Machine Learning-Assisted Automated Parametric Modeling, Simulation, and Structural Optimization of On-Chip Inductors

Aiming at the challenges of complex structural parameter spaces, excessively time-consuming full-wave electromagnetic simulations, and the tendency of traditional optimization algorithms to fall into local optima in radio frequency integrated circuit (RFIC) on-chip passive inductor design, this paper proposes an automated modeling, surrogate model, and PSO-GA hybrid optimization framework based on full-wave simulation. First, automated parametric modeling and efficient interaction with full-wave solvers are achieved through dedicated scripting interfaces. Second, a high-precision surrogate model is constructed using a physical component decoupling mechanism and Gaussian Process Regression (GPR), with the coefficient of determination (R2) for each physical component exceeding 0.970. Finally, a hybrid optimization strategy integrating Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) is implemented to perform rapid inverse design under target inductance and quality factor constraints. The experimental results show that the error between the inductance and quality factor values predicted by the surrogate model and the target inductance and quality factor values is less than 1%.