On May 14th 2025, the 3rd seminar of the JOSEPHINE project took place : Emilie Jué (National Institute of Standards and Technology, Colorado, USA) presented her work in a talk entitled “Superconducting artificial synapses integrated into a self-training neuromorphic architecture”.

Superconducting electronics is a compelling technology for designing neuromorphic computing systems. The technology uses Josephson junctions (JJ), which have a natural spiking behavior and can transmit voltage spikes over long distances with near-zero loss. One important building block of a neural network that still needs to be developed in superconducting electronics is the synapse, whose role is to adjust the strength of the connection between two neurons by changing the synaptic weight. In this work, we propose a new superconducting synapse developed at NIST and demonstrate its integrations in neural networks. The artificial synapse is obtained with a SQUID-based circuit using a flux storage loop for the synaptic weight. Using SPICE simulations, we demonstrate that the synaptic
circuit can be tuned using digital single flux quantum pulses.

In the second part of this presentation, we demonstrate the implementation of the synapses in a neural network. We integrate the SQUID-based artificial synapses into a small-scale self-training neural network architecture using SPICE simulations. The network follows reinforcement learning rules that update local weights internally. This property allows the network to learn new functions by changing the target output for a given set of inputs without needing any external adjustments. Finally, using the same constraint as for the network mentioned above, we extend our simulations in Python to a larger problem to classify the handwritten digits from the MNIST dataset and show that the MNIST network can be trained in about 100 ms.

You can access the full recording of this talk on the JOSEPHINE website, at this address, or under Seminars & Workshop > See previous seminars.