PCB Factory Process Control: SPC, Limits and Reaction Plans

What should a PCB factory control plan actually specify?

PCB factory process control connects a measurable requirement to a repeatable check and a defined response when something changes. A machine recipe or a final pass label is not enough. The plan needs to say what is measured, where and when it is measured, which limit applies, who responds and what happens to potentially affected boards.

Statistical process control, or SPC, helps distinguish normal variation from signals that deserve investigation. It is one part of that plan, not a replacement for product acceptance. Start with one critical characteristic, such as finished thickness, and make its control record complete before adding more charts.

This is a template to complete with the manufacturer, not an assertion that every PCB uses the same instrument, sampling frequency or process limit. Relate the plan to the actual board construction and the consequences of a missed defect.

How do control limits differ from specification limits?

Specification limits define acceptable product; control limits help detect a change in process behavior. A board can meet its drawing requirement while its process shows a warning signal. A process can also be stable yet consistently too variable or off-center to meet the requirement.

The NIST explanation of variables control charts separates these two purposes. Do not turn drawing tolerances into control-chart limits simply because both are upper and lower numbers.

For illustration, imagine a thickness requirement of 1.50 to 1.70 mm and an established individual-value chart with an upper control limit of 1.66 mm. A reading of 1.68 mm is within the stated product range but above that chart limit. The right response is to investigate and follow the reaction plan, not to claim that nothing changed. These numbers are invented for teaching, not a PCBtry tolerance specification.

Chart rules can also flag non-random patterns while points remain inside the limits, as described by NIST’s control-chart overview. Select rules in advance. Adding more rules increases sensitivity but can also increase false alarms; it does not automatically improve decisions.

What do Cp and Cpk tell you about a PCB process?

Cp compares the available tolerance width with process spread. Cpk also reflects how close the process mean is to the nearer specification boundary. Neither number replaces a time-ordered stability review or proves every board is acceptable.

The NIST capability reference gives the conventional two-sided formulas. For their usual interpretation, establish a stable process, adequate representative data and a suitable distribution model. Define how sigma is estimated; mixing overall variation with within-process estimates can change the meaning of a reported index.

Cp = (USL - LSL) / (6 × sigma)
Cpk = min[(USL - mean) / (3 × sigma), (mean - LSL) / (3 × sigma)]

Worked teaching example: assume an appropriate study estimates mean thickness as 1.62 mm and sigma as 0.02 mm, with LSL = 1.50 mm and USL = 1.70 mm.

• Cp = (1.70 – 1.50) / (6 × 0.02) = 1.67.
• Upper-side index = (1.70 – 1.62) / 0.06 = 1.33.
• Lower-side index = (1.62 – 1.50) / 0.06 = 2.00.
• Cpk is the smaller result: approximately 1.33.

The example shows why a favorable Cp can conceal off-center operation. It is not a measured factory result or a universal pass threshold. Agree capability targets with the customer and identify the characteristic, technology, sample period and estimation method behind the number. A factory-wide “Cpk” without those details is not enough to evaluate your particular board.

How should measurements and sampling be organized?

Organize data around the variation you need to detect. Record location and time, verify the measurement method, and avoid pooling unlike processes into a convenient average.

NIST’s gauge R&R guidance addresses measurement errors in production. A calibration label alone does not show that operator technique, fixturing or resolution is adequate for the specific tolerance. For example, if contact force or measurement location changes the thickness reading, standardize those conditions before interpreting a small shift as a manufacturing change.

1. Name the population. Identify the board family, construction, line and operation. Do not combine unrelated stackups or recipes and call the result one controlled process.
2. Choose meaningful locations. If panel-position variation matters, preserve position identifiers rather than averaging center and edge readings into one unlabeled result.
3. Define the time and subgroup logic. Decide which observations belong together and what changes should appear between groups. Use a chart suitable for that sampling design; an average chart and an individual-reading chart do not use interchangeable limits.
4. Retain raw observations. Keep readings, units, rejected measurements and reasons for exclusions. Do not retain only a rounded mean and a green status icon.
5. Review assumptions and baseline changes. Check whether dependence, mixed populations or a changed method undermine the analysis. A new baseline requires documented review, not an automatic reset after a bad result.

A published Uyemura microetch case study demonstrates why data properties matter: different bath measurements did not support the same statistical treatment. Its lesson is to check the model against the data, not to copy its chemical limits or sampling counts into another line.

Small prototype runs need an honest alternative. There may not be enough representative history to establish reliable capability for a new construction. Agree first-article evidence, enhanced checks or a qualification plan instead of presenting a handful of measurements as a mature Cpk study.

What happens when the chart signals a problem?

A signal should activate a written reaction plan that protects product and guides investigation. It is neither automatic proof of a defective board nor permission to continue without review.

NIST describes an out-of-control action plan as the route for investigating assignable causes. For a PCB order, make the operational decisions explicit before a signal occurs:

1. Verify the observation. Check identity, units, transcription and the measurement setup while preserving the original reading. Do not repeatedly measure until one result passes.
2. Contain the potentially affected scope. Follow the approved hold or stop criteria. Use time, panels, lots and process records to identify material requiring review, including downstream inventory.
3. Investigate and authorize correction. Look for a change in equipment, material, chemistry, tooling or method. Avoid changing several settings at once without a controlled investigation.
4. Verify recovery and disposition. Check the corrected process and separately decide what happens to held boards. Restarting a machine does not automatically release earlier product.
5. Close the record. Record the cause, evidence, action, release authority and any revised instruction. Preserve the link to affected lots.

Keep this evidence alongside the agreed PCB testing process. Process monitoring and final acceptance answer different questions. If boards will proceed to PCB assembly, resolve open fabrication holds before treating them as released incoming material.

To define evidence for your order, send PCBtry the fabrication drawing, stackup, critical characteristics and requested reports. Ask which controls and records are included, which require additional agreement, and how exceptions will be handled. The useful outcome is a clear control and acceptance plan, not an unsupported claim that every parameter is statistically monitored.

Frequently asked questions

Is SPC the same as final inspection?

No. SPC monitors process behavior over time. Final inspection and testing evaluate product against the required criteria. Both may be needed for different reasons.

Does an in-control process guarantee conforming PCBs?

No. A repeatable process can still be off target or too variable for the specified tolerance. Review stability and conformance separately.

Is Cpk of 1.33 always acceptable?

No universal PCB acceptance threshold follows from that number. The required target, characteristic, analysis method and evidence must be agreed for the application.

Can one good panel prove capability?

No. It may provide useful first-article information, but it cannot represent variation across the production conditions and time that a capability claim concerns.

Should a factory chart every machine setting?

Not automatically. Select measurements that inform a meaningful decision and have a defined response. More charts do not compensate for weak measurement or unclear ownership.

Can operators reset control limits after a signal?

Not simply to make the signal disappear. Baseline changes need technical justification and approval under the quality system, with the previous evidence retained.

What if data from several PCB types are mixed?

First determine whether those types share a justified statistical population. If not, separate the data or use an appropriate method rather than interpreting the pooled average as one process.

What should an RFQ request besides a capability number?

Specify critical characteristics, acceptance limits, measurement and reporting needs, traceability, and the intended response to deviations. Ask for the context behind any capability claim.