The circuit implementation of invertible neural networks has never been achieved. To realize it, the use of the Reversible Quantum Flux Parametron (RQFP) [1], based on the Adiabatic Quantum Flux Parametron (AQFP) [2], can be considered. RQFP is a logic circuit that has logical and physical reversibility but does not have reconfigurability. To be used as neurons in invertible neural networks, it must be able to change its logic function. A previous study proposed that connecting logic switching circuits to RQFP enables logic reconfiguration [3]. However, the logic switching circuit in that study produced a small output current, resulting in a critical margin of approximately 2% for the subsequent AQFP [3]. It made experimental demonstration difficult.
In this study, to demonstrate the operation of a reconfigurable reversible logic circuit, we designed a circuit that switches between buffer and not functions based on whether a phase shift is induced by magnetic flux. We analytically derived the parameter conditions for correct operation and modified the circuit structure to increase the output current. As a result of parameter optimization considering the parameter conditions and energy consumption, the critical margin reached 14.17% for buffer operation and 19.30% for not operation. We confirmed that the circuit operates correctly and adiabatically through simulation and by calculating energy consumption.