Memristors for brain-inspired computing: the synapse and its electronic twin

A mirrored graphical abstract: the biological synapse on the left, the memristor on the right, and a central spine that pairs them row by row. Synaptic strength maps to conductance, spike-timing learning to voltage pulses, and a neural network to a crossbar array that multiplies in memory.

1. Synaptic strength ↔ conductance: A spike makes vesicles release transmitter in under a millisecond (Südhof 2013) and a synapse can be persistently strengthened or weakened (Bi & Poo 1998); a memristor is a conductor/oxide/conductor stack whose resistance, set by ion motion and a filament that forms or dissolves, remembers its history (Yang et al. 2013; Kumar et al. 2016). 2. Learning ↔ pulses: Post spiking within 20 ms after pre gives LTP, within 20 ms before gives LTD (Bi & Poo 1998); voltage pulses tune a memristor’s conductance, and memristor synapses show spike-timing-dependent plasticity (Jo et al. 2010). 3. Neural network ↔ crossbar array: Network computation rests on synapses and connectivity (London & Häusser 2005); a crossbar with a memristor at every crosspoint does the vector–matrix product by Ohm’s and Kirchhoff’s laws, I = Σ G·V (Mahmoodi et al. 2019; Xia & Yang 2019).

Starting materials: A biological synapse whose strength changes with activity. Product: A two-terminal memristor whose conductance plays the synaptic weight, in crossbars that compute in memory.

Strukov, Snider, Stewart & Williams, “The missing memristor found”, Nature 2008 (doi:10.1038/nature06932); Jo et al., “Nanoscale memristor device as synapse in neuromorphic systems”, Nano Lett. 2010 (doi:10.1021/nl904092h); Bi & Poo, “Synaptic modifications in cultured hippocampal neurons: dependence on spike timing, synaptic strength, and postsynaptic cell type”, J. Neurosci. 1998 (doi:10.1523/JNEUROSCI.18-24-10464.1998); Kumar et al., “Conduction channel formation and dissolution due to oxygen thermophoresis/diffusion in hafnium oxide memristors”, ACS Nano 2016 (doi:10.1021/acsnano.6b06275); Prezioso et al., “Training and operation of an integrated neuromorphic network based on metal-oxide memristors”, Nature 2015 (doi:10.1038/nature14441); Mahmoodi, Prezioso & Strukov, “Versatile stochastic dot product circuits based on nonvolatile memories for high performance neurocomputing and neurooptimization”, Nat. Commun. 2019 (doi:10.1038/s41467-019-13103-7); Yao et al., “Fully hardware-implemented memristor convolutional neural network”, Nature 2020 (doi:10.1038/s41586-020-1942-4); Sharma et al., “Linear symmetric self-selecting 14-bit kinetic molecular memristors”, Nature 2024 (doi:10.1038/s41586-024-07902-2); Yang, Strukov & Stewart, “Memristive devices for computing”, Nat. Nanotechnol. 2013 (doi:10.1038/nnano.2012.240); Xia & Yang, “Memristive crossbar arrays for brain-inspired computing”, Nat. Mater. 2019 (doi:10.1038/s41563-019-0291-x); Südhof, “Neurotransmitter release: the last millisecond in the life of a synaptic vesicle”, Neuron 2013 (doi:10.1016/j.neuron.2013.10.022); London & Häusser, “Dendritic computation”, Annu. Rev. Neurosci. 2005 (doi:10.1146/annurev.neuro.28.061604.135703).

Yes. Open it in the LabCanvas editor: all 113 layers are editable, including 2 generated objects (structures, plots, equipment) with their own settings. Change the text, colours and layout, then export PNG, TIFF, SVG or PDF.

Not to view it here. To open and edit it, sign in to LabCanvas or create a free account (the free plan keeps up to 3 designs). Downloading the figure is part of LabCanvas Pro (US$19 a month or US$199 a year).