On Test Data Compression and nDetection Test Sets

1
On Test Data Compressionand n-Detection Test Sets

• Irith Pomeranz, Purdue University
• Sudhakar M. Reddy, University of Iowa

2
Overview

• Generic Scheme of Input Test Data Compression
• General n-Detection Test Sets
• n-Detection Test Sets Suitable for Compression
• Test Generation
• Experimental Results

3
Input Test Data Compression
NS
D
CUT
NI


D Combinational Decompressor Encode NS-bit
vectors as NI-bit vectors NIltltNS provides data
compression
4
The Size of NIReddy, Miyase, Kajihara and
Pomeranz, VTS-02

• A test set T detects a set of faults F.
• Using a set of CUT input vectors Vit is possible
  to create tests to detectall the faults in F.
• NIlog2V.
• V needs to be a minimal set in orderto minimize
  NI.

5
n-Detection Test Sets

• A test set that detects each fault n times,by n
  different tests.
• Shown to result in improved defect coverage and
  improved coverageof unmodeled faults.
• The size of a compact n-detection testset grows
  approximately linearly with n.

6
n-Detection Test Sets and Compression

• Formulation 1 Given an n-detection testset Tn
  find a minimal set of decompresser output vectors
  Vn such that using Vnit is possible to detect
  the faults detectedby Tn the same numbers of
  times.
• Difficulties
• Vn for Tn may be larger than V for T.
• The decompresser design needs to be modified for
  Tn.

7
Formulation 2

• Constrained n-detection test generation Given a
  test set T with a set of vectors V, generate a
  test set Tn such that
• (1) every test t?Tn is comprised only ofvectors
  out of V and
• (2) every target fault is detected n times by Tn.
• The numbers of detections are increased without
  changing the decompresser.

8
Test Generation Procedure

• Set TnT.
• For m1,2,,n-1
• For every fault f detected m times by Tn
• Attempt to generate a test tw for f such that
  tw?Tn. If tw is generated add it to Tn(numbers
  of detections are updated by fault simulation).
• Generation of tw is done by a simulation-based
  procedure.

9
Example
10
Simulation-Based TG

• Fault f1 is detected only once by t3 0111 0000.
• Tests considered for f1

11
Experimental Results

• Test set T is a compacted 1-detection test set
  modified so as to minimize V.
• First number of scan chains for every circuit
  power of two, at least 10 flip-flopsper scan
  chain.
• Second number of scan chains power of two, at
  least 20 flip-flops per scan chain.
• n3 and n10.

12
Results for n3
13
Normalized Run Times
Test generation time is divided by the time it
takes to fault simulate T.