Process Capability

Process capability analysis is the statistical evaluation of whether a manufacturing process can consistently produce output within its specification limits, using a specific family of indices — Cp, Cpk, Pp, and Ppk — to convert raw process variation data into a single quantitative answer about how reliably that process will actually meet spec in production. These indices are the mathematical foundation underneath process qualification decisions throughout semiconductor manufacturing: a process doesn't get judged ready for volume production on intuition or a few good-looking sample runs, it gets judged ready based on what its capability indices actually say about long-run defect probability.

Cp measures a process's potential capability by comparing the width of the specification window to the width of the process's natural variation, deliberately ignoring where the process distribution actually sits relative to those specification limits — which is exactly why a process can show an excellent Cp while still producing real defects, if that otherwise well-contained distribution happens to sit shifted off-center. This is precisely the gap Cp alone can't reveal: a narrow, tightly controlled process distribution that's drifted toward one specification limit will show the same high Cp it would show if perfectly centered, because Cp by design only evaluates spread, never position.

Cpk closes that exact gap by incorporating both spread and centering simultaneously, taking whichever specification limit the process distribution sits closest to and computing capability against that nearer limit specifically — which is exactly why Cpk, not Cp, is the index that actually predicts real-world defect rate, since it's sensitive to precisely the centering problem Cp was blind to. A process showing Cp of 2.0 but Cpk of only 0.8 is telling a very specific, actionable story: the underlying process variation is genuinely tight and would support excellent capability if centered correctly, but something — a miscalibrated target, a systematic bias in the process — has shifted that otherwise-excellent distribution dangerously close to one specification limit.

Pp and Ppk extend this same spread-versus-centering logic from short-term, within-subgroup variation to the full long-term variation observed across many separate production lots or many separate time periods, which is exactly why Cpk and Ppk can diverge meaningfully even for the identical nominal process. Cpk and Cp are calculated from within-subgroup variation — essentially, how consistent the process is within a single short run — while Pp and Ppk are calculated from the total variation observed across the complete dataset, including whatever lot-to-lot or day-to-day shifts occurred between subgroups; when Cpk equals Ppk, the process shows no meaningful between-subgroup shift, but when Cpk is noticeably higher than Ppk, that gap itself is a direct, quantitative signal that the process is drifting meaningfully between subgroups even though each individual subgroup looks tightly controlled in isolation.

The sigma-level-to-defect-rate conversion table — Cpk of 1.0 corresponding to roughly 2,700 defective parts per million, Cpk of 1.33 to roughly 63 DPPM, Cpk of 2.0 to a small handful of defects per billion — is what actually makes capability indices operationally meaningful rather than purely abstract statistics, translating a dimensionless index directly into the defect rate a customer or downstream process step should actually expect to see. This conversion is exactly why specific Cpk thresholds — commonly 1.33 as a baseline production-readiness bar, and 1.67 or higher for safety-critical or extremely high-reliability applications — get written directly into process qualification criteria and customer quality agreements: the index itself is just the quantitative mechanism, but the defect rate it predicts is the thing that actually matters to the business.

Process capability analysis functions as the quantitative foundation underneath equipment, process, and product qualification decisions precisely because it converts the qualitative question "is this process good enough" into a specific, defensible number tied to a specific, predictable defect rate — exactly the kind of objective evidence a qualification decision needs rather than relying on subjective engineering judgment alone. This is exactly why capability indices appear so consistently across qualification runbooks, customer quality agreements, and ongoing process-monitoring programs throughout semiconductor manufacturing: they provide the one statistical language that process engineers, quality teams, and customers can all agree on to answer the same underlying question — will this specific process reliably produce acceptable product once it's running at volume.