Test Coverage and Yield Balance

Test coverage and yield balance describe the same tension from two directions. Coverage is what the test programme can detect; balance is whether the effort spent detecting it matches the risk of missing it. A line with heavy inspection and light process control will report good outgoing quality and still ship escapes, while a line with excellent process control and no independent verification has no way to prove it. The practical task is to place each test step where it earns its cost.

Two Numbers That Must Move Together

Coverage without balance produces cost. If every board passes through three inspection steps that all look for the same faults, the third one adds almost nothing while consuming floor space, labour and cycle time. Balance without coverage produces risk: a strategy built entirely on process control depends on the control charts being honest, and the moment a parameter drifts unnoticed the escape leaves the building without ever being screened.

Both numbers need a common reference, and the fault spectrum is that reference. It lists what can go wrong for this product, how often each fault has occurred and how likely it is to escape each test step. With that table in hand, coverage decisions become arguments about frequency and detection rather than about preferences, and the cost of a step can be weighed against the value of the escapes it prevents.

Mapping the Fault Spectrum

The fault spectrum should be built from data rather than from imagination: field returns, internal scrap, rework records and the defect codes collected at each inspection point. Where the product is new, the spectrum can be borrowed from a similar assembly and then corrected as real data arrives. The list is usually short, with a handful of mechanisms accounting for most of the loss, and those are the ones the test strategy has to address.

For each mechanism, the spectrum should record where it is created and where it can first be detected. A missing component is created at placement and detectable at optical inspection, at X-ray or at electrical test, at increasing cost and decreasing convenience. Writing those options down makes it obvious that some faults are best caught close to the step that created them, while others are only visible once the assembly is complete.

What Each Test Step Covers

Optical inspection sees presence, position and solder appearance, and it is fast and inexpensive per board. X-ray sees joints that optical inspection cannot, particularly under area array packages, and costs more per board and much more in equipment. In-circuit test measures nets and values and can localise a fault to a specific component. Functional test proves the product works, which is the only step that catches design and firmware interactions. Inspection choice is therefore about which faults each step can see.

Flying probe sits between in-circuit test and functional test in cost and coverage, and it needs no fixture, which makes it useful for low volume and for prototypes. Flying probe versus fixture test is a comparison of volume, coverage and change cost rather than of accuracy. Bed-of-nails fixtures remain the fastest way to test a mature, high-volume product, provided the design offers the test points and the product lives long enough to amortise the fixture development.

Cost of Test per Board

The cost of a test step is not just the machine time. It includes the amortised equipment, the fixture, the programming and maintenance effort, the floor space, the operator time and the cost of the false failures that the step generates. A step with a two percent false failure rate consumes repair capacity and can cost more than the escapes it prevents, particularly when the failures are intermittent and the board passes on retest.

Expressing all of this as a cost per board makes the comparison honest and lets the test strategy be reviewed on the same basis as any other expenditure. It also exposes steps that exist for historical reasons. Most mature lines contain at least one test that was added to solve a defect that was subsequently fixed at source, and that test now costs money every board without earning anything.

Balance and Escape Rate

Escape rate is the ultimate measure of whether the balance is right. It counts the defects that reach the customer, and it should be reviewed alongside the coverage data, because a step that is nominally in place but not detecting anything may have quietly stopped working. A rising escape rate with stable internal defect rates points at detection rather than at creation, and the investigation should begin with the test steps the fault passed through.

Detection should also be verified deliberately. Injecting a known defect, or building a board with a deliberate fault, confirms that the step catches what it claims to catch. This practice is common in safety-related industries and useful everywhere, because a program that has been edited many times often no longer tests what its documentation asserts, and nobody discovers that until a customer does.

Adjusting Coverage Over Time

Coverage should be reviewed whenever the product changes, the defect spectrum shifts or a test step is added or removed. A change in component package may remove an optical inspection capability and add an X-ray requirement; fixing a placement defect at source may make a downstream step redundant. Reviewing the strategy on a schedule prevents the accumulation of steps that no longer correspond to any risk.

Reducing coverage is harder than adding it, because the decision has to be defended. The defence is the data: a step that has detected nothing for six months, on a process that is statistically controlled and whose fault mechanism has been eliminated, can be moved to audit frequency or removed. Documenting that reasoning means the decision can be revisited if the escape rate moves, and it keeps the strategy honest rather than merely frugal. A reduction made without records is indistinguishable from neglect when the escapes return.

Reviewing the Test Strategy

A useful review has three inputs: the fault spectrum, the cost per board of each step and the escape rate over the last year. It produces three outputs: steps to add, steps to reduce and steps whose capability needs to be verified. Keeping the review short and repeating it annually is more effective than a comprehensive study that happens once and is never updated.

The strategy should also be written down in one place, with each step, its fault coverage, its cost and the reason it exists. That document is what allows a new engineer to understand the line without reverse engineering it, and it is what makes a change in document control meaningful when a test step is added or removed. Without it, the test floor becomes a museum of past problems.

FAQ

Is more test coverage always better? No. Coverage that duplicates another step adds cost and cycle time without reducing escapes. Coverage should map to a fault the process can actually produce.

How often should the escape rate be reviewed? Monthly with the defect data, and in detail whenever a customer return cannot be explained by the internal records.

Should every test step have a documented reason? Yes, with the fault it targets. A step whose reason has been forgotten cannot be defended, reduced or improved.