Browse Prior Art Database

Improved Propagation of Detection Probabilities

IP.com Disclosure Number: IPCOM000061793D
Original Publication Date: 1986-Sep-01
Included in the Prior Art Database: 2005-Mar-09
Document File: 3 page(s) / 38K

Publishing Venue

IBM

Related People

Smith, GL: AUTHOR

Abstract

Self-test based on hardware-generated pseudo-random patterns and hardware signature registers is a known process for detecting hardware faults in digital logic. Prior to the use of self-test on a part, however, at least a lower bound on the expected coverage of the self- test must be determined in order to establish the adequacy of the test. There are known methods of generating a lower bound on the detection probability for each individual fault (LB(P)), from which it is possible to generate a lower bound on expected coverage for that part. Determination of LB(P) for all faults of a part is likely to result in excessive costs, however. Therefore, any means for reducing the number of faults that must be processed is useful.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 3

Improved Propagation of Detection Probabilities

Self-test based on hardware-generated pseudo-random patterns and hardware signature registers is a known process for detecting hardware faults in digital logic. Prior to the use of self-test on a part, however, at least a lower bound on the expected coverage of the self- test must be determined in order to establish the adequacy of the test. There are known methods of generating a lower bound on the detection probability for each individual fault (LB(P)), from which it is possible to generate a lower bound on expected coverage for that part. Determination of LB(P) for all faults of a part is likely to result in excessive costs, however. Therefore, any means for reducing the number of faults that must be processed is useful. The present invention provides a means for derivation of LB(P) values for all faults based on the LB(P) values provided for a subset of faults of the part. The derived LB(P) values are superior to those provided by previously described means. Surprisingly, the values of LB(P) generated by the invention are often better than the values of LB(P) in the subset. It is well known that stuck-at-zero faults (s-a-0) at all of the inputs and at the output of an AND block are all equivalent. It is also well known that stuck-at-one faults (s-a-1) at any input of AND block dominate a s-a-1 fault at the output. Any test for a stuck- at- one (s-a-1) at one of the inputs results in the imposition of a value of zero at that input and values of one at all other inputs. It follows that a test for a s-a-1 at one input cannot be a test for a s-a-1 at a second input because the required values are incompatible. Therefore, the tests for s-a-1 at the various inputs of an AND block are all disjoint. Clearly, faults at the input of an invert block are equivalent to the opposite faults at the output. The following can be concluded from the above: 1. A value of LB(P) for any one of the following faults of an AND block equally applies to all of these faults: s-a-0 at each of the inputs, s-a-0 at the output. 2. The sum of LB(P) values for s-a-1 for all inputs to an AND block defines a valid value for LB(P) for s-a-1 at the output. 3. A value of LB(P) for either of the following faults equally applies to the other: s-a-1 at an invert block input and s-a-0 at the output. The same holds for a s-a-0 at an invert...