The News Observer has an article summarizing the work of Adrian Thompson at the University of Sussex who has implemented genetic algorithms directly in hardware using Xilinx FPGAs. The result is a chip that evolves to perform required tasks. This approach seems faster than software based evolutionary computing but may have one potential problem: the results aren't limited by human design rules. In one case an FPGA layout which was evolved to distinguish between two audio tones worked better with far fewer components than human-designed layouts but appears to be using magnetic properties of unconnected cells in ways that are not yet understood.