AI Accelerated Computational Design of The Freeform Solar Cell Structures

Artificial intelligence generated configurations (AIGC) have emerged as a game-changing approach for inverse design, demonstrating remarkable potential in the realm of free-form device structures. This innovation is especially valuable in the context of thin-film solar cell (SC) devices, where patterned meta-surfaces offer the potential to enhance both their electrical and optical properties.<br/><br/>Traditional human-designed periodic meta-surfaces, while effective, are inherently constrained by topological limitations and a limited set of parameters. The advent of free-form design liberates us from these constraints, allowing for the creation of intricate and unconventional shapes, as well as the exploration of a vastly expanded design landscape. This newfound design freedom holds the promise of unlocking unprecedented functional capabilities, potentially surpassing human intuition. Nevertheless, the efficient exploration of this expansive design space presents a significant computational challenge.<br/><br/>One of the primary obstacles to realizing high-throughput free-form designs is the computational cost associated with conventional numerical simulations. In our study, we tackle this challenge head-on by introducing a fully automated system that harnesses the power of high-speed Deep Learning (DL) surrogate solver alongside intelligent configuration optimizer. Compared to standard numerical methods, our DL surrogate model accelerates the prediction of outcomes by 22,700 times, resulting in a 98.47% reduction in computation costs, all while maintaining an average accuracy of approximately 99%.<br/><br/>Our extensive evaluation, encompassing a dataset of 600,000 configurations generated by the intelligent configuration optimizer, has led to the discovery of an optimized SC device design boasting a power conversion efficiency of 17.58%. This performance greatly surpasses the 14.70% baseline efficiency of SC device lacking patterns on the substrate layer. These findings underscore the immense potential of AIGC techniques in efficiently enhancing the performance of photovoltaic devices, paving the way for a brighter future in renewable energy applications.