A superconducting synapse circuit, which allows digitally variable synaptic weights, is presented. Each bit of the synapse circuit consists of 2 m (m=0, 1, 2,.) double-junction SQUIDs (superconducting quantum interference devices). Numerical simulation shows that the output current of the circuit can be changed digitally by switching the bias currents. To demonstrate this ability of the synapse experimentally, a superconducting neural network which functions as a 3-bit A/D (analog-to-digital) converter has been fabricated using a niobium integration technology. The network contains three neurons composed of coupled SQUIDs and three synapses composed of double-junction SQUIDs. This is the first successful implementation of such an active synapse into a superconducting neural network.
ASJC Scopus subject areas
- Physics and Astronomy (miscellaneous)