OptoGPT: New AI designs perfect light traps for solar cells in 0.1 seconds

University of Michigan engineers have developed OptoGPT. It is an AI algorithm that leverages the architecture behind ChatGPT to rapidly design optical multilayer film structures.

These structures, composed of thin layers of different materials, have applications in solar cells, telescopes, semiconductor manufacturing, and smart windows.

The study notes that current inverse design methods struggle to adapt to diverse design targets or various structure types.

OptoGPT significantly streamlines the design process, producing designs in a mere 0.1 seconds, making it significantly faster than conventional methods.

One of OptoGPT’s key advantages lies in its ability to generate designs with fewer layers than previous models.

This simplification not only reduces manufacturing complexity but also lowers production costs, making advanced optical technologies more accessible.

Utilization of transformer architecture

“Designing these structures usually requires extensive training and expertise as identifying the best combination of materials, and the thickness of each layer, is not an easy task,” said L. Jay Guo, U-M professor of electrical and computer engineering.

OptoGPT democratizes this process by providing a user-friendly tool that automates the design, making it easier for researchers and engineers to explore new possibilities.

The core of OptoGPT is a transformer architecture, a machine learning framework renowned for its effectiveness in natural language processing.

“Similar to how large language models are able to respond to any text-based question, OptoGPT is trained on a large amount of data and able to respond well to general optical design tasks across the field,” says the media release.

By treating materials and their thicknesses as words and encoding their optical properties as inputs, OptoGPT identifies patterns and relationships between these “words.”

This enables the algorithm to predict the next word, effectively constructing a “phrase” – a design for an optical multilayer film structure – that fulfills the desired optical properties .

“In a sense, we created artificial sentences to fit the existing model structure,” Guo explained.

Accuracy and validation of OptoGPT

To assess OptoGPT’s accuracy, researchers tested it against a dataset of 1,000 known design structures. The results were impressive, with OptoGPT’s designs deviating from the validation set by a mere 2.58%, underscoring its remarkable precision. Further refinement through local optimization improved accuracy by 24%.

The researchers also used statistical techniques to map OptoGPT’s workings. They discovered that materials naturally cluster by type, such as metals and dielectrics, and that all dielectrics converge as their thickness approaches 10 nanometers.

This observation aligns with the behavior of light at such small scales, further validating OptoGPT’s accuracy.

OptoGPT’s flexibility sets it apart from previous inverse design algorithms, which were often tailored for specific tasks. As a general-purpose tool, OptoGPT empowers researchers and engineers to design optical multilayer film structures for a broad spectrum of applications.

This versatility has the potential to accelerate innovation in fields ranging from renewable energy to telecommunications.

Limitations demand collaborative efforts

However, there are certain limitations with OptoGPT. “The current framework still lacks explain ability and does not allow users to directly understand the physical principles involved in its designs,” read the study .

Moreover, its dataset covers only a small fraction of the large canvas of optical multilayer thin film structures. Due to this, OptoGPT may not be able to find a design that lies outside the sampled dataset.

“Close collaboration across multiple research groups is needed to obtain a better model for a more general and better photonic inverse design that expands to more complicated structures,” concluded the research paper.