TY - JOUR
T1 - Instability-driven electrohydrodynamic neurons for probabilistic computing
AU - Sun, Miao
AU - Wang, Yongchang
AU - Li, Wenfeng
AU - Gu, Yewen
AU - Feng, Jiajie
AU - Sun, Chao
AU - Xie, Yanbo
N1 - Publisher Copyright:
© The Author(s) 2026.
PY - 2026/12
Y1 - 2026/12
N2 - Fluidic memristors provide a liquid-state alternative to semiconductor hardware for energy-efficient neuromorphic computing. Yet, despite progress in synaptic plasticity, fluidic platforms have lacked a threshold-activated leaky integrate-and-fire (LIF) spiking primitive-the core building block of neuronal computation. Here we introduce an electrohydrodynamic neuron based on a floating liquid bridge, where resistance switching is driven by reversible interfacial instability rather than ionic migration or defect evolution in prior devices. This instability-driven mechanism implements LIF integration with a well-defined activation threshold, generates rhythmic spiking, and exhibits an excitation-refractory cycle. The threshold-integration spike encoding retrieves image content from low-signal-to-noise inputs by converting noisy analog signals into sparse spike events, and its intrinsic fluctuations further enable stochastic firing for probabilistic computation. Together with its enhanced performance, these results establish interfacial hydrodynamics as a device-physics route to fluid-based spiking neuromorphic hardware.
AB - Fluidic memristors provide a liquid-state alternative to semiconductor hardware for energy-efficient neuromorphic computing. Yet, despite progress in synaptic plasticity, fluidic platforms have lacked a threshold-activated leaky integrate-and-fire (LIF) spiking primitive-the core building block of neuronal computation. Here we introduce an electrohydrodynamic neuron based on a floating liquid bridge, where resistance switching is driven by reversible interfacial instability rather than ionic migration or defect evolution in prior devices. This instability-driven mechanism implements LIF integration with a well-defined activation threshold, generates rhythmic spiking, and exhibits an excitation-refractory cycle. The threshold-integration spike encoding retrieves image content from low-signal-to-noise inputs by converting noisy analog signals into sparse spike events, and its intrinsic fluctuations further enable stochastic firing for probabilistic computation. Together with its enhanced performance, these results establish interfacial hydrodynamics as a device-physics route to fluid-based spiking neuromorphic hardware.
UR - https://www.scopus.com/pages/publications/105051148798
U2 - 10.1038/s41467-026-77036-8
DO - 10.1038/s41467-026-77036-8
M3 - 文章
C2 - 42768020
AN - SCOPUS:105051148798
SN - 2041-1723
VL - 17
JO - Nature Communications
JF - Nature Communications
IS - 1
M1 - 10068
ER -