Session 3.2 - Applying “Ready-to-Use” Machine Learning to Improve Production & Yield for Semiconductor Fabrication

Room 20 Thursday, July 11
10:25am to 10:50am

Semiconductor manufacturing requires high precision. When precision is off, yield is reduced and expensive raw materials are wasted. Production throughput, a function of equipment availability, is equally as important as precision. Maintenance of high-precision equipment impacts equipment availability. Every day lost to maintenance of high-precision equipment can cost hundreds of thousands of dollars in unfilled orders. 
The challenge of maintaining high-precision equipment is in predicting when to calibrate the fabrication tools and schedule maintenance Periodically-scheduled maintenance is non optimal as the need for calibration does not follow a strict time-based cadence. A predictive operations maintenance approach, however, can trigger calibration maintenance exactly when it is needed, thus reducing or eliminating unexpected downtime. Predicting calibration maintenance needs optimizes maintenance activities and yields substantial operational and process benefits over the lifetime of the high-precision equipment. 
Algorithms for continuously monitoring equipment to predict unexpected downtime and schedule predicted maintenance; however, have traditionally required significant effort by data scientists and cost to develop by software development teams. A less expensive and shorter time-to-value approach is to empower operations teams with “ready-to-use” predictive operations machine learning technology. Providing “a data scientist in a box” to subject matter experts who know and understand the equipment, as well as who have direct ownership over its performance, enables rapid predictive model development and ownership of the monitoring technology which remains in the hands of the operations teams who best understand the processes and equipment. 
This talk will discuss how semiconductor customers are taking this approach to predict the need for fabrication tool calibration, before a loss of production, by using ready-to-use operational machine learning. By using a predictive operations learning system to automatically detect patterns in operating tools and recognize early warning conditions of calibration problems, customers are now able to schedule calibration maintenance in advance of production losses while optimizing production uptime and yield.
Share page with AddThis