Nondestructive monitoring of annealing and chemical-mechanical planarization behavior using ellipsometry and deep learning

This study presents a novel nondestructive defect inspection method for through-silicon via (TSV-Cu) structures using Mueller matrix spectroscopic ellipsometry and deep learning. Three typical defects—overdishing (defect-OD), protrusion (defect-P), and defect-free—were identified in 3-µm-diameter and 8-µm-deep Cu filling TSV-Cu structures. Single-channel deep learning using a Mueller matrix element (MME) achieved a 99.94% accuracy rate in distinguishing defect types. A multichannel approach utilizing seven effective MMEs quantified the height variation in the Cu filling with 98.92% accuracy and 1 nm resolution. This method offers a rapid and nondestructive evaluation of annealing and chemical-mechanical planarization (CMP) processes, improving the reliability of high-density integration.

Qimeng Sun, Dekun Yang, Tianjian Liu, Jianhong Liu, Shizhao Wang, Sizhou Hu, Sheng Liu, Yi Song