Placing trainable nonlinear functions on connections in analogue networks enables efficient representation of smooth continuous targets with hardware transfer at projected 30 microwatt power.
The missing memristor found.Nature, 453(7191):80–83
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
citation-role summary
background 3
citation-polarity summary
fields
cond-mat.dis-nn 1 cond-mat.mes-hall 1 cond-mat.mtrl-sci 1 cs.AR 1 cs.LG 1 cs.NE 1 physics.app-ph 1 q-bio.QM 1roles
background 3polarities
background 3representative citing papers
A neuron-astrocyte network with dual-timescale memory reduces median path lengths up to sixfold in partially observable grid-world navigation tasks.
BMRUs enable analog recurrent neural network hardware via discrete outputs that suppress noise 20-fold, with one-to-one parameter-to-circuit mapping and linear power scaling for recurrence.
Rigid single-molecule junctions show memristive hysteresis from extrinsic mechanical rearrangements at contacts, not internal molecular states.
Annealing-optimized Ag/HZO memristors demonstrate artificial neurons with TTFS, spike-count, and firing-rate coding modes using minimal circuitry.
Gate-tunable analog memcapacitance is demonstrated in LaAlO3/SrTiO3 interface devices, originating from lateral floating gate charge localization with a supporting model.
Self-organising memristive networks exhibit collective nonlinear dynamics that can support physical learning with parallels to biological plasticity and potential for energy-efficient edge intelligence.