I remember reading about combining genetic algorithms and fpgas in the 90's. The end circuits were over-tuned and could stop working if the temperature changed by even a few degrees. Or it would work only on FPGAs from a single output batch. Thanks a bunch for the names, I wouldn't have remembered the people doing the research otherwise.