operational solar energy capacity, including self-consumption. Illa was speaking during the inauguration of the 300 MW Alcarràs Solar Park. It is to be expanded by 100 MW along with the addition of batteries with up to 4 hours duration.
TaiyangNews 2024

source

First-of-its-kind research balances the needs of solar farms and grazing  phys.org
source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 16, Article number: 23222 (2026) Cite this article
1230 Accesses
Metrics details
Energy retention from losses is the primary goal of fault detection methodology for photovoltaic (PV) solar systems. A fault detection model should be designed effectively to minimize power and cost waste. We propose a novel fault detection and localization method that leverages deep learning techniques for PV systems. The model is a hybrid semantic segmentation method that combines Convolutional Neural Networks (CNN) and multi-level transformer neural networks. We examine our model on four different segmentation datasets, which have varied characteristics and different capturing conditions, to ensure our model generalization. Using aerial inspection, the first thermal dataset images give a high performance in locating the faulty cells in PV arrays with an accuracy and global precision of 99.89%, a mean average precision (mAP) of 88.73%, a mean intersection over union (mIoU) of 76.29%, and 84.48% of mean dice (mDice, or mean/unweighted F1 score). We use three electroluminescence datasets to investigate the precise location and class of different fault types and minor cracks at the cell level. The second dataset result achieved perfect detection and segmentation of anomaly areas in cells with 99% accuracy, 96.36% mIoU, 98.13% mDice, and 98.41% mAP. The first two datasets contain binary segmentation images with fault and no-fault classes; to accommodate multiple classes, we have utilized the third and fourth datasets. The third dataset, comprising five PV failure classes, is evaluated against other models, yielding a superior performance of 96.27% precision, 77.64% mAP, 57.3% mIoU, and 68.45% mDice. The final dataset has 29 segmentation classes for testing 25 classes, for which it achieves a precision of 95.05%, 66.7% mAP, 49.3% mIoU, and 59.3% mDice.
Renewable energy resources have grown increasingly vital recently, as they mitigate environmental degradation. One of the most crucial resources is solar energy, which is expected to overtake the most common renewable energy sources, such as wind and hydropower, by the year 2030¹. It is also anticipated to supplant traditional resources (e.g., natural gas and oil) in the foreseeable future due to its affordability, safety, and cleanliness. More than 1.64 TW of photovoltaic plants have been installed to produce solar energy in 2023, with a generated capacity of approximately 1.294 TWh, while over 50% were installed in the past three years. PV is the fastest-growing renewable energy source, with a year-over-year increase of more than one quarter (25.6%), resulting in over 35 countries having a gigawatt-scale annual market^2,3. PV systems encounter numerous energy losses due to multiple defects. Particular strategies must be implemented to prevent or mitigate losses, depending upon the severity and cause of the defects. Temporary failures, such as soiling and shading, can be readily addressed; however, permanent ones, including cracks and delamination, may necessitate professional attention⁴. Consequently, a robust fault detection and diagnosis system (FDD) is necessary for monitoring photovoltaic (PV) systems to enhance their performance and dependability by promptly identifying and locating failures as they occur. The FDD system should detect a fault, identify its type and exact location and promptly isolate it to prevent the risk of fire in the event of serious failures. This process of isolation is exceedingly difficult and necessitates additional information and specialized expertise from Operation and Maintenance (O&M) services, who determine whether the fault poses a threat and if immediate intervention is necessary⁵.
Artificial intelligence (AI) technologies are involved in a significant number of aspects and fields that facilitate dealing with hard or complex problems that help the machine to act like humans and see what they see, especially in computer vision (CV) applications that would assist fault detection systems for automatic supervision and monitoring of PV systems. FDD systems based on AI methods such as deep learning (DL) and machine learning (ML) can be designed using historical data-driven information from actual PV systems. Other model-based FDD systems depend on the mathematical principles of the PV system or model simulation, which compares readings from sensors with normal operation to find out the problem but not its precise location. DL has become an effective tool for diagnosing and detecting faults in PV modules rather than traditional techniques that may be insufficient, particularly for large-scale PV arrays⁶, because DL can adapt to new knowledge and handle unpredictable conditions, making it effective for lengthy and complex tasks as the linkages between inputs and outputs become clear.
Computer vision techniques enable computers to analyze, interpret, and process the visual world through the use of artificial intelligence and deep learning. One common method for detecting faults in solar modules is visual and thermal imaging technologies. Thus, a data-driven operating FDD system depends on collecting a large number of images under many situations and various scenarios to give the machine the opportunity to learn. In the thermography (infrared IR) technique, the images are collected remotely, likely by drones to cover big areas, without any intervention or system interruption as one of the nondestructive methods. It is a quick, straightforward, dependable, precise, and cost-effective method that requires just an infrared camera to distribute the distinctive properties of PV modules in two dimensions.
One of the recent deep learning techniques, the transformer neural network (TNN), was originally developed for natural language processing (NLP)⁷. It was considered a revolution in sequential processing since it could handle complex dependencies for large tasks employing attention mechanisms by processing sequential input in parallel. In 2020, the authors of the vision transformer model⁸ changed the transformer NN’s design to make it more practical for computer vision applications such as image recognition and classification procedures. Since then, Transformer NN models have been widely adopted in numerous fields, such as autonomous driving, medical image analysis, remote sensing, land coverage analysis and recently PV fault detection^{9,10,11,12,13,14,15}.
The FDD system of⁹ uses a new deep learning model based on a modified multiscale Vision Transformer (ViT) neural network technique to improve the globality and robustness of the system. Preprocessing methodologies are applied to enhance and increase 20,000 thermography images of the dataset for detecting and classifying eleven PV module anomalies, e.g., soiling, bypass diode, cracking, shadowing, and hotspots.
Also, in¹⁰, a Vision Transformer model and five hybrid ViT models, which combine the ViT model with different kinds of Machine Learning (ML) algorithms, are the methods that were developed and compared in this study to handle the problem of classifying and comprehending the nature of faults in thermal images.
To facilitate dynamic communication and data exchange between dispersed energy resources, such as PV systems, and the grid infrastructure, smart grids and Internet of Things technologies (IoT) are essential. Real-time monitoring, adaptive load management, and predictive defect detection are all made possible by IoT-enabled smart grids, which improve grid reliability. The work by¹⁶ identifies PV system problems using PV arrays’ on-board devices (IoT modules) that apply a fault detection algorithm based on traditional I-V curve electrical measurement analysis and then notify a number of unmanned aerial vehicles (UAVs) with RGB and infrared cameras to do thermal and visual examinations of the damaged photovoltaic panels. The IoT modules receive the inspection data and take the necessary action.
One of the key deep learning tasks in computer vision is the semantic segmentation method, which assigns a class (label) to each pixel in an image rather than classifying the entire image (classification) or just a portion of it (object recognition). The model finally produces a color copy of the original image based on the PV defects classification. This algorithm can manage the diverse, intricate backgrounds and textures of PV panel surfaces, precisely delineating distinct parts of the panels from images and recognizing the location and kind of failures. The DeepLab series, Mask R-CNN, PSPNet, SegNet, and U-Net are some examples of popular semantic segmentation techniques.
The method in¹⁷ introduced a segmentation methodology in which thermal images are collected from a UAV fitted with infrared sensors to identify faulty panels on large solar plants. Segmentation models with varying encoder architectures are applied, which include DeepLabV3+, Feature Pyramid Network (FPN), and U-Net. However, it just identifies the module’s cells as either faulty or non-faulty.
The study by¹⁸ presents a deep learning framework that integrates Attention Mechanisms, Residual Blocks, and Atrous Spatial Pyramid Pooling (ASPP) to improve the U-Net architecture after the image preprocessing stage for IR images. Together, these improvements enhance contextual comprehension, fault localization, and feature extraction, thereby overcoming the limitations of conventional segmentation techniques.
On the other hand, electroluminescence (EL) imaging technologies are needed to overcome the drawbacks of low resolution in RGB and thermal images and also the challenge of detecting small (often invisible) defects. EL imaging is an essential diagnostic technique for evaluating the quality and functionality of PV modules. In this technique, PV modules are provided with direct current (DC) to facilitate radiative recombination in the solar cells during the capturing of the images. Numerous defects, including cracks, finger interruptions, delamination, and ablation, can be seen in these images; each one could require a different solution for repairing or replacing it.
Authors of¹⁹ implement a semantic segmentation model for EL images to detect irregularities in photovoltaic panels at the cell level using Convolutional Block Attention Module (CBAM), Attention Refinement Module (ARM), and ASPP modules and employ K-Net as a baseline to improve network performance, especially in identifying small-area connected faults and defect edges.
Another study²⁰ is able to distinguish between cracks, contact interruptions, cell connection failures, and contact corrosion for both multi-crystalline and monocrystalline silicon cells. The suggested model makes use of a DeepLabv3 segmentation model with a ResNet-50 backbone. In order to address class imbalance, it was trained using 17,064 EL pictures, including 256 physically realistic simulated images of PV cells.
Based on EL polarization imaging and the fundamental mechanism of PV cell crack generation, the work in²¹ suggests a new semantic segmentation method for PV cells. Three single-channel pictures of PV cells are obtained, which are polarization intensity (I), degree of polarization (DOP), and polarization quadrature (Q), which are then stacked into several channels to produce an extensive dataset. After the dataset has been annotated for training purposes, the texture features linked to microcrack faults are analyzed. In order to optimize the utilization of the unique characteristics of the polarized three-channel images, they improve the conventional U-Net design by incorporating the Efficient Channel Attention module, depth-wise separable convolutions, and the Squeeze-and-Excitation attention mechanism.
A new architecture has been developed²² for semantically segmenting 29 different characteristics and failures in PV panel EL pictures by replacing the SegNet architecture encoder with a pre-trained VGG16 encoder and using a CBAM block to improve the decoder’s capacity for producing fine-grained segmentations.
Also, a ViT SegFormer-based PV cell segmentation system for fault identification is employed¹¹ to automate the visual examination of defects in photovoltaic modules of EL images, incorporating fault pseudo-colorization.
In order to address the problem of extreme imbalance between abnormal and background pixel distributions, the authors of²³ presented a method for detecting micro-crack anomalies in photovoltaic modules based on an M-shape structure and attention module for classification and segmentation network. They showed superior performance for segmenting micro-cracks and more effectively extracted and fused shallow-level and deep-level features in the segmentation network.
In²⁴ they present an end-to-end deep learning pipeline that uses EL pictures to identify, locate, and segment cell-level anomalies from complete solar modules using weakly supervised segmentation (autoencoder), image classification (EfficientNet), and object detection (modified Faster-RNN).
In the field of industrial defect detection, there are also promising technologies using model optimization, attention mechanisms, and data augmentation. The YOLOv5-based lightweight algorithm in²⁵ prioritizes computational efficiency and multi-class defect recognition, making it suitable for edge deployment (resource-constrained). In contrast, the MSWindD-YOLO model in²⁶ specifically enhances feature discrimination by integrating attention mechanisms to capture subtle or irregular surface defects better. Meanwhile, the EP-YOLOv12n in²⁷ with improved ant colony optimization uniquely couples adaptive defect detection with optimized inspection path planning, addressing both perception and navigation challenges in real-world environments.
Recent literature has emphasized lightweight architectures for solar PV defect detection under edge deployment constraints. For object detection, in²⁸, they use a Deep Convolutional Generative Adversarial Network (DCGAN)-based minority class augmentation coupled with lightweight YOLO variants, which have been proposed for edge-deployable PV fault localization. Similarly, the model in²⁹ of YOLOv8n-GBE integrates ghost convolutions with BiFPN-ECA attention to reduce parameters while maintaining localization accuracy. For the classification task, the authors in³⁰ depend on FPGA-orientated designs using patch-wise reusable CNN IP cores, which have demonstrated high-speed PV module defect classification. In³¹, the model of AAPN-Tiny offers a compact adaptive attention pyramid architecture specifically for multi-class fault diagnosis on edge devices, while in³², pruned and low-rank-optimized tiny residual architectures have been tailored for Edge TPU acceleration. For segmentation of faults, the model in³³ utilizes a lightweight SegFormer architecture featuring a modified Mix Transformer encoder and an optimized Multi-Layer Perceptron decoder. These works priorities inference speed, power efficiency, and model compactness, critical for large-scale or real-time deployments.
Traditional FDD methods, which include continuous systems monitoring using various statistical analyses, require significant effort to achieve acceptable results for finding PV failures. It may identify the failed element or module, but it does not specify the affected regions, types of cracks, or the exact defective parts of the cells. Semantic segmentation as one of the FDD methods is designed to locate the affected parts of PV elements, enabling efficient handling and maintenance. Therefore, it is imperative to develop efficient methodologies to assess the health state of photovoltaic plants by detecting defects, determining their type and location and predicting failure patterns of potentially affected combining usage components³⁴. To rapidly and effectively identify, locate, and classify solar PV system faults’ exact location automatically, we suggest a reliable FDD system built on a hybrid transformer neural network that is applied for both thermal and EL imaging to ensure the system’s robustness under various imaging and capturing conditions. We aim to achieve this goal by combining IR image segmentation that is acquired contactlessly by UAVs to identify the impacted cells in large PV systems, followed by a detailed examination of microcracks and minor defects through EL image semantic segmentation, which identified the precise location and classified fault regions in the cell. Our main contributions in this work can be summarized as follows:
A novel automated fault detection and diagnosis (FDD) system is proposed, built upon a hybrid semantic segmentation framework that integrates Convolutional Neural Networks (CNNs) with multi-level Transformer Networks. This hybridization leverages the strengths of both architectures: CNNs effectively capture local spatial features with fewer layers, while Transformers, through their attention mechanisms, learn global contextual relationships across the entire image to accurately associate class types with their corresponding locations.
An enhanced U-Net architecture is developed by embedding multi-scale Transformer layers within the encoder. This modification enables more comprehensive multi-level feature extraction, substantially improving the classical U-Net’s segmentation precision and reliability. Consequently, the proposed PV FDD system achieves more accurate fault localization and classification, contributing to enhanced solar system protection and reduced energy and maintenance costs.
Extensive validation and evaluation of the proposed PV FDD model are conducted using diverse photovoltaic datasets, including one thermal dataset and three electroluminescence (EL) datasets. To the best of our knowledge, such a comprehensive cross-dataset evaluation has not been reported previously. Comparative results with several state-of-the-art models, including published articles, demonstrate the superior segmentation accuracy and robustness of the proposed approach in both binary and multi-class fault scenarios.
The remainder of this paper is structured as follows: Sect. 2 introduces materials and methodology. Section 3 presents the training process, evaluation results, and comparative analysis with other state-of-the-art models. Finally, Sect. 4 provides the conclusions and outlines potential directions for future work.
To comprehensively evaluate the robustness and generalization capability of the proposed hybrid CNN–Transformer fault detection and localization framework, four distinct photovoltaic (PV) datasets were employed, encompassing both thermal and electroluminescence (EL) imaging modalities. Each dataset represents a distinct imaging condition, defect type, and level of segmentation complexity.
This dataset consists of 1,009 infrared thermographic images³⁵ captured by a UAV-mounted Flir Tau 2 640 thermal camera with a resolution of 640 × 512 pixels during the inspection of a 66 MW PV plant located in Tombourke, South Africa³⁶. Each image is paired with a binary ground truth mask indicating faulty and non-faulty cells, as shown in Fig. 1. The dataset was split into 70% training, 20% validation, and 10% testing subsets. It serves to evaluate the system’s ability to detect defective cells at the module level under large-scale, real-field conditions.
First PV thermal dataset samples.
The PVEL-S dataset comprises 1,200 EL images captured by a 1024 × 1024 cooled CCD camera (WP-US146) in a darkroom under 24 V DC and 8 A excitation current³⁷. The dataset includes twelve types of micro-defects, such as grid cracks, black cores, and thick lines³⁸, grouped into two classes: Defective and Non-defective¹⁹, as illustrated in Fig. 2. Images were divided into 70% for training and 30% for validation. This dataset is utilized to assess fine-grained fault localization at the cell level.
Second PV electroluminescence (PVEL-S) dataset samples.
Provided by the University of Central Florida Photovoltaics Lab, this dataset contains 11,851 EL images (are used from total 17,064), including both real and simulated images of crystalline silicon (c-Si) PV cells³⁹. It originally included nine defect types (e.g., corrosion, cracks, and interconnect failures), which were merged into five main categories (including the no-fault class) to address data imbalance. Ground-truth masks were annotated using the VGG Image Annotator²⁰; some images are shown in Fig. 3. The dataset was divided into 80% training, 10% validation and 10% testing subsets. It is used to evaluate the model’s multi-class segmentation capability.
Third PV electroluminescence (UCF-EL) dataset samples.
This dataset integrates 2,354 EL images collected from five different public and private sources, representing both mono- and multi-crystalline PV modules⁴⁰. It contains 29 segmentation classes, including 16 defect-related classes (e.g., cracks, corrosion, inactive areas) and 13 structural or contextual classes (e.g., busbars, ribbons, frame edges) in Fig. 4. Ground-truth masks were created using the GNU Image Annotator, and data augmentation was applied as described in previous works^41,42. The dataset is divided into 2212 training, 70 validation, and 72 testing images, serving as a challenging benchmark for large-scale multi-class PV defect detection and localization.
Fourth PV electroluminescence (Benchmark EL) dataset samples.
Proposed Hybrid Multi-Scale Transformer U-Net Model.
Photovoltaic (PV) systems are susceptible to various failures that can significantly impact performance and overall efficiency post-installation. Continuous monitoring and maintenance are therefore essential to minimize potential faults and performance degradation. However, merely detecting the presence of a fault is insufficient; it is equally important to accurately localize the defect and identify its shape and type.
Semantic segmentation is crucial in deep learning-based fault detection and diagnosis (FDD) systems for PV applications, as it enables pixel-level localization of defects without interfering with system operation. This process produces an annotated version of the original PV image, clearly indicating the faulty regions.
As illustrated in Fig. 5, the proposed methodology starts by collecting diverse PV image datasets containing annotated anomalies. These datasets are then used to train the proposed hybrid multi-scale Transformer-U-Net model, which performs automatic segmentation and localization of PV faults. The model is trained without data enhancement or augmentation to objectively evaluate its inherent performance and robustness across different imaging techniques.
Overall workflow of the proposed PV fault localization framework.
The proposed architecture is based on the U-Net framework but incorporates multiple Transformer layers within the encoder section to capture both local and global feature representations. This hybrid design leverages the spatial feature extraction ability of convolutional neural networks (CNNs) along with the long-range dependency modeling capability of Transformers, enabling more accurate and reliable localization of PV faults under various imaging conditions.
The detailed framework of the proposed hybrid deep learning model based on an encoder-decoder U-Net architecture⁴³ is shown in Fig. 6, which combines convolutional NNs with transformer NNs.
Detailed framework of the proposed hybrid deep learning model.
In our model, the cascaded CNN layers are connected by skip connections (residual blocks) to preserve low-level spatial information for improving finer segmentation details and to mitigate the vanishing gradient problem. In the encoder stages, it begins with CNN layers that are down-sampled sequentially and then integrated with multi-level transformer layers to extract relations with unique information from the entire image, resulting in enhanced performance compared to the classical U-Net. The model starts with one convolution NN layer that applies 3 × 3 filters on the dataset images and produces 32 diverse features using the rectified linear unit (ReLU) activation function, followed by the residual block that consists of two 3 × 3 CNN layers separated by batch normalization and ReLU, and also one 1 × 1 CNN that creates a skip connection between the input and the second CNN output (the 1 × 1 CNN is used at down-sampling or up-sampling only; else it is unity), as illustrated in Fig. 6 at the top left. Equation 1 shows the output of the residual block y, and x denotes its input that resulted from the previous layer⁴⁴.
where σ denotes the RELU activation function, W₁ and W₂ denote the weight matrices of the two cascaded CNN layers, and W_s is a square matrix for the 1 × 1 CNN layer.
The self-attention mechanism in the transformer NN identifies intricate patterns and long-range dependencies due to its global perspective. The multi-head attention mechanism processes data in parallel, allowing them to capture information from various areas of the images and establish a correlation between them. This capability facilitates more accurate predictions through extraction of discriminative features from various areas of the image, enabling effective handling of both local and global features.
By adding three parallel branches of multi-level transformer layers that receive feature images from CNN layers of different depths, the model can preserve the detailed image information and integrate the high-level features of the deep layers with the low-level features of the shallow layers. Each transformer branch comprises three sequential layers with distinct configurations. The main components of the transformer NN layer are detailed on the bottom left side in Fig. 6.
The first branch employs 6 heads in its multi-head self-attention (MHA) layers, and its multi-layer perceptron (MLP) has 128 and 64 dense layers. The second branch layers have 4 heads, with an MLP having 64 and 32 dense layers. The third branch uses 2 heads with MLP dimensions of 32 and 16 and a dropout rate of 0.1 in all transformer layers, along with a normalization layer following each skip connection addition.
The self-attention layer converts the input image features into matrices V, Q, and K. Here, V represents the value associated with a query Q, which corresponds to an image feature, and K denotes the key representing all other image features. This process utilizes a scaled dot product, as shown in Eq. 2, to determine the degree of importance. Equation 3 shows the multi-head layer when concatenating the result of the attentions of all h heads.
where d_model represents the dimension of each head h, and(::{W}^{O}in:{mathbb{R}}^{{d}_{model}times:{d}_{model}}). The outputs of the transformer branches are fused again in CNN residual blocks, acting as a bridge and bottleneck that incorporates positional information to the previous transformer layers without the need for a positional encoding embedding at the beginning of the transformer layers, as in the ViT model¹¹. Then, the outputs are concatenated to initiate up-sampling in the decoder stages, which conclude with the final CNN layer using the SoftMax activation function to determine the correct pixel class decisions.
The proposed model benefits from merging CNN, which identifies the fine-grained defects, and the transformer NN, which analyzes the entire image to contextualize identified defects within the global scene due to its global contextual information ability, enabling our system to accurately detect, localize, and classify all defect types. Using a CNN with a transformer TNN reduces our model complexity. It eliminates the need for a large model size, substantial image data, or a pre-trained model that pure transformer NNs would require to enhance their precision.
We were inspired by the idea of using residual blocks in the encoder from ResNet18⁴⁴ owing to their high efficiency, ease of implementation, strong generalization capability, fast training with limited parameters (sort of lightweight design), and consistent high accuracy. Our compact model can be helpful for IoT and edge device deployment for online PV fault localization implementation.
The proposed model, in its current configuration, is more suitable for offline high-precision inspection or cloud/edge-server cooperative deployment due to its slightly higher computational complexity and inference time. Real-time edge-only applications require less than about 30 ms latency; a pruned version of our model, then, would be preferred, which may be achieved by removing one Transformer branch and reducing the channels by 1.5, sacrificing approximately 2.5% mIoU to double the speedup. Preliminary profiling on NVIDIA Jetson Orin Nano yields ~ 45 ms per 512 × 512 image (22 FPS), with a model size of 168 MB and peak RAM of ~ 420 MB. While feasible for edge GPU platforms, this exceeds the memory constraints of low-power IoT devices (e.g., ARM Cortex-M class with less than 512 MB of total memory). Future work must therefore explore: quantization-aware training to reduce memory and latency; structured pruning of Transformer heads with minimal accuracy loss; TinyML-friendly architectures (e.g., MobileNet + linear attention); and hardware-software co-design using FPGA or Edge TPU accelerators. Such optimizations would enable deployment on less than100 MB memory footprints and less than100 ms latency targets essential for practical large-scale PV monitoring networks.
Table 1 summarizes our model configuration and compares it with the traditional U-Net architecture using ResNet18 as one of the pretrained backbones (encoders) in the segmentation models of the Keras library that are trained on the ImageNet dataset. The compared model is the closest architecture to ours, and it has an approximate number of stages and total parameters, demonstrating that our model represents an enhancement over that model.
Additionally, the U-Net architecture is recognized for its precision in delineating object boundaries and its exceptional performance on small and costly-to-create datasets. Skip connections are critical components in this architecture, which combines the high-resolution, fine-grained spatial information from the encoder with the highly processed, semantically rich feature maps from the decoder. This allows the decoder to make precise, pixel-accurate localization decisions while having a deep understanding of the context. Collectively, these advantages render our proposed model more accurate, robust, and reliable for automated PV failure detection and localization.
Experiments for the proposed model were conducted on the following hardware platform: NVIDIA GeForce RTX 3050 with 4 GB of GPU, Intel Core i7-11800 H at 2.30 GHz, and 16 GB of RAM. Windows 10 Pro, a 64-bit operating system, and the Keras framework with the TensorFlow 2.10.1 backend supported GPUs with CUDA toolkit 11.2 and CUDNN 8.1.0.
To assess the proposed strategy across several datasets, we compared our model’s performance with the following pretrained models: ResNet18 and VGG16 (Visual Geometry Group)⁴⁵ as a backbone for the U-Net architecture to determine which one is best, as well as ResNet18 for the PSP-Net (Pyramid Scene Parsing Network)⁴⁶, FPN (Feature Pyramid Networks)⁴⁷, and LinkNet⁴⁸ architectures. VGG16 is a strong baseline and simple architecture, which stacks multiple convolutional layers (with 3 × 3 filters) to increase depth, but it is significantly large, slow, and prone to the vanishing gradient problem in very deep networks; therefore, we limited its comparison to the U-Net architecture. PSP-Net architecture relies on building a pyramid of pooling at the end of the backbone to capture context at multiple scales, while LinkNet is designed to be efficient and fast because it uses additive skip connections (instead of the U-Net’s concatenation) between the encoder and decoder, which reduces computational cost. FPN is constructed as a top-down pathway with lateral connections to build a rich, multi-scale feature pyramid. The compared models and their descriptions are summarized in Table 2. Ultimately, recently published models utilizing identical datasets are introduced to demonstrate that our model surpasses their performance. For comparison results with published papers across all datasets, we did not use the same image size, batch size, or loss function. Some authors modify dataset images through preprocessing or by using different test sets, introducing variations that complicate direct comparison.
A loss function prevents the model from growing overconfident or depending too much on specific training samples, serving as a kind of regularization. The model learns to produce increasingly accurate predictions by reducing the difference between its outputs and the actual values through an iterative process of minimizing the loss. Weighted cross-entropy is a modified version of cross-entropy loss (traditional cross-entropy loss shown in Eq. 4) that assigns different weights to different classes, as in Eq. 5. It has widespread applications in semantic segmentation with imbalanced datasets. The Dice loss in Eq. 6 focuses on region-based optimization, as opposed to cross-entropy loss (which operates at the pixel level), which calculates how much the expected and ground truth masks overlap, making it well-suited for unbalanced datasets^49,50,51.
where y and ŷ are the true and predicted values of a sample, respectively, i denotes a sample image of total N, and j denotes a class of total C for multiclass problems.
where β ε [0, 1] is the factor value that penalizes for a specific class, and w is the weight assigned to a class j.
All dataset images are resized to 160 × 160 × 3 pixels, and with batch sizes of 2, we start training our model using a 0.0001 learning rate and the optimizer of Adam, which is able to modify the learning rate for every single parameter. The size of images and batches are chosen according to the hardware limitations.
Evaluation metrics are employed to evaluate the suggested methodology’s performance. The predictions of testing the suggested model are assessed using measures such as global pixelwise accuracy, mIoU, precision, recall⁵², and F1-score (weighted and unweighted). All of these pertain to true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN). A TP indicates that the model accurately classifies a certain PV fault class in all pixels, whereas a TN indicates that the model correctly classifies the opposite PV faults (or classes) in pixels. Conversely, FP indicates that the model classifies the class wrongly, while FN denotes the inaccurate classification of the opposite classes.
In semantic segmentation (pixelwise classification), the accuracy statistic indicates how often a model is correct overall (predicted true pixels) in Eq. 7. The precision metric indicates how often a model properly predicts a target class in Eq. 8, and the recall metric evaluates the percentage of correctly real positive cases that a model properly detects in Eq. 9. Equation 10 has the model performance parameter of F1-score (Dice coefficient), which integrates precision and recall, evaluates the correctness of both positive and negative pixel classifications in the segmentation outcomes (calculates the images’ similarity). The weighted F1-score (wDice/wF1) was employed to assess the models, accounting for class imbalance in the dataset, which computes the F1-score for each class and subsequently averages them using weights proportionate to the number of true cases for each class. The Intersection over Union (IoU) (Jaccard index) in Eq. 11 quantifies the overlap between expected and actual segmentation masks, assessing the precision of fault localization. We have used three types of values: the mean/unweighted (mAP) or macro values, which sum the individual values for all samples in each class and calculate their average without accounting for unbalanced classes; the weighted values (wIoU), which incorporate the ratio of the number of pixels in each class as a weight in calculations; and the global values (gF1) or micro, which compute totals across all classes at once.
The thermal dataset that indicates the faulty cells in the PV system has a highly imbalanced pixel distribution, as the faulty class areas (faulty PV cells) are fewer than the non-faulty ones due to the relatively small area occupied by faulty regions. In that case, the model training is significantly challenged, resulting in biased predictions towards the majority class towards the non-faulty class. Our model, along with all the other models being compared, is trained, validated, and tested using a combined loss function of weighted categorical cross-entropy and Dice loss. This strategy enhances model performance by combining the two losses, as shown in Eq. 12⁵⁰. Also, combining unweighted cross-entropy loss with dice loss when training the second dataset enhances the results (replacing L_WCE with L_CE in Eq. 12).
where α controls the amount of contribution of both losses.
All the trained hyperparameters of the model according to the four investigated datasets are illustrated in Table 3.
Training and evaluation procedures were implemented independently for each of the compared models to establish comprehensive fault localization models employing diverse AI approaches. This validation strategy confirms the efficacy of the proposed model and enables comparative performance analysis across all models. The performance of training and validation of our model for the first thermal imaging dataset is shown in Fig. 7, which indicates that the accuracy improves and stabilizes around epoch 50. A testing phase was conducted to assess the proposed methodology on unseen images from the dataset, as summarized in Table 4 and visualized in the confusion matrix (Fig. 8). We also compared it with other models, including published papers using the same dataset. All the results are combined in Table 5.
Training and validation performance of thermal dataset.
Confusion matrix for thermal dataset.
The results indicate that our proposed model achieves the highest accuracy of 99.89% among all the compared models, demonstrating its efficacy in accurately localizing faulty cells with a mean precision (mAP) of 88.73%, a mean dice (mDice) of 84.48% and a mean recall (mRecall) of 81.07%. While traditional ResU-Net or FPN models are the second-best results, PSP-Net exhibited training difficulties, resulting in the poorest performance.
Our model demonstrated superior performance compared to other models, although when comparing our model with the model in¹⁸, the comparison may not be equitable because the authors applied image enhancements (Contrast Limited Adaptive Histogram Equalization) and data augmentation that resulted in increasing performance of mIoU and mAP by about 3% more than ours, suggesting that thermal images require additional preprocessing due to the huge misbalancing of classes. The test set contains over 2.5 million pixels, but only about 5 thousand of these are faulty, representing approximately 0.2% of the total, which highlights the unbalancing problem present in the rest of the dataset. Also, the quality of the thermal image needs to increase defect visibility without creating noticeable artifacts, as some regions exhibit less variation¹⁸. Dataset enhancements would be considered in the future approach of our fault detection and localization system.
The performance for segmentation is also measured using the Receiver Operating Characteristic Curve (ROC) and area under the ROC curve (AUC) in Fig. 9. An AUC of 81% indicates satisfactory discrimination capability and separability.
ROC and AUC curves for the thermal dataset.
Figure 10 displays the predicted masked images and their Gradient-Class Activation Map (Grad-CAM) heatmaps for a few tested images. Heatmaps are superimposed with the original images to illustrate the dependencies from the segmented classes that allow us to gauge how sensitive our model is to the input images. The figure shows a perfect predicted segmentation, which ensures our system’s ability to aid in fault localization of cells.
Visualizing some predicted images and their Grad-CAM of the tested images for the thermal dataset.
The PVEL-S dataset, which identifies defective regions in the PV cell, was investigated and also exhibits class imbalance, but with a moderate ratio; we employed a combined loss function to address this challenge. Model accuracy decreased when combining Dice loss with weighted cross entropy loss, so we combined the unweighted categorical cross entropy at label smoothing of 0.0001 with the Dice loss, which improved model performance without needing any data augmentation, as illustrated in Fig. 11. The effect of label smoothing improves the accuracy by more than 1% and increases model robustness against label noise, also preventing the model from becoming overconfident. For future work, we would be investigating an adaptive soft labelling method, which might better handle fine-grained defect boundaries for other datasets.
Training and validation performance of PVEL-S dataset.
To evaluate our model for the second dataset, testing was performed to validate model training effectiveness, as illustrated in Fig. 12; Table 6, which show that defective area segmentation achieved satisfactory results with 96.88% precision. Furthermore, the comparison results demonstrate our model’s superiority over all benchmark models, as shown in Table 7.
Confusion matrix for PVEL-S dataset.
With an accuracy of 99% and an mDice of 98.13%, our model has the best performance ever, even compared to the published paper in¹⁹ that used data augmentation. The second-best performance is the FPN and LinkNet models. Although our model mean recall (mRecall) does not have the best score, the weighted and global recall of 99% is the highest value of the rest of the models. Heatmaps for the encoder layers in Fig. 13, which show the important defective areas, have been highlighted perfectly. This confirms that the transformer layers added more global details to the encoder, resulting in impressive performance against all the compared models.
Visualizing some predicted images and their Grad-CAM for the encoder for the PVEL-S dataset.
With a fabulous discrimination AUC of 98% in Fig. 14, and observing the predicted image results in Fig. 13, we find that the proposed model has localized the PV defective regions on cells for the PVEL-S dataset with excellent performance and minor errors.
ROC and AUC curves for PVEL-S dataset.
The UCF-EL dataset comprises five PV fault classes with an unbalanced number of pixels; thus, a weighted cross-entropy loss function with customized weights was employed to address the class imbalance and enhance model performance, as illustrated in Fig. 15. We observed that the weighted loss function improves the overall results of our model but had a more pronounced impact on the benchmark models. This can be attributed to the multi-scale architecture of our Transformer layers, which results in more enhancement using the unweighted loss function, as in the second dataset results.
Training and validation performance of UCF-EL dataset.
Testing our model results can be found in Table 8; Fig. 16, which give the details of the segmentation accuracy, showing that the cracks and corrosion are the most efficient PV failure localization results.
Confusion matrix for UCF-EL dataset.
Comparative results with state-of-the-art models are presented in Table 9. The results show that our model’s performance has improved, especially compared to published papers using the same dataset, with an accuracy of 96.27% and a mean precision of 77.64%, achieving the highest precision among all models. However, the precision improvement came at the expense of a reduced Dice score.
The VGG-U-Net and FPN models also demonstrated competitive performance for mean recall, dice and IoU scores. Future work will explore alternative architectures or backbone encoders that can be improved by incorporating transformer neural networks to enhance dice performance in model segmentation for EL images containing minor PV faults, which are challenging to detect even for experienced human observers or through image enhancement prior to training.
The mean and global AUC, and also their values per class, are shown in Fig. 17, which confirms the model’s good overall discrimination ability of 97.7% for classes and excellent performance on most of them. The predicted image samples and their Grad-CAM can be found in Fig. 18, which shows a perfect multiclass segmentation demonstrating that our model is perfect for localizing almost all PV fault classes.
ROC and AUC curves for the UCF-EL dataset.
Visualizing some predicted images and their Grad-CAM for the encoder for the UCF-EL dataset.
The benchmark dataset also has unequal numbers of classes, and it requires a weighted entropy loss function with equal weights of 1, as the model’s precision is affected by customized weights. The model training performance is illustrated in Fig. 19.
Training and validation performance of Benchmark EL dataset.
Model testing results are shown in Fig. 20; Table 10. The dataset unbalancing here is a big issue due to the number of classes and their various sizes. The results of only 25 classes have the highest defect segmentation of scuff and corrosion ribbon and the highest cell features of padding and junction box.
Confusion matrix for Benchmark EL dataset.
According to Table 11, which compares our model performance with other models, with 95% accuracy and 66.7% mean precision, ours has the best results among the other comparative models and a close performance to the published one in²². To show the results as the predicted images and their heatmaps in Fig. 21 and, moreover, in Fig. 22, which shows the global AUC of 97.4%.
Visualizing some predicted images and their Grad-CAM for the encoder for the Benchmark EL dataset.
ROC and AUC curves for the Benchmark EL dataset.
We have added a supplement annotation consistency analysis using Cohen’s Kappa scores in Eq. 12 to validate a classifier against a baseline of random guessing, ensuring the model’s predictive power is genuine, which yields 0.6897 for the Thermal, 0.9626 for the PVEL-S, 0.7455 for the UCF-EL, and 0.8928 for the Benchmark EL datasets, demonstrating overall model reliability. These numbers indicate that the first and third datasets could need to be enhanced using different data augmentation methods, appropriate filters, or labelling methods such as adaptive soft labelling.
where P_e and P_o are the expected agreement (the probability of the model agreeing purely by chance, calculated using the proportion of times the model assigns each) and the observed agreement class (the actual percentage of times the model and the truth agree), respectively.
Supplemental relevant experiments are added to enhance the interpretability and reproducibility of the model design, which provide deeper insight into the architectural design choices. Due to computational constraints, ablation experiments were conducted on the PVEL-S and Benchmark EL datasets as representative cases. Table 12 shows hyperparameter sensitivity and architectural contribution analysis for each component individually.
The ablation study confirms that the proposed configuration (B = 3, TL = 3, with residuals and skip connections) achieves the best performance on both datasets. Several observations emerge for PVEL-S binary dataset first, all configurations achieve high performance (> 98.88% accuracy, > 95.94% mIoU), confirming the model’s robustness to hyperparameter variations. Second, increasing the number of Transformer branches from 1 to 2 yields the largest gain (+ 0.09% accuracy, + 0.31% mIoU), while the third branch provides marginal additional improvement (+ 0.03% each). Third, adding Transformer layers from 1 to 3 progressively improves performance, with TL = 3 achieving the best results. Fourth, removing residuals or skip connections degrades performance by 0.07–0.08% in accuracy and 0.26–0.30% in mIoU, confirming their contributions. The effect on performance is not evident from increasing the number of heads and MLPs only; it seems to be affected more by i/p depth. Finally, the proposed configuration (three branches, three Transformer layers per branch, with residuals and skip connections) achieves the highest accuracy (99.00%) and mIoU (96.36%), outperforming all alternatives. The improvements are marginally but consistently outperforming all variants. The more challenging Benchmark EL multi-class dataset achieves 95.00% accuracy and 49.26% mIoU, improving upon the best single-branch baseline (94.87% accuracy, 46.37% mIoU) by + 0.13% accuracy and + 2.89% mIoU. Removing residuals or skip connections degrades performance on Benchmark EL substantially (mIoU drops by 5.7–7.4%), confirming their critical role. Given the extreme challenge of micro-defects in this dataset (per-class analysis shows sub-600px defects fail), these results confirm the optimality of our design. The small performance differences across configurations on PVEL-S (≤ 0.12% accuracy) further demonstrate model robustness.
To demonstrate the robustness and stability of the proposed model, we have conducted a 3-fold cross-validation method with approximate 40 epochs per fold (due to hardware limitation) using the training data of the four datasets. The thermal dataset gives a mean validation accuracy of 99.85% (individual folds: 99.86%, 99.84%, 99.86%) with standard deviation (std) ± 0.01% and a mean validation loss of 0.0263 ± 0.0069 also mean validation IoU of 71.74 ± 1.2623%. The PVEL-S dataset gives mean validation accuracy of 98.81% ± 0.06% (98.74%, 98.80%, 98.89%) and mean validation loss of 0.0298 ± 0.0016 also mean validation IoU of 95.54%±0.169%. The UCF-EL dataset mean accuracy of 95.95%±0.2095% (96.17%, 96.02%, 95.67%), mean loss 0.2384 ± 0.01736, and mean IoU of 52.85 ± 0.6719%. The Benchmark EL dataset gives an accuracy of 95.36% ± 0.16% (95.44%, 95.14%, 95.50%), loss of 0.1313 ± 0.0053, and IoU of 46.53% ± 0.99%. The overall 3-fold cross-validation across the four datasets in Table 13 reveals consistent patterns. The model achieves near-perfect stability on all datasets, as evidenced by low standard deviations in accuracy (≤ 0.21% for all). However, segmentation precision (IoU) varies substantially: PVEL-S shows excellent boundary alignment (95.54%), while Thermal (71.74%), UCF-EL (52.85%), and Benchmark EL (46.53%) reveal increasing challenges in defect boundary localization (due to minor PV defects or classes and data imbalance, especially for a large number of classes (29 in Benchmark EL)). This discrepancy between high accuracy and moderate IoU suggests that while the model reliably classifies defect regions, precise delineation of irregular defect boundaries remains the key bottleneck. The low cross-fold variance across all metrics (accuracy std < 0.21%, IoU std < 2% for most datasets) strongly confirms the model’s robustness and insensitivity to data partitioning.
Additional analysis based on detailed test results in Table 10 of the fourth Benchmark EL dataset to identify the reasons for the data’s challenges: we have analyzed the relationship between defect size and detection success using a size threshold. F1 vs. log pixel-count calculations reveal a clear threshold: all classes with > 10,000 pixels achieve F1 > 0.88, while all classes with < 600 pixels have F1 < 0.13 (mean F1 = 0.06 ± 0.06). Between 600 and 10,000 pixels, identifying the model’s effective receptive field limit. For failing classes, precision (P) exceeds recall (R) substantially: text (P = 0.36, R = 0.05), star (P = 0.33, R = 0.015), rings (P = 0.62, R = 0.07). This indicates the model can identify some tiny defect pixels only (a recall-dominated problem). Complete failures (crack rbn edge, dead cell with < 10 pixels) show P = R=0. The recall gap suggests that high-resolution features are lost in the encoder-decoder pathway. This directly motivates: (1) Feature Pyramid Networks to preserve spatial resolution via top-down pathways (create multi-scale feature pyramids with top-down pathways), and (2) Atrous Spatial Pyramid Pooling (ASPP) to expand receptive fields without reducing resolution as a future work (with various dilation rates to capture multi-scale context). (3) class-balanced sampling or focal loss to address extreme pixel imbalance (e.g., dead cell: 2 pixels vs. background: 1.29 M). Preliminary analysis suggests such modules could improve F1 on tiny defect classes from near-zero to 0.4–0.6. Performance analysis across all four datasets shows that the PVEL-S dataset achieves excellent and balanced performance across both classes (defect F1 = 0.968, non-defect F1 = 0.994), confirming strong generalization without class bias. The Thermal dataset has overall accuracy of 99.85%, but this is dominated by the background class (99.8% of pixels). The faulty cell class (5,064 pixels, 0.2% of dataset) achieves only F1 = 0.690, revealing that high accuracy masks moderate defect detection due to extreme class imbalance. That dataset would need one of the data augmentation methods to overcome this limitation. The Benchmark EL dataset shows a strong size-performance correlation (Pearson correlation coefficient (r) = 0.7123 with P-value = 6.4828 × 10⁻⁵ < 0.001). All defect classes with < 600 pixels have F1 < 0.13, including three classes with F1 = 0 with < 10 pixels. This systematic failure on tiny defects motivates multi-scale feature enhancement (e.g., FPN, ASPP). The UCF-EL has the poorest class performance for the Interconnect class (F1 = 0.232, 20,989 pixels), while similarly sized Corrosion class achieves F1 = 0.808. This indicates a semantic bottleneck (class distinguishability) rather than a size limitation, suggesting contrastive learning or refined class definitions as future solutions. While the model reveals three distinct bottlenecks on other datasets: class imbalance (Thermal), defect size (Benchmark EL), and class distinguishability (UCF-EL), it excels in PVEL-S dataset. These dataset-specific limitations provide clear, actionable pathways for future work.
In general, the proposed hybrid enhanced Transformer model outperforms existing methods, achieving high accuracy and good precision for PV failure localization across four diverse datasets, while revealing clear pathways for addressing remaining micro-defect challenges.
This study introduces a new deep learning approach that presents a data-driven approach for PV fault localization and detection through image semantic segmentation. We developed and evaluated a hybrid Transformer neural network model with four different datasets. First, the thermal dataset was utilized to demonstrate large-scale PV plant inspection in detecting and locating failure cells, achieving exceptional accuracy of 99.89%. For cell-level fault analysis, electroluminescence (EL) images were employed for both binary and multi-class segmentation. The second dataset gives a superior result with 99% accuracy and 96.36% mIoU for locating the defective areas in the cell. For multiclass segmentation the third and fourth datasets are investigated with an accuracy of 96.27% and 95%, respectively, which demonstrates that our FDD and localization model can perform robustly across diverse datasets, demonstrating robust PV failure localization under various challenging conditions.
This research serves as a valuable reference for researchers and the PV industry to increase the likelihood of problem localization and detection in solar PV systems. We recommend evaluating the proposed methodology on additional large-scale datasets with various fault class segmentations to identify any potential limitations of our approach.
Future work should explore sophisticated data augmentation methods like instance-aware copy-paste or online hard-example mining augmentations to artificially increase the presence of rare defect patterns (like some classes in the fourth dataset) for improving its performance. Furthermore, the use of conditional GANs or latent diffusion models specifically designed for semantic segmentation in synthetic data generation enables the generation of realistic defect examples without artefacts. Also, we may explore adaptive patch partitioning, hybrid local-global attention reweighting or multi-scale feature enhancement (e.g., FPN, ASPP). These approaches may lead to better detection performance for the hardest minority classes than what is possible with only fine-tuning the loss function. Future work also will focus on developing a simple, lightweight real-time implementation in the field with the help of the Internet of Things (IoT) to help speed maintenance, like complementary knowledge distillation techniques, where the current model serves as a teacher to train a compact student network model (e.g., feature-based alignment or attention map transfer) or developing lightweight transformer versions (e.g., MobileViT, EfficientFormer, shifted window and linear attention). All are promising possibilities, evaluating trade-offs between segmentation accuracy, inference latency, and model size on embedded devices for real-time, large-scale smart PV monitoring. Additionally, examining diverse hybridization architectures utilizing the transformer neural network for the localization of small and minor faults and exploring full hyperparameter grid search and cross-dataset ablation for more reliability in future.
The data used in this study are four different datasets:- The first **Photovoltaic Thermal Dataset** Images are available from the Universit`a Politecnica delle Marche – DII – VRAI Group repository at [http://vrai.dii.univpm.it/content/photovoltaic-thermal-images-dataset](http:/vrai.dii.univpm.it/content/photovoltaic-thermal-images-dataset) after completing the request form and contract conditions, in which the candidate (researcher with an email address affiliated with an institution or university) outlines their goals for the study. A download link and credentials are sent after the owners’ approval with the condition that they are for research purposes only, with no other use. Notice: “That data are not publicly redistributable and were used under license for the current study; however, by following the repository’s request procedure, the access to the data for replication or verification reasons can be acquired by a researcher.“- The second **PVEL-S Electroluminescence Dataset** Images are publicly available at [https://www.kaggle.com/datasets/yaozhang01182010/dataset-of-solar-cells-defect-segmentation](https:/www.kaggle.com/datasets/yaozhang01182010/dataset-of-solar-cells-defect-segmentation).- The third **UCF-EL Dataset** Images are publicly available at [https://github.com/ucf-photovoltaics/UCF-EL-Defect](https:/github.com/ucf-photovoltaics/UCF-EL-Defect).- The fourth **Benchmark EL Dataset** Images are publicly available at [https://github.com/TheMakiran/BenchmarkELimages.git](https:/github.com/TheMakiran/BenchmarkELimages.git).The generated data during the course of the research, as software algorithm code, is owned privately by the corresponding author, who can be contacted with any questions or concerns regarding the algorithm or data.
International Energy Agency (IEA. ). https://www.iea.org/energy-system/renewables/solar-pv
IEA report. https://iea-pvps.org/wp-content/uploads/2024/10/IEA-PVPS-Task-1-Trends-Report-2024.pdf
International Renewable Energy Agency (IRENA). Renewable energy statis-tics 2024. https://www.irena.org/Publications/2024/Jul/Renewable-energy-statistics-2024
El-Banby, G. M., Moawad, N. M., Abouzalm, B. A., Abouzaid, W. F. & Ramadan, E. A. Photovoltaic system fault detection techniques: a review. Neural Comput. Appl. 35(35), 24829–24842. https://doi.org/10.1007/s00521-023-09041-7 (2023).
Article  Google Scholar 
Mellit, A. & Kalogirou, S. Artificial intelligence and internet of things to improve efficacy of diagnosis and remote sensing of solar photovoltaic systems: Challenges, recommendations and future directions. Renew. Sustain. Energy Rev. 143, 110889. https://doi.org/10.1016/j.rser.2021.110889 (2021).
Article  Google Scholar 
Masita, K., Hasan, A., Shongwe, T. & Hilal, H. A. Deep learning in defects detection of PV modules: A review. Solar Energy Advances 100090. https://doi.org/10.1016/j.seja.2025.100090 (2025).
Article  Google Scholar 
Vaswani, A. et al. NIPS. Attention is all you need, In: 31st Conference on Neural Information Processing Systems, Long Beach, CA, USA, ArXiv abs/1706.03762, (2017). https://doi.org/10.48550/arXiv.1706.03762
Dosovitskiy, A. et al. An image is worth 16×16 words: Transformers for image recognition at scale, (2020). https://doi.org/10.48550/arXiv.2010.11929
Ramadan, E. A. et al. An innovative transformer neural network for fault detection and classification for photovoltaic modules. Energy Convers. Manag. 314, 118718. https://doi.org/10.1016/j.enconman.2024.118718 (2024).
Article  Google Scholar 
Kellil, N., Mellit, A., Rus-Casas, C. & Benghanem, M. Hybrid vision transformer model for defect classification in photovoltaic modules using thermographic imaging: Leveraging self-attention mechanisms for enhanced accuracy. Renewable Energy 124138. https://doi.org/10.1016/j.renene.2025.124138 (2025).
Article  Google Scholar 
Mahboob, Z., Khan, M. A., Lodhi, E., Nawaz, T. & Khan, U. S. Using segFormer for effective semantic cell segmentation for fault detection in photovoltaic arrays. IEEE Journal of Photovoltaics 15(2), 320–331. https://doi.org/10.1109/JPHOTOV.2024.3450009 (2025).
Article  Google Scholar 
Dwivedi, D., Babu K V S M, Yemula, P. K., Chakraborty, P. & Pal, M. Identification of surface defects on solar pv panels and wind turbine blades using attention based deep learning model. Eng. Appl. Artif. Intell. 131, 107836. https://doi.org/10.1016/j.engappai.2023.107836 (2024).
Article  Google Scholar 
Khan, U., Khan, S. F., Khalil, I. U. & Ahsan, M. Deep learning-based vision transformer approach for detecting overlapping PV faults using multi labeling. Measurement 118537. https://doi.org/10.1016/j.measurement.2025.118537 (2025).
Article  Google Scholar 
Demirci, M. Y., Beşli, N. & Gümüşçü, A. Deep learning-based fully automatic system for segmentation and defect classification of the solar modules using electroluminescence images. Sol. Energy Mater. Sol. Cells. 295, 113974. https://doi.org/10.1016/j.solmat.2025.113974 (2026).
Article  Google Scholar 
Yu, S. et al. Multimodal fault detection model for photovoltaic modules. Sci. Rep. 16 (1), 3278. https://doi.org/10.1038/s41598-025-28603-4 (2026).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Tradacete-Ágreda, M. et al. Framework for autonomous inspection of PV plants using IoT electronics on each PV panel and UAV collaboration. Energy Conversion and Management: X 100878. https://doi.org/10.1016/j.ecmx.2025.100878 (2025).
Article  Google Scholar 
Jumaboev, S., Jurakuziev, D. & Lee, M. Photovoltaics plant fault detection using deep learning techniques. Remote Sens. 14(15), 3728. https://doi.org/10.3390/rs14153728 (2022).
Article  Google Scholar 
Awedat, K., Comert, G., Ayad, M. & Mrebit, A. Advanced fault detection in photovoltaic panels using enhanced U-Net architectures. Mach. Learn. Appl. 20, 100636. https://doi.org/10.1016/j.mlwa.2025.100636 (2025).
Article  Google Scholar 
Duan, R., Wang, Y., Chen, X. & Li, S. An enhanced algorithm for cell-level anomaly segmentation in photovoltaic solar panels using electroluminescence imaging. Energy 136711. https://doi.org/10.1016/j.energy.2025.136711 (2025).
Article  Google Scholar 
Fioresi, J. et al. Automated defect detection and localization in photovoltaic cells using semantic segmentation of electroluminescence images. IEEE Journal of Photovoltaics 12(1), 53–61. https://doi.org/10.1109/JPHOTOV.2021.3131059 (2022).
Article  Google Scholar 
Li, W., Wang, F. & Sun, Z. Semantic segmentation method of photovoltaic cell microcracks based on EL polarization imaging. Sol. Energy 291, 113364. https://doi.org/10.1016/j.solener.2025.113364 (2025).
Article  Google Scholar 
Mazen, F. M. A., Shaker, Y. & Seoud, R. A. A. Attention-Based SegNet: Toward Refined Semantic Segmentation of PV modules Defects. IEEE Access https://doi.org/10.1109/ACCESS.2024.3431098 (2024).
Article  Google Scholar 
Jiang, Y. & Zhao, C. Attention classification-and-segmentation network for micro-crack anomaly detection of photovoltaic module cells. Sol. Energy 238, 291–304. https://doi.org/10.1016/j.solener.2022.04.012 (2022).
Article  ADS  Google Scholar 
Otamendi, U. et al. Segmentation of cell-level anomalies in electroluminescence images of photovoltaic modules. Sol. Energy. 220, 914–926. https://doi.org/10.1016/j.solener.2021.03.058 (2021).
Article  ADS  Google Scholar 
Liu, Y., Zheng, Y., Wei, T. & Li, Y. Lightweight algorithm based on you only look once version 5 for multiple class defect detection on wind turbine blade surfaces. Eng. Appl. Artif. Intell. 138, 109422. https://doi.org/10.1016/j.aei.2023.102292 (2024).
Article  Google Scholar 
Liu, Y. H. et al. Defect detection of the surface of wind turbine blades combining attention mechanism. Adv. Eng. Inform. 59, 102292. https://doi.org/10.1080/10589759.2025.2604578 (2024).
Article  Google Scholar 
Liu, Y., Zheng, Y., Wei, T. & Zhang, Y. Lightweight algorithm based on you only look once version 5 for multiple class defect detection on wind turbine blade surfaces. Nondestructive Testing and Evaluation https://doi.org/10.1016/j.engappai.2024.109422 (2025).
Article  Google Scholar 
Maddileti, N. S., Namburi, R., Raj, R. D. A., Yanamala, R. M. R. & Pallakonda, A. DCGAN-driven minority class augmentation for lightweight YOLO-based photovoltaic defect localization suitable for edge deployment. IEEE Trans. Device Mater. Reliab. https://doi.org/10.1109/TDMR.2025.3592416 (2025).
Article  Google Scholar 
Yeddula, L. R. et al. YOLOv8n-GBE: A hybrid YOLOv8n model with ghost convolutions and BiFPN-ECA attention for solar PV defect localization. IEEE Access. https://doi.org/10.1109/ACCESS.2025.3584249 (2025).
Article  Google Scholar 
Vinod, G. et al. FPGA-powered solar photovoltaic module defect classification: Patch-wise reusable CNN IPs for high-speed edge processing. Journal of Computing and Information Science in Engineering https://doi.org/10.1115/1.4070331 (2025).
Article  Google Scholar 
Raj, R. D. A., Yanamala, R. M. R., Pallakonda, A. & Kanasottu, A. N. AAPN-Tiny: A Compact Edge-Deployable Adaptive Attention Pyramid Architecture for Multi-Class Fault Diagnosis in Solar Photovoltaic Modules. IEEE Lat. Am. Trans. 23 (12), 1284–1296. https://doi.org/10.1109/TLA.2025.11231229 (2025).
Article  Google Scholar 
Yanamala, R. M. R., Pallakonda, A. & Amar Raj, R. D. Pruned and low-rank optimized tiny residual architecture for solar photovoltaic fault classification on edge TPU. J. Energy Resour. Technol. Part. A: Sustainable Renew. Energy. 1 (6), 061305. https://doi.org/10.1115/1.4069350 (2025).
Article  Google Scholar 
Joshi, D. & Pal, M. Lightweight transformer driven segmentation of defects in solar PV thermal imagery. Sol. Energy. 308, 114402. https://doi.org/10.1016/j.solener.2026.114402 (2026).
Article  Google Scholar 
Mellit, A. & Kalogirou, S. Recent advances in the application of infrared thermographic imaging and embedded artificial intelligence for fault diagnosis and predictive maintenance of photovoltaic plants: Challenges and future directions. Renew. Sustain. Energy Rev. 223, 116057. https://doi.org/10.1016/j.rser.2025.116057 (2025).
Article  Google Scholar 
http://vrai.dii.univpm.it/content/photovoltaic-thermal-images-dataset http:///vrai.dii.univpm.it/content/photovoltaic-thermal-images-dataset
Pierdicca, R., Paolanti, M., Felicetti, A., Piccinini, F. & Zingaretti, P. Automatic faults detection of photovoltaic farms: solAIr, a deep learning-based system for thermal images. Energies 13 (24), 6496. https://doi.org/10.3390/en13246496 (2020).
Article  ADS  Google Scholar 
https://www.kaggle.com/datasets/yaozhang01182010/dataset-of-solar-cells-defect-segmentation
Su, B., Zhou, Z. & Chen, H. PVEL-AD: A large-scale open-world dataset for photovoltaic cell anomaly detection. IEEE Trans. Ind. Inf. 19 (1), 404–413. https://doi.org/10.1109/TII.2022.3162846 (2022).
Article  Google Scholar 
https://github.com/ucf-photovoltaics/UCF-EL-Defect
https://github.com/TheMakiran/BenchmarkELimages.git
Pratt, L., Govender, D. & Klein, R. Defect detection and quantification in electroluminescence images of solar PV modules using U-net semantic segmentation. Renew. Energy. 178, 1211–1222. https://doi.org/10.1016/j.renene.2021.06.086 (2021).
Article  Google Scholar 
Pratt, L., Mattheus, J. & Klein, R. A benchmark dataset for defect detection and classification in electroluminescence images of PV modules using semantic segmentation. Syst. Soft Comput. 5, 200048. https://doi.org/10.1016/j.sasc.2023.200048 (2023).
Article  Google Scholar 
Ronneberger, O., Fischer, P. & Brox, T. U-net: Convolutional networks for biomedical image segmentation. In: International Conference on Medical image computing and computer-assisted intervention (234–241). Cham: Springer international publishing. (2015). https://doi.org/10.1007/978-3-319-24574-4_28
He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition (770–778). (2016). https://doi.org/10.1109/CVPR.2016.90
Simonyan, K. & Zisserman, A. Very deep convolutional networks for large-scale image recognition, conference paper at ICLR 2015, arXiv preprint arXiv:1409.1556. (2014). https://doi.org/10.48550/arXiv.1409.1556
Zhao, H., Shi, J., Qi, X., Wang, X. & Jia, J. Pyramid scene parsing network. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) https://doi.org/10.1109/CVPR.2017.660 (2017).
Article  Google Scholar 
Lin, T. Y. et al. Feature pyramid networks for object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2117–2125). (2017). https://doi.org/10.1109/CVPR.2017.106
Chaurasia, A. & Culurciello, E. Linknet: Exploiting encoder representations for efficient semantic segmentation. In: 2017 IEEE visual communications and image processing (VCIP) (1–4). (IEEE, 2017). https://doi.org/10.1109/VCIP.2017.8305148
Jadon, S. A survey of loss functions for semantic segmentation. In: 2020 IEEE conference on computational intelligence in bioinformatics and computational biology (CIBCB) (1–7). (IEEE, 2020). https://doi.org/10.1109/CIBCB48159.2020.9277638
Sudre, C. H., Li, W., Vercauteren, T., Ourselin, S. & Jorge Cardoso, M. Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations. In: International Workshop on Deep Learning in Medical Image Analysis ( 240–248). Cham: Springer International Publishing. (2017)., September https://doi.org/10.1007/978-3-319-67558-9_28
Ho, Y. & Wookey, S. The real-world-weight cross-entropy loss function: Modeling the costs of mislabeling. IEEE access. 8, 4806–4813. https://doi.org/10.1109/ACCESS.2019.2962617 (2019).
Article  Google Scholar 
Fikry, W. M., Heliel, A. F. A., EL-SEHIEMY, R. A. G. A. B. & Rezk, N. Accurate, tiny size and low power consumption framework for controlling physiological humans states in smart home systems. J. Contemp. Technol. Appl. Eng. 4 (1), 1–16. https://doi.org/10.21608/jctae.2025.371243.1047 (2025).
Article  Google Scholar 
Eesaar, H., Joe, S., Rehman, M. U., Jang, Y. & Chong, K. T. SEiPV-Net: An efficient deep learning framework for autonomous multi-defect segmentation in electroluminescence images of solar photovoltaic modules. Energies 16 (23), 7726. https://doi.org/10.3390/en16237726 (2023).
Article  Google Scholar 
Download references
On behalf of all authors, the corresponding author states that there is no conflict of interest.
Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with the Egyptian Knowledge Bank (EKB).
Department of Industrial Electronics and Control Engineering, Faculty of Electronic Engineering, Menoufia University, Menouf, 32952, Egypt
Ebrahim A. Ramadan, Belal A. Abouzalam & Ghada M. El-Banby
Department of Electrical Engineering of Computer and Control Systems, Faculty of Engineering, Kafrelsheikh University, Kafrelsheikh, 33516, Egypt
Nada M. Moawad & Wessam F. Abouzaid
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Nada M. Moawad wrote the original draft of the main manuscript text and methodology design. Ebrahim A. Ramadan, Ghada M. El-Banby, and Belal A. Abouzalam did the text review and editing. Wessam F. Abouzaid and Ghada M. El-Banby revise the methodology, formal analysis, and conceptualization. Ebrahim A. Ramadan and Belal A. Abouzalam revise visualization and validation. All authors reviewed the final manuscript.
Correspondence to Nada M. Moawad.
The authors declare no competing interests.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
Ramadan, E.A., Moawad, N.M., Abouzalam, B.A. et al. A hybrid convolutional-transformer neural network model for photovoltaic fault detection and localization. Sci Rep 16, 23222 (2026). https://doi.org/10.1038/s41598-026-57859-7
Download citation
Received: 11 February 2026
Accepted: 10 June 2026
Published: 24 July 2026
Version of record: 24 July 2026
DOI: https://doi.org/10.1038/s41598-026-57859-7
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Scientific Reports (Sci Rep)
ISSN 2045-2322 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

source

Brazilian researchers have found that delayed photovoltaic inverter startup can result in annual energy losses of 1.88% at a utility-scale solar plant. They analyzed 2023 operational data from a PV plant in the western part of the Brazilian state of São Paulo and developed a methodology to quantify losses associated with delayed inverter startup after sunrise.
The scientists said inverter startup can be delayed even when irradiance levels are sufficient for power generation. At the plant analyzed, they recorded delays of up to four hours after sunrise.
The proposed methodology primarily relies on sunrise times and the plant’s alternating-current (AC) active power data. Plane-of-array (POA) irradiance measurements are required only during the configuration and validation stages, reducing the instrumentation needed for the analysis.
The results showed that more than 55% of annual losses associated with inverter startup delays occurred within the first 90 minutes after sunrise. The researchers also found that losses varied throughout the year.
A quarterly analysis showed that the first half of the year, corresponding to summer and autumn in the Southern Hemisphere, accounted for 68.8% of the recorded losses. The researchers said the results indicate that assessing inverter behavior at the beginning of the daily generation period could help identify opportunities to improve plant performance.
The study also revealed significant differences among the eight inverters analyzed. Monthly losses associated with startup delays ranged from 0.14% to 5.02% of each inverter’s energy generation.
According to the researchers, the variability indicates that startup delays do not necessarily affect all inverters at a plant equally. Identifying the units responsible for the largest losses could therefore enable more targeted operational interventions.
The researchers said losses caused by inverter startup delays can be difficult to distinguish from other sources of performance losses at PV plants. Their methodology uses sunrise as a reference point to determine when inverters should begin operating and compares the expected startup behavior with recorded power output.
The approach makes it possible to distinguish startup losses from other factors affecting energy generation, including system unavailability, component mismatch, and weather conditions.
At the plant analyzed, inverter startup delays resulted in annual energy losses equivalent to 1.88% of total generation. The researchers said the figure is comparable to losses typically considered in PV system performance and availability assessments.
They said the methodology can be applied to utility-scale PV plants using existing operational data, helping operators identify inverter startup issues and improve plant performance.
Researchers from the Marcelo Villalva Photovoltaic Energy and Systems Laboratory (LESF-MV) at the University of Campinas (Unicamp) conducted the study. The research team included João Frederico Souza de Paula, Marcel Veloso Campos, João Lucas de Souza Silva, Marcelo Vinícius de Paula, Valentin Ferrer, Gustavo Fraidenraich, and Tárcio André dos Santos Barros.
The study, “A method for quantifying availability losses caused by PV inverter start-up delays in a large-scale PV power plant: A case study in Southeastern Brazil,” was published in Electric Power Systems Research.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Researchers have assessed the reliability of models used to estimate direct and diffuse irradiance from global horizontal irradiance (GHI) data at two sites in southern Peru. The study identifies a systematic, altitude-dependent divergence between the model results and Solargis satellite data, with potential implications for estimating photovoltaic (PV) yields in Andean regions.
The study, “GHI-based PV yield assessment in high-altitude Andean sites: workflow performance and systematic limitations of irradiance decomposition models,” compares two sites with contrasting climates: Cota Cota, located 4,240 meters above sea level on the Peruvian Altiplano, and Pampas de Majes, at 1,498 meters above sea level in an arid coastal valley in southern Peru.
The researchers used GHI and ambient temperature measurements recorded between 2021 and 2023. They used the data to adapt ERA5 meteorological reanalysis data with multilayer perceptron models, a type of neural network, to reconstruct a 25-year hourly climatology for the 2000-24 period.
Based on these time series, they generated representative meteorological years and estimated diffuse horizontal irradiance (DHI) and direct normal irradiance (DNI).
For the irradiance decomposition, the researchers used the Engerer2 model with two clear-sky schemes. The first was Threlkeld-Jordan, which was used in the original calibration of the Engerer2 model. The second was ARGPv2, an empirical model developed specifically for high-altitude sites in the Andes of northwestern Argentina, in the provinces of Salta and Jujuy, and recalibrated using the McClear clear-sky model.
Engerer2 is an empirical decomposition model that separates GHI into its direct and diffuse components. It uses a clear-sky model as a reference to describe expected irradiance under cloud-free conditions. Threlkeld-Jordan and ARGPv2 provide this clear-sky reference, with ARGPv2 designed to better represent the atmospheric conditions found at high-altitude Andean sites.
The researchers then fed the resulting irradiance components into a Liu-Jordan transposition model to calculate irradiance on the plane of the PV modules. This model is a widely used solar-energy model for estimating how much solar irradiance reaches a tilted surface, such as a PV module, from irradiance measured or estimated on a horizontal surface.
They subsequently applied a PV production model based on standard test conditions (STC) to calculate annual generation across a full matrix of module tilt and orientation combinations.
The procedure allowed the researchers to assess how differences introduced during irradiance decomposition ultimately affect energy yield estimates for PV installations.
The limited availability of direct DNI and DHI measurements is one of the main challenges facing solar resource assessments in many parts of the Andes. Where only GHI measurements are available, both components must be estimated using models, introducing an additional source of uncertainty.
Because direct DNI and DHI measurements were unavailable at the two study sites, the researchers compared their decomposition results with Solargis data, an independent dataset based on satellite observations and atmospheric models.
The comparison revealed a systematic divergence between the two approaches that, according to the researchers, varies with altitude.
The finding is particularly relevant to PV projects on the Altiplano and in other high-altitude areas of the Andes, where atmospheric conditions differ from those at the locations used to calibrate many conventional irradiance decomposition models.
The study highlights a limitation of solar resource assessment methods based exclusively on GHI. Although these approaches can reconstruct long-term time series and estimate PV production at sites without comprehensive solar radiation measurements, uncertainty associated with separating direct and diffuse irradiance can become systematic at high altitudes.
The researchers said the findings also point to the need for local validation of models used in PV resource and yield assessments in the Andes, particularly when the results are used for project sizing or long-term generation forecasts.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Advancing the commercial and industrial energy transition.
Solar power plants in Qinghai Province, China.
China is celebrating a milestone after its installed solar power generating capacity has officially surpassed coal power for the first time, aligning with its broader goals to peak carbon emissions by 2030 and achieve total carbon neutrality by 2060.
According to multiple news reports citing China’s National Energy Administration (NEA), the country’s installed solar power capacity reached 1,286 GW as of the end of July, pulling ahead of the country’s coal-fired power installed capacity, which rose to 1,285 GW. Coal has been a key power generation source for China’s electricity grid for decades, making the country the world’s largest annual emitter of greenhouse gases (GHG) and carbon dioxide (CO2).
Data published in February by global research agency Wood Mackenzie highlighted that the country’s coal-fired power generation fell by nearly 2% in 2025, despite rising energy demand. Senior research analyst Sharon Feng for Wood Mac noted that “China’s wind and solar capacity had risen more than ten-fold” to 1,842 GW over the past decade.
Engineering and technical personnel installing solar photovoltaic glass in Hebei, China.
China added 315 GW of solar power and 119 GW of wind power capacity last year, with two sources combining to account for over 80% of the country’s total newly installed power generation capacity, according to secondary reports like the Global Times citing data from the China Electricity Council.
“For the first time, photovoltaic installed capacity surpassed coal-fired power, becoming the largest power source category in China,” the NEA said on Tuesday, according to reports.
The H1 2026 report marked the first time coal’s share of China’s total electricity resource mix dropped below 50%, according to the energy administration. The installations of solar farms in China reportedly overtook coal plants in July, becoming the country’s top power capacity source.
Newly installed solar photovoltaic (PV) power capacity in July totaled 13.73 GW, up 43% year-over-year (YOY). The Chinese NEA notes that PV power generation amounted to 802.4 billion kWh (802,000 GWh), representing a 15.5% YOY increase.
The Global Times, citing the NEA, reported that this marks 10.8 percentage points higher than the total social electricity consumption growth rate, which measures changes in the total amount of electricity used over specific periods by an entire society. However, since solar doesn’t perform 24/7 around the clock when the sun isn’t shining, it remains behind coal in terms of ​power generation.
According to the London-based Energy Institute (EI), U.S. emissions grew four times the rate of China’s emissions growth, representing 40% of the total global emissions increase. U.S. emissions rose by roughly 147 million metric tons over the full year of 2025.
The EI’s Statistical Review of World Energy report highlighted that China’s rush to achieve economic dominance has forced it to build out its energy portfolio. Demand signals have grown among hyperscalers and AI data centers as developers race to meet supercomputing load with more accelerated deployment targets.
In 2025 alone, China’s electricity generation reportedly increased by almost 500 TWh. For comparison, scientific researchers from the nonprofit Our World in Data point out that Germany generates almost exactly that amount, effectively signaling China added a Germany-sized grid to its electricity system in just one year.
China is now outpacing the U.S. on electrification progress and dominates the world in newly installed renewable energy capacity, according to the latest EI statistical review report. China’s massive buildout of large-scale wind and solar projects accounts for 74% globally, compared to roughly 6% of U.S. global utility-scale wind and solar projects.

source

Subscribe
Today’s print edition
Home Delivery
Japan’s agrivoltaics projects, which use land for both solar power generation and agriculture, are finding themselves at a crossroads after regulations are set to be tightened following a series of inappropriate cases, including failures to meet required crop yield standards.
Some oppose the tighter regulations, arguing they could exclude proper operators that seek to preserve farmland and promote smart agriculture using revenue from electricity sales.

The first of agrivoltaics projects started in Japan in 2013. Operators were granted approval to install solar panels on farmland as a temporary conversion of land use, on the condition that crop yields exceed 80% of the local average.
The projects attracted attention as a way to boost farmers’ income, as the same land can be used for both agriculture and power generation without losing its status as farmland, to which lower property tax rates apply.
A total of 6,137 projects had been approved as of the end of fiscal 2023.
However, there were many cases in which solar power companies rented abandoned farmland and failed to engage sufficiently in agricultural activities. Crop yields fell short of targets or crops showed poor growth in 1,221 cases, about one-fourth the number of projects in which panels were installed as of the end of fiscal 2023.
In response to the situation, the agriculture ministry drafted a set of new requirements, including the minimum production or sales of at least ¥500,000 and keeping the shading rate caused by solar panels below 30%.
Keisuke Obikawa, 30, a part-time rice farmer in a hilly area close to mountains in Chino, Nagano Prefecture, introduced smart farming technology to remotely control water supply by using revenue increased by selling electricity generated by solar panels.
Obikawa warned the uniform tightening of regulations will narrow the options for maintaining paddy fields in hilly areas near mountains.
“I hope projects will be evaluated based on their contribution to farmland conservation,” especially in areas where farmland consolidation is difficult, he said.
Masaya Ishida, director at the Renewable Energy Institute, said there is “too little basis” for applying the shading rate rule to all projects.
Ishida is calling for a review of the rule, saying that power generation can be combined with the cultivation of coffee beans, which prefer semishaded conditions, and tea, which requires shading for a certain period.
In a time of both misinformation and too much information,
quality journalism is more crucial than ever.
By subscribing, you can help us get the story right.
With your current subscription plan you can comment on stories. However, before writing your first comment, please create a display name in the Profile section of your subscriber account page.
Your subscription plan doesn’t allow commenting. To learn more see our FAQ
Sponsored contents planned and edited by JT Media Enterprise Division.
広告出稿に関するおといあわせはこちらまで
Read more

source

The Jersey Shore’s Hit Music Channel
New Jersey residents have a new option to save on their electric bills.
Portable solar panels that meet safety standards are now legal to buy online or at local home improvement stores, Gov. Mikie Sherrill announced on Tuesday.
The law exempts these plug-in energy generators under 1,200 watts from utility interconnection and metering requirements. Depending on wattage, models can range in price between a few hundred dollars and $1,500.
"These units cost a fraction as much as rooftop solar, but they can still shave up to $50 off the typical monthly bill," Sherrill said. That's savings of $600 a year. She said there are more than 1 million of these panels in Germany alone.
Often placed on balconies or backyards, most portable solar panels have attached microinverters that go out to a regular plug. They plug into standard 120-volt wall outlets, allowing other appliances and devices in the home to draw from that power.
Under the new law, residents who use balcony solar panels don't have to notify or get approval from their power companies. The law also stops landlords and homeowner associations from banning them. Municipalities can't ban or require permits for them either.
The change gives New Jersey homeowners and renters a new option for generating some of their own electricity without taking on the cost and complexity of a traditional rooftop solar installation.
It also removes several potential roadblocks that could otherwise prevent residents from using the systems.
Only six months passed between when the bill (S2368) was introduced and its signing on Tuesday, a relatively speedy journey in Trenton. It passed unanimously. This makes New Jersey the 9th state to legalize balcony solar, according to PlugInSolarUS.
The Garden State Balcony Solar Act is one of several laws the Sherrill administration has passed to counter the spike in New Jersey energy bills, which was the highest increase in the nation last year.
Gallery Credit: New Jersey 101.5
Report a correction 👈 | 👉 Contact our newsroom
Gallery Credit: Erin Vogt

source

A research group in China has fabricated a perovskite solar cell using a new molecular design approach aimed at reducing defects in the perovskite film and increasing device efficiency and stability.
“Rapid film crystallization can generate subtle defects that become concentrated at surfaces, grain boundaries, and buried interfaces,” the research’s lead author, Cong Chen, told pv magazine. “Although molecular additives can effectively mitigate these imperfections, their design requires a careful balance: strong interactions favor defect passivation, but excessive molecular packing or the formation of insulating interfacial layers can hinder charge transport.”
“Additive engineering offers a practical strategy to address these challenges by regulating crystallization while simultaneously passivating electronic defects,” he went on to say. “However, effective additives must balance strong defect binding with unhindered charge transport while ideally providing additional protection against environmental stress.”
The proposed molecular design approach is described as a steric-gated dual-site chelation (SGDC) strategy. It consists of a rigid three-dimensional adamantane scaffold bearing two carboxymethyl coordination arms that enable strong chelation of undercoordinated lead ions (Pb²⁺), while sterically regulating local molecular packing.
The scientists used an adamantane-based molecular additive known as ADA-DA, which was designed to passivate defects in perovskite films. Its two carboxymethyl groups can simultaneously coordinate with undercoordinated Pb²⁺, while its rigid three-dimensional adamantane core controls molecular packing and prevents excessive interfacial crowding.
“This combination enables steric-gated dual-site chelation, providing strong defect passivation without hindering charge transport,” Chen said. “Experimental measurements confirmed strong interactions of ADA-DA with both lead and iodide ions.”
The perovskite film resulting from the molecular strategy exhibited larger grains, fewer grain boundaries, and a more compact and smoother morphology, according to the research team. ADA-DA was distributed throughout the perovskite but preferentially accumulated near surfaces and grain boundaries, where defect passivation is most needed. Electrical measurements further confirmed a substantial reduction in both electron and hole trap densities.
The proposed perovskite solar cell was based on an inverted positive-intrinsic-negative (p-i-n) architecture consisting of an indium tin oxide (ITO) substrate, a hole transport layer (HTL) made of nickel oxide (NiOₓ) and a self-assembled monolayer, the perovskite film, a buckminsterfullerene (C60) electron transport layer (ETL), a bathocuproine (BCP) buffer layer, and a silver (Ag) metal contact. The perovskite layer incorporating ADA-DA was deposited using a vacuum-flash crystallization process.
Under standard illumination conditions, the device achieved a power conversion efficiency of 27.39%, compared with 26.24% for a reference cell built without the proposed molecular strategy. The result was verified by an undisclosed independent third-party certification body, the researchers said.
In a further step, the academics scaled up the ADA-DA-modified device from small-area cells to bifacial mini-modules using either transparent ITO or indium zinc oxide (IZO) electrodes on both sides. Under glass/ITO-side illumination, a 13-cell module with an area of 163.93 cm² reached a peak power conversion efficiency of 22.21%, while under IZO-side illumination, a 12-cell module with an area of 151.32 cm² achieved an efficiency of up to 22.33%.
“The bifacial architecture combines high efficiency with transparency and illumination from either side, making it attractive for building-integrated photovoltaics (BIPV),” Chen stated. “More importantly, the 151.32 cm² ADA-DA module maintained nearly constant power output for over 5,000 h under continuous white light-emitting diode (LED) illumination. Its output decreased only from 1,028.5 to 1,024.1 mW, corresponding to 99.6% retention, demonstrating exceptional long-term operational stability.”
Both the cells and modules were described in “Steric-gated dual-site chelation enables 27.3% efficient perovskite solar cells and ultra-stable 22% bifacial modules”, published in Joule. The research team included scientists from Hebei University of Technology, Jiaxing Nanhu University, and Tianjin University.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
Martedì, 22 Settembre 2026
11:00 – 12:00 CEST, Roma
Monday, October 26, 2026
10:30 am – 11:30 am CEST, Berlin, Paris, Madrid
Thursday, September 10, 2026
2:00 pm – 3:00 pm CEST, Berlin, Paris, Madrid
Tuesday, September 15, 2026
5:00 pm – 6:00 pm CEST, Berlin, Paris, Madrid
Our special edition for Intersolar South America 2026 is here!
Discover the latest insights into the Brazilian solar market – in Portuguese.
A two-day conference in Austin, Texas, bringing together leaders in US solar manufacturing, equipment specification, and factory execution.
Saudi Arabia is accelerating its clean energy transition—join the SunRise Arabia Clean Energy Conference 2026 in Riyadh to explore how solar PV and energy storage are powering its digital economy.
pv magazine USA hosts its multi-day virtual event on U.S. solar and energy storage, covering domestic manufacturing, distributed energy and the growing role of solar-plus-storage in meeting AI-driven power demand.
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Aligned Climate Capital Raises $500 Million for U.S. Distributed Solar  energynews.pro
source

As farmland shrinks and solar demand soars, one Indian state found a way to generate power without sacrificing crops—by harvesting sunlight above water instead of earth.
By Ainura Kalau
Published September 4, 2026 · How we edit
Here’s the puzzle a lot of solar planning eventually runs into. A meaningful amount of solar power takes a meaningful amount of land, somewhere between four and five acres for every megawatt of ground-mounted panels. In a country with a huge population to feed and a limited amount of arable land to feed them from, that math turns into a real conflict fast: every acre under panels is an acre not growing food, and every acre spared for food is an acre not generating electricity.
So where do you put a solar array that isn’t already spoken for by farming, and isn’t just empty desert either?
Gujarat’s answer, tested first on the Sanand Branch Canal near the village of Chandrasan in Mehsana district on April 24, 2012, was to stop thinking of land as the only available surface. A canal is infrastructure, not farmland. It’s already claimed, already fenced off from crops, and it runs for hundreds of kilometers in a state crisscrossed by the irrigation network built to move water from the Sardar Sarovar dam.
Someone had to notice that a canal’s surface could hold a solar array the same way a roof holds one, and that a 1-megawatt pilot project could test the idea without waiting for a bigger, riskier bet to prove it first.
The obvious win is power generation on land nobody had to give up. The less obvious win showed up in the water itself. A canal running through Gujarat’s heat loses a real volume of its water to evaporation before it ever reaches a farmer’s field, and shading even a portion of that surface slows the loss down.
A study led by researcher Sagarkumar Agravat at the Gujarat Energy Research and Management Institute found that canal-top installations save roughly 90 lakh litres of water per megawatt per year from evaporation, north of 9 million litres annually for a project the size of the original Chandrasan pilot. The same study found a second, unplanned benefit: panels mounted over flowing water ran about 10 degrees Celsius cooler than equivalent panels mounted on dry ground, which pushed their solar efficiency up by roughly 2.5 percent. Water that was already just sitting there, doing its regular job of getting crops watered, turned out to be a decent air conditioner for a solar panel too.
The pilot’s basic claim, first canal-top solar installation of its kind in the country, has held up well enough that NITI Aayog, the Indian government’s policy think tank, cites the Narmada branch canal system project as the template other Indian states have since copied, from New Town near Kolkata to canal networks well outside Gujarat. Rangan Banerjee, director of the Centre for Technology Alternatives for Rural Areas at IIT Bombay, frames the appeal in blunt planning terms: “Decentralized solar solutions, especially those integrated with existing infrastructure, can address local energy needs while minimizing environmental impact.” India’s canal network runs past 120,000 kilometers nationwide, which is a lot of unclaimed rooftop hiding in plain sight if a state actually wants to use it.
Gujarat didn’t stop at the pilot. A later, considerably larger installation strung panels across 3.5 kilometers of canal between the Sama and Chhani areas of Baroda, a 10-megawatt project that, according to SS Rathore, chairman and managing director of Sardar Sarovar Narmada Nigam Ltd, generates 16.2 million units of power every year while sparing roughly 20 acres of land that an equivalent ground-mounted array would have needed. Scale that reasoning up further, to the more than 2,000 megawatts Gujarat has floated in planning documents, and the land spared runs into the tens of thousands of acres, none of it taken from a single working farm.
None of this makes canal-top solar the obvious choice for every new project, and it’s worth being honest about why growth has slowed since the original pilot. Mounting a panel array over a moving canal costs real money that a flat field doesn’t. Kaushik Patel, of the same Gujarat Energy Research and Management Institute that ran the water-savings study, put a number on it: “Of the entire capital cost of a CETP, 40% is that of the mounting structure, more than the cost of panels,” which pushes a canal-top megawatt roughly 15 to 20 million rupees above a ground-mounted one. Utilities pay the same rate for electricity either way, so developers have little financial incentive to pick the harder, pricier build. Yogesh Sehgal, director of the solar firm SAM Solar Private Ltd, framed the incentive problem plainly: “The rate of solar paid by the discoms is the same for both ground-mounted and CTPV, so why will a developer invest more? Unless the government provides subsidies, CTPVs will not be economically viable.” Repairs are harder too. Jaideep Parmar, a deputy executive engineer at Sardar Sarovar Narmada Nigam Ltd, described what a simple wiring fault costs on a canal-top array: “2-3 hours are lost in rectifying it as the technician can’t go below the arrays without safety equipment. All this while, the entire unit has to be shut off.”
The 2012 pilot proved the physics and the water savings were real. The economics of building more of them, at least so far, have proved slower to catch up.
I keep coming back to how unglamorous the actual insight was. Nobody invented a new kind of solar cell or a new kind of canal. Someone just looked at infrastructure that already existed, already had a job, and asked whether it could do a second one without asking anything new of the land around it. My own read on ambition has always leaned toward that same instinct: people who want something tend to go looking for the opportunity sitting in front of them, rather than waiting for permission to build something bigger and more complicated. A canal that was only ever asked to move water is now also making electricity and losing less of itself to the sun in the process. Real tricks like that tend to hide in plain sight, right up until someone actually looks.
Standards
Articles on Science Blog are edited and fact-checked before publication. We use AI tools in the newsroom. See our editorial standards and masthead.
Written by
Ainura was born in Central Asia, spent over a decade in Malaysia, and studied at an Australian university before settling in São Paulo, where she’s now raising her family. Her life blends cultures and perspectives, something that naturally shapes her writing. When she’s not working, she’s usually trying new recipes while binging true crime shows, soaking up sunny Brazilian days at the park or beach, or crafting something with her hands.
Clear science for curious people
A Brown Brothers Media publication. All rights reserved © 2026.
Clear science for curious people — free PDF, then our science email.
Also available via RSS.

source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Materials (2026) Cite this article
Regulation of electron- or hole-selective interfacial properties has traditionally relied on extrinsic chemical doping. Here we report a structurally driven strategy for modulating carrier-selective electronic characteristics through the organic-cation-induced reconfiguration of lead iodide octahedral connectivity. Two closely related imidazoline-based cations, which differ by only a single heteroatom, drive the formation of one-dimensional organic lead triiodide phases with distinct octahedral-sharing patterns. Single-crystal analysis combined with theoretical calculations reveals that the variation in connectivity reshapes orbital coupling, leading to pronounced changes in the work function and absolute band-edge positions that give rise to distinct carrier-selective interfacial behaviour without the introduction of extrinsic dopants. Leveraging this structural reconfiguration-induced electronic modulation, we realize complementary charge-selective contacts in perovskite solar cells, achieving a power conversion efficiency of 27.61% (certified steady-state 27.19%), together with excellent operational stability and scalability, including 22.26% efficiency in 655-cm² modules.
This is a preview of subscription content, access via your institution
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$32.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$259.00 per year
only $21.58 per issue
Buy this article
USD 39.95
Prices may be subject to local taxes which are calculated during checkout
The data that support the findings of this study are available within the Article and its Supplementary Information. Source data are provided with this paper.
Kim, H.-S. et al. Lead iodide perovskite sensitized all-solid-state submicron thin film mesoscopic solar cell with efficiency exceeding 9%. Sci. Rep. 2, 591 (2012).
Article  PubMed  PubMed Central  Google Scholar 
Xiong, Z. et al. Homogenized chlorine distribution for >27% power conversion efficiency in perovskite solar cells. Science 390, 638–642 (2025).
Article  CAS  PubMed  Google Scholar 
Zhang, Y. et al. Dynamical transformation of two-dimensional perovskites with alternating cations in the interlayer space for high-performance photovoltaics. J. Am. Chem. Soc. 141, 2684–2694 (2019).
Article  CAS  PubMed  Google Scholar 
Liang, Z. et al. Homogenizing out-of-plane cation composition in perovskite solar cells. Nature 624, 557–563 (2023).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Cho, S.-H. et al. Solution-phase PbI₂ coordination controls perovskite film formation and photovoltaic performance. ACS Energy Lett. 10, 3931–3940 (2025).
Article  CAS  Google Scholar 
Canil, L. et al. Tuning halide perovskite energy levels. Energy Environ. Sci. 14, 1429–1438 (2021).
Article  CAS  Google Scholar 
Peng, Y. et al. Enlarging moment and regulating orientation of buried interfacial dipole for efficient inverted perovskite solar cells. Nat. Commun. 16, 1252 (2025).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Zhang, C. et al. Work function tuning of a weak adhesion homojunction for stable perovskite solar cells. Joule 8, 1394–1411 (2024).
Article  CAS  Google Scholar 
Zhang, J., Sun, Y. & Yu, H. Reducing energy loss via adjusting the anode work function and perovskite layer morphology for the efficient and stable hole transporting layer-free perovskite solar cells. Chem. Eng. J. 431, 133948 (2022).
Article  CAS  Google Scholar 
Ren, B. et al. A universal p-type heterointerface for inverted perovskite solar cells. Angew. Chem. Int. Ed. 64, e202507950 (2025).
Article  CAS  Google Scholar 
Liu, S. et al. Buried interface molecular hybrid for inverted perovskite solar cells. Nature 632, 536–542 (2024).
Article  PubMed  Google Scholar 
Zhou, C. et al. Dual interfacial modification with 1D perovskite for self-assembled monolayer based inverted perovskite solar cells. Nano Energy 128, 109811 (2024).
Article  CAS  Google Scholar 
Jeon, Y.-S. & Park, N.-G. Nickel oxide thin films formed from the pH-adjusted solutions for inverted perovskite solar cells. ACS Energy Lett. 10, 1266–1274 (2025).
Article  CAS  Google Scholar 
Liu, G. et al. Mixed-phase low-dimensional perovskite-assisted interfacial lead directional management for stable perovskite solar cells with efficiency over 24%. ACS Energy Lett. 6, 4395–4404 (2021).
Article  CAS  Google Scholar 
Xue, J. et al. Reconfiguring the band-edge states of photovoltaic perovskites by conjugated organic cations. Science 371, 636–640 (2021).
Article  CAS  PubMed  Google Scholar 
Li, M., Huang, P. & Zhong, H. Current understanding of band-edge properties of halide perovskites: Urbach tail, Rashba splitting, and exciton binding energy. J. Phys. Chem. Lett. 14, 1592–1603 (2023).
Article  CAS  PubMed  Google Scholar 
Daboczi, M. et al. Optimal interfacial band bending achieved by fine energy level tuning in mixed-halide perovskite solar cells. ACS Energy Lett. 6, 3970–3981 (2021).
Article  CAS  Google Scholar 
Kutkan, S. et al. Impact of the organic cation on the band-edge emission of two-dimensional lead–bromide perovskites. Nanoscale 15, 12880–12888 (2023).
Article  CAS  PubMed  Google Scholar 
Li, X. et al. Constructing heterojunctions by surface sulfidation for efficient inverted perovskite solar cells. Science 375, 434–437 (2022).
Article  CAS  PubMed  Google Scholar 
Monika, Verma, A., Tiwari, M. K., Show, B. & Saha, S. Modulation of weak interactions in structural isomers: positional isomeric effects on crystal packing and physical properties and solid-state thin-film fabrication. ACS Omega 5, 448–459 (2019).
Article  PubMed  PubMed Central  Google Scholar 
Zeng, M. et al. Tailored supramolecular interface enables efficient and stable tin halide perovskite photovoltaics. ACS Energy Lett. 10, 1357–1365 (2025).
Article  CAS  Google Scholar 
Dahlman, C. J., Kubicki, D. J. & Reddy, G. N. M. Interfaces in metal halide perovskites probed by solid-state NMR spectroscopy. J. Mater. Chem. A 9, 19206–19244 (2021).
Article  CAS  Google Scholar 
Hui, W. et al. Stabilizing black-phase formamidinium perovskite formation at room temperature and high humidity. Science 371, 1359–1364 (2021).
Article  CAS  PubMed  Google Scholar 
Kang, D.-H., Park, Y.-J., Jeon, Y.-S. & Park, N.-G. Extended X-ray absorption fine structure (EXAFS) of FAPbI₃ for understanding local structure–stability relation in perovskite solar cells. J. Energy Chem. 67, 549–554 (2022).
Article  CAS  Google Scholar 
Zhang, B. et al. Inhibiting ion migration and stabilizing crystal-phase in halide perovskite via directly incorporated fluoride anion. Angew. Chem. Int. Ed. 137, e202413550 (2025).
Article  Google Scholar 
Wen, Y. et al. Amorphous (lysine)₂PbI₂ layer enhanced perovskite photovoltaics. Nat. Commun. 15, 7085 (2024).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Sanson, A. EXAFS spectroscopy: a powerful tool for the study of local vibrational dynamics. Microstructures 1, 2021004 (2021).
Google Scholar 
Ibaceta-Jaña, J. et al. Vibrational dynamics in lead halide hybrid perovskites investigated by Raman spectroscopy. Phys. Chem. Chem. Phys. 22, 5604–5614 (2020).
Article  PubMed  Google Scholar 
Zheng, J. et al. Incorporation of CoO nanoparticles in 3D marigold flower-like hierarchical architecture MnCo₂O₄ for highly boosting solar light photo-oxidation and reduction ability. App. Catal. B 237, 1–8 (2018).
Article  CAS  Google Scholar 
Yang, Y. et al. Inverted perovskite solar cells with over 2,000 h operational stability at 85 °C using fixed charge passivation. Nat. Energy 9, 37–46 (2024).
Article  CAS  Google Scholar 
Ji, R. et al. Spontaneous formation of 1D/3D perovskite heterojunctions for efficient inverted perovskite solar cells. Adv. Energy Mater. 14, 2304126 (2024).
Article  CAS  Google Scholar 
Wei, N. et al. Post-treatment-free dual-interface passivation via facile 1D/3D perovskite heterojunction construction. JACS Au 3, 3324–3332 (2023).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Chen, R. et al. Conformal imidazolium 1D perovskite capping layer stabilized 3D perovskite films for efficient solar modules. Adv. Sci. 9, 2204017 (2022).
Article  CAS  Google Scholar 
Jiao, H. et al. Perovskite grain wrapping by converting interfaces and grain boundaries into robust and water-insoluble low-dimensional perovskites. Sci. Adv. 8, eabq4524 (2022).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Xiong, W. et al. Controllable p-and n-type behaviours in emissive perovskite semiconductors. Nature 633, 344–350 (2024).
Article  CAS  PubMed  Google Scholar 
Xiong, S. et al. Direct observation on p- to n-type transformation of perovskite surface region during defect passivation driving high photovoltaic efficiency. Joule 5, 467–480 (2021).
Article  CAS  Google Scholar 
Zheng, Y. et al. Towards 26% efficiency in inverted perovskite solar cells via interfacial flipped band bending and suppressed deep-level traps. Energy Environ. Sci. 17, 1153–1162 (2024).
Article  CAS  Google Scholar 
Al-Ashouri, A. et al. Monolithic perovskite/silicon tandem solar cell with >29% efficiency by enhanced hole extraction. Science 370, 1300–1309 (2020).
Article  CAS  PubMed  Google Scholar 
Cacovich, S. et al. Imaging and quantifying non-radiative losses at 23% efficient inverted perovskite solar cells interfaces. Nat. Commun. 13, 2868 (2022).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Xu, H. et al. Constructing robust heterointerfaces for carrier viaduct via interfacial molecular bridges enables efficient and stable inverted perovskite solar cells. Energy Environ. Sci. 16, 5792–5804 (2023).
Article  CAS  Google Scholar 
Ye, F. et al. Overcoming C₆₀-induced interfacial recombination in inverted perovskite solar cells by electron-transporting carborane. Nat. Commun. 13, 7454 (2022).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Warby, J. et al. Mismatch of quasi-Fermi level splitting and V_oc in perovskite solar cells. Adv. Energy Mater. 13, 2303135 (2023).
Article  CAS  Google Scholar 
Caprioglio, P. et al. Open-circuit and short-circuit loss management in wide-gap perovskite p-i-n solar cells. Nat. Commun. 14, 932 (2023).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Duan, C. et al. Scalable fabrication of wide-bandgap perovskites using green solvents for tandem solar cells. Nat. Energy 10, 318–328 (2025).
Article  CAS  Google Scholar 
Sarritzu, V. et al. Optical determination of Shockley–Read–Hall and interface recombination currents in hybrid perovskites. Sci. Rep. 7, 44629 (2017).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Kresse, G. & Hafner, J. Ab initio molecular dynamics for open-shell transition metals. Phys. Rev. B 48, 13115–13118 (1993).
Article  CAS  Google Scholar 
Kresse, G. & Nafner, J. Ab initio molecular-dynamics simulation of the liquid-metal–amorphous-semiconductor transition in germanium. Phys. Rev. B 49, 14251–14269 (1994).
Article  CAS  Google Scholar 
Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996).
Article  CAS  PubMed  Google Scholar 
Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations. Phys. Rev. B 13, 5188–5192 (1976).
Article  Google Scholar 
Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H–Pu. J. Chem. Phys. 132, 154104 (2010).
Article  PubMed  Google Scholar 
Dronskowski, R. & Blöchl, P. E. Crystal orbital Hamilton populations (COHP): energy-resolved visualization of chemical bonding in solids based on density-functional calculations. J. Phys. Chem. 97, 8617–8624 (1993).
Article  CAS  Google Scholar 
Deringer, V. L., Tchougréeff, A. L. & Dronskowski, R. Crystal orbital Hamilton population (COHP) analysis as projected from plane-wave basis sets. J. Phys. Chem. 115, 5461–5466 (2011).
Article  CAS  Google Scholar 
Maintz, S., Deringer, V. L., Tchougréeff, A. L. & Dronskowski, R. LOBSTER: a tool to extract chemical bonding from plane-wave based DFT. J. Comput. Chem. 37, 1030–1035 (2016).
Article  CAS  PubMed  PubMed Central  Google Scholar 
Hoover, W. G. Canonical dynamics: equilibrium phase-space distributions. Phys. Rev. A 31, 1695–1697 (1985).
Article  CAS  Google Scholar 
Download references
This work was supported by National Research Foundation of Korea (NRF) grants funded by the Korean government (MSIT and MOE) under contracts RS-2026-25501632 (NRL 2.0), NRF-2021R1A3B1076723 (Research Leader Program), RS-2025-02316700 (Carbon-free Energy Core Technology Program), RS-2023-00259096 (GRDC Cooperative Hub) and the Ministry of Trade, Industry and Energy (MOTIE) of Korea (P0022336). We thank the Shanghai Synchrotron Radiation Facility for use of the BL17B1 and BL13SSW beamlines (cstr.cn/31124.02.SSRF.BL13SSW) and for support with the XAFS experiments.
These authors contributed equally: Yalan Zhang, Zheng Liang.
School of Chemical Engineering and Center for Antibonding Regulated Crystals, Sungkyunkwan University, Suwon, Republic of Korea
Yalan Zhang, Zheng Liang, Seong Chan Cho, Seong-Ho Cho, Guiming Fu, Sang-Uk Lee, Sanwan Liu & Nam-Gyu Park
School of Chemical Engineering, Sungkyunkwan University, Suwon, Republic of Korea
Seong Chan Cho, Jae Hun Seol & Sang Uck Lee
Key Laboratory of Applied Surface and Colloid Chemistry, National Ministry of Education, Shaanxi Key Laboratory for Advanced Energy Devices, Shaanxi Engineering Lab for Advanced Energy Technology and School of Materials Science and Engineering, Shaanxi Normal University, Xi’an, People’s Republic of China
Xin Chen & Kui Zhao
University of Science and Technology of China, Hefei, People’s Republic of China
Boyuan Liu, Hui Zhang & Xu Pan
School of Microelectronics, Hefei University of Technology, Hefei, People’s Republic of China
Xu Pan
SKKU National Lab for Intelligent Energy Solution Technology (SIEST), Sungkyunkwan University, Suwon, Republic of Korea
Nam-Gyu Park
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Y.Z., Z.L. and N.-G.P. conceived the study. N.-G.P. supervised the project. Y.Z. and Z.L. fabricated the perovskite devices, performed the photovoltaic characterizations and analysed the experimental data. Y.Z. and Z.L. wrote the paper. N.-G.P. revised and edited the paper. G.F. performed the KPFM measurements. B.L. and H.Z. assisted with the EXAFS analysis. X.C. conducted and analysed the GIWAXS measurements under the supervision of K.Z. The XRD measurements were performed by S.-U.L. The DFT calculations were carried out by S.C.C. and J.H.S. under the supervision of S.U.L. Analysis of the NMR data was assisted by S.-H.C. and S.L. All authors discussed the results and contributed to the paper.
Correspondence to Sang Uck Lee, Xu Pan or Nam-Gyu Park.
The authors declare no competing interests.
Nature Materials thanks Milos Dubajic and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Figs. 1–39; Tables 1–11, Notes 1–4 and refs. 1 and 2.
Statistical source data.
Statistical source data.
Statistical source data.
Statistical source data.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Reprints and permissions
Zhang, Y., Liang, Z., Cho, S.C. et al. Octahedral connectivity reconfigures interfacial carrier-selective properties for efficient perovskite solar cells. Nat. Mater. (2026). https://doi.org/10.1038/s41563-026-02722-3
Download citation
Received: 03 January 2026
Accepted: 28 July 2026
Published: 03 September 2026
Version of record: 03 September 2026
DOI: https://doi.org/10.1038/s41563-026-02722-3
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.