LEAP: Closed-Loop AI Framework Boosts Perovskite Solar Cell Efficiency to 21.32%

Overview

A research team proposed LEAP: A closed-loop framework for perovskite precursor additive discovery, combining a domain-specific LLM with Bayesian optimization in an expert-in-the-loop workflow. The resulting perovskite solar cells reached a champion power conversion efficiency of 21.32%, up from a 19.25% control baseline.

| Attribute | Detail | | :--- | :--- | | Paper title | LEAP: A closed-loop framework for perovskite precursor additive discovery | | Authors | Xin-De Wang, Zhi-Rui Chen, Ze-Feng Gao, Peng-Jie Guo, Cheng Mu, Zhong-Yi Lu | | arXiv ID | 2605.20242 | | Submitted | May 18, 2026 | | Categories | cs.LG; cond-mat.mtrl-sci; cs.AI; physics.chem-ph | | Keywords | perovskite solar cells, AI materials discovery, Bayesian optimization, expert-in-the-loop, closed-loop framework, active learning |

Key points

Why additives matter

• Perovskite thin films are cheap to process (solution methods at room temperature) but contain defects—grain boundaries, vacancies, impurities—that trap photogenerated electrons.
• Additive engineering: tiny amounts of organic molecules passivate defects and improve charge transport. But chemical space contains tens of millions of candidates; intuition-driven trial and error is slow and cannot be parallelized.

How LEAP works

1. AI recommendation: a domain-specific LLM trained on perovskite-additive literature proposes candidates with human-readable mechanistic justifications (e.g., "strong electron-withdrawing group, likely passivates positively charged iodine vacancies"). Benchmarks show it outperforms general-purpose LLMs on mechanistic-consistency reasoning. 2. Bayesian uncertainty: an optimization model prioritizes molecules that are untested but potentially promising—well suited to expensive, small-data experiments. 3. Expert filtering: human chemists reject roughly half the AI suggestions (hard synthesis, costly precursors, unstable structures). 4. Real experiments: synthesis, solution prep, spin-coating, electrode evaporation, and current–voltage measurements on actual devices. 5. Data feedback: results update the models, and the loop repeats over three rounds.

Results

| Round | Control efficiency | Best additive | Average efficiency | Champion efficiency | | :--- | :--- | :--- | :--- | :--- | | Control | 19.25% | — | 19.25% | — | | Round 2 | — | 6-CDQ | 20.13% | — | | Round 3 | — | 2-CNA | 20.87% | 21.32% |

• Absolute gain of 2.07 percentage points (~10.7% relative); neither additive had previously been tried as a perovskite additive.

Stated limitations

• Only a tiny fraction of chemical space was explored; 6-CDQ and 2-CNA may not be global optima.
• 21.32% is a champion device, not an average—the 20.87% average is the more reproducible figure.
• Synthesis costs of the additives were not reported.
• Expert-in-the-loop filtering reduces false positives but may discard unconventional yet effective candidates (false negatives are unquantifiable within the framework).
• Three lab rounds prove concept within one lab and device architecture; generalization across labs, compositions, and environments remains unverified.

Broader significance

• The formula—AI extracts mechanistic knowledge from literature → generates candidates → experts filter → experiments validate → data feeds back—is transferable to catalyst design, drug discovery, and alloy development.
• LEAP demonstrates that expert-in-the-loop, closed-loop AI discovery works in the physical world, not just in simulation: AI does not replace chemists, but decides what to try next far more accurately than intuition alone.

*Tags: AI, MaterialsScience, Perovskite, Solar, ActiveLearning, BayesianOptimization*