Intel has embedded agentic AI into the management of critical equipment at its semiconductor fabrication plants, with actual test results showing that the detection speed for equipment anomaly symptoms has increased by 30% compared to before, saving approximately $7.5 million in costs per year for just a single use case.
According to reports, Intel (INTC) has spent the past 8 months deploying multiple groups of AI agents using the agentic AI platform Seeq as an orchestrator, achieving the aforementioned results. The Seeq platform adopts a multi-AI collaborative model, where a single "orchestrator" coordinates all individual AI agents. Once an anomaly signal is detected, the system can automatically complete cause analysis and generate a draft work order, which is ultimately reviewed and confirmed by engineers. The core value lies in significantly eliminating the process where engineers previously had to switch back and forth between multiple systems and manually investigate. The collaboration between the two parties began about 3 years ago, when Intel (INTC) was looking for a self-service analytics platform capable of developing and deploying analytical capabilities at scale. As Intel (INTC) pushes forward with comprehensive AI implementation, the partnership has now expanded into the agentic AI direction.
The deployment targets core utility equipment that supports fab operations, including ultrapure water systems, chillers, specialty gases, and chemical supply units. Although such equipment does not directly participate in wafer manufacturing, a failure in even one of them can easily cause wafer contamination or production line shutdowns. The ultrapure water system in particular does not tolerate even trace amounts of impurities, and quickly capturing anomalies is directly related to yield levels. In the past model, engineers would open multiple interfaces to retrieve data and investigate root causes after an alarm was triggered; now, after the system captures an anomaly signal, it automatically aggregates related data, historical similar cases, and recommended handling plans and pushes them to engineers. Engineers only need to verify the content and approve it to generate a work order.
The benefits go beyond faster anomaly identification. Automatically generated work orders also reduce the workload of technical personnel by 15% to 20%, saving more than 120,000 work hours annually; the energy efficiency of the chiller plant has improved by about 15%. Seeq has been connected to systems at approximately 70 Intel (INTC) sites worldwide, integrating 65 million to 70 million tags (time-series signal data) that can be directly accessed. Currently, about 1,000 Intel (INTC) engineers use the software each week.
Comments