Semiconductor fabs are pushing yield detection earlier in the manufacturing flow, a “shift left” strategy that relies on real-time equipment sensor data and AI/ML models. This allows process modules to identify yield excursions in situ, rather than waiting for downstream electrical test. The result is faster advanced process control and quicker corrective action, but it comes with a steep data-engineering cost.
The approach is dramatically increasing data volumes. Suppliers have added hundreds of sensor parameters, sampling rates have risen from about once per second to every tens of milliseconds, and engineers are increasingly working with raw trace data instead of summaries. Sources note data variety has grown at least 10× from a decade ago, and some fabs are generating hundreds of petabytes per year.
That flood creates hard trade-offs. Real-time control requires sub-second processing, but not all data can be captured: equipment makers sometimes withhold data as intellectual property, and fabs may lack the capacity to handle the full stream. Sources agree the trend is pushing fabs to decide between optimizing for context and connectivity or simply managing the sheer volume, with AI/ML inference models helping to turn the firehose of sensor data into faster yield learning and advanced process control.