Improving Energy Efficiency of Field-Coupled Nanocomputing Circuits by Evolutionary Synthesis

Moore's law provoked decades of advances in computer's performance due to transistor's evolution. Despite all success in its improvement, current technology is reaching its physical limits and some replacements are the focus of investigations, such as the Field-Coupled Nanocomputing devices. These devices achieve information transfer and computation via local field interactions, reaching ultra-low power consumption. Nevertheless, there exists a hard energy limit related to the Laws of Thermodynamics that bounds any digital evaluation. To reduce the impact of this restriction, we propose a fitness function to improve the energy efficiency on a given circuit implementation. We embed our fitness function on an evolutionary method known as Cartesian Genetic Programming to iteratively modify the circuit, searching for a new valid configuration that dissipates less energy. To assess our method, we use it on pre-optimized circuits from benchmarks and compare our results with the ones from the classic Cartesian Genetic Programming. Based on the outcome, we show that our method outperforms the latter, achieving, on average, 15% gain in energy efficiency.