A Unified Framework for Epilayer Thickness Inversion Based on Lorentz–Drude Modeling and Particle Swarm Optimization

To address the challenge of accurately inverting epitaxial layer thickness under complex conditions, this paper proposes a comprehensive computational framework that integrates physical modeling with intelligent optimization. Methodologically, a basic model is first established based on optical interference theory, and the Lorentz–Drude composite model is introduced to characterize the frequency-dependent properties of refractive index. This approach corrects the traditional assumption of constant parameters, thereby enhancing the model’s physical consistency and broadening its applicability. At the data processing level, sliding average filtering and differential screening strategies are employed to enhance the stability and noise resistance of spectral data. Furthermore, to address the nonlinear offset issues caused by multi-beam interference, a full-frequency-domain reflectance model is constructed, transforming the thickness estimation problem into a parameter optimization problem. By utilizing a particle swarm optimization algorithm for global search, the method effectively avoids the shortcomings of traditional gradient methods, such as sensitivity to initial conditions and the tendency to get stuck in local optima. Experimental results demonstrate that this method exhibits excellent stability and high accuracy under various incident conditions and across multiple datasets, significantly improving the consistency and reliability of thickness estimates. Overall, the proposed model integrates physical mechanisms with intelligent algorithms, possesses strong general applicability, and can be extended to other scenarios involving the inversion of complex optical parameters and big data modeling, thereby providing an effective analytical and optimization approach for related fields.