Information Technology Reference
In-Depth Information
(b)
C17 with output selection.
(a)
C17 benchmark circuit.
Fig. 1.
Output selection implementation on C17 benchmark circuit
because the available analysis tools only handle single stuck-at faults. The di-
agnosed single stuck-at faults then are not real but are “surrogates” meaning
that they have some, but not all, characteristics of the actual defect in the cir-
cuit. The term “surrogate fault” has been used before in the literature [10,13,16].
2 Preliminaries
A fault simulator reports all single stuck-at faults that can be detected by an
input pattern on all primary outputs (POs). To use this information for dis-
tinguishing among several faults that could have caused the failure we employ
output selection
. AND gates are added in the simulation netlist at each PO, with
the other input of the AND gate being a new primary input (PI). The failing test
pattern is duplicated as many times as the number of POs, activating exactly
one PI at a time. Thus, new PIs that directly go to the added AND gates are all
forced to 0 except for one PI to a transparent AND gate to find the detectable
faults at the corresponding PO. Consider C17 benchmark circuit of Figure 1a. A
test pattern “abcde” produces good circuit responses 'f1' and 'f2'. Assume this
circuit has a failure only at the second output. A typical fault simulator may
identify detectable faults without associating them to any PO. With output se-
lection of Figure 1b, the test pattern is duplicated as “abcde
10
” and “abcde
01
”.
3 Diagnosis Algorithm
The diagnosis algorithm relies on a basic concept that a test pattern fails because
a detectable fault is present in the circuit or a test pattern passes because none
of the detectable faults is present. For this to be effective, we assume that there
is no circular fault masking present in the circuit. Let
'passing set'
be the set
of passing test patterns,
'failing set'
be the set of failing test patterns,
'sus flts'
be the suspected fault list,
'set1 can flts'
be
prime suspect candidate faults
and
'set2 can flts'
be
surrogate candidate faults
. For simplicity, we will refer
to
'set1 can flts'
as SET1 and
'set2 can flts'
as SET2.
The algorithm has four phases [6] as shown in the flowchart of Figure 2.
Initially, Phase 1 takes the union of all faults detectable by all failing patterns as
a list of suspects. Since this set can be large, we need to reduce the list. In Phase 2,
