Deep Generative Models Regulate Ultralow-Pt Catalyst Layer in Fuel Cells

Reducing the precious Pt loading in catalyst layers (CLs) while maintaining fuel cell performance is critical for cost reduction, necessitating the development of an optimized CL microstructure, but conventional empirical expertise-dependent trial-and-error optimization faces severe efficiency limitations when dealing with multiple transfers coupled with electrochemical reactions inside the multi-component hierarchical porous architecture. Here, we present Generative Artificial Intelligence for Controllable Electrode Synthesis (GAI4CES), a framework that enables rapid CL microstructure synthesis, achieving nearly 500x speedup for constructing a representative CL microstructure of 128 nm³ (in 0.36 s compared to 168 s required by traditional numerical methods). It can tailor key property constraints, such as Pt loading and electrochemical surface area (ECSA), to generate statistically representative microstructures satisfying target specifications. This capability enables an exhaustive search for the optimal CL structural design through large-scale geometry‑constrained synthesis coupled with fast performance prediction. The optimized CL microstructure achieves an improvement of 11.6% in ECSA and a reduction of 32.9% in average ionomer thickness surrounding the Pt/C at ultralow Pt loading (e.g., 0.05 mg cm⁻²), and contributes to a maximum 70.2% voltage increase at 3 A cm⁻², indicating that the GAI4CES is powerful in promoting the development of advanced low-cost and high-performance fuel cells.