Multi-weight adiabatic quantum-flux-parametron for superconducting neural network circuits

Abstract Adiabatic quantum-flux-parametron (AQFP) is a promising superconducting logic family for energy-efficient neural network circuits thanks to its ultralow switching energy. Conventional AQFP logic, however, naturally provides binary output currents with identical amplitudes and opposite directions, which limits its direct use in neural circuits requiring multiple synaptic weight values. To overcome this, we demonstrate in this work a multi-weight AQFP logic cell based on a three-inductor transformer. By engineering the intermediate inductance of the transformer, multi-level output current amplitudes are obtained without modifying the core AQFP structure. AQFP neurons adopting the proposed cells are constructed to perform weighted current summation and threshold-based activation, and their operation is verified through numerical simulations. The proposed circuits are further fabricated using a superconducting thin-film process and measured at 4.2 K. The experimental results confirm the operation of multi-weight synaptic cells and threshold-controlled functionality, indicating the potential of the proposed approach as a building block for integrated superconducting neural network hardware.