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.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Nature Materials (Nat. Mater.)
ISSN 1476-4660 (online)
ISSN 1476-1122 (print)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Sept. 3: Grand Wayne Convention Center completes roof and solar project  Fort Wayne Business Weekly
source

China-aided Laos’ first integrated photovoltaic-storage-charging “zero-carbon library” opens  Yahoo Finance
source

Global Solar PV O&M Market Reaches 348 GWdc As Vendor Consolidation Accelerates—Wood Mackenzie Report  SolarQuarter
source

Developed and constructed by Greece-headquartered energy company Metlen, the projects are now fully commissioned and operational, with a combined capacity of 236 MVA.
The portfolio of projects includes in the Riverina region of southern New South Wales (NSW), the 30 MW Corowa Solar Farm, 30 MW Junee Solar Farm, and 30 MW Wagga Wagga Solar Farm, 75 MW Wyalong Solar Farm, and Queensland-based 40 MW Kingaroy Solar Farm.
All six solar farms have been energised using Ingeteam’s central inverter technology, to deliver high efficiency and grid compatibility for utility-scale solar applications. 
Ingeteam Australia Managing Director Juan Miguel Gutierrez said the partnership with Metlen has been key to the successful delivery of Ingeteam’s inverters.
“We are proud to contribute our technology and expertise to these significant solar projects,” Gutierrez said.
Ingeteam has been present in Australia for over a decade through its subsidiary in North Wollongong.
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
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Pro Football Hall of Fame running back Emmitt Smith, his CRE company 4 13 Solutions Inc. and his business partners are facing a lawsuit over allegations that they exploited an investor in a Texas solar project scheme.
Kituwah LLC, a tribally owned Native American economic development agency, filed a lawsuit in Delaware on Monday alleging that Smith and business partner David Mosley promised to use $2.5M to acquire an interest in Project Exodus, a solar farm project in Texas, but didn't. Darrel Wilson and his company, Wilson Holdings of North America, are also named as defendants in the lawsuit.
Kituwah's investigation determined that "4 13 Solutions’ representations were part of Smith’s and Mosley’s scheme to cheat Kituwah out of $2.5 million dollars," according to the lawsuit.
Kituwah claims that Smith and Mosley persuaded the development arm to form a joint venture with their company and to loan $2.5M to the JV. Smith and Mosley allegedly said developer Genesis Consolidated Industries was in the process of acquiring land in Texas for the solar farm but didn't have enough money to acquire it, according to the lawsuit.
The lawsuit claims that 4 13 Solutions, through Smith and Mosley, made multiple misrepresentations and false promises to facilitate the investment, including that other investors were "clamoring" to finance the project. Kituwah also claims that 4 13 Solutions said it would obtain a Department of Energy loan to provide permanent financing and that the solar farm would be functional by the end of 2024 and generate millions in income.
Kituwah alleges that instead of investing in the renewable project, Smith and Mosley took the money and used it to "improperly pay Wilson Holdings, with whom they had partnered on other ventures."
4 13 Solutions did not respond to a request for comment.
The economic development agency claims that by the end of 2024, there was no sign that 4 13 Solutions had made any progress in acquiring the rights to the solar farm project. When its loan came due on Feb. 1, 2024, it wasn't repaid, according to the lawsuit.
"Kituwah’s promissory note remains unpaid years after maturity, and it has not recovered a penny from its $2.5 million loan," the lawsuit reads.
Smith is the National Football League's all-time leader in career rushing yards and rushing touchdowns, winning three Super Bowls with the Dallas Cowboys. He founded 4 13 Solutions in 2020 and has also formed other real estate organizations since his retirement.
Contact Ryan Wangman at ryan.wangman@bisnow.com
Becknell Taps Industrial Veteran To Boost REIT's Selling Group
American Healthcare REIT On A $1.5B Senior Housing Buying Spree
Joss Realty CEO, Executives Join Another Company's Real Estate Division
Crow Holdings Raising Nearly $3.3B Fund, Its Largest Ever
California Duo's Ponzi-Like Investment Scheme Unravels With Bankruptcy, SEC Charges
Sea Wall Rehabilitation: Strengthening Hudson River Walkway For Next 100 Years
The 'Fourth Utility': Why Connected Digital Infrastructure Is Essential To CRE
CRE's Clearance Rack Is Empty, Leaving Landlords To Create Their Own Value
Ares And Scion's Student Housing JV Acquires 4 Communities For $435M
Anthropic Inks $35B Deal With Nvidia-Backed AI Cloud Provider
Deutsche Finance Alleges BVK Is Responsible For CRE Portfolio Losing $1B In New Lawsuit
Chobani To Invest $1.2B In Manufacturing, Warehouse Campus In Pennsylvania
Company Info
Newsletters
Other Stuff
© Copyright 2026 Bisnow. All Rights Reserved
We’re not asking for your money. This is not a step towards a paywall. Our news is free and we intend to keep it that way.
Forgot your password? We can help!

source

Chinese researchers set 24% efficiency world record for large perovskite solar module  Yahoo Tech
source

Chiral Molecules Revolutionize Perovskite Solar Cells  Mirage News
source

SECI Invites Bids For 700 MW ISTS-Connected Solar PV Projects For C&I Consumers  SolarQuarter
source

Cambria Heights School Board President Ken Vescovi speaks at the district's ribbon cutting ceremony Thursday at the Elementary School in Carrolltown with his grandson, fifth grader William Petre. Mirror photo by Matt Churella
Cambria Heights officials cut the ribbon on energy and sustainability improvements Thursday outside the elementary school's library. They were joined by local and state officials, as well as members of the fifth grade class. Mirror photo by Matt Churella
Cambria Heights High School student Aubrey Ranck gives an ice cream treat to fifth grader Riley Couturiaux at the conclusion of Thursday's ribbon cutting ceremony. Schneider Electric purchased ice cream for the entire elementary school. Mirror photo by Matt Churella
Cambria Heights fifth grader Michael Hendrix tests his solar racer outside during Thursday's ribbon cutting ceremony where district officials celebrated the completion of major solar energy and sustainability improvements to the elementary school building. Mirror photo by Matt Churella
Cambria Heights School Board President Ken Vescovi speaks at the district's ribbon cutting ceremony Thursday at the Elementary School in Carrolltown with his grandson, fifth grader William Petre. Mirror photo by Matt Churella
Cambria Heights officials cut the ribbon on energy and sustainability improvements Thursday outside the elementary school's library. They were joined by local and state officials, as well as members of the fifth grade class. Mirror photo by Matt Churella
Cambria Heights High School student Aubrey Ranck gives an ice cream treat to fifth grader Riley Couturiaux at the conclusion of Thursday's ribbon cutting ceremony. Schneider Electric purchased ice cream for the entire elementary school. Mirror photo by Matt Churella
Cambria Heights fifth grader Michael Hendrix tests his solar racer outside during Thursday's ribbon cutting ceremony where district officials celebrated the completion of major solar energy and sustainability improvements to the elementary school building. Mirror photo by Matt Churella
CARROLLTOWN — When Cambria Heights Elementary School students returned to school last Thursday, the district’s new rooftop solar array was already running and generating electricity for their classrooms, Superintendent Ken Kerchenske said.
During a ribbon-cutting ceremony Thursday, district officials celebrated the completion of various energy and sustainability projects they’ve pursued over the last two years — including the new rooftop array, replacing the school’s boilers, upgrading heating, ventilation and air conditioning systems and adding energy-efficient LED lighting.
The projects were funded, in part, through nearly $1.5 million in state grants the district received — a $1 million Public School Facility Improvement Grant and a $437,522 Solar for Schools program grant awarded by the Pennsylvania Commonwealth Financing Authority.
Larry Myers, Schneider Electric’s northeast sales team leader, said the rooftop array is estimated to generate about 215,000 kilowatt hours per year. The array is designed to basically offset the electric load that will be generated from having air conditioning in the building, he said.
“We’re actually driving the school to net zero electric use. It’s going to satisfy about 80% to 90% of the electric load,” Myers said, adding the district’s budget should be reduced by about $2.5 million over the next 20 years due to the energy cost savings at the school.
“It’s an efficient use of taxpayer dollars,” Myers said, noting two-thirds of the roughly $6 million was funded through either state grants, federal monies and by offsetting costs.
Kerchenske said the district was able to install the panels in time to get federal tax credits. It’s estimated about 40% of the district’s cost will be reimbursed from the federal government after one year, he said, noting Cambria Heights owns its solar panels.
The district’s payback period — the amount of time it takes to recover the cost of an investment — was initially believed to be about eight or nine years. However, district officials believe it might be as quick as four years now, not only because the solar panels are efficient, but because prices have gone up so much since Cambria Heights started the project, Kerchenske said.
“As energy costs have skyrocketed, this is really going to help us offset those costs,” Kerchenske said.
Cam Willison of Envinity, one of the project’s partners, said the array has already produced enough energy to power a house for an entire year in its first week in operation.
“That is one week of generation, and we’ve got all the weeks in the year and the whole life of the system,” Willison said.
Cambria County Commissioner Tom Chernisky, a 1983 graduate of Cambria Heights, credited Kerchenske and the Cambria Heights school board of directors for preparing the district’s children for a changing world.
“Projects like this provide an opportunity to show students that innovation, conservation and fiscal responsibility go hand in hand,” Chernisky said, adding communities don’t have to be large to think big and find solutions that make sense for the district, its students and taxpayers.
“Cambria Heights School District is helping to make Cambria County a great place to live, work, volunteer, invest and play,” Chernisky said.
Commissioner Keith Rager noted the county is about halfway through pursuing a solar project to offset costs at the Cambria County Prison.
“I don’t think this is a party issue,” Rager said, adding he will always support solar energy projects. “Solar makes sense, and if you can cut your energy costs and keep things down, that’s a good thing for your community.”
Kerchenske thanked state Rep. Jim Rigby, R-Cambria/Somerset, state Rep. Dallas Kephart, R-Cambria/Clearfield, and state Sen. Wayne Langerholc Jr., R-Cambria, for their support in helping the district receive state funding for the projects.
Schneider Electric hosted solar energy activities with the school’s students ahead of the ceremony. (See related story)
Fourth graders built solar-powered bugs, and fifth graders built racer kits in the school’s cafeteria while learning about renewable energy.
Schneider Electric also sponsored ice cream treats for all of the school’s students — grades pre-K through fifth grade.
Mirror Staff Writer Matt Churella is at 814-946-7520.

301 Cayuga Ave., Altoona, PA 16602 – Copyright © Altoona Mirror

source

Zepp Health is getting much more specific about how its upcoming Amazfit T-Rex Dual Solar will work. The latest build of its software reveals that the watch is being built with photovoltaic panels on both the front and the case back. And surprisingly, the panel underneath is the larger of the two.
So now we know what “Dual Solar” actually means. This is not just a solar ring around the display with some clever branding attached to it, Zepp Health is working on two separate surfaces that can generate power.
We first started seeing signs of this setup earlier in the summer, when Zepp Health was working with separate measurements for power coming from the watch face and the case back. At the time, that was interesting but not enough to say with confidence that the T-Rex Dual Solar would actually have photovoltaic panels on both sides.
That ambiguity has now disappeared. The latest APK describes the watch face and case back as both having built-in photovoltaic panels. And each is getting its own charging measurement.
Even more interesting is the difference between them. The photovoltaic panel on the case back is larger than the panel around or beneath the watch face. So Zepp Health says its charging index will normally be higher when both receive the same amount of light.
That is a pretty unusual way to build a solar watch. Most solar wearables concentrate on getting light through or around the display. That’s because the display is obviously the part of the watch exposed to the sun while you are wearing it.
The rear panel creates a slightly different proposition. While the watch is strapped to your wrist, it is not going to see much sunlight at all. The obvious use would be taking the watch off and leaving it somewhere with both surfaces exposed, although exactly how Zepp Health expects owners to position it remains to be seen.
There is also a new way of measuring how much useful light each panel is receiving.
Zepp Health is calling it the Solar Charging Index. The watch can show separate values for the watch-face panel and the case-back panel, rather than simply telling you that solar charging is taking place.
The reference point is 50 kLux. When the watch face receives that level of light, Zepp Health defines its Solar Charging Index as 100%. That does not mean the battery is charging at 100% or that the watch has reached full solar output, it is essentially a reference scale for the amount of usable solar energy reaching the panel.
The larger case-back panel changes the numbers. Under identical lighting, Zepp Health says the back should normally report a higher Charging Index because it has a greater photovoltaic area.
This should make the solar system much more transparent than simply sticking a little sun icon on the screen. Owners should be able to see whether moving the watch, changing its angle or exposing the rear panel is actually improving the amount of energy coming in.
This also adds some context to the Solar Boost Mode we previously uncovered. When activated, the watch cuts back to essentials such as the time, battery level and solar information, while temporarily disabling a number of normal functions.
The idea appears to be simple. Reduce how much power the watch is consuming while those two photovoltaic panels are bringing energy in, giving solar charging a better chance of making a significant difference to the battery level.
Zepp Health is also building temperature protection into the system. Solar charging can pause when the watch becomes too hot, which is particularly relevant if someone leaves it sitting directly in strong sunlight to take advantage of that larger rear panel.
There are dedicated solar statistics as well, including separate information for the two charging surfaces. So this is looking less like the passive solar top-up used on some watches and more like a system Zepp Health wants owners to actively interact with.
The big unanswered question is how much extra battery life all of this actually produces. Two photovoltaic panels sound good on paper, but panel size, efficiency, power consumption and the amount of time the rear of the watch spends exposed to light will ultimately decide whether Dual Solar makes a noticeable difference.
At least the hardware idea is much clearer now. The Amazfit T-Rex Dual Solar has already surfaced under model A2570, and Zepp Health is now putting together a fairly elaborate solar system around it, complete with two panels, individual charging measurements, Solar Boost Mode and temperature management.
Don’t miss the latest from Gadgets & Wearables
Subscribe to our monthly newsletter and check out our YouTube channel.
You can also follow Gadgets & Wearables on Google News and add us as a preferred source in Google Search.

Marko Maslakovic founded Gadgets & Wearables in 2014 after more than 15 years working in the City of London’s financial sector. He has spent more than a decade testing and writing about smartwatches, fitness trackers, sports watches and connected health devices. His reviews are based on hands-on use, including real-world GPS, heart-rate, battery and workout testing. Marko personally tests and writes most of the product reviews published on the site.
Marko Maslakovic has 3245 posts and counting. See all posts by Marko Maslakovic
Your email address will not be published. Required fields are marked *
Comment *
Name
Email
Website


Δ
This site uses Akismet to reduce spam. Learn how your comment data is processed.
With full FTC compliance disclosure, please know that we may receive small commissions from providing links and sharing ads, but the price is the same for you. The team has your best interest at heart, we care as much about your health as you do.
Gadgets & Wearables is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites. As an Amazon Associate we earn from qualifying purchases.
We are independently owned and all opinions expressed here are our own. View our affiliate disclosure.
Acemate tennis robot S10 8.7 Design 8.5/10 Ease of use 9.0/10 Use of information 9.0/10 Value for money 8.0/10 Pros
AFFILIATE DISCLOSURE
Gadgets & Wareables Ltd is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites. As an Amazon Associate we earn from qualifying purchases. View our affiliate disclosure.
Amazon and the Amazon logo are trademarks of Amazon.com, Inc. or its affiliates.

source

485 megawatts of AES solar projects set to come online  San Juan Daily Star
source

Brazil – Salesians have new photovoltaic energy system thanks to donor funding from Salesian Missions  ANS – Agenzia iNfo Salesiana
source

What started as a weed-control problem is becoming a thriving ecosystem where renewable energy and regenerative agriculture graze side by side.
By Ainura Kalau
Published September 3, 2026 · How we edit
A solar farm is, underneath all the technology, just a field that refuses to stop being a field. Grass keeps growing between twenty thousand panels the same way laundry keeps piling up in a house with two small kids in it, not because anyone is careless, but because the job simply doesn’t have an off switch.
For years the industry’s answer was gas mowers and weed trimmers, crews driving between rows of hardware every few weeks to keep the grass from shading the lower edge of a panel. Now, in more states than you’d guess, the answer has four legs and a wool coat.
I don’t have a solar farm or a flock of sheep. What I have is a household that runs on the same basic math: some tasks never disappear, they just need a system that doesn’t burn you out repeating them by hand. So when I came across how many solar operators have quietly swapped mowers for sheep, the appeal wasn’t the novelty of it. It was how unglamorous and obvious the fix turned out to be once someone actually tried it.
Sheep have grazed under solar panels informally for years, but the practice has an actual name now: solar grazing. It works because sheep are short enough to fit under the lowest panel racking without touching the equipment, and because they’d rather eat grass than chew on cables.
Lewis Fox, a sheep farmer who runs the vegetation program at Agrivoltaic Solutions in Vermont, put it plainly to public radio this spring: “We’re in charge of keeping the vegetation within certain limits, and the sheep are the tools that we use to do it.” He also pushed back on the idea that solar and farmland are natural enemies. “It’s often difficult for people to see solar being built on ag land, for various reasons, and I think you could argue the merits either way,” he said. “But we can help bridge the gap in that. What we’re able to do is have solar production coexist with agriculture. And it’s not just window dressing. It’s real agriculture.”
Sheep work under panels in a way cattle or goats generally don’t, for a practical reason. Fox has also described sheep as “pretty short stature, so they can really fit into nooks and crannies,” adding that “they’re also not really interested in chewing on wires or jumping on panels,” which is the entire job in one sentence: get the grass without touching anything electrical.
That framing matters more than it sounds like it should. A lot of the public pushback against utility-scale solar comes from the sense that panels take farmland out of production for good. Solar grazing doesn’t erase that tension entirely, but it does something more useful than arguing about it: it puts working livestock back on the acreage and pays a farmer to be there. Chad Farrell, co-CEO of Encore Renewable Energy, one of the companies that hires grazing flocks for its sites, frames the arrangement as a long-term commitment rather than a marketing line. “At the end of the useful life of the project, we’re actually able to return that land in a better condition than what we found,” he said. That’s a promise that only gets tested decades from now, but it’s a different kind of promise than “we’ll mow it and hope for the best.”
How big is this, really? Bigger than the industry itself had assumed. The American Solar Grazing Association, working with the Department of Energy’s National Renewable Energy Laboratory, ran a census of solar grazing sites and found that as of October 2024, 113,050 sheep were grazing 129,261 acres across 506 solar sites in 30 states. The association’s own summary of the results doesn’t hedge: “The scale of solar grazing in the U.S. is much larger than previously understood and undergoing rapid growth.” For comparison, the industry’s working estimate just a few years earlier had been closer to 15,000 acres, so the real footprint turned out to be nearly nine times bigger than assumed.
Thirty states is not a rounding error. The same census found solar grazing acreage in the South now outpaces the Midwest, Northeast, and West, and that more than 40 percent of the reporting sites were utility-scale installations rather than small community solar arrays. That’s the difference between a regional curiosity and a genuinely national practice, and it happened quietly enough that even people inside the industry were surprised by their own numbers.
Vermont’s own solar-grazing story predates the national count by years. Agrivoltaic Solutions, the operation Fox co-founded with Niko Kochendoerfer, started with a small flock in 2017 as a side project meant to bring in some extra income. Vermont’s Green Energy Times reported that by 2022, partnering with farms in New York, the operation had grown to 1,700 sheep rotating through 24 solar sites across Vermont, New York, and Pennsylvania. That’s not a demonstration flock brought out for a press photo. That’s a working seasonal operation, moving sheep from array to array as the grass in each one grows back, the same way a rancher rotates cattle between pastures.
What strikes me about that number isn’t its size on its own. It’s how ordinary the underlying method actually is. Nobody invented a new breed of sheep or a new kind of solar panel to make this work. Someone noticed that a very old form of land management could solve a very new problem, and then did the unglamorous work of scaling it up, one contract and one flock at a time.
I’m generally suspicious of any story that treats a new idea as a total reinvention of how something works, because in my own experience the reinvention is rarely the part that holds up. What actually works is a plain, repeatable habit done consistently, long after the novelty has worn off. Solar grazing fits that pattern better than almost any other clean-energy story I’ve come across this year. Nobody involved is claiming sheep will transform the electric grid. They’re just quietly doing the same job, panel row after panel row, cheaper and more reliably than a mower crew ever did.
Part of what makes it work, too, is that nobody is trying to solve the whole 129,000-acre problem at once. Each flock handles one site, then moves to the next, then comes back around when the grass has grown in again. Break a genuinely large task into small, checkable moves and it stops feeling impossible, whether the task is vegetation management across 30 states or just getting through a full day of errands with two kids in tow. The scale only looks overwhelming from a distance. Up close, it’s just one paddock at a time.
If nothing else, it’s a decent argument for asking whether the fix you actually need is the exciting one, or just the one nobody bothered to try yet.
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

A group of Cambria Heights fourth graders get their solar-powered bugs activated from artificial light inside the elementary school’s cafeteria Thursday since it was raining outside. Once powered, the bugs vibrated and many of the students’ faces lit up with smiles. Mirror photo by Matt Churella
Cambria Heights senior Mitchell Weiland (standing) helps a group of fourth graders build their solar bugs Thursday in the elementary school's cafeteria. Mirror photo by Matt Churella
Cambria Heights second grade teacher Tasha Paronish helps her students assemble their solar bugs Thursday in her classroom. From left are Wolfgang Howlett-Oliver, John Reasbeck, William Anna and Bryson Warner. Mirror photo by Matt Churella
Cambria Heights Elementary School Principal Eric Nagel visits second grade students in Tasha Paronish's homeroom Thursday as they learned about solar energy and built solar-powered bugs in class. Mirror photo by Matt Churella
Cambria Heights Superintendent Ken Kerchenske speaks with a group of fourth grade students Thursday in the elementary school's cafeteria. The students built solar-powered bugs with the help of volunteers from Schneider Electric. The fifth grade class later built solar racers with the volunteers. Mirror photo by Matt Churella
A group of Cambria Heights fourth graders get their solar-powered bugs activated from artificial light inside the elementary school’s cafeteria Thursday since it was raining outside. Once powered, the bugs vibrated and many of the students’ faces lit up with smiles. Mirror photo by Matt Churella
Cambria Heights senior Mitchell Weiland (standing) helps a group of fourth graders build their solar bugs Thursday in the elementary school's cafeteria. Mirror photo by Matt Churella
Cambria Heights second grade teacher Tasha Paronish helps her students assemble their solar bugs Thursday in her classroom. From left are Wolfgang Howlett-Oliver, John Reasbeck, William Anna and Bryson Warner. Mirror photo by Matt Churella
Cambria Heights Elementary School Principal Eric Nagel visits second grade students in Tasha Paronish's homeroom Thursday as they learned about solar energy and built solar-powered bugs in class. Mirror photo by Matt Churella
Cambria Heights Superintendent Ken Kerchenske speaks with a group of fourth grade students Thursday in the elementary school's cafeteria. The students built solar-powered bugs with the help of volunteers from Schneider Electric. The fifth grade class later built solar racers with the volunteers. Mirror photo by Matt Churella
CARROLLTOWN — Cambria Heights Elementary School students assembled solar-powered bugs and racers Thursday morning as part of an activity with representatives of Schneider Electric, the company that installed solar panels on the school’s rooftop earlier this year.

In the school’s cafeteria, fourth graders built solar bugs with help from the Schneider Electric team and some Cambria Heights seniors who volunteered to help celebrate the district’s renewable energy and sustainability projects with the students.
Since it was raining outside at the time, students took their bugs to Drew Thomas, the district’s maintenance director, who used artificial lighting to power the solar batteries attached to the paper bugs.
One by one, as the bugs would vibrate in their hands, the students’ faces lit up with joy.
Fifth graders also put their knowledge of solar energy to the test as they built solar racers with the Schneider Electric team. As the sun came out later in the day, fifth grader Michael Hendrix was able to play with his racer outside during the district’s ribbon-cutting ceremony, which was held just outside of the school’s library, at the playground area where students enjoy recess.
District officials celebrated the completion of several energy-saving and renewable energy improvements that will reduce the school’s utility costs, improve building comfort and support long-term sustainability goals while providing students with opportunities to learn about clean energy technologies.
“It’s just to emphasize the project that we did here with the solar panels on the school building to get (students) to understand more about solar energy and the benefits of it as a renewable energy source,” said Donna Byrd, Schneider Electric’s regional client coordinator.
Byrd said it’s always fun to watch the students’ reaction to seeing solar energy in action with their bugs or racers.
“When it goes as planned and comes together perfectly like that, it just means a lot to see them learning about the solar measures that we have implemented here with the school system,” she said.
Mitchell Weiland, one of the seniors who volunteered to help with the day’s activities, said it was a cool experience to be a part of.
“They’re learning about new technology that’s coming around, and it’s supposed to be a big thing. So, I think it’s very important for them to get early access to learning about it,” Weiland said.
Principal Eric Nagel said it’s important to show the students how solar power works, so they have a better understanding of how it’s being used to produce energy for the school.
“It’s a great hands-on activity for the kids to understand the importance of what we have here and how it’s going to improve the school,” Nagel said.
Nagel also visited second grade teacher Tasha Paronish’s homeroom, who also made solar bugs and learned about solar energy in class. The students designed their own bugs and then had help from Paronish, who attached solar batteries to the bugs.
Shannon Crooker of the Generation 180 nonprofit organization was there to document the day and celebrate with the students.
According to Crooker, Pennsylvania is among the top 10 states with the most solar power projects, and school districts like Cambria Heights are leading the way.
“They’re really making a difference here,” Crooker said. “We want to celebrate that and lift that up to say, look what’s happening in rural areas just as much as urban areas.”
Mirror Staff Writer Matt Churella is at 814-946-7520.

301 Cayuga Ave., Altoona, PA 16602 – Copyright © Altoona Mirror

source

Jay Select Board sends solar farm moratorium to November ballot  The Maine Monitor
source

By Hou Liqiang | China Daily | Updated: 2026-09-03 20:15
A recent study published in the journal Science on the impact of China’s solar power expansion on bird diversity has triggered a heated discussion. Based on data covering 2,344 county-level areas from 2014 to 2023, the study argues that photovoltaic development was associated with a decline in local bird diversity, mainly because some projects were converting cropland and grassland into developed areas, thereby reducing vegetation.
Although some observers have criticized the report as one that depends heavily on bird-watchers’ records, the study deserves attention. No form of human development leaves nature untouched, and solar power is no exception. Almost every modern infrastructure — be it roads, cities, farms, mines and dams — reshapes ecosystems in one way or another. Therefore, it is not whether solar energy development has any adverse ecological effect, but how that effect should be assessed and reduced in a broader balance sheet.
That is where media coverage has fallen short. Some Western media outlets have framed the study in highly negative terms, with one headline saying solar panels in China are “killing” bird diversity. The wording might be eye-catching, but it risks turning a complex governance issue into a simplistic accusation. The study itself does not conclude that solar power is inherently harmful. Nor does it deny the significant role renewable energy plays in mitigating the threat from climate change that has been looming larger.
After all, climate change itself is triggering extreme weather events and destroying habitats, exacerbating global biodiversity loss. By mitigating climate change, the development of solar energy can play an important role in promoting biodiversity conservation.
A balanced assessment will show that China’s solar panel boom has made a major contribution to clean energy development, lowered the cost of renewable technologies worldwide and helped reduce dependence on fossil fuels.
In fact, in some arid regions, solar projects have even reversed desertification, and enabled the recovery of vegetation and new models of rural development, showing that the climate and ecological benefits of solar energy development far outweigh its costs.
In discussions on social networking sites in China, many, including industry insiders, have raised questions about the study. They noted, for example, that the bird-watchers’ records may reflect uneven observation intensity that deserves professional scrutiny.
That does not mean the concerns raised by the study should be dismissed. On the contrary, green development must be judged not only by installed capacity, but also by planning quality, ecological management and long-term monitoring.
Greater emphasis should be placed on biodiversity assessment in the planning phase, especially in areas with complex habitats, rich bird diversity or important migratory routes. Sensitive habitats and migration corridors should be avoided. More attention should be given to solar projects that do not invade natural spaces, such as rooftop solar, distributed solar and agrivoltaic projects. In large solar farms, ecological restoration should focus on diverse native vegetation rather than replacing complex habitats with single cash crops or monotonous greenery.
The government is already treading cautiously. In late 2024, the Ministry of Ecology and Environment released a draft notification to solicit public opinion on strengthening ecological and environmental protection for land-based wind and photovoltaic power projects.
The document calls for full life-cycle environmental management of such projects while advancing the green and low-carbon energy transition and supporting the country’s climate goals of peaking carbon dioxide emissions before 2030 and realizing carbon neutrality before 2060. It includes stronger policy and planning-level environmental impact analysis, improved project-level environmental impact assessment, enhanced supervision during and after construction, and ecological restoration after decommissioning.
The green transition in China will not happen overnight. And there will be challenges, including how to ensure biodiversity conservation while developing renewable energy. But the transition to renewable energy is not a choice between climate mitigation and biodiversity protection. Done well, it can and should serve both.

source

Victoria’s State Electricity Commission (SEC) has connected the 119MW solar PV plant component of its SEC Renewable Energy Park in Australia to the grid for the first time.
The milestone follows the completion of all 212,296 PV modules across the site in late 2025, and means the solar PV plant has now finished every major construction task, including the tracker system, a 162.5-tonne transformer, a switch room substation and high-voltage infrastructure, cables and earth grid.
Get Premium Subscription
SEC executive general manager of assets Lane Crockett said the step marks one of the most important milestones on any renewable energy project.
“Before this stage, a project must complete thousands of installation, construction, and safety activities, which must then pass inspections, testing, and safety checks to ensure they operate correctly and safely,” Crockett said.
The solar PV plant will now proceed through hot commissioning and hold point testing, a staged process that involves gradually switching on and testing the substations, transformers, underground cables, inverters and other electrical equipment.
The 100MW/200MWh battery energy storage system (BESS) has not yet arrived on site, with the SEC stating that its next step is preparing for the system’s arrival and installation later this year.
Once operational in 2027, Crockett said the project would support the state’s planned retirement of coal assets, including the Yallourn power station, scheduled to close in 2028.
The Horsham project’s ownership history traces back to Victoria’s second Renewable Energy Target auction.
Under its original name, Horsham Solar Farm won VRET2 in October 2022 under ESCO Pacific ownership, one of six successful projects that together added 623MW of solar and 365MW/600MWh of battery storage across the state, backed by AU$1.48 billion (US$0.95 billion) in investment.
At that stage, Horsham was specified at 118.8MW of solar paired with a comparatively modest 50MW/100MWh battery.
The project later changed hands to Swedish developer OX2 before the SEC acquired it in September 2024, doubling the planned battery storage component to its current 100MW/200MWh specification as part of a broader AU$370 million investment.
OX2 has continued as a development partner following the ownership transfer, with construction officially beginning in April 2025, following pre-construction works that started in February.
The SEC’s return to energy generation follows the Victorian government’s revival of the entity, historically the state’s public electricity provider before privatisation in the 1990s, as a vehicle to drive the renewable energy transition while retaining public ownership of the resulting assets.
The Horsham development will operate alongside the SEC’s larger Melbourne Renewable Energy Hub, a 600MW/1.6GWh facility which became operational in December 2025, with the two projects together positioning the SEC to supply approximately 5% of Victoria’s electricity market once it re-enters the retail sector.

source

From Mercom India
ROSI, a France-based solar module recycling company, has secured over €20 million (~$23 million) in funding to support its expansion across Europe.
The funding includes a Series B round along with French and European grants. The company said the capital will support the rollout of its industrial projects, including a planned facility in Teruel, Spain.
New international investors and existing shareholders participated in the Series B round, led by InnoEnergy, CMA CGM, the European Innovation Council, and Spanish family office G3T. Finadvice, a Zurich-based corporate finance advisory firm focused on deep tech, acted as a strategic financial advisor and also participated in the investment round, alongside Swiss and Polish family offices.
The company has also appointed Thierry Galvez as Production Site Director of ROSI Alpes. Galvez assumed the role on April 1, 2026.
“This funding marks an important milestone for ROSI,” said Yun Luo, President and Co-founder of ROSI. “It gives us the means to accelerate our industrial deployment, strengthen our operational execution, and prepare for a new phase of growth in Europe. Our ambition is to build a European-scale industrial platform for circular management and the production of strategic raw materials, transforming end-of-life solar panels into a reliable source of high-purity materials for the European industries of tomorrow.”
The company plans to develop a photovoltaic module recycling facility in Teruel with an annual processing capacity of 10,000 tons. The plant will use an integrated and automated production line designed for large-scale deployment.
According to the company, the facility will process end-of-life solar panels and recover materials, including silver, silicon, copper, aluminum, and glass.
ROSI said the project builds on its first industrial site, ROSI Alpes. The company added that the Spain facility is intended to support a scalable recycling model for photovoltaic modules in Europe and reduce reliance on imported raw materials.
According to Mercom’s Annual and Q4 2025 Solar Funding and M&A report, Global VC and private equity funding in the solar sector in 2025 totaled $3.5 billion across 75 deals, 22% lower than the $4.5 billion raised in 60 deals in 2024. There were eight VC funding deals of $100 million or more in 2025.
In 2025, OnePlanet Solar Recycling (OnePlanet), an advanced materials recovery processor specializing in end-of-life solar modules, secured $7 million in a seed financing round led by Khasma Capital, a low-carbon infrastructure investment firm.

NEWSLETTER

Get our leading market intelligence
report covering solar, storage & the grid.
© 2026 by Mercom Capital Group, LLC
All Rights Reserved.
Terms And Conditions. Privacy Policy

source

Aresearch group led by scientists in Saudi Arabia and Greece has developed a stabilizer-free, seed-assisted growth strategy to produce the pure α-phase of formamidinium lead iodide (α-FAPbI₃) perovskite.
FAPbI₃ is one of the leading absorber candidates for single-junction perovskite solar cells. However, it is unstable under ambient conditions, typically requiring chemical stabilizers that can widen the bandgap and limit its photovoltaic potential.
“This research presents an exciting approach to one of the biggest challenges facing FAPbI₃ perovskite solar cells: stabilizing the highly efficient α-phase without relying on compositional additives that compromise the material’s ideal bandgap,” corresponding author Essa A. Alharbi told pv magazine. “Rather than using conventional α-phase stabilizers such as cesium (Cs), rubidium (Rb), or methylammonium (MA), we introduce a seeded-growth strategy in which α-FAPbI₃ seed crystals are incorporated directly into the precursor solution to guide crystallization.”
Alharbi added that “the work combines experimental characterization with multiscale simulations to reveal the underlying mechanism of seeded growth. This provides a scientific explanation for the exceptional efficiency and long-term stability achieved.”
The study consisted of two parts – experimental work and simulations. In the first, the researchers fabricated control and seed-assisted perovskite solar cells with an n-i-p architecture. For the seed-assisted devices, instead of using conventional chemical stabilizers, they incorporated preformed α-FAPbI₃ seeds directly into the PbI₂ precursor to guide crystallization toward the desired α-phase.
The target devices were then produced through a second deposition step using formamidinium iodide (FAI) and methylammonium chloride (MACl), followed by annealing at 150 C for 20 minutes. The control devices were fabricated under the same conditions but without the seeds.
“The pre-existing α-FAPbI₃ seeds lower the nucleation barrier and direct the growth of the desired photoactive α-phase while suppressing the formation of the photoinactive δ-phase. This results in highly crystalline, compact films with larger grains, fewer defects, lower surface roughness, and significantly reduced non-radiative recombination,” Alharbi said.
“As a result, the devices achieve a power conversion efficiency of 23.51%, compared with 15.5% for conventionally processed control devices, while maintaining 99% of their initial performance after 3,000 hours of continuous operation under ambient conditions and one-sun illumination without encapsulation.”
In the second part of the study, the researchers used multiscale simulations to investigate how the seeds influence crystallization and device performance. Density functional theory (DFT) calculations compared the energetics of α- and δ-phase growth on an existing α-FAPbI₃ seed, while molecular dynamics simulations tracked the dissolution of a 10 nm α-FAPbI₃ seed in the precursor solution and compared it with a seed-free solution.
The researchers also used metadynamics to examine nucleation pathways, while separate optical-electrical simulations assessed the effects of carrier mobility, lifetime, and non-radiative recombination on solar cell performance.
“The multiscale simulations showed that dissolving α-FAPbI₃ seeds retain structural motifs that preferentially promote α-phase nucleation while suppressing the competing δ-phase. This reveals that seeded growth not only improves film morphology but fundamentally alters the crystallization pathway, resulting in fewer defects, reduced non-radiative recombination, negligible hysteresis, and more balanced charge transport,” the academics explained.
The research team now plans to extend the seeded-growth strategy from laboratory-scale devices to large-area perovskite solar modules using scalable sequential deposition processes.
“We will optimize seed concentration, size, and processing conditions to ensure uniform crystallization over large substrates while maintaining high efficiency and long-term operational stability,” Alharbi said.
“The seeded-growth strategy is fully compatible with sequential deposition, making it well suited for large-area manufacturing and commercial-scale perovskite photovoltaics,” he concluded. “The approach also has potential to improve other perovskite optoelectronic devices, including light-emitting diodes, where suppressing defect-assisted recombination is essential for high performance.”
The research, “Stabilizer-free pure α-phase FAPbI3 perovskite through seed-assisted growth yields efficient photovoltaics,” was published in Materials Horizons.
Researchers from Saudi Arabia’s King Abdulaziz City for Science and Technology (KACST), Princess Nourah bint Abdulrahman University and Taibah University; Greece’s Foundation for Research and Technology – Hellas (FORTH), Hellenic Mediterranean University (HMU) and University of Ioannina; and the UK’s University College London (UCL) contributed to the 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].
Comments
Please login to comment
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source

Alberta’s solar recycling program imposes a $14 environmental fee on eligible panels
CanREA says the charge is much higher than the estimated future cost of recycling a panel
The association is also questioning the fee’s timing and its impact on renewable energy costs
Starting October 1, 2026, Canada’s Alberta will launch a new solar panel recycling program aimed at keeping end-of-life panels out of landfills. However, the ‘inflated’ recycling fee under the program has drawn criticism from the renewable energy industry. 
The Canadian province plans to collect solar panels for recycling to recover valuable components such as glass, aluminum, silicon, silver, copper and other metals to be reused in new products while supporting local collection and processing capacity.
It will come at a cost since it plans to impose a CAD 14 environmental fee to each new solar panel supplied, starting next month. For a typical 20-panel residential system for instance, the fee would total CAD 280.
This money will cover future recycling costs to collect, transport and recycle solar panels when they reach the end of their lives. It prevents municipalities and taxpayers from paying the bill decades from now.  
The government specifies that the fee will be imposed on new panels when the program begins, and won’t be imposed retroactively on panels already installed.
Alberta’s Minister of Affordability & Utilities RJ Sigurdson stressed that this will protect taxpayers from future cleanup costs and keep power affordable and sustainable for generations to come.
Administered by the Alberta Recycling Management Authority, the CAD 14 fee will apply to solar panels measuring at least 1 square meter and will cover both crystalline silicon and thin-film technologies used in residential, commercial, industrial and utility-scale projects.
The program follows a solar panel recycling pilot conducted between 2022 and 2025. Alberta said the pilot was designed to examine future solar waste streams and work with industry, recyclers and communities on end-of-life management.
Under the rollout, suppliers are required to report and remit the environmental fee from next month. The program will also begin collecting end-of-life panels from October 1, 2026. Alberta plans to engage stakeholders on panel reuse from 2027 and explore investment to expand local recycling capacity.
However, the government’s CAD 14 environmental fee is too high, according to the Canadian Renewable Energy Association (CanREA) that questions both the size and timing of the fee.
In a July 17, 2026 opinion piece, CanREA Director of Policy for Alberta Radha Rajagopalan noted that recycling a television in Alberta costs CAD 2.75, compared with CAD 14 for a solar panel. She argued that the difference is difficult to justify, particularly because most solar panels installed in recent years are not expected to reach end of life for decades.
The government expects more than 95% of solar panels currently installed in Alberta to reach the end of their lives by 2045, generating 72,700 tons of material.
CanREA said independent research commissioned by the association estimates the present-value cost of recycling a solar module at about CAD 5, after accounting for cost increases over 20 years and potential panel reuse. It therefore considers the CAD 14 charge significantly higher than the estimated cost of recycling.
The association has also raised concerns about the treatment of renewable energy compared with other forms of power generation. CanREA said solar and wind are the only electricity generation technologies required to pay a recycling surcharge in Alberta. This will be on top of existing reclamation security requirements for utility-scale solar and wind projects, which include recycling and disposal costs.
The association calls for greater focus on panel reuse, certification standards and collection infrastructure before imposing what it considers an inflated fee.
“Imposing an inflated, single-industry fee now, decades before most panels need recycling and at nearly three times the estimated cost, is so far off the mark that it will ultimately take away consumer choice and drive up electricity costs,” stated Rajagopalan.
TaiyangNews 2024

source

Ultrasonic Sprayer and Coater: JH’s Precision Ultrasonic Coating Systems for Thin Film and Solar Cell Manufacturing  EIN News
source

Ver en vivo: Conexión Local
Sigue las noticias en Conexión Local

Mainly clear. Low 48F. Winds S at 5 to 10 mph..
Mainly clear. Low 48F. Winds S at 5 to 10 mph.
Updated: September 3, 2026 @ 9:52 pm

NonStop Local Reporter
Northwest Harvest solar project aims to save money for meals
YAKIMA, Wash. — Northwest Harvest said a new solar project at its Yakima distribution center is expected to lower operating costs and send more money back into food support programs.
The statewide hunger relief organization invited media to the site to show the construction and explain how the project ties into its larger food distribution work. In Yakima, the group runs a program that helps students across the county get food they need.
Mike Doonan, the supply chain operations manager, said the student support program had been in place for more than a decade.
“It’s usually a little breakfast, lunch, and dinner. There’s a snack. Just some things that get them over,” he said.
Doonan said the food is meant to help students who may not have enough to eat at home.
“We know that some kids don’t get food at home. If that’s the only place to get food is at school. So this is a supplemental thing that they can use,” he said.
Northwest Harvest partnered with Ellensburg Solar to install more than 300 solar panels at the facility. The organization said the added renewable energy will reduce power use.
Acting CEO Gary Newte said the savings could top $1 million and would be put back into the food distribution system.
“Times are tough, and expenses are up. So any opportunity that we have to save on operational costs and redirect them back into direct services is always a great opportunity,” he said.
Northwest Harvest says it wants the project to stand for more than solar energy. The organization said it also saw the effort as part of building resiliency in the food system.
NonStop Local Reporter
{{description}}
Email notifications are only sent once a day, and only if there are new matching items.
Currently in Kennewick
Your browser is out of date and potentially vulnerable to security risks.
We recommend switching to one of the following browsers:
Get up-to-the-minute news sent straight to your device.

source

A group of Cambria Heights fourth graders get their solar-powered bugs activated from artificial light inside the elementary school’s cafeteria Thursday since it was raining outside. Once powered, the bugs vibrated and many of the students’ faces lit up with smiles. Mirror photo by Matt Churella
Cambria Heights senior Mitchell Weiland (standing) helps a group of fourth graders build their solar bugs Thursday in the elementary school's cafeteria. Mirror photo by Matt Churella
Cambria Heights second grade teacher Tasha Paronish helps her students assemble their solar bugs Thursday in her classroom. From left are Wolfgang Howlett-Oliver, John Reasbeck, William Anna and Bryson Warner. Mirror photo by Matt Churella
Cambria Heights Elementary School Principal Eric Nagel visits second grade students in Tasha Paronish's homeroom Thursday as they learned about solar energy and built solar-powered bugs in class. Mirror photo by Matt Churella
Cambria Heights Superintendent Ken Kerchenske speaks with a group of fourth grade students Thursday in the elementary school's cafeteria. The students built solar-powered bugs with the help of volunteers from Schneider Electric. The fifth grade class later built solar racers with the volunteers. Mirror photo by Matt Churella
A group of Cambria Heights fourth graders get their solar-powered bugs activated from artificial light inside the elementary school’s cafeteria Thursday since it was raining outside. Once powered, the bugs vibrated and many of the students’ faces lit up with smiles. Mirror photo by Matt Churella
Cambria Heights senior Mitchell Weiland (standing) helps a group of fourth graders build their solar bugs Thursday in the elementary school's cafeteria. Mirror photo by Matt Churella
Cambria Heights second grade teacher Tasha Paronish helps her students assemble their solar bugs Thursday in her classroom. From left are Wolfgang Howlett-Oliver, John Reasbeck, William Anna and Bryson Warner. Mirror photo by Matt Churella
Cambria Heights Elementary School Principal Eric Nagel visits second grade students in Tasha Paronish's homeroom Thursday as they learned about solar energy and built solar-powered bugs in class. Mirror photo by Matt Churella
Cambria Heights Superintendent Ken Kerchenske speaks with a group of fourth grade students Thursday in the elementary school's cafeteria. The students built solar-powered bugs with the help of volunteers from Schneider Electric. The fifth grade class later built solar racers with the volunteers. Mirror photo by Matt Churella
CARROLLTOWN — Cambria Heights Elementary School students assembled solar-powered bugs and racers Thursday morning as part of an activity with representatives of Schneider Electric, the company that installed solar panels on the school’s rooftop earlier this year.

In the school’s cafeteria, fourth graders built solar bugs with help from the Schneider Electric team and some Cambria Heights seniors who volunteered to help celebrate the district’s renewable energy and sustainability projects with the students.
Since it was raining outside at the time, students took their bugs to Drew Thomas, the district’s maintenance director, who used artificial lighting to power the solar batteries attached to the paper bugs.
One by one, as the bugs would vibrate in their hands, the students’ faces lit up with joy.
Fifth graders also put their knowledge of solar energy to the test as they built solar racers with the Schneider Electric team. As the sun came out later in the day, fifth grader Michael Hendrix was able to play with his racer outside during the district’s ribbon-cutting ceremony, which was held just outside of the school’s library, at the playground area where students enjoy recess.
District officials celebrated the completion of several energy-saving and renewable energy improvements that will reduce the school’s utility costs, improve building comfort and support long-term sustainability goals while providing students with opportunities to learn about clean energy technologies.
“It’s just to emphasize the project that we did here with the solar panels on the school building to get (students) to understand more about solar energy and the benefits of it as a renewable energy source,” said Donna Byrd, Schneider Electric’s regional client coordinator.
Byrd said it’s always fun to watch the students’ reaction to seeing solar energy in action with their bugs or racers.
“When it goes as planned and comes together perfectly like that, it just means a lot to see them learning about the solar measures that we have implemented here with the school system,” she said.
Mitchell Weiland, one of the seniors who volunteered to help with the day’s activities, said it was a cool experience to be a part of.
“They’re learning about new technology that’s coming around, and it’s supposed to be a big thing. So, I think it’s very important for them to get early access to learning about it,” Weiland said.
Principal Eric Nagel said it’s important to show the students how solar power works, so they have a better understanding of how it’s being used to produce energy for the school.
“It’s a great hands-on activity for the kids to understand the importance of what we have here and how it’s going to improve the school,” Nagel said.
Nagel also visited second grade teacher Tasha Paronish’s homeroom, who also made solar bugs and learned about solar energy in class. The students designed their own bugs and then had help from Paronish, who attached solar batteries to the bugs.
Shannon Crooker of the Generation 180 nonprofit organization was there to document the day and celebrate with the students.
According to Crooker, Pennsylvania is among the top 10 states with the most solar power projects, and school districts like Cambria Heights are leading the way.
“They’re really making a difference here,” Crooker said. “We want to celebrate that and lift that up to say, look what’s happening in rural areas just as much as urban areas.”
Mirror Staff Writer Matt Churella is at 814-946-7520.

301 Cayuga Ave., Altoona, PA 16602 – Copyright © Altoona Mirror

source

Something went wrong
By Sethuraman N R
NEW DELHI, May 20 (Reuters) – A solar industry group has asked India’s power market regulator ‌to increase the cap on electricity prices on power ‌exchanges, saying the current limit is hurting companies and slowing investment, with demand ​at record highs.
• The National Solar Energy Federation of India told the Central Electricity Regulatory Commission that the existing cap of 10 Indian rupees ($0.1032) per unit makes it difficult for some players, including ‌energy storage firms, to ⁠operate profitably.
• The petition comes as India’s power demand has surged to record levels over the ⁠past two days due to heat waves.
• Peak power demand jumped to 260.45 gigawatts on Tuesday, according to the country’s power ministry, ​surpassing Monday’s ​record of 257.37 GW.
• A ​separate market segment created to ‌allow higher prices has not worked well because there are very few buyers willing to purchase power at those levels, the group said in a regulatory filing made public late Tuesday.
• India’s power producers are often forced to sell electricity at low ‌prices during periods of weak demand, ​but cannot make up for those ​losses when demand rises ​because of the price ceiling, the industry body ‌told the regulator.
• Keeping the price ​cap unchanged is ​discouraging investment in areas such as energy storage, which could help manage supply and demand swings in the future, ​the industry body ‌said in the filing.
• The regulator has heard the matter ​and reserved its order.
($1 = 96.9100 Indian rupees)
(Reporting by Sethuraman ​NR; Editing by Sonali Paul)
Sign in to access your portfolio

source

Botley West Solar Farm’s deadline extension follows a lengthy approval process, with the site initially proposed in November 2022.
September 3, 2026
The deadline for a decision on Botley West Solar Farm’s development consent has been extended, as the government has allowed more time to consider the application.
The news comes as part of an Energy Infrastructure Planning Projects statement released by Martin McCluskey, minister for local energy and jobs.
The application for development consent was submitted by Photovolt Development Partners (PVDP) on behalf of SolarFive Ltd, under the Planning Act 2008. The examination of the application closed on 13 November 2025; on 24 March 2026, the government set the decision deadline to 10 September, which has now been extended to 10 November.
McCluskey explained: “I have decided to allow an extension and set a new deadline… for deciding this application. This is to enable my department and other interested parties to consider further information received from the applicant.”
On completion, Botley West could deliver 840MW of renewable power to the national grid. It will connect to a National Grid substation near Farmoor reservoir in Oxfordshire and has an anticipated grid connection date of Autumn 2029.
Related:New South Yorkshire solar and BESS site breaks ground
As of the 2024 proposals, Botley West’s development site covers 1,300 hectares – 890 of which would be covered by installed panels. While this site plan was a reduction from its original proposed size of 1,400 hectares with 1,000 hectares of panels, it is still classed as a nationally significant infrastructure project (NSIP) under the Planning Act 2008.
The site plan reduction was issued in March 2024, when the local council requested revisions to minimise the farm’s scale and impact on Oxfordshire’s environment.
For PVDP, the project is a £800 million investment in both renewable energy and the local area. The site proposals include new foot and cycle paths across the farm, community benefits, and a minimum biodiversity net gain of 70%.
Regarding the site’s future following the application deadline extension, the planning inspectorate stated that the decision was made “without prejudice” to whether the proposed solar farm would be granted or refused development consent.

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

Solar energy: record growth thanks to technology and research  Il Sole 24 ORE
source

We use cookies to improve your experience and for marketing. Read our cookie policy Manage cookies
Search across reports, market insights, and blog stories.
Victoria’s State Electricity Commission (SEC) has achieved a first-time grid connection for the 119MW solar PV array at its SEC Renewable Energy Park in Australia. This achievement comes after the installation of all 212,296 PV modules was completed across the site in late 2025, signifying that all major construction work on the solar plant is now finished, including the tracker system, a 162.5-tonne transformer, a switch room substation, high-voltage infrastructure, cabling, and the earth grid.
Lane Crockett, SEC’s executive general manager of assets, noted that this event is among the most significant milestones for any renewable energy venture. He explained that reaching this point requires completing thousands of installation, construction, and safety tasks, all of which must undergo inspections, testing, and safety verifications to confirm proper and safe operation. The solar facility is now set to move into hot commissioning and hold point testing, a phased procedure that entails progressively energizing and evaluating the substations, transformers, underground cables, inverters, and other electrical components.
The 100MW/200MWh battery energy storage system (BESS) has yet to be delivered to the site, with the SEC indicating that its immediate focus is on preparing for the system’s delivery and installation in the latter part of this year. Once the project becomes operational in 2027, Crockett stated it would aid the state’s planned phase-out of coal generation, which includes the Yallourn power station, slated for closure in 2028.
The Horsham project’s origins can be traced to Victoria’s second Renewable Energy Target auction. Initially named Horsham Solar Farm, it secured a VRET2 contract in October 2022 while under ESCO Pacific’s ownership, being one of six winning projects that collectively contributed 623MW of solar capacity and 365MW/600MWh of battery storage, supported by AU$1.48 billion (US$0.95 billion) in funding. At that time, the project was configured for 118.8MW of solar generation alongside a smaller 50MW/100MWh battery system.
The project subsequently transferred to Swedish developer OX2 before the SEC took ownership in September 2024, at which point the battery storage capacity was doubled to its present 100MW/200MWh configuration, part of a larger AU$370 million investment. OX2 has remained involved as a development partner after the ownership change, with construction formally commencing in April 2025, following preliminary works that began in February.
The SEC’s re-entry into power generation comes after the Victorian government revived the entity, which had historically served as the state’s public electricity provider until its privatisation in the 1990s, positioning it as a tool for advancing the renewable energy transition while keeping assets under public control. The Horsham development is set to function alongside the SEC’s larger Melbourne Renewable Energy Hub, a 600MW/1.6GWh facility that came online in December 2025, with both projects together enabling the SEC to provide roughly 5% of Victoria’s electricity demand when it returns to the retail market.
Making Data-Driven Decisions to Grow Your Business
A Quick Overview of Market Performance
Understanding the Current State of The Market and its Prospects
Finding New Products to Diversify Your Business
Choosing the Best Countries to Establish Your Sustainable Supply Chain
Choosing the Best Countries to Boost Your Export
The Latest Trends and Insights into The Industry
The Largest Import Supplying Countries
The Largest Destinations for Exports
The Largest Producers on The Market and Their Profiles
The Largest Markets And Their Profiles
Instant access. No credit card needed.
Online access to 2M+ reports, dashboards, and tables. Trusted by Fortune 500 teams.
IndexBox, Inc.
2093 Philadelphia Pike #1441
Claymont, DE 19703, USA
Contact us
© 2026 IndexBox, Inc

source

We use cookies to improve your experience and for marketing. Read our cookie policy Manage cookies
Search across reports, market insights, and blog stories.
Pennar Industries has deployed INR 5.8 crore, roughly equivalent to US$700,000, into ZAP91 Solar India Pvt Ltd, according to a report published by pv magazine on May 27, 2026. This entity is a joint venture with Zetwerk dedicated to solar PV module production in Telangana, India.
The capital injection, scheduled as a single payment, will secure Pennar Industries a 45% ownership position in the venture. The transaction is anticipated to finalize by June 30, 2026. ZAP91 Solar India was officially incorporated on December 27, 2024.
Pennar Industries entered into a joint venture agreement with Zetwerk Manufacturing Businesses and additional partners on December 31, 2024, to establish the solar manufacturing unit. Within the joint venture, Zetwerk Manufacturing Businesses controls a 50.1% stake, Pennar Industries holds 45%, and other shareholders possess the remaining 4.9%.
ZAP91 Solar India is currently building a solar module manufacturing plant in Sadashivpet, Telangana. The company noted that the investment will aid in completing the factory, encompassing commissioning and the initiation of commercial production. At present, the facility remains under development and has not yet started commercial output.
The joint venture intends to produce and sell solar PV modules to both domestic and international markets. Pennar Industries offers renewable energy services across the entire value chain, including solar engineering, procurement and construction (EPC) services, along with the production of solar modules and module mounting structures.
Interactive table based on the Store Companies dataset for this report.
Report Scope and Analytical Framing
Concise View of Market Direction
Market Size, Growth and Scenario Framing
Commercial and Technical Scope
How the Market Splits Into Decision-Relevant Buckets
Where Demand Comes From and How It Behaves
Supply Footprint and Value Capture
Trade Flows and External Dependence
Price Formation and Revenue Logic
Who Wins and Why
How the Domestic Market Works
Commercial Entry and Scaling Priorities
Where the Best Expansion Logic Sits
Leading Players and Strategic Archetypes
How the Report Was Built
Major integrated solar manufacturer
India's largest solar module manufacturer
Part of Adani Group, integrated manufacturing
Leading manufacturer, part of Tata Group
Major PV module and cell producer
Historical leader in solar manufacturing
Makes solar cells, modules, encapsulants
Module and cell manufacturer
Solar PV module manufacturer
Solar panel manufacturer and distributor
Manufactures solar modules and inverters
Solar panel manufacturer
Solar panel manufacturer
Solar panel manufacturer
Solar cell and module manufacturer
Major LED lighting products manufacturer
Leading electrical goods co, major LED player
Major manufacturer of LED lights and fixtures
Major player in LED lighting segment
LED lighting manufacturer
Manufactures LED displays and lighting
Indian subsidiary, major LED mfg in India
Manufactures LED lights and fixtures
Major Indian electrical brand, produces LEDs
LED lighting products manufacturer
Manufactures LED bulbs and lighting
Major player in consumer LED lighting
Leading LED lighting solutions provider
Manufactures LED lights under Finolex brand
Wires & cables major, also manufactures LEDs
Instant access. No credit card needed.
Online access to 2M+ reports, dashboards, and tables. Trusted by Fortune 500 teams.
IndexBox, Inc.
2093 Philadelphia Pike #1441
Claymont, DE 19703, USA
Contact us
© 2026 IndexBox, Inc

source

Solex Energy Secures INR 74.77 Crore Solar PV Module Supply Orders  SolarQuarter
source

Sign up for the wires and see archived wires
Browse experts available to comment on breaking news
Request an expert contact, get responses directly to your inbox
Find an expert by topic in a comprehensive database
Perovskite/CIGS tandem solar cell developed by the KIER research team
NLR’s Best Research-Cell Efficiencies Chart (KIER result highlighted)
Sort Images/Video
Newswise — The Photovoltaic Research Department of the Korea Institute of Energy Research (KIER) achieved a certified world-record efficiency of 26.7% for perovskite/CIGS tandem solar cells, opening a new chapter in next-generation thin-film photovoltaics.
 This achievement was officially certified by the Fraunhofer Institute for Solar Energy Systems (ISE) in Germany and listed in the Best Research-Cell Efficiencies Chart published by the US National Laboratory of the Rockies (NLR, formerly NREL).
 The previous world-record efficiency of 26.3%, set a year earlier by a joint research team from Seoul National University and the Korea Institute of Science and Technology (KIST), was surpassed by another Korean research team at KIER, demonstrating the country’s leading role in next-generation thin-film solar cell technology. 
 Silicon solar cells, currently the most widely used solar cell technology, have already reached technological maturity, leaving limited room for further efficiency improvements due to fundamental physical limitations. Against this backdrop, tandem solar cells are emerging as a promising next-generation solution for high-efficiency photovoltaics. This technology uses multiple solar cells with different characteristics stacked in layers to capture a broader range of sunlight wavelengths.
 The perovskite/CIGS tandem solar cells developed by the KIER research team have a perovskite cell at the top and a CIGS cell at the bottom. This unique configuration enables the two cells to absorb different wavelengths of sunlight simultaneously. Since both perovskite and CIGS are well suited for thin-film processing, the technology combines high efficiency, light weight, and flexibility.
 Assembling the two cells may, however, degrade the perovskite light-absorbing layer. In addition, some cell layers may absorb unwanted light, reducing the overall efficiency. To address these challenges, the KIER research team conducted a comprehensive analysis of the root causes of such efficiency losses. As a result, they developed an advanced interfacial layer material and processing technology to mitigate potential damage to the perovskite layer. The team also successfully minimized undesired light absorption and potential photocurrent loss by optimizing the structure of the top transparent electrode and charge transport layer.
 This approach resulted in a laboratory-measured efficiency of 27% and a Fraunhofer ISE-certified efficiency of 26.7%.
 The developed technology is expected to increase electricity generation per unit area and thereby expand the potential applications of photovoltaic power generation. Furthermore, owing to their lightweight and flexible thin-film design, they are promising not only for buildings and automobiles but also as power sources for small satellites and space-based data centers for future space applications, where weight and space constraints are critical.
 Inyoung Jeong, a senior researcher at KIER who led the research, said, “This achievement is significant in that both cell efficiency and stability can be enhanced by minimizing potential interfacial and optical losses during the integration of perovskite and CIGS. The resulting efficiency was also officially certified by a world-renowned institute and recognized as a world-record performance, underscoring Korea’s technological competitiveness.” 
 Going forward, the research team will focus on ensuring that large-area modules achieve the same efficiency as the small-area devices developed in this study. They will collaborate with industry partners interested in mass production and commercialization and pursue technology transfer. In the long term, they will expand the technology to next-generation space solar cells capable of reliable operation in space due to their light weight and high efficiency.

Credit: KOREA INSTITUTE OF ENERGY RESEARCH
Caption: Perovskite/CIGS tandem solar cell developed by the KIER research team
Credit: KOREA INSTITUTE OF ENERGY RESEARCH
Caption: NLR’s Best Research-Cell Efficiencies Chart (KIER result highlighted)
Connecting Research and Experts with Journalists
Unlock Your Access to Newswise Research News including Embargoed News and Expert Pitches
Used only to deliver research news. Unsubscribe anytime.
Journalists use Newswise as a source for research news, experts, ready-to-use content and story ideas. Media relations professionals can connect with reporters and share their organization’s news with a wider audience. Public readers discover the latest research news in science, medicine, social sciences, environment, technology, factchecks and business news from the world’s most credible universities and research organizations. More than 7,000 email wires go to journalists from more than 2,400 media outlets around the globe.
2026 Newswise, Inc

source

Technical advisory firm Intertek CEA has released its Q2 2026 PV Price Forecasting Report, projecting a strategic realignment across global solar manufacturing hubs. While Chinese suppliers push to restore profit margins following extended price compression, module pricing in the United States, India, and other major rest-of-world markets is expected to hold relatively flat through 2027.
Annual global solar installations are forecast to remain constrained in the low-600 GW range in 2026 and 2027, down from roughly 650 GW in 2025. This slowdown is primarily driven by the stagnating domestic Chinese market, reinforced by the phase-out of demand-side subsidies, tighter energy consumption rules, and new efficiency standards, said the report.
Chinese suppliers pivot to margin expansion
Domestic policy in China is accelerating domestic price increases, which are expected to spill over into international markets, said the report. Major Chinese manufacturers are guiding toward reduced export volumes while actively pursuing higher-margin international sales.
According to Intertek CEA’s regional cost modeling, integrated production costs globally show a massive spread. Fully integrated production costs for TOPCon modules in China remain the global floor at under $0.12/W. In Southeast Asia and India, regional manufacturing costs hover near $0.17/W for TOPCon technology.
Meanwhile, unsubsidized all-in U.S. manufacturing costs for TOPCon modules using U.S. cells exceed $0.37/W prior to incentives. However, factoring in Section 45X Advanced Manufacturing Production Credits brings net U.S. TOPCon production costs down to approximately $0.21/W.
The Section 45X subsidies effectively eliminate much of the historical cost penalty for domestic U.S. manufacturing, narrowing the net cost gap between U.S.-made modules and non-Chinese imports from Southeast Asia or India to just $0.01/W to $0.03/W.
Trade policy and policy mandates dictate regional pricing
U.S. module prices are projected to stay elevated as buyers await final clarity on the tariff structures emerging from the ongoing polysilicon Section 232 investigation. While operational cell capacity outside duty-subject nations remains tight, expanding non-duty ingot, wafer, and cell capacity throughout 2026 and 2027 is expected to alleviate acute procurement bottlenecks.
In India, pricing dynamics are increasingly governed by domestic procurement mandates. The Approved List of Models and Manufacturers (ALMM) List-II, which requires domestic module makers to utilize domestic cells for public tenders, is officially in effect.
While Indian module prices are expected to linger near $0.20/W due to grandfathered 2026 projects, developers face near-term cell supply shortages for late-2026 and 2027 deliveries. A secondary cost adjustment is anticipated in 2028 when ALMM List-III mandates the use of domestically produced wafers.
Across all international sea lanes, elevated freight costs continue to compound baseline module pricing, said the report. Logistics disruptions tied to ongoing Middle East conflict and early peak-season surcharges have pushed ocean freight rates above $0.01/W, adding cost pressures to cross-border deliveries through 2027.
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
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid

You have no items in your basket.

source