Performance macromodelling and optimization of regular VLSI structures

Abstract

The objective of this work was to develop an environment for the automated performance optimization of regular VLSI structures. The environment should enable designers with little or no knowledge of optimization theory or IC design technology to specify, create and optimize the performance of regular VLSI structures. The SRAM was initially used as the regular VLSI structure with which to develop such an environment. Macromodelling was used to increase the level of design abstraction at which the performance of a VLSI structure could be specified and new accurate, fast performance models developed. A novel numerical optimization technique which enabled a weighted sum formulations to search nonconvex design surfaces was then developed and shown to be applicable to IC performance optimization in general. Several different SRAM designs were then used to demonstrate the effectiveness and efficiency of the macromodels and optimizing environment.<>