Page 104 - Building A Succesful Board-Test Strategy

90 BUILDING A SUCCESSFUL BOARD-TEST STRATEGY

of the sequence, as close to the misprobe as possible, reduces diagnostic time con-
siderably. Using automated probe handlers, similar to conventional x-y probers.
during repair can speed diagnosis and minimize probing errors.
Some manufacturers construct functional tests in sections that test parts of
the logic independently. In this way, GFI probing chains are shorter, reducing both
time and cost.
As parts, board traces, and connections have shrunk, concern has mounted
that physical contact with the board during failure analysis may cause circuit
damage. The proliferation of expensive ASICs and other complex components and
the possibility that probed boards will fail in the field have increased the demand
for less stressful analysis techniques.
Fault dictionaries address some of these concerns. A fault dictionary is merely
a database containing faulty input and output combinations and the board faults
that cause them. Fault simulators and automatic test-program generators can
create these databases as part of their normal operation.
A complete dictionary would contain every conceivable input and output
pattern and, therefore, pinpoint all possible failure modes. As a practical matter,
manufacturers have experienced mixed success with this technique because few dic-
tionaries arc so comprehensive. The analysis may identify a fault exactly, but it
more often narrows the investigation to a small board section or a handful of
circuit nodes, then reverts to GFI for confirmation and further analysis. Even more
often than with GFI techniques, a fault dictionary depends on dividing circuit logic
to "'narrow the search" for the fault source.
This method is very fast and requires no operator intervention except during
supplemental GFI. It is, therefore, appropriate for high-volume applications. Test
programming is generally faster than with GFI because the test-program genera-
tor does much of the work. For specific faults that it has seen before, the technique
is quite accurate.
The need to subdivide board logic represents a design constraint. In addi-
tion, not all logic subdivides easily. The method is deterministic-—that is, an unfa-
miliar failure pattern generally reverts to GFI. Revising board designs may
necessitate manually updating the dictionary, which can be a significant headache
during early production.
An interesting technique that has fallen into disuse in the past few years
involves expert systems (also known as artificial-intelligence techniques) -essen-
tially smart dictionaries. Like conventional fault dictionaries, they can examine
faulty outputs from specific inputs and identify failures that they have seen before.
Unlike their less-flexible counterparts, expert systems can also analyze a nevtr-
before-seen output pattern from a particular input and postulate the fault's loca-
tion, often presenting several possibilities and the probability that each is the
culprit. When an operator or technician determines the actual failure cause, he or
she informs the tester, which adds that fault to the database.
Expert systems do not require conventional test programming. Instead, the
tester executes a set of input vectors on a good board and reads the output pat-
terns. Then, a person repeatedly inserts failures, allowing the tester to execute the