The design of solar cells, telescopes and other optical and photonic components is set to benefit from a new algorithm developed at the University of Michigan. The approach is based on the ChatGPT computer architecture to engineer materials with desired optical properties.

The transformer architecture-based framework of Opto Generative Pretrained Transformer (OptoGPT) generates designs for multilayer film structures within 0.1 seconds. The resulting optimized designs typically include six fewer layers than previous models, which translates into more streamlined manufacturing processes.

OptoGPT treats materials at a certain thickness as words and encodes the associated optical properties as inputs. It seeks out correlations between these terms and predicts the next word to create a “phrase” that achieves the desired property. A large-scale dataset with 10 million samples was used for training.

According to the researchers, OptoGPT can “effectively deal with the non-trivial inverse design problem in multilayer structure … Combined with many proposed techniques, our model can unify the inverse design under different types of input targets under different incident angle/polarization, be versatile to different types of structures, as well as facilitate the fabrication process by providing diversity and flexibility.”

The research is published in Opto-Electronic Advances.