Abstract:
Intel has introduced intelligent AI into the core equipment management of semiconductor factories (wafer fabs). Actual test results show that the detection speed of equipment abnormality signs is increased by 30% compared with before, and a single scenario alone can save approximately US$7.5 million in costs per year. According to reports, Intel has used the agent AI platform Seeq as a coordinator in the past eight months to deploy multiple groups of AI agents and achieved the above results.
The Seeq platform adopts a multi-AI division of labor and collaboration model, with a "coordinator" uniformly scheduling each AI agent. Once an abnormal signal is detected, the system can automatically complete the cause analysis and generate a first draft of work instructions, which will ultimately be reviewed and approved by engineers. The core value lies in greatly eliminating the previous process of engineers switching back and forth between multiple systems and manually troubleshooting.
The cooperation between the two parties began about three years ago, when Intel was looking for a self-service analytics platform that could develop and deploy analytics capabilities on a large scale. As Intel comprehensively promotes the implementation of AI, cooperation has now expanded to the direction of intelligent AI.
The implementation targets are ultrapure water equipment, chillers, special gases, chemical supply devices and other core public engineering equipment that support the operation of the wafer fab. Although this type of equipment is not directly involved in wafer manufacturing, as long as one of them fails, it can easily cause wafer contamination or production line shutdown. Especially in ultrapure water systems, even trace amounts of impurities are not allowed. Quickly catching abnormalities is directly related to the yield level.
In the past, after an alarm was triggered, engineers would open multiple interfaces to retrieve data and investigate the root cause; now, after the system captures abnormal signals, it will automatically summarize related data, historical similar cases, recommended solutions, and push them to engineers. Engineers only need to verify the content and approve it to generate a work order.
The benefits are more than just speeding up anomaly identification. The automatic generation of work orders also reduces the workload of technicians by 15% to 20%, saving more than 120,000 man-hours every year; the energy efficiency of the cold water plant increases by about 15%.
Seeq has been connected to the systems of about 70 Intel factories around the world, integrating 65 million to 70 million tags (time series signal data), and can be called directly. About 1,000 Intel engineers currently use the software every week.
Comments