Evolvable Hardware with Genetic Learning

Neural networks can recognize patterns very flexibly as humans do, even if inputs with noise or incompleteness are given. We aim at the implementation of this flexible mechanism for association by taking a completely different approach from neural networks. We name this approach “Evolvable Hardware” (EHW). EHW is a hardware device which can adapt its own hardware structure to the environment to give the best performance. EHW has the pattern recognition capability like neural network. In addition to it, EHW can attain real-time performance which could not be obtained by neural networks. Specifically, EHW is built on Programmable Logic Device (PLD) or Field Programmable Gate Array (FPGA) where a software bit string determines the hardware structure of the device. EHW finds out by GAs such a bit string which adapts best to the environment and then reconfigures its own hardware structure according to the bit string. In EHW, adaptation takes the form of direct modification of the hardware structures according to rewards received from the environment. This results in a number of advantages. Adaptation in real-time is feasible due to a speed-up by many orders of magnitude. The system will be flexible and fault-tolerant since EHW can change its own structure in the case of environmental change or hardware error. Our long-term goal is to implement EHW on one chip so that it can be utilized as an “off-the shelf” device. This paper reports on the ongoing EHW-research project.