Designing the Solar Cell the World Can Actually Build
Overview
Almost all solar panels today are made from polysilicon manufactured overwhelmingly in China's Xinjiang province. The US government has documented that this production involves forced labor. Our research focuses on perovskite solar cells — a next-generation photovoltaic technology — that can be made without polysilicon at all.
In our most recent work (Solar Energy, 2026), we combined a machine learning model trained on 26,000 experimental data points with direct laboratory fabrication to optimize a lead-free, bismuth-based perovskite solar cell. The result is a device with a verified power conversion efficiency of 3.73% — achieved without lead, without polysilicon, and with a supply chain that has no exposure to conflict minerals or sanctioned entities.
Technical Highlights
- Active material: Cs₃Bi₂I₉ — bismuth-based, fully lead-free
- ML model: Random Forest trained on 26,440 experimental data points from the Perovskite Database Project; 10 algorithms benchmarked; R² = 0.88 on training, 0.71 on test
- SHAP analysis: identified device architecture (ETL/HTL selection) and quenching method as the primary drivers of cell performance
- Champion device: PCE = 3.73%, VOC = 0.69 V, JSC = 12.22 mA/cm², FF = 43.94%
- Film stability: unchanged color for 100+ days in ambient conditions
Supply Chain & Policy Relevance
Bismuth (Bi): primary sources are Vietnam and Mexico. Cesium (Cs): primarily from Canada. Iodine (I): Chile and Japan. None appear on the USGS Critical Minerals list. None are sourced from Foreign Entities of Concern. A commercial solar technology built on this chemistry would satisfy the IRA's FEOC requirements automatically — unlike current silicon-based panels.
Key Publication
Choudhary, Rumman, Sahriar, Islam, Ahmed*, Efstathiadis. "Machine learning assisted development of lead-free Cs₃Bi₂I₉ perovskite solar cells." Solar Energy 315 (2026) 114743.
doi.org/10.1016/j.solener.2026.114743