By the time a yield report shows a problem, the lots that caused it went through the tool days ago.
Semiconductor cleanrooms are specified by ISO class, and the numbers are extreme: ISO Class 1 to 3, sub-Class 1 in the most critical zones. Holding them takes multi-stage HEPA and ULPA filtration (U15 is at least 99.9995% efficient at MPPS), chemical filtration for AMC, micro-environments at ±2% RH, vibration isolation and ultra-high-purity utilities. Any of these can drift without tripping an alarm.
The design question is whether anyone sees the drift before the wafers do. That puts the facility monitoring system in the same design as the HVAC and the utilities:
- Measure at the tool, not only in the room: particles and humidity where the wafer is exposed.
- Trend UHP water and gas quality, and power quality on supplies up to 22 kV, on the same system.
- Keep everything on one time base, so a humidity step, a pressure dip and a lot's process time can be laid side by side.
- Set alert limits inside the action limits, so a trend gets looked at while it is still a trend.
When yield moves, the first question is what changed at the tool, and when. A monitoring system designed with the fab can answer that in minutes.
Humiley engineers semiconductor cleanrooms from HVAC through UHP utilities, power and FMS/SCADA as one design.
When yield drifts, which data do you pull first?
Industries


