Design For Testability: Design Rules and Process Limits

Design for testability is the practice of making a board easy to verify. It is a layout decision that is made once, early, and that decides whether the product can be tested in a minute or debugged for an hour. A design that ignores it still works, and the cost appears later on the test floor, in the fixture budget and in the number of units that escape to the customer with a fault that testing could have caught.

What Testability Means

A testable board has access to the nets that matter, a defined way to power it, and a defined way to observe the outputs. The test can then be written from the netlist rather than from a description of the circuit.

Test coverage is the fraction of the possible faults that the test detects, and it is the number that matters. A board with a hundred test points and poor coverage is less testable than one with twenty points placed on the critical nets.

The cost of testing is set by the fixture, the test time and the rework that follows a failure. All three are influenced by the layout, and all three are easier to influence before the artwork is released.

The release checklist is a good place to record the coverage figure, because a release without it is a release that will be tested by whatever the fixture happens to reach.

Access and Test Points

Access is provided by test points, by vias that are left open, and by connector pads that a probe can reach. The choice depends on the fixture and on the density of the board.

A test point on the solder side is the most useful, because the component side is populated. Where the layout does not allow it, a via with an opened mask is the standard alternative.

The probe needs a clear vertical path, so a tall component beside a test point blocks the fixture. The clearance is measured from the point to the nearest obstruction, and it is usually a few millimetres.

The pad size and the finish also matter, since a probe needs a contact area and a surface that it can penetrate. A flying probe has the same requirements and does not remove the need for access.

Coverage and Net Selection

A net that carries a high speed signal is also a candidate for access, not to measure the signal but to confirm that the trace is continuous and that the driver is present. The probe loading is a concern at speed, so the measurement is usually made with the board unpowered or with the probe removed after the continuity check.

The nets that matter are the ones whose failure would not be caught by a functional test, and the ones that are hard to reach by other means. Power rails, references and the nodes of a feedback loop are typical examples.

A net that is invisible to the test is a net that has to be trusted. On a safety related product that trust is not acceptable, and the layout has to provide access even at the cost of area.

Coverage is usually calculated by the test engineer from the netlist and the access list, and the result should be reviewed with the layout engineer before the artwork is finalised. A coverage figure of ninety per cent is a common target.

Some nets cannot be probed at all, such as a very fine pitch ball grid array, and those are covered by a boundary scan or by a functional test rather than by a probe.

Fixture Cost and Constraints

A bed of nails fixture is a machined plate carrying a spring probe for every access point, and its cost grows with the number of probes and with their density. A board with a fine pitch access pattern needs a more expensive fixture than one with a coarse pattern.

The fixture also imposes rules on the layout. The board has to be supported so that it does not flex under the probe force, the edges have to give the fixture something to clamp, and the tooling holes have to match the fixture plate.

A double sided fixture costs more than a single sided one, so access on one side is preferable where the density allows it. The decision should be made with the test engineer rather than after the layout.

The test coupon area of the panel is not part of the fixture, which is a point of confusion in some reviews. The coupon is for the process, the test points are for the product.

Boundary Scan and Alternatives

Boundary scan uses the devices on the board to test the connections between them. It needs a device that supports the standard and a small number of access pins, and it covers nets that a probe cannot reach.

The method is a complement to a fixture rather than a replacement. It detects an open or a short between two devices, while a probe can also measure a component value.

Other alternatives include a functional test with the product running, a built in self test in the firmware, and an optical inspection for the joints. Each covers a different class of fault.

The short circuit inspection methods are a useful companion to the coverage discussion, since a short is the fault that most often escapes a functional test.

Documentation for the Test House

The test house needs the netlist, the access list with coordinates, the power requirements and the expected values. A package without them is a package that will be tested by hand.

The access list should come from the same data as the artwork, so that a change in the layout is reflected in the fixture. A separate list that is maintained by hand drifts within one revision.

The documentation should also state the test strategy, because a fixture, a flying probe and a boundary scan each need different preparation. A board that is handed over without a strategy is tested three times.

The final item is the acceptance criteria: what constitutes a pass, what is reworked and what is scrapped. Those rules belong in the documentation, and the quality record should be built around them.

Practical Rules

Decide the test strategy before the layout, and write it down. A strategy that appears after the placement is a compromise with whatever space is left.

Place the test points early, on the nets that matter, with the clearance that the fixture needs. Adding them at the end forces a placement change.

Keep the coverage figure with the artwork so that a later revision can be checked against it. Coverage that silently falls is a common way for a product to become untestable.

Review the fixture cost as part of the product cost, because a design that saves a connector and doubles the fixture cost has not saved anything.

FAQ

Does a flying probe remove the need for test points? No. It still needs access to the net, and a via under a mask cannot be probed. It removes the fixture cost rather than the access requirement.

What coverage should a board have? A common target is around ninety per cent of the structural faults, with the remainder covered by boundary scan or by a functional test.

When should test points be placed? During the placement, before the routing is complete. Adding them at the end forces a layout change and often reduces the coverage.