Automated synthesis of standard cells using genetic algorithms

Abstract

We have demonstrated the possibility of applying a genetic algorithm (GADO) to automate the design of a standard cell given only a behavioral description and optimization criteria such as power, area, speed or their combination. GADO maintains a population of potential standard cell designs. The designs are evaluated by (1) internal rules, (2) MAGIC for design rule checks and (3) SPICE for correctness of circuit behavior. Connections between transistors are encouraged by an influence check, which checks for broken connections and floating I/O nodes. Working inverters with arbitrary label placements are designed by evolution as a proof of concept.

A genetic approach to gateless custom VLSI design flow

“…This automatic transistor level optimization of the topology starting from truth tables, completely avoiding the Boolean simplification approach is the first known methodology to the best of our knowledge. The transistor sizing [2] and placement optimization [1], [4] and [5] can easily be incorporated in the current model for the complete genetic custom design flow.…”

“…In literature, much attention has been given to the sizing [2], [3] and placing of transistors [1], [4], [5] in custom and semi-custom circuits. Using Genetic Algorithms (GAs), in [4], Murphy, et.…”

“…Their major aim is the use of primitive components and reduction of the parasitic capacitance rather than a topological optimization. In [5], Bahuman, et. al, describe a GADO model for a custom cell.…”