Using SDPC for Visual Exploratory Analysis of Semiconductor Production Line Sensor Data

Vast amounts of data are continuously collected through sensors fitted into various pieces of equipment and processes in semiconductor production lines. These integrated datasets often encompass tens of thousands of dimensions, making it challenging to identify complex relationships among data dimensions for diagnosing defects and achieving high yield rates. Parallel Coordinate Plots (PCPs) are effective for visually analyzing multi-dimensional data, but traditional axis reordering methods struggle with superhigh-dimensional datasets. To address these challenges, we propose SDPC, an interactive PCP-based visual analysis system specifically tailored to the unique requirements of semiconductor production lines. SDPC employs a server-client architecture that efficiently visualizes sensor data in real time by dynamically selecting dimensions and down-sampling data based on user interactions. This enables engineers to explore high-dimensional sensor data without noticeable delays, enhancing their ability to identify defects quickly. By integrating user-defined filter conditions and focusing on defect-relevant dimensions, SDPC enhances interpretability and accelerates root cause identification. An evaluation with semiconductor production engineers demonstrated SPDC's ability to facilitate real-time exploratory analysis, boost operational efficiency, reduce visual analysis time by two-thirds for on-site engineers, and ultimately lead to more effective production processes.