Artificial intelligence-based optimization for high-performance optoelectronic nanowires solar cells

We present an advanced reinforcement learning (RL)-driven optimization approach employing neural networks to boost the optical and electrical performance of nanowire (NW) silicon solar cells (SCs). The proposed assembly consists of 2 × 2 periodic cylindrical (C) NWs with different diameters which disrupt structure mirror's symmetry. Our results show that the RL methodology successfully improves the optoelectronic performance of the suggested design. In this study, the RL-optimized asymmetric array design increased average absorption to 0.65-an improvement of 12% over the symmetric C-NWs configuration. The photo-generated current density (J_ph) is elevated by 7%, reaching 13.6 mA/cm², while reducing the filling area by nearly 50%. Incorporation of a bottom Si substrate and back metal further boosted the J_ph and ultimate optical efficiency to 19.3 mA/cm² and 21.9%, respectively. After taking into consideration recombination effects, an extensive electrical simulation revealed an optical-to-electrical conversion efficiency (PCE) of 6%, which is 13.2% higher than that of a traditional C-NWs array. This work highlights the potential of RL-based design methodologies in improving NW architectures in addition to providing a path towards affordable and highly efficient silicon NW SC technologies.