This is a master thesis in computer science from the Department of Computer Science (Daimi) at the University of Aarhus.
The thesis is based on evolvable hardware experiments, i.e automatic hardware design by an evolutionary approach. In these experiments artificial evolutions of hardware circuits are conducted in simulations and in real hardware, i.e. extrinsic and intrinsic Evolvable HardWare (EHW). These two different experiments are selected to cover the field of EHW as broad as possible, and to investigate many aspect of the field.
Logic synthesis of combinatorial CMOS circuits using building blocks like AND, NAND, OR, INV etc. The extrinsic EHW experiments will investigate the possibility of lowering the level of abstraction to switch transistors, and evolve combinatorial circuits on this level. This case study will emphasise on the construction of a good functional descriptive language for the circuitry, and use this as a genetic encoding. Two descriptive languages are defined, one constrained and one unconstrained. It is shown that by constraining the descriptive language the obtained results are improved.
In the intrinsic EHW experiment the practical problem of setting up an evolutionary experiment in a physical circuitry environment will be studied. A Field Programmable Analog Array (FPAA) will be investigated as an example of an analog reconfigurable hardware platform. The benchmark problem is the search for circuits that work as an alternative for artificial neural networks for classification. The focus of these experiments is the hardware platform and it's interconnection with the evolutionary algorithm. It is shown that it is possible to automatically evolve circuits in this setting.
The thesis will also contain an overview of the current research results and techniques in artificial evolution and their applications in hardware design. The research field is often classified by its surroundings, a section will briefly comment on the related fields of EHW.
Please send comments to email@example.com.