Predicting VCSEL Emission Properties Using Transformer Neural Networks

TL;DR

This paper introduces a transformer neural network model tailored for predicting VCSEL emission properties, achieving high accuracy and faster training than traditional methods, with potential applications across physics fields.

Contribution

The paper adapts transformer neural networks for physics-based predictions, specifically for VCSEL emission characteristics, demonstrating improved accuracy and training speed over traditional neural networks.

Findings

High accuracy in predicting VCSEL parameters

Faster training compared to traditional neural networks

Transformer architecture applicable to other physics problems

Abstract

This study presents an innovative approach to predicting VCSEL emission characteristics using transformer neural networks. We demonstrate how to modify the transformer neural network for applications in physics. Our model achieved high accuracy in predicting parameters such as VCSEL's eigenenergy, quality factor, and threshold material gain, based on the laser's structure. This model trains faster and predicts more accurately compared to traditional neural networks. The transformer architecture we propose is also suitable for applications in other fields. A demo version is available for testing at https://abelonovskii.github.io/opto-transformer/.