Defect Diagnosis of Digital Circuits Using Surrogate Faults - VLSI Design and Test - page 376

Defect Diagnosis of Digital Circuits
Using Surrogate Faults
Chidambaram Alagappan and Vishwani D. Agrawal
Auburn University
ECE Dept., 200 Broun Hall, Auburn, AL 36849 USA
[e-mail] ,
[e-mail]
http://www.eng.auburn.edu/ ~ vagrawal
Abstract. Classical single stuck-at faults are analyzed as surrogates
for any non-classical fault that may have caused an observed failure.
Although multiple stuck-at faults are used as an illustrative example of
non-classical faults, proposed algorithms are applicable to any other type
of fault. Our effect-cause analysis is less complex than existing methods.
The diagnostic procedure adds or removes faults from a set of candi-
date faults based on the observed circuit outputs, using minimal fault
simulation, to obtain a small set of suspected faults.
Keywords: Dictionary-less fault diagnosis; fault simulation; multiple
stuck-at faults; stuck-at faults; surrogate faults.
1 Introduction
An ideal fault diagnosis procedure should report true failures with accuracy,
i.e., resolution (the number of true failures reported among the total number of
faults reported) and diagnosability (the percentage of correctly identified fail-
ures) of the diagnosis result should be high [8]. Previous research on fault diag-
nosis attempts trade-offs between the resolution, diagnosability and CPU time,
but the algorithms become increasingly complex. Two major classes of algo-
rithms are cause-effect and effect-cause types. Cause-effect analysis has a stored
simulated response database of modeled faults. The faulty circuit response is
compared against this database to find out which fault might have caused the
failure [5,7,12,15]. This database, called dictionary, is memory intensive and im-
practical for large circuits. Effect-cause analysis works on the observed failing
signals and searches for the cause by tracing back the error propagation path
from the failing primary outputs to identify faults likely to have produced the
failure [3,4,9]. Backward implication and forward propagation are used for this
purpose [9]. Such procedures use moderate amount of memory.
Although a real defect is rarely a classical single stuck-at fault, diagnostic
procedures match observed symptoms to closest single stuck-at faults. This is
Research supported in part by the National Science Foundation Grant CCF-1116213.