Superconducting Probabilistic Neural Synapse With Spike-Timing-Dependent Plasticity

Brain-inspired computing paradigms offer a more energy efficient approach for artificial intelligence. The spike-timing-dependent plasticity of biological neural synapses reveals the potential learning mechanisms of the brain. This work proposes a synaptic circuit design based on superconducting Josephson junctions. The circuit achieved the functions that include spike sequence judgment of pre-synaptic and post-synaptic neurons, weight storage, and probabilistic-coded weight calculation of spike counts. Simulations show that the combined functions of each part of the circuit can emulate the spike-timing-dependent plasticity of biological neural synapses. When integrated with existing superconducting single-flux-quantum circuits, it successfully models the “winner-takes-all” behavior of the biological brain. Embedding this synaptic circuit into superconducting integrated circuits could be an important step towards the future development of brain-inspired superconducting computational systems.