Dual-mode neuron design with deterministic and non-deterministic operations using adiabatic superconductor devices

Abstract

Abstract In this research, we unveil an innovative strategy in neuromorphic computing by developing a neuron model tailored for the energy-efficient Adiabatic Quantum-Flux-Parametron (AQFP) logic. This model is particularly aimed at enhancing neural network accelerators. Our design of the AQFP-based neuron operates effectively in both deterministic and non-deterministic modes. In deterministic mode, the design relies on superconducting inductive coupling to activate neurons by comparing the sum of AQFP signal currents against a tunable threshold. For non-deterministic operation, we demonstrate how altering specific circuit parameters can correlate these aggregated currents with the non-deterministic operational range of an AQFP current comparator. We verified its versatility and functionality by fabricating varied circuits and conducting extensive tests, confirming its practical application potential. Our work not only showcases the practical implementation of AQFP in neuromorphic computing but also sets a foundation for future advancements in energy-efficient AI hardware.