Information Technology Reference
In-Depth Information
Tabl e 2.
Single fault diagnosis with 2-detect tests
Circuit No. of No. of DC Diagnosis CPU* Fault ratio
name outputs patterns % % s SET1 SET2
C499 32 3872 98.400 100 1.025 1.029 7.970
C1908 25 6425 86.203 100 2.242 1.379 14.798
C7552 108 27756 86.750 100 16.076 1.281 8.023
∗
PC with Intel Core-2 duo 3.06GHz processor and 4GB memory
faults in SET1 and SET2 are reported in columns 7 and 8, respectively. This is
the ratio of the total number of faults reported in each set to the number of faults
expected in that set. The expected number of faults includes the actual fault,
its equivalent faults and the opposite polarity faults for all equivalent faults,
including the actual fault. This ratio denotes the diagnostic resolution of the
procedure. The closer the fault ratio is to 1.0, better is the resolution. For single
stuck-at faults, the ratio of SET1 faults is almost 1.0 in all cases. Hence, when
the faults identified in SET1 are probed (by electron beam or other failure mode
analysis procedures), one would locate the actual fault and it will unnecessary
to probe the faults in SET2. But in a real situation since we would not know
whether the actual fault is a single stuck-at fault or a non-classical fault, the
SET2 surrogate faults should not be disregarded.
For circuits C499, C1908 and C7552, the ratio of faults in SET2 is high. This
is due to the fact that the diagnostic coverage of the test pattern set is not high
enough. To examine the effect of improving diagnostic coverage of the test pat-
tern set on diagnostic resolution, 2-detect test patterns were used to diagnose
these three circuits. The results are shown in Table 2. Note that 2-detect patterns
provide a marginal, though definite, increase in diagnostic coverage (
DC
). Most
increase occurred for C1908, which is only 1.016%. Still, the resolution improved
as SET2 ratio dropped to about 50%. So, for patterns with even higher diag-
nostic coverage, the resolution will be further improved. An utmost eciency of
the diagnosis algorithm can be expected from higher diagnostic capability test
pattern set than from just the detection test pattern set.
To verify the relevance of the reported surrogate faults to actual non-classical
faults, we examined multiple stuck-at faults by introducing two stuck-at faults
simultaneously. One hundred failure cases were generated for each circuit. In
each case, two stuck-at faults were chosen in such a way that they are close
to each other in the circuit. The reason for considering only two simultaneous
faults is that the probability of fault masking is maximum when there are just
two faults and this probability keeps reducing as the number of faults present in
the circuit increases. This increased chance of fault masking created a pessimistic
environment for the algorithm. All diagnosis results are averaged over 100 cases
for each circuit. Table 3 summarizes the multiple fault diagnosis experiment with
1-detect test patterns.
Column 4 of Table 3 shows the percentage of cases where both faults were
diagnosed. Column 5 shows the percentage of cases where only one of the actual
faults present was diagnosed. The sum of these two percentages subtracted from
