Accelerating discovery of high-efficiency donor-acceptor pairs in organic photovoltaics via SolarPCE-Net guided screening

Abstract

Organic photovoltaic (OPV) materials hold great potential in accelerating solar energy conversion. Rapid screening of high-performance donor-acceptor (D-A) materials helps reduce the cost and time consumption associated with traditional experimental trial-and-error methods. However, for predicting the power conversion efficiency (PCE) of D-A in OPV, the existing approaches focus on efficiency prediction of single-component material and neglect synergistic D-A coupling effects critical to device performance. Here, we propose the Solar Power Conversion Efficiency Network (SolarPCE-Net), a novel deep learning-based framework for OPV material screening that captures the intricate dynamics within D-A pairs. By integrating a residual network with self-attention mechanism, SolarPCE-Net employs a dual-channel architecture to process molecular descriptor signatures of D-A while quantifying interfacial coupling effects through attention-weighted feature fusion. We apply the proposed method to the HOPV15 dataset. Experimental results show that our proposed SolarPCE-Net exhibits certain advantages in terms of accuracy and generalization ability compared to traditional methods. Interpretability analysis by attention weighting reveals key molecular descriptors that influence performance. Our work screens undeveloped D-A combinations, demonstrating its potential to accelerate high-performance OPV material discovery.