Manufacturing

Cpk

Q: What is the difference between Cp and Cpk?

Cp measures only the potential capability of a process: the ratio of the specification width to the process spread, computed as (USL minus LSL) divided by 6 times sigma. It assumes the process is perfectly centered between the limits. Cpk adds the effect of centering by taking the smaller of the distances from the process mean to each specification limit, each divided by 3 times sigma. When a process is perfectly centered, Cp equals Cpk. As the mean drifts toward one limit, Cpk falls below Cp while Cp stays unchanged. For example, an RF amplifier gain process with USL of 22 dB, LSL of 18 dB, mean of 20 dB, and sigma of 0.5 dB has Cp = 1.33 and Cpk = 1.33. If the mean drifts to 21 dB, Cp is still 1.33 but Cpk drops to 0.67, predicting far higher fallout. Engineers track both: Cp shows whether the spread is tight enough, and the gap between Cp and Cpk shows how much yield is being lost to poor centering.

Q: What Cpk value corresponds to a six sigma process?

A six sigma process has six standard deviations between the process mean and the nearest specification limit, which corresponds to a short-term Cpk of 2.0. With the conventional 1.5 sigma long-term mean shift, this delivers about 3.4 defects per million opportunities. More common manufacturing targets are lower: a Cpk of 1.0 places the nearest limit three sigma away and yields roughly 1,350 defects per million on one side, a Cpk of 1.33 is the classic four sigma capable threshold at about 32 defects per million, and a Cpk of 1.67 is a five sigma target near 0.3 defects per million. RF foundries and module houses typically require Cpk of at least 1.33 on critical parameters such as gain, noise figure, and output power before a process is released to volume production.

Q: How many samples are needed to estimate Cpk reliably?

Cpk is a point estimate from sampled data, so it carries statistical uncertainty that shrinks as the sample size grows. With only 30 measurements, the 95 percent confidence interval on a reported Cpk of 1.33 can span roughly 1.0 to 1.66, wide enough that the true capability could be marginal. Increasing to 100 samples narrows that interval to about 1.15 to 1.51, and 300 or more samples are typical for a formal process capability study tied to a production release. The data must also come from a process that is in statistical control, confirmed with a control chart, before Cpk is meaningful, because the calculation assumes a stable, approximately normal distribution. For RF parameters that are non normal, such as return loss expressed in dB, a transformation or a non normal capability method should be applied first.

/see-pee-kay/

A process capability index that quantifies how well a stable manufacturing process holds a measured parameter inside its specification limits, while penalizing any offset between the process mean and the center of those limits. Unlike Cp, which sees only the spread, Cpk takes the smaller of the two one-sided distances from the mean to each limit and divides it by three standard deviations. A Cpk of 1.33 corresponds to a four sigma capable process at roughly 32 defects per million, while a value of 2.0 marks a true six sigma process. RF module houses commonly gate gain, noise figure, and output power on a minimum Cpk before releasing a process to volume production.

Category: Manufacturing

Typical Target: ≥ 1.33

Six Sigma: Cpk = 2.0

How Cpk Predicts RF Manufacturing Yield

Cpk is the workhorse metric for judging whether an RF production line can repeatably ship parts that meet a numeric specification. For a parameter like small-signal gain, insertion loss, or return loss, an engineer defines a lower specification limit (LSL) and an upper specification limit (USL). Cpk then compares the distance from the process mean to whichever limit is closer against three times the process standard deviation. Because it always reports the worse of the two one-sided margins, Cpk reflects the side of the distribution most likely to produce a reject, which is exactly the side that drives fallout.

The index is deliberately conservative. A process can have a tight spread yet still fail Cpk if its mean has drifted toward one limit, a common situation when an amplifier bias point or a laser-trim target wanders over a production run. This is what separates Cpk from Cp: Cp measures only potential capability assuming perfect centering, so the gap between Cp and Cpk is a direct, dollar-relevant measure of yield lost to mis-centering. Tracking both lets a process engineer decide whether to tighten the spread or simply recenter the mean.

Cpk is only valid once a process is in statistical control, confirmed on a control chart, and is approximately normally distributed. RF parameters expressed in logarithmic units, such as return loss in dB, are often skewed, so a transform or a non normal capability method may be required before the index can be trusted. Capability is also a sampled estimate, so the reported number should always be paired with a sample size and ideally a confidence interval.

The Cpk Formula and Sigma Level

Process Capability Index:
C_pk = min [ (USL − μ) / (3σ) , (μ − LSL) / (3σ) ]

Potential Capability (for comparison):
C_p = (USL − LSL) / (6σ)

Sigma level from Cpk:
Sigma level ≈ 3 × C_pk

Where USL and LSL are the upper and lower specification limits, μ is the process mean, and σ is the process standard deviation. Example: an amplifier gain process with USL = 22 dB, LSL = 18 dB, μ = 21 dB, σ = 0.5 dB gives C_p = 1.33 but C_pk = (22 − 21) / (3 × 0.5) ≈ 0.67, predicting heavy fallout from poor centering.

Cpk, Sigma Level, and Defects Per Million

Cpk	Sigma Level	Approx. DPMO (one-sided)	Yield	Manufacturing Interpretation
0.67	2σ	22,750	~97.7%	Not capable; redesign or 100% screen
1.00	3σ	1,350	~99.87%	Marginal; tightening required
1.33	4σ	32	~99.997%	Common minimum for release
1.67	5σ	0.3	~99.99997%	High-reliability / space target
2.00	6σ	0.002	~99.9999998%	True six sigma capability

Common Questions

Frequently Asked Questions

What is the difference between Cp and Cpk?

C_p measures only potential capability, the spec width over the process spread, (USL − LSL) / 6σ, and assumes perfect centering. C_pk adds centering by taking the smaller of the two one-sided margins, each over 3σ. They are equal when the process is centered. For a gain process with USL = 22 dB, LSL = 18 dB, μ = 21 dB, σ = 0.5 dB, C_p = 1.33 but C_pk = 0.67, so the gap between them is yield lost to mis-centering.

What Cpk value corresponds to a six sigma process?

A six sigma process places the nearest spec limit six standard deviations from the mean, a short-term C_pk of 2.0, which with the usual 1.5σ long-term shift yields about 3.4 defects per million. More typical targets are C_pk = 1.0 (3σ, ~1,350 DPMO), 1.33 (4σ, ~32 DPMO), and 1.67 (5σ, ~0.3 DPMO). Most RF foundries require at least 1.33 on critical parameters before volume release.

How many samples are needed to estimate Cpk reliably?

C_pk is a sampled estimate with real uncertainty. From 30 measurements, the 95% confidence interval on a reported 1.33 can span roughly 1.0 to 1.66; 100 samples narrows it to about 1.15 to 1.51, and 300 or more is typical for a formal study tied to a release. The process must first be in statistical control and approximately normal; skewed RF data such as return loss in dB needs a transform or a non normal method.

Related Terms

← CPI (Communications & Power Industries) CPML →

Cpk