The semiconductor and other high-tech manufacturing industries have been on a continuous improvement journey for decades, producing ever more powerful devices at giga-factory scales while achieving near-perfect yields. This is partly achieved through almost full automation of the physical manufacturing process, but also by incorporating domain experts across a variety of disciplines into the associated data analysis processes that support manufacturing technology development and production operations. As the scope and scale of semiconductor manufacturing increase, the data analysis processes must become more efficient, combining the power of semi-automated analysis workflow engines, scalable multi-protocol data access platforms, computational notebooks, and advanced visualization tools for use by “citizen data scientists” who possess deep manufacturing domain knowledge without being computer scientists or professional programmers. To this end, this presentation summarizes aspects of the current research on the use of these techniques and tools to analyze an important but under-utilized subset of manufacturing data, namely event sequence data and its associated context.
This topic is important because event and event sequence data is currently an underutilized class of information in semiconductor manufacturing, and other disciplines have developed techniques for capturing, representing, visualizing, and analyzing event data to recognize patterns, identify anomalies, and leverage human expertise to benefit their operations.
The topic is challenging because of the wide variety of event types, the volume and density of events, the need for time synchronization across disparate and distributed data sources, the range of stakeholders and their careabouts, and finally, the period over which analysis insights may be sought.
This presentation highlights the application of several innovative and powerful solution technologies, including (but not limited to): a multi-protocol equipment data collection and application integration platform which can scale seamlessly from an equipment/process development lab setting to a gigafab production environment; an abstraction layer that makes all equipment capabilities available via an open RESTful API; and an analysis workflow application that enables the creation of no code/low code custom dashboards from standard panes, incorporates commercial computational notebooks for more complex experimentation and data visualization, and supports the automated and/or manual creation of data collection plans (DCPs) that can stream data from equipment and other potential sources to repositories linked to the dashboards and notebooks.
The resulting environment provides the power and flexibility needed to effectively implement "human in the loop" analysis systems for the growing community of citizen data scientists in the semiconductor manufacturing industry.
