Inverse Design of Photonic Crystal Surface Emitting Lasers is a Sequence Modeling Problem

TL;DR

Problem: Designing PCSELs is challenging due to coupled physics and high domain expertise requirements. Approach: frame inverse design as a sequential decision-making problem and introduce PiT, a Transformer-based model that uses offline trajectories and return conditioning to predict actions. Contributions: formalizes PCSEL inverse design as sequence modeling, provides an offline data pipeline with roughly 16k trajectories, demonstrates data-efficient performance surpassing a behavior cloning baseline and prior methods, and analyzes the impact of dataset quality and Transformer choices. Impact: enables faster, data-efficient PCSEL laser design and potentially generalizes to other photonic devices.

Abstract

Photonic Crystal Surface Emitting Lasers (PCSEL)'s inverse design demands expert knowledge in physics, materials science, and quantum mechanics which is prohibitively labor-intensive. Advanced AI technologies, especially reinforcement learning (RL), have emerged as a powerful tool to augment and accelerate this inverse design process. By modeling the inverse design of PCSEL as a sequential decision-making problem, RL approaches can construct a satisfactory PCSEL structure from scratch. However, the data inefficiency resulting from online interactions with precise and expensive simulation environments impedes the broader applicability of RL approaches. Recently, sequential models, especially the Transformer architecture, have exhibited compelling performance in sequential decision-making problems due to their simplicity and scalability to large language models. In this paper, we introduce a novel framework named PCSEL Inverse Design Transformer (PiT) that abstracts the inverse design of PCSEL as a sequence modeling problem. The central part of our PiT is a Transformer-based structure that leverages the past trajectories and current states to predict the current actions. Compared with the traditional RL approaches, PiT can output the optimal actions and achieve target PCSEL designs by leveraging offline data and conditioning on the desired return. Results demonstrate that PiT achieves superior performance and data efficiency compared to baselines.

Figures (4)

• Figure 1: PCSEL structure schematic, with the active layer and PhC cavity layer shown. Based on the band-edge mode, a large area lasing resonance mode is formed within the PhC resonance cavity. Lasing arises from the evanescent coupled MQW gain medium in the active layer. The inverse design problem therefore focuses on optimizing the designs of the active layer and the PhC cavity as a whole. The size of the device is about 200 microns in side lengths and 400 microns in height and solely manufactured with semiconductor materials.
• Figure 2: The architecture of PiT.
• Figure 3: The frequency of the total return of the trajectories used to train PiT.
• Figure 4: Training loss curve of both BC and PiT under the whole and rising dataset.