Chapter 33 - Evolvable FPGAs

Abstract

This chapter observes the phylogenetic axis of hardware bioinspired systems, most known as evolvable hardware (EHW). The scope of EHW covers diverse areas ranging from analog circuits to antenna design, but this chapter focuses on the evolution of digital circuits using reconfigurable computing devices, more precisely, field-programmable gate arrays (FPGAs). EHW has been shown to be effective at finding solutions for real-world applications. Additionally, some solutions have proven to perform better than their engineered counterparts. On the other hand, EHW generally performs poorly, as a system-level solution: Microprocessor architectures, for example, are not among evolution results. As a matter of fact, evolution works better when the target is a complex cellular architecture: cellular automata, neural networks, or gate arrays. This chapter focuses on evolving silicon circuits, which constitute the main developments achieved by the EHW community. However, other types of substrates have been evolved that extend the domain and represent new directions for evolvable hardware. FPGAs present two major problems that include the genome's length on the order of tens of thousands of bits, rendering evolution practically impossible using current technology, and within the circuit space, consisting of all representable circuits, many circuits are invalid. With the introduction of the Xilinx XC6200 family of FPGAs, these problems are reduced. As with previous FPGA families, there was a direct correspondence between the bit string of a cell and the actual logic circuit. However, because the XC6200 was completely multiplexer-based, the result was always a viable system with no short circuits. Moreover, as opposed to previous FPGAs where the entire system had to be configured, the XC6200 family permitted the separate configuration of each cell, which was markedly faster and more flexible.