Apart from maximization of parametric yield, minimization of the spread in performance functions due to process variation is of extreme importance in very large scale integrated circuit design. To achieve efficient minimization of the spread, a novel algorithm based on the genetic algorithm and global approximation methods is proposed. The algorithm operates in two stages designated as coarse and fine optimization stages and adjusts design parameter set to simultaneously achieve the target performance and reduction in performance spread. The algorithm has distinctive features, such as global optimum design, subexponential complexity algorithm for N-P complete problem of global optimization, and simultaneous optimization of many functions. The algorithm is demonstrated using four design examples.

Process variations invariably give rise to a parametric yield below 100% for VLSI circuits. Improving the yield by choosing a set of optimum parameter values does not incur any extra cost, and it is a preferred method as it directly translates into profits. The paper presents an efficient and novel method to improve the VLSI parametric yield by selecting optimum parameter values. This method utilizes the worst-case distance analysis, design centering and gradient-dependent techniques. One circuit example is presented to demonstrate the optimization scheme.