Browse Prior Art Database

# Weighted Test Pattern Generation in Monte Carlo Testing of Integrated Circuits

IP.com Disclosure Number: IPCOM000078043D
Original Publication Date: 1972-Nov-01
Included in the Prior Art Database: 2005-Feb-25
Document File: 2 page(s) / 14K

IBM

## Related People

Schnurmann, HD: AUTHOR

## Abstract

The present development relates to random or Monte Carlo testing of integrated circuits. Such testing is generally described in U. S. Patents 3,614,608 and 3,633,100.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 53% of the total text.

Page 1 of 2

Weighted Test Pattern Generation in Monte Carlo Testing of Integrated Circuits

The present development relates to random or Monte Carlo testing of integrated circuits. Such testing is generally described in U. S. Patents 3,614,608 and 3,633,100.

The advantages of weighting random test pattern inputs has been recognized. The present development is a technique whereby with a judicious use of good machine simulation in random testing, coupled with suitable feedback techniques, one may obtain the desired ratio between the various primary inputs.

The basic concept centers around determining the rate of change of the activity of the different inputs, instead of measuring activity in absolute terms.

Assume that AC, the activity vector, is measured by counting the number of lines that are set to 0 or 1 as the result of a change of state in one of the inputs. An equal weight (=W) good machine simulation (GMS) is performed with AC traced every few patterns. Upon completion of the GMS run, a cumulative count of AC is displayed. If by adding the values of all the components of AC (as many as there are inputs) one measures at least some activity, then this AC is normalized and used as W(1), the set to 0 or 1 is kept, so that only new lines not previously set (in the =W run) will be counted to update the new AC. The process is repeated with W(2) W (3) etc., until no new activity is recorded.

A problem arises if the first adaptive run (with W(1)) does not create new activity. This may be due, for instance, to an unsatisfactory clock/ reset ratio. This problem is solved by rerunning the =W GMS once again. When a decrease of the rate of change of the reset line is noticed, the simulation is interrupted, AC is reset back to 0 (AC = 0, 0,.....,0), the GMS is restarted from the point of
interruption, and any new activity is recorded in the same manner as prior to the interruption. Thus, one obtains a new W different from the straight adaptive Monte Carlo, with the difference that the activity assigned to the reset is now much lower than it would otherwise be. (Note: in almost all cases, the rate of change...