Non-Invasive Quantification of Corrosion in Cu Interconnects Using Regression Models With S-Parameters

Copper electrodes are crucial in electronic systems but are prone to corrosion, which can degrade performance and lead to system failures and safety risks. Existing fault diagnosis methods are limited in their ability to effectively and non-destructively evaluate corrosion due to constraints in size resolution and the lack of effective indicators for corrosion evolution in electronic packages. This study presents a non-destructive quantification method using regression models with scattering parameters, offering early detection and high diagnostic accuracy. We gathered a comprehensive database of scattering parameter patterns reflecting various levels of corrosion to enhance the reliability of fault detection and diagnosis in electronic devices. To evaluate the feasibility of scattering parameters for non-destructive and quantitative corrosion assessment in interconnects, we conducted a comparative study across ten categories of models: Basic Linear Regression, Regularized Linear Models, Advanced Linear Models, Bayesian Models, Robust Models, Kernel-based Models, Neighbor-based Models, Tree-based Models, Gradient Boosting Models, and Neural Networks. Despite testing various advanced models, the Basic Linear Regression model outperformed all of them, achieving a Mean Absolute Error as low as 0.214% (Evaluated corrosion range: 0 - 100%). This superior performance is attributed to the close and linear relationship between scattering parameters and the degree of corrosion, demonstrating that scattering parameters are effective indicators for corrosion. Additionally, integrating independent scattering parameter channels with a data augmentation method using interpolation further enhanced the diagnostic accuracy of the regression models.