UCLA

The rapid advancement of modern technology is driven largely by improvements in semiconductor devices, which require manufacturing techniques capable of producing increasingly complex nanoscale three-dimensional structures with high precision, high throughput, and reasonable cost. Atomic layer deposition (ALD) and atomic layer etching (ALE) have emerged as critical techniques in advanced semiconductor manufacturing because their self-limiting surface reactions enable atomic-scale control of material deposition and removal, resulting in highly uniform thin films and precise surface modification. To better understand and optimize these processes, this dissertation develops multiscale modeling frameworks that integrate reactor-scale transport phenomena with surface reaction mechanisms to analyze atomic layer processes and identify optimal operating conditions, reactor geometries, and control strategies that improve precursor utilization and overall manufacturing efficiency. First, a multiscale simulation framework is developed for the Al2O3 ALE process to investigate process mechanisms and optimize operating conditions and reactor configurations. The framework is then extended to study temperature management and multiple control strategies that enhance process stability, efficiency, and robustness under disturbances by run-to-run control, model predictive control, and machine learning-based endpoint detections. Next, a multiscale model is developed for a novel area-selective ALD process for SiO2, where internally integrated etching steps enable selective growth on SiO2 surfaces while suppressing deposition on Al2O3 regions. This bottom-up selective deposition strategy provides a potential solution to lithography alignment challenges and offers guidance for optimizing etching durations to maintain high selectivity while maximizing process efficiency. In addition, this dissertation investigates machine learning-based soft sensing methods for semiconductor manufacturing processes using large-scale industrial data from Seagate Technology. A systematic framework for data cleaning, data aggregation, and ML modeling is developed. The proposed soft sensing models are applied to multiple semiconductor manufacturing processes, including plasma dry etching and slider milling processes, demonstrating their ability to predict wafer pass/fail outcomes and reduce reliance on costly offline metrology.