A new efficient approach to VLSI yield improvement via yield/specification trade-offs

A method is described for making changes to specifications to optimize yield using information derived from a Monte Carlo analysis. Since nominal values and tolerances are fixed, only one Monte Carlo analysis is required for the entire optimization, i.e. the exploration of performance space is extremely inexpensive. In addition to yield/performance tradeoffs the method can be applied to (1) the definition of performance specifications (often a nontrivial task), and (2) to identification of redundant specifications and performance functions for testing.<>