XRL-QNet: an explainable reinforcement learning framework for optimizing and evaluating quantum dots fabrication

Abstract The development of quantum photonics relies on the availability of high-quality, on-demand single-photon sources. Most appropriate are semiconductor quantum dots (QDs) with tunable emission spectra and discrete energy levels. Due to the spontaneous occurrence of growth being based on an intrinsic randomness in their spectral and spatial properties, their practical implementation is challenging. Scalability and efficiency have been compromised by the existing state of QD measurement techniques, which rely heavily on manual spectrum analysis and lack real-time control over fabrication. explainable reinforcement learning quantum network (XRL-QNet), a machine learning-based platform, is proposed here that leverages the current state-of-the-art reinforcement learning algorithm, proximal policy optimization (PPO), to optimize substrate temperature, material flux, and growth time, among other fabrication parameters, in real-time applications like molecular beam epitaxy for semiconductor device manufacturing. Latent spectral features are obtained through the use of a CNN autoencoder and subsequently scored using a neural regression model that provides a confidence level and appropriateness score. The majority of determining QD assessment spectral features are obtained via explainable AI explainable artificial intelligence techniques, such as SHAP and LIME, to provide interpretability and transparency. High-performance QDs tailored to specific quantum applications can be synthesized using convergence-time, improvement in quantum yield and real-time adjustment of fabrication parameters during the synthesis process methods, as XRL-QNet with PPO optimization closes the loop between fabrication process control and emission-spectrum analysis.