Enhanced Control of WLCSP Wafer Dicing through Big Data Automation and Analysis

Wafer dicing is a crucial step in the semiconductor industry, particularly with blade sawing leading to small PPM (Parts Per Million) of chip and crack defects. This issue is intensified in Wafer Level Chip Scale Package (WLCSP) devices as die sizes decrease and thicknesses become thinner. Ensuring quality without compromising device shipment requires meticulous control over the wafer dicing process. While Fault Detection Classification (FDC) is a common method for enhanced control, its high setup costs and capital requirements can be prohibitive. This paper proposes a cost-effective alternative, demonstrating the team's approach of automating machine data logs from the wafer dicing machine for monitoring. Control is achieved through statistical analytical techniques, offering a proactive methodology for problem-solving and process improvement. The paper outlines a Proof of Concept with four main steps: automated machine data log, data pre-processing, algorithm investigation and visualization/package solution for ad-hoc analysis. The resulting dashboard facilitates monitoring of key dicing and blade input /output responses, incorporating statistical analyses such as outlier analysis and comparative methods for mean and variance.