Test Limits and Guard Bands: Setting an Acceptance Window

A test limit that is set equal to the design specification will fail good units. The measurement has its own uncertainty and the process has its own spread, and both must be accounted for before a limit is written into a test program.

What the Limit Has to Separate

A test limit separates units that will work in the field from units that will not. The two populations overlap, and the limit is placed where the cost of a false failure balances the cost of an escape.

Where the cost of a field failure is high, the limit is set conservatively, which produces more false failures. Where the cost of a false failure is high, the limit is set loosely, which produces more escapes.

The decision is therefore economic as well as technical, and it should be made explicitly rather than by taking the specification value as the limit.

Measurement Uncertainty

Every measurement has an uncertainty, which is the range within which the true value lies. The uncertainty comes from the instrument, the fixture, the environment and the operator.

Where the uncertainty is comparable to the tolerance, the limit cannot distinguish a good unit from a marginal one. The measurement system analysis is the tool that quantifies this, and it should be performed before the limit is used. Our coupon notes describe the reference structures used to assess the measurement.

Where the uncertainty is too large, the answer is a better measurement rather than a wider limit. A limit that is wider than the specification accepts units the design did not intend.

Guard Banding

A guard band is the amount by which the limit is moved inside the specification to account for the uncertainty. A unit that measures within the guard band is accepted, and one that measures outside is rejected.

The guard band is normally set to the uncertainty, so that a unit that is accepted is almost certainly within the specification. The price is that units within the guard band that are actually good are also rejected.

Where the process capability is high, the guard band costs nothing because no unit approaches the limit. Where the capability is marginal, the guard band turns a yield problem into a visible one. Our quality notes describe how the capability is reported.

Process Capability and the Limit

The process capability index expresses the tolerance in units of the process spread. A high index means the process is well inside the limits and the test rarely rejects a good unit.

A low index means the distribution touches the limits, and the test rejects a proportion of good units along with the bad. The remedy is a process improvement or a wider specification, not a looser test.

The capability should be measured on the production process, not on a prototype sample. A capable prototype process can become marginal when the equipment or the supplier changes.

False Calls and Their Cost

A false call consumes an operator, a repair slot and a re-test. Where the false call rate is high, the operators begin to accept marginal results without investigation, and the test loses its value.

The false call rate should be measured, and the causes should be addressed: a marginal measurement, a limit that is too tight, or an inspection program that cannot see the feature reliably.

Where the false call rate cannot be reduced, the limit should be revisited. A test that is trusted is worth more than one that is theoretically more sensitive.

Different Limits for Different Purposes

A limit used for process monitoring can be different from a limit used for shipping. A monitoring limit detects drift earlier and can be tighter without rejecting units.

Where a unit is measured at several points in the process, the limits at each point can differ. A tight limit at an intermediate step catches the drift before the value is fixed by a later process.

The limits should be documented with their basis, so that a later change is a decision rather than an adjustment. Our design release checklist notes where the limits are recorded.

When a Limit Is Changed

A limit should be changed when the capability improves, when the measurement improves or when the cost of an escape changes. It should not be changed to make a yield number look better.

The change should be supported by data showing the effect on the escape rate, which requires the field return data or an accelerated test. Without it, the change is a guess.

The version of the test program and its limits should be recorded with the lot, so that a unit can be traced to the limits that applied when it was tested. Our solderability notes describe the same principle applied to a material test.

Process Control and Verification

On a design of this kind, test limit is the item that decides how the rest of the board is arranged. Reviewing the design before the data is released is cheaper than correcting it after the panel is in the tank, because every step downstream inherits the decision made at the front end. Documenting the assumption is part of the design work, and a short note on the drawing prevents a question that would otherwise arrive a day later and cost a day of schedule.

The process window is set by the narrowest step in the flow, so an improvement anywhere else shows up as margin rather than as yield until that step is addressed. A short note on the drawing about handling, storage or packaging is often worth more than an extra decimal place on a tolerance.

Where a value sits close to a process limit, the drawing should say so, since the shop can then open the process window rather than working to a nominal figure that carries no tolerance.

Process Control and Verification

On a design of this kind, test limit is the item that decides how the rest of the board is arranged. Reviewing the design before the data is released is cheaper than correcting it after the panel is in the tank, because every step downstream inherits the decision made at the front end. Documenting the assumption is part of the design work, and a short note on the drawing prevents a question that would otherwise arrive a day later and cost a day of schedule.

The process window is set by the narrowest step in the flow, so an improvement anywhere else shows up as margin rather than as yield until that step is addressed.

Process Control and Verification

On a design of this kind, test limit is the item that decides how the rest of the board is arranged. Reviewing the design before the data is released is cheaper than correcting it after the panel is in the tank, because every step downstream inherits the decision made at the front end. Documenting the assumption is part of the design work, and a short note on the drawing prevents a question that would otherwise arrive a day later and cost a day of schedule.

FAQ

Should the limit equal the specification? No. It should be inside it by the measurement uncertainty at least, unless the measurement is very much better than the tolerance requires.

What if the guard band rejects too many units? That is a capability problem, and the answer is to improve the process or widen the specification with justification.

What does gopcb provide for test limits? We provide measurement system analysis for the gauges used, capability studies on the production process, guard band recommendations with their basis, and records that tie each lot to the test program version and its limits.