Machine learning assisted optimization of surface roughness on silicon

In recent years, semiconductor thinning and IC probing have become crucial for detecting and preventing vulnerabilities in electronic devices. Optimizing surface quality is essential for reliable analysis but can be time-consuming due to numerous experimental parameters. Machine learning can address these challenges by analyzing complex data patterns. We developed an automatic acquisition process using a confocal profilometer to gather several hundred ablation images on silicon and steel samples. Regression algorithms and neural networks predict ablation depth, surface roughness (Sa), and large-scale surface patterns. Preliminary results show a strong correlation between predicted and actual values, validating the model's accuracy and generalizability. Future work will focus on streamlining data acquisition to enable dynamic optimizations in semiconductor processes.