The application of evolution-inspired strategies to hardware design and circuit self-configuration leads to the concept of evolvable hardware (EHW). EHW refers to self-configuration of electronic hardware by evolutionary/genetic algorithms (EA and GA, respectively). Unconventional circuits, for which there are no textbook design guidelines, are particularly appealing for EHW. Here we applied an evolutionary algorithm on a configurable digital FPGA chip in order to evolve analog-behavior circuits. Though the configurable chip is explicitly built for digital designs, analog circuits were successfully evolved by allowing feedback routings and by disabling the general clock. The results were unconventional circuits that were well fitted both to the task for which the circuits were evolved, and to the environment in which the evolution took place. We analyzed the morphotype (configuration) changes in circuit size and circuit operation through evolutionary time. The results showed that the evolved circuit structure had two distinct areas: an active area in which signal processing took place and a surrounding neutral area. The active area of the evolved circuits was small in size, but complex in structure. Results showed that the active area may grow during evolution, indicating that progress is achieved through the addition of units taken from the neutral area. Monitor views of the circuit outputs through evolution indicate that several distinct stages occurred in which evolution evolved. This is in accordance with the plots of fitness that show a progressive climb in a stair-like manner. Competitive studies were also performed of evolutions with various population sizes. Results showed that the smaller the size of the evolved population, the faster was the evolutionary process. This was attributed to the high degeneracy in gene variance within the large population, resulting in a futile search.
|Number of pages||21|
|Journal||Physica A: Statistical Mechanics and its Applications|
|State||Published - 1 Aug 2003|
- Evolution of hardware
- Fitness landscape
- Genetic algorithm
- Programmable chips