Advanced deep learning-based strategy for optical inversion engineering of optical coatings

Optical inversion engineering is crucial for the precise manufacturing of optical coatings. We present a fast-analytical model to generate a set of simulated datasets for training the deep learning model. Subsequently, a deep learning strategy based on the transformer framework for inversing errors in the manufacturing of optical coatings is proposed. After several rounds of training, the model achieves a spectral difference of less than 1% between the inverse spectrum and the measured spectrum from an actual deposition process, with each computation completed in just tens of milliseconds. This level of spectral accuracy, combined with the rapid computation speed, highlights the model's exceptional capability to precisely and efficiently inverse thickness and refractive index errors for actual production.