Semiconductor manufacturing has two distinct measurement traditions, and people outside the industry routinely confuse them.
SEMI E10 — Specification for Definition and Measurement of Equipment Reliability, Availability and Maintainability — defines the six basic equipment states (productive, standby, engineering, scheduled downtime, unscheduled downtime, non-scheduled) and the metrics derived from them: MTBF, MTTR, availability, utilisation. SEMI E79 builds on it to define Overall Equipment Efficiency (OEE) for semiconductor equipment: the availability × performance × quality product, with the semiconductor-specific definitions that make one fab's OEE comparable with another's.
Neither is a statistical process control standard. They are about the tool, not the process running on it. But they meet SPC in an important place: every one of those metrics is a time series, and every one of them is routinely reported as a monthly number compared with last month — which is exactly the mistake control charts exist to prevent. Charting availability or MTBF on an individuals chart tells you whether a tool's reliability actually changed or whether you are explaining noise to a manufacturing manager.
Process control in the fab, meanwhile, follows the general framework of ISO 7870 and a large body of practice that is mostly not written down as a standard at all.
Where fab practice goes beyond textbook SPC
This is the honest part of the page: semiconductor manufacturing outgrew univariate Shewhart charts decades ago, and any vendor telling a fab otherwise has not been in one.
- Multivariate SPC. A plasma chamber has dozens of correlated sensors. Watching each on its own chart produces a false alarm every few minutes and still misses the failure mode that only appears as a change in the correlation between them. Hotelling's T² and PCA-based monitoring are the standard answer.
- FDC — fault detection and classification. Sensor traces sampled at high rate, summarised into features per wafer or per step, modelled, and used to interdict a tool before it processes scrap.
- Run-to-run control. Feedback and feedforward recipe adjustment between lots — an EWMA controller, most often. Note what this does to your chart: a controlled process is supposed to move, and applying ordinary control limits to the controlled variable will flag the controller doing its job.
- Sampling strategy. Not every wafer is measured. What you sample, and whether the sample is representative of a lot with a systematic within-lot pattern, determines what the chart can possibly see.
- Spatial analysis. A wafer is not a set of independent measurements; it has a geometry, and radial or azimuthal signatures carry most of the diagnostic information.
Where a univariate SPC tool still earns its place
Not in a 300 mm production fab's process module. Where it does:
- R&D and pilot lines before the enterprise system is configured.
- Compound semi, MEMS, photonics, packaging and specialty houses, where full yield-management stacks were never affordable and the real incumbent is a spreadsheet.
- Equipment, materials and parts suppliers into fabs, who must demonstrate control of their own processes to a customer who takes SPC extremely seriously.
- Equipment metrics themselves — charting availability, MTBF or OEE as a time series with frozen limits, which is a genuinely useful and widely skipped application of E10/E79 data.
What this platform does and does not do
Does: subgrouped charts with the subgroup you choose (including across sites on a wafer, which is the single most instructive fab subgrouping), limits frozen to a baseline and versioned, the eight Nelson rules with per-rule severity, alerting, an API for automated ingest, and a read-only wall display.
Does not: multivariate anything, FDC, run-to-run control, wafer maps, spatial statistics, or millions of points a day. See SPC for semiconductor manufacturing for a worked chart where the range chart carries the signal and the mean shows nothing.
The SEMI standards themselves are available from SEMI.