Statistical Circuit Design: Large Change Sensitivities for Statistical Designs
01 April 1971
Realistic system and circuit design must account for the fact that exact realizations of paper designs are seldom achieved. The Bell System is particularly sensitive to this problem not only because of physical and economic constraints in manufacture, but also because of the varied field environments in which the system must operate. The effects of variations in design parameters, which are usually modeled as random variables, can be investigated via a Monte Carlo study. However, a Monte Carlo study is an analysis in the sense that for specified probability density functions of design parameters (specified by "nominal" value, tolerance, correlation, etc.) an empirical distribution for various outputs or performance measures is found. The inverse problem, that of finding nominal values, tolerances, and correlation in order to obtain an acceptable performance distribution, has received 1209 1210 T H E BELL SYSTEM T E C H N I C A L J O U R N A L , APRIL 1971 relatively little attention. This paper describes an approach addressed to this problem of "closing the loop around tolerance analysis." * There are three significant points about this approach. First, the approach does not rely on first- or second-order approximations.* Like Monte Carlo, no attempt is made to approximate measures of performance. Second, the approach can accommodate multiple-specifications. Third, the implementation of the approach is feasible with a computer and so the techniques may be thought of as computer aids to statistical design.