Automatic design of digital electronic circuits via evolutionary algorithms is a promising area of research. When evolved intrinsically on real hardware, evolved circuits are guaranteed to work and the emergence of novel and unconventional circuits is likely. However, evolution of digital circuits on real hardware can cause various reliability issues. Thus, key mechanisms that produce reliable evolution of digital circuits on a hardware platform are developed and explained in the first part of this thesis.
On the other hand, the evolution of complex and scalable designs without any assistance is thwarted due to increasingly large genomes. Using traditional circuit design knowledge to assist evolutionary algorithms, the evolution of scalable circuits becomes feasible, but the results found in such experiments are neither novel anymore nor are they competitive with engineered designs.
A novel, biologically inspired gene regulatory network based multicellular artificial developmental model is introduced in this thesis. This developmental model is evolved to build digital circuits that can automatically scale to larger designs. However, the results achieved still remain inferior to engineered digital circuit designs.
Evolving a developmental system for the design of engineering systems or computational paradigms provides a variety of desirable properties, such as fault tolerance, adaptivity, and scalable designs automation. However, developmental systems in their role as computational networks are as yet poorly understood. Many mechanisms and parameters that a developmental system comprises are based on various assumptions, their biological counterparts, or educated guesses. There is a lack of understanding of the roles of these mechanisms and parameters in forming an evolvable platform for evolutionary computation.
Initially, various experiments are shown to demonstrate the evolvability of the new developmental system. A thorough investigation is then undertaken in order to obtain large amounts of empirical data that yields a better understanding of some of the crucial developmental mechanisms and parameters on the evolvability of multicellular developmental systems.
