ICT Test Coverage: What the Fixture Actually Proves

Test coverage is the fraction of the circuit that a test can actually reach and measure, and it is fixed long before the fixture is built. It is decided by the layout, by the test points that were placed and by the nets that were left without access.

The number matters because it bounds what any fixture can do. A well built fixture on a board with poor access produces a clean report and a real escape rate, and the reports look identical.

What Coverage Means

Coverage is usually expressed in two ways: the share of nets with a usable test point, and the share of the fault spectrum that the test can detect. The first is a layout property and the second is a measurement property, and a board can score well on one and poorly on the other. A board that reports high access and low fault coverage is usually one where the program measures too coarse a quantity, such as a resistance that cannot reveal a wrong value.

The fault spectrum includes opens, shorts, wrong values, missing parts and out of tolerance parts. A fixture with access to every net detects opens and shorts well and detects a wrong value only if the measurement is accurate enough to distinguish it. Fault detection also depends on the tolerance band loaded into the program, and a band that is too wide hides a genuine drift while a band that is too narrow produces false calls.

Node Access and the Ceiling It Sets

Node access is the percentage of electrical nodes that carry a test point of adequate size. A typical mixed technology board reaches 85 to 95 percent, and the missing few percent are usually the ones that matter, because they are the nets that were too dense or too sensitive to bring out.

The access figure should be produced from the netlist rather than estimated, and it should be reviewed before the layout is released. Adding a test point at that stage costs nothing, and adding one after the fixture exists costs a new fixture. The placement rules are set out in the notes on test point design.

<img src="https://www.gopcba.com/wp-content/uploads/2026/08/hdi-pcb-design-guidelines-11-b6900f71.webp" alt="ICT fixture pressing probes onto PCBA test points” />

Nets That Cannot Be Probed

Some nets resist access for good reasons. A pad under a package has no exposed copper, a high speed differential pair cannot tolerate the stub of a test point, and a net that carries a very small signal may be disturbed by probe capacitance.

The design response is to bring the net out to a via or a dedicated pad rather than to leave it untested, and the placement of that pad has to respect the fixture as well as the circuit, which is the subject of the notes on test point placement for fixtures.

What an Analog Test Can and Cannot Detect

In-circuit test measures impedance between nodes with the board unpowered, and it compares the result with a stored value. It detects a missing component, a wrong value that is well outside tolerance and most opens and shorts with high confidence.

It cannot see a component that is within tolerance but wrongly specified, a joint that is mechanically sound and electrically marginal, or a failure that only appears under load or at temperature. Those are the areas where functional test, boundary scan or a powered check has to take over.

Limited Access and Vectorless Methods

Where access is poor, vectorless techniques infer the presence of components from the signatures of the surrounding network rather than measuring each one directly. They extend coverage where a test point cannot be placed, at the cost of some diagnostic resolution.

Flying probe is the other answer, since it needs access to only one side at a time and can reach features a bed of nails cannot. It is slower than a fixture, which changes the economics rather than the coverage, as the comparison in the notes on flying probe and ICT sets out.

Coverage Reports and How to Read Them

A coverage report should list every net, the access it has, the test applied and the measurement accuracy. Summaries hide the detail that matters, and a single percentage tells the reader nothing about which nets are missing.

The useful view groups the untested nets by function. Untested power nets are usually tolerable, untested signal nets in a control loop are not, and the report should make that distinction visible without an engineer having to reconstruct it from the netlist.

Design Changes That Raise Coverage

Three changes raise coverage cheaply: adding test points on nets that currently have none, enlarging pads that are too small for a reliable probe, and removing components that block access to a pad underneath. Each is a layout change with a small cost at design time.

The larger change is the test strategy itself. Where a board has a high component density and no room for test points, designing for boundary scan on the digital devices or for a flying probe operation may raise coverage further than any number of added pads.

Coverage, Escapes and Field Returns

The value of coverage is measured against the escapes that reach the field. Where a returned assembly fails on a net that the fixture never tested, the coverage gap is the direct cause, and the return data is the argument for closing it. Each escape should be traced back to the net and to the test that should have caught it, and that trace is usually conclusive within a few hours of work.

Comparing the fault distribution of field returns with the coverage report is therefore a routine task, not an annual review. The mismatch is usually concentrated on a few nets, and those nets can often be given access by a small layout change at the next revision. The correction is then a design change rather than a program change, and it belongs in the next layout revision instead of in a workaround at test.

Records and Review

Coverage belongs in the release package with the test program, and it should be re-issued whenever the layout or the program changes. A coverage figure from three revisions ago describes a board that no longer exists. The report should also record the version of the program that produced it, because a coverage figure is valid only for one version of the fixture data.

The review should include the customer where the coverage figure is a contractual requirement, and it should state the measurement accuracy assumed for each test. Without that number the coverage claim is not verifiable, and the discussion reverts to opinions. Where coverage is traded against cost, the decision should be documented with the accepted escape risk rather than left implicit in a meeting. Probe and fixture selection is covered in the notes on test probe selection.

FAQ

What coverage should a new design target? Node access above 90 percent is a reasonable aim for a mixed technology board, with the untested nets identified and justified rather than merely counted.

Does a better fixture raise coverage? No. Coverage is set by access and by the measurement method. A better fixture reduces false calls and improves repeatability within the coverage that already exists.

How is coverage verified? From the netlist against the fixture data, then confirmed by the escape rate in test and by comparing field returns with the untested net list.