On Diagnosis of PatternDependent Delay Faults

1
On Diagnosis of Pattern-Dependent Delay Faults

• Irith Pomeranz and Sudhakar M. Reddy
• Electrical and Computer Eng. Dept.
• University of Iowa
• Iowa City, IA 52242

2
Overview

• Motivation for Considering Pattern Dependent
  Delay Faults
• Definition of Pattern Dependent Delay Faults
• Diagnosis of Pattern Dependent Transition Faults
• Experimental Results
• Conclusion

3
Motivation

• Delay defects may show pattern-dependent
  behavior.
• Two tests detecting the same delay fault may
  result in different observed responses if
  pattern-dependent defects are present.
• Franco McCluskey, VTS-94
• Pierzynska Pilarksi, EURO-DAC 95, ITC-95

4
Modeling Issues

• Physical effects causing pattern dependence are
  complex.
• Accurate modeling is difficult, if not
  impossible.
• Solution Use an abstract model to augment the
  existing delay fault models.

5
The Proposed Model

• Every faulty value obtained under the
  conventional fault model may become fault free
  because of pattern-dependence.
• New faulty values cannot be created since the
  conventional fault model already captures all the
  cases where fault effects are created and
  propagated.
• Any faulty value may disappear.

6
Example
R
R1
R2
R3
R4
R5
R6
00
00
00
00
00
00
00
01
00
01
01
00
00
01
11
11
01
10
01
10
00
7
Diagnosis

• To say that faults f1 and f2 are distinguished
• Assume that R1 is the observed response.
• Check whether a pattern-dependent version of f2
  can explain this observed response.
• Yes f1,f2 are not distinguished.
• No f1,f2 are distinguished.

8
Example

• f1,f2 are not distinguished.
• f2,f1 are distinguished.

• R1
• 000
• 010
• R2
• 110
• 011

9
Diagnostic Measures

• Percentage of distinguished fault pairs.
• For fi, find the set of undistinguished faults
  Fi.
• Find the maximum set Fi.
• Find the average size of a set Fi.

10
Results for s1423

• Pattern-dependent
• 1-detection test set
• dist 99.94
• max.und 20
• ave.und 1.57
• 2-detection test set
• dist 99.95
• max.und 10
• ave.und 1.46

• Pattern-independent
• 1,2-detection test set
• dist 99.99
• max.und 3
• ave.und 0.28

11
Diagnosing Observed Responses

• Finding observed responses
• Randomly select a transition fault.
• Randomly change 1 values in its response to 0s.
  Probability of change p0.5 (0.1, 0.01).
• Repeat until 100 non-zero responses are obtained.
• Find the set of faults that explain each observed
  response.

12
Experimental Results p0.5
1-detect
2-detect
circuit
max
ave
max
ave
S400
18
3.64
11
3.50
S526
13
3.58
10
3.24
S641
24
6.00
32
5.15
S953
36
4.64
23
3.88
S1423
14
3.14
9
2.63
13
Results for Various Probabilities of
Pattern-Dependent Effects
p0.01
p0.1
p0.5
circuit
max
ave
max
ave
max
ave
s400
10
3.12
10
3.12
18
3.64
s526
9
2.94
9
2.98
13
3.58
s641
15
5.51
18
5.54
24
6.00
s953
13
3.31
13
3.44
36
4.64
s1423
6
2.85
6
2.85
14
3.14
14
Conclusion

• We defined a model for pattern-dependent delay
  faults.
• We defined conditions under which two faults are
  diagnosed.
• Experimental results show that the resulting
  resolutions are reasonable.