Solar cell, telescope and other optical component manufacturers may be able to design better devices more quickly with AI.
An innovative technology called OptoGPT, created by University of Michigan engineers, utilizes the same computer framework as ChatGPT to develop solutions for optical components.
OptoGPT focuses on designing optical multilayer film structures, which are thin layers of different materials stacked together to enhance specific properties. These structures can improve light absorption in solar cells, optimize reflection in telescopes, enhance semiconductor manufacturing using extreme UV light, and regulate building heat through smart windows that adjust transparency based on temperature.
OptoGPT can generate designs for these multilayer structures nearly instantaneously, within 0.1 seconds. Moreover, its designs typically require six fewer layers than previous models, making them more feasible for production.
Professor L. Jay Guo from the University of Michigan highlighted the traditional challenges in designing these structures and emphasized the expertise needed to identify the right materials and layer thicknesses.
To streamline the design process, the research team adapted a transformer architecture, similar to those used in language models like ChatGPT and Google’s Bard.
The model compares materials to words and their optical properties to inputs, predicting the next “word” or material thickness to create effective multilayer film structures with desired properties.
In testing, OptoGPT’s designs closely matched a validation dataset of known structures, demonstrating high accuracy compared to existing models.
Researchers can further enhance accuracy through local optimization, adjusting variables to achieve optimal outcomes. Fine-tuning improved accuracy by 24%, reducing discrepancies with the validation dataset.
By examining the associations identified by OptoGPT, the researchers revealed clusters of materials based on type and thickness, validating the model’s effectiveness in design accuracy.
As an inverse design algorithm, OptoGPT offers greater flexibility than previous approaches, enabling the creation of various optical multilayer film structures for different applications.
This research received partial funding from the National Science Foundation.
Additional contributors to the study include Taigao Ma and Haozhu Wang from the University of Michigan. Professor L. Jay Guo holds multiple faculty positions at the university.
