Fault-simulation programs for integrated-circuit yield estimations

Abstract

Three programs are described here which have been used for integrated-circuit yield modeling at the IBM facility in Essex Junction, Vermont. The first program generates negative binomial distributions which are used to represent the frequency distribution of the number of faults per chip. Calculations with the generalized combination function A ! B in APL are limited to simulations of up to 99,999 faults, and can take too much computer time to run. These limitations are eliminated when the calculations make use of the scan function. The second program simulates clustered fault locations on a map. The clusters are initially generated using a radial Gaussian probability distribution. Each fault location is stored as a complex number, which facilitates the use of cluster-shaping programs that are also described. In a third program, another simulator of fault maps, faults are added as a function of time. This program also results in fault distributions that are clustered. In addition, it produces frequency distributions that very closely approximate negative binomial distributions.