Automata of Evolvable Computational Machines

Evolutionary algorithms (EA) were successfully applied to design of various computational machines in recent years. Some evolved designs have outperformed human designs especially in cases where no or weak design methodology is known for a given problem. Cellular automata [5] or digital circuits [6] are typical examples. In case of evolvable hardware, circuit implementation is usually evolved in a reconfigurable device like FPGA (Field Programmable Gate Array). There is also a chance to design adaptive hardware since the circuit specification (fitness function) can vary in time. The idea of evolvable component have been introduced in [1] to make the design of evolvable hardware-based applications easier. In case of evolvable components, genetic operators and a reconfigurable device are encapsulated and they operate as a single subsystem while the process of the fitness calculation remains outside and thus it is opened for a designer. To give an example, let us mention the evolvable component for image pre-processing. Various image operators may be evolved only by redefinition of the fitness function while the whole component is fixed [2]. Such evolvable component could be implemented as an IP (Intellectual Property), specialized chip or a software component in future. In this paper, we introduce a mechanism of communication between the evolvable component and its environment and also a mechanism of internal communication of the evolvable component. We claim that these mechanisms that are described by Moore’s automata are independent of a given evolutionary algorithm, reconfigurable device and application and thus they are common for all evolvable components. From practical viewpoint, a design template can be prepared beforehand.