Multiple criterion optimization and statistical design for electronic circuits

M Ligntner, Stephen W. Director

PhDT · 1979 · 20 citations · 15 references

Concepts

Abstract

In this paper we examine the problem of designing electronic circuits using Multiple Criteria Optimization where one of the competing criteria is the circuit yield. A new technique for generating solutions to the MCO problem based upon a family of weighted p-norms is presented. We concentrate on the max norm member of this family (this gives a minimax problem) and propose a method of solution based upon a new constrained optimization method due to Powell. The yield and gradient of yield are estimated using a method based upon Simplicial Approximation which is used to form a piecewise linear approximation to the probability density function of the designable parameters. An example illustrates that it may be possible to significantly alter the values of various circuit criteria, over their value at the maximum yield point, with very little change in the circuit yield. £• Introduction Historically circuit design can be viewed as consisting of two broad methodologies: performance design and statistical design. In performance design the circuit designer chooses a circuit configuration, adjusts parameters to attain a desired performance and then tests the circuit yield. If the yield is too small the parameters are re-adjusted. Statistical design arose mainly in response to integrated circuit design problems. In statistical design a circuit configuration is chosen and then the parameters are adjusted to achieve maximum circuit yield (worst case design being the extreme of 100% yield). These two methodologies can be unified by considering circuit design as a Multiple Criteria Optimization (MCO) problem with yield as one of the competing objectives. Specifically, after a circuic configuration is chosen, we consider the following MCO problem:

References

15