Fig. 5: Fully memristive spiking neural network. | Nature Communications

Fig. 5: Fully memristive spiking neural network.

From: Spiking neurons with spatiotemporal dynamics and gain modulation for monolithically integrated memristive neural networks

Fig. 5

a Scanning electron microscopy (SEM) image of the monolithically integrated memristive neural network. Scale bar, 100 μm. b SEM image of the Pt/Ti/NbOx/Pt threshold switching device. Scale bar, 20 μm. c Cross-sectional transmission electron microscopy (TEM) image of the Pt/Ti/NbOx/Pt threshold switching device. Scale bar, 50 nm. d SEM image of a Pt/Ta/TaOx/Pt synapse device. Scale bar, 10 μm. e Cross-sectional TEM image of the Pt/Ta/TaOx/Pt device. Scale bar, 20 nm. f, g Neuronal outputs when presented with different input patterns. f Neuron 1 learnt to recognize pattern “0101”, while neuron 2 learnt to recognize pattern “1100” g. The input pulses are 1 V in amplitude, 1 μs in width, 0.1 μs in interval and repeated for 10 cycles. h Schematic illustration of supervised learning in a fully memristive neural network. i Evolutions of synaptic weights over time during online learning, where the pattern “1010” is repeatedly applied, with an initial state of “0110”.

Back to article page