FPGA-Based Streaming Processors for Tree-Based Genetic Programming

In this chapter, a specialized processor architecture for program evaluation in tree-based genetic programmingtree-based genetic programmingis introduced, and we coin the term “evolution processing unit (EPU)” in order to succinctly refer to evolutionary computation systems built with specialized hardwarehardware, rather than with general-purpose CPU/GPUGPUsystems. We design our initial EPU for a high-end 7nm field-programmable gate arrayfield-programmable gate array(FPGAFPGA) device, and when compared to prior hardwarehardwareaccelerators, the presented system allows for higher throughput while also being simple and scalable. To provide effective comparisons, we implement a state-of-the-art interpreter for CPU devices, and we leverage a 16-core, 7nm CPU system equipped with AVX2 instructions. Across a set of performance simulations, we find that the EPU is up to \(161.80\times \) faster than the CPU, and \(2.89\times \) faster on average. Overall, our results showcase the potential of FPGAFPGA-based GP systems, and they give further motivation for the explorationexplorationof EPUs, especially by way of specialized chips, which can have significantly fewer resource constraints when compared to FPGAs.