Page 122 - Building A Succesful Board-Test Strategy

108 BUILDING A SUCCESSFUL BOARD-TEST STRATEGY

changes is awkward at best. Success of HVI depends a great deal on the experi-
ence and diligence of the inspectors. Difficult boards or subtle problems can slow
the process and reduce the technique's accuracy and effectiveness. Also, whereas
machines make a "yes/no" decision based on established specifications and heuris-
tics, people's judgments are often more subjective. As a result, the consistency of
manual inspection leaves much to be desired. According to one study from AT&T,
now more than a decade old, two inspectors examining the same boards under the
same conditions agreed only 28 percent of the time. With three inspectors, agree-
ment dropped to 12 percent, and with four inspectors, to 6 percent. Even the same
inspector examining a board twice came up with an identical diagnosis only 44
percent of the time. With today's smaller feature sizes, the situation would likely
be worse. In many (if not most) cases, one of the automated techniques would work
considerably better.
Human inspection also suffers from inconsistency based on the time of day
or the day of the week—the previously mentioned "Monday/Friday syndrome."
named after an admonition by Ralph Nader in the 1970s never to buy a car man-
ufactured on Monday morning or Friday afternoon. Manual inspectors often miss
failures, while flagging and unnecessarily touching up good joints.
The last drawback to this technique applies to all the visual methods, as well
as to laser and white-light approaches when manufacturers use them on loaded
boards. They require line-of-sight access to the features they are inspecting. Since
one of the reasons for turning to inspection instead of conventional bed-of-nails
test is lack of access, the implications of this limitation are significant.

3.3.2 Automated Optical Inspection (AOI)

Automated optical inspection consists of a camera or other image input and
analysis software to make the pass/fail decision. Implementations include the fol-
lowing range of applications:
* A spot check of critical board areas
* A cursory check for component existence and orientation
* Comprehensive analysis of the entire board surface
AOI systems use several techniques to identify failures. Template matching
compares the image obtained from a theoretical "golden" image (assuming one is
available either from a good board or a CAD simulation). Template matching is
somewhat unforgiving of deviations from perceived good-board specifications
and of ECOs and other board modifications. The latter remain very common
during the early stages of production ramp-up. Pattern matching stores examples
of both good and bad boards, comparing the board under test to these standards.
Statistical pattern matching works similarly, except that the pattern represents
a compendium of a number of boards, so minor deviations will less likely cause
false failures. In fact, its proponents contend that statistical pattern matching can
produce orders-of-magnitude fewer false calls than its simpler siblings do.
Humans perform better than machines on recognizing image patterns. Nev-
ertheless, even when features are large enough to be detected by human inspectors,