Accurate approximation to the probability of critical performance

Statistical analysis has to be taken into consideration in the design process of submicron integrated circuits. The variation of characteristic performance values depend upon the variation of many design parameters. In this paper we present a scheme for accurately and efficiently determining such dependencies at cell level. Additionally we employ an approximation for the performance distribution which does not only rely on a few low-order moments and therefore also performs well in the tails of the distribution. A typical problem that can be handled in this way is for instance the computation of the probability that the leakage current of a cell or circuit is greater than a given thershold value. The same question occurs for power and delay dependencies. Furthermore, sensivities of such marginal probabilities can be determined. Thus, this approach may potentially assist parameter monitoring and optimisation in the fabrication process.