Scholarlink

<p>Indoor photovoltaic deployment of CsPbI<sub>2</sub>Br perovskite solar cells imposes stringent demands on transport-layer selection, because the microampere-level short-circuit currents under indoor illumination amplify band-offset mismatches that are negligible under AM1.5G. Here we report a machine-learning-augmented SCAPS-1D screening of 15 electron transport layer/hole transport layer (ETL/HTL) combinations (TiO<sub>2</sub>, SnO<sub>2</sub>, ZnO × CuI, Cu<sub>2</sub>O, Spiro-OMeTAD, PTAA, P3HT) under cool- and warm-white light-emitting diodes (LEDs) (900 lux). Parallel gradient-boosted-tree and Transformer-based surrogates trained on 60,000 SCAPS-1D samples achieve <i>R</i><sup>2</sup> &gt; 0.996 and converge on absorber bulk-defect density and shallow-donor density as the dominant device drivers. A simple irradiance-ratio calibration enables spectrum-transferable screening without retraining. Multi-objective Pareto screening across efficiency, defect-tolerance area under the curve (AUC), and process-robustness coefficient of variation (CV) identifies TiO<sub>2</sub>/Cu<sub>2</sub>O as Pareto-optimal, jointly delivering the highest AUC (0.858), low CV (4.2%), and a two-stage-optimized power conversion efficiency (PCE) of 34.07% under cool-white illumination. Across all Pareto candidates the optimization converges to identical absorber-quality targets, indicating absorber crystallinity rather than transport-layer chemistry as the dominant lever for further improvement. Because SCAPS-1D omits Auger recombination and series resistance, these efficiencies should be interpreted as idealized upper-bound estimates intended to prioritize experimental effort. A SCAPS-1D reconstruction of the four Pareto-optimal candidates incorporating literature Auger coefficients (Cn = Cp = 1.5 × 10<sup>−28</sup>&#xa0;cm<sup>6</sup>/s) confirms this: the predicted efficiencies fall by 17.5–20.8 percentage points once Auger recombination is included, to a corrected range of 13.6–16.9%, while TiO<sub>2</sub>/Cu<sub>2</sub>O remains the best-performing candidate. Experimental validation of the predicted transport-layer rankings will be an important direction for future work.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transport-Layer Screening for Indoor CsPbI2Br Perovskite Solar Cells via Machine-Learning-Augmented SCAPS-1D Simulation

  • 期刊论文

  • 发表日期：

    2026年8月4日

查看全文
  • Chih-Hsi Peng,
  • Yi-Cheng Lin

摘要

Indoor photovoltaic deployment of CsPbI2Br perovskite solar cells imposes stringent demands on transport-layer selection, because the microampere-level short-circuit currents under indoor illumination amplify band-offset mismatches that are negligible under AM1.5G. Here we report a machine-learning-augmented SCAPS-1D screening of 15 electron transport layer/hole transport layer (ETL/HTL) combinations (TiO2, SnO2, ZnO × CuI, Cu2O, Spiro-OMeTAD, PTAA, P3HT) under cool- and warm-white light-emitting diodes (LEDs) (900 lux). Parallel gradient-boosted-tree and Transformer-based surrogates trained on 60,000 SCAPS-1D samples achieve R2 > 0.996 and converge on absorber bulk-defect density and shallow-donor density as the dominant device drivers. A simple irradiance-ratio calibration enables spectrum-transferable screening without retraining. Multi-objective Pareto screening across efficiency, defect-tolerance area under the curve (AUC), and process-robustness coefficient of variation (CV) identifies TiO2/Cu2O as Pareto-optimal, jointly delivering the highest AUC (0.858), low CV (4.2%), and a two-stage-optimized power conversion efficiency (PCE) of 34.07% under cool-white illumination. Across all Pareto candidates the optimization converges to identical absorber-quality targets, indicating absorber crystallinity rather than transport-layer chemistry as the dominant lever for further improvement. Because SCAPS-1D omits Auger recombination and series resistance, these efficiencies should be interpreted as idealized upper-bound estimates intended to prioritize experimental effort. A SCAPS-1D reconstruction of the four Pareto-optimal candidates incorporating literature Auger coefficients (Cn = Cp = 1.5 × 10−28 cm6/s) confirms this: the predicted efficiencies fall by 17.5–20.8 percentage points once Auger recombination is included, to a corrected range of 13.6–16.9%, while TiO2/Cu2O remains the best-performing candidate. Experimental validation of the predicted transport-layer rankings will be an important direction for future work.

Graphical Abstract