Optimization of Quantum Cellular Automata Majority Gate Using Multiobjective Reinforcement Learning

The Quantum Cellular Automata (QCA) majority block gate, while more robust than the simple or classic version of the gate, is still affected by lithographic manufacturing errors that affect the polarization of each individual QCA and the block circuit. Analysis and correction of these errors has been done using Bayesian, Markovian, or neural network methodologies. The problem with learning an objective to minimize errors a priori is the inherent multiobjective nature of optimization in a network of QCA composing the majority gate. Our work attempts to solve this problem by maximizing utility to achieve Pareto optimal evaluation and correction of these objectives in order to better evaluate and correct Gaussian and non-Gaussian distribution of errors. Simulation results show greater Bayesian decision-making approach has greater reduction in standard deviation of error versus Pareto-optimization of the function for Gaussian noise; in non-Gaussian noise the Pareto-optimization method performs better for a positively charged majority gate, but worse for a negatively charged majority gate.