Information Technology Reference
In-Depth Information
Tabl e 6.
Multiple (two) fault diagnosis with diagnostic patterns
Circuit No. of DC % of cases diagnosed CPU* Fault ratio
name patterns % Both faults One fault No fault s SET1 SET2
C17 12 100 80.952 19.047 0.000 0.067 0.489 2.102
∗
PC with Intel Core-2 duo 3.06GHz processor and 4GB memory
Single fault diagnosis with 100% diagnostic coverage vector set produced a
perfect diagnostic resolution '1.0' as expected in SET1 and a slightly improved
resolution in SET2. Multiple fault diagnosis with this test pattern set improved
the resolution in SET1 by a very small amount and decreased the resolution
of SET2 by the very same amount. Also, the diagnostic coverage was improved
by a very small percentage. Since the 1-detect test pattern set already had a
diagnostic coverage of 95.454, there was very little left to improve.
To sum up, the proposed diagnostic procedure, given a failing vector and the
cause of failure a single stuck-at fault, will always come up with the actual fault,
irrespective of the detection or diagnostic coverage of the test pattern set. If
the detection coverage of the test pattern is higher, better will be the resolution
of the faults reported. Provided with 100% diagnostic coverage, the maximum
resolution can be achieved. If the actual fault is a multiple stuck-at fault with-
out circular fault masking, the diagnostic procedure will come up with surrogate
faults that represent the actual faults or the behavior of the actual faults, with
higher resolution as the diagnostic coverage of the pattern set increases.
5Con lu on
We have proposed a lower complexity fault diagnosis algorithm that is based
on effect-cause analysis. The algorithm has higher diagnosability and resolution
for the surrogate faults identified to represent multiple stuck-at faults without
circularly masking, even if provided just with a high detection coverage test
pattern set. The same trend is exhibited when the diagnostic coverage of the
test pattern set is increased. The algorithm is memory ecient, since it does not
require a dictionary and also has reduced diagnostic effort (CPU time), since it
works on relatively smaller number of fault suspects and does not require re-
running simulations after frequently moving faults to and from the suspected
fault list based on heuristics.
In the future, we should examine the performance of the diagnosis algorithm
on other non-classical faults by using appropriate fault models and their simula-
tors. Also, redundant faults as one of the interfering fault in fault masking may
be examined. Considering that fault simulation tools will always be limited to a
few fault models (e.g., single stuck-at or transition faults), we should explore the
relationships between non-classical faults (bridging, stuck-open, coupling, path
delay, etc.) and the corresponding surrogate classical representatives. For exam-
ple, some non-classical faults like stuck-open or bridging require an initialization
pattern to precede a stuck-at test pattern. Thus, the test result for the non-
classical fault agrees with a single stuck-at fault only on a subset of patterns.
