Comparison of genetic algorithms to other optimization techniques for raising circuit yield in superconducting digital circuits
Novel logic devices in the RSFQ and COSL superconducting logic families are most often sub-optimal. Before such devices can be incorporated into physical designs, they have to be optimized for high theoretical yield, and preferably for highest possible yield. Even simple logic gates can contain numerous inductors, resistors and Josephson junctions. During optimization, it is often needed to adjust all the element values. The search space is therefore very large, and genetic algorithms have been used with success to optimize such gates. The conversion of circuit file to genome for the genetic algorithms is discussed, as well as fitness evaluation through Monte Carlo analysis. Results with both novel and existing logic gates are presented. Other optimization techniques are also discussed in comparison to genetic algorithms.