Intelligent Design of Silicon-based Spiral Inductor for High-Speed Electro-Optic Modulators: A Transformer-Genetic Algorithm Approach

With the exponential growth of data traffic demanding 100 Gbps+ optical interconnects, silicon-based electro-optic modulators face critical bandwidth limitations due to parasitic effects in driver circuits. Traditional design relies on time-consuming electromagnetic simulations and empirical tuning, leading to suboptimal solutions and long development cycles. This paper presents a novel intelligent design framework that integrates a Transformer-based deep learning predictor with a multi-objective genetic algorithm, achieving a 177% bandwidth improvement (from 26 GHz to 72 GHz) in silicon Mach-Zehnder modulators. We construct a high-fidelity dataset of over 10,000 inductor samples using Advanced Design System (ADS) simulations. A Transformer-based neural network is trained to predict key performance metrics with average error below 5%, outperforming traditional physical models and other neural architectures. This surrogate model enables rapid design space exploration within an improved genetic algorithm, generating Pareto-optimal inductor geometries. System-level simulations under 100 Gbps NRZ modulation demonstrate 50% larger eye opening and one-order reduction in bit error rate. The proposed "data-driven modeling + intelligent optimization" framework offers an efficient, accurate, and scalable solution for high-performance passive component design in silicon photonics, with potential extension to other micro/nano photonic devices.