Statistical Circuit Design: The Optimum Assignment of Component Tolerances for Electrical Networks

Monte Carlo tolerance analysis has proven to be a useful tool in evaluating the effects of component tolerances and environmental variations on electrical circuit performance. The method involves "constructing" samples of the circuit inside the computer using element values that obey the manufacturing statistics, analyzing these samples, and forming empirical distributions of performance. One common outcome of the process is the prediction of yield. Monte Carlo tolerance analysis, henceforth referred to as TAP 1225 1226 THE BELL SYSTEM TECHNICAL JOURNAL, APRIL 1971 (Tolerance Analysis Procedure), is an open-loop structure. If we examine the way in which it is used, we find that for discrete circuits the designer supplies the T A P program with a set of component tolerances he has chosen based on breadboard measurements, linear sensitivity or worst-case analysis, or some other approximate technique. At the conclusion of the T A P run, he observes yield and is faced with one of two situations: (i) Yield is too low. With this result the designer knows he must change his tolerances. Unfortunately, TAP gives him little information as to which tolerances to change and by how much. (it) Yield is adequate. Here the designer may be satisfied by the design but he obtains little help in determining whether a cheaper (looser) set of tolerances might not give equally satisfactory yield. TAP, then, is open-loop in that it is a tool for predicting yield given a set of tolerances; both the initial set of tolerances or any changes to that set must be provided by the designer without assistance from TAP.