neuromorphic-computing

Layer 1 — Physics24 concepts in this subtree

Device-level analog / mixed-signal computation modelled on biological neural circuits. Canonical substrate elements: leaky-integrate-and-fire (LIF) neurons — Lapicque 1907, Stein 1965 stochastic generalisation; spike-timing-dependent…

LIF neuron: τ_m·dV/dt = -(V-V_rest) + R·I(t); fire when V≥V_th, reset to V_r
STDP: Δw = A₊ exp(-Δt/τ₊) for Δt>0 (LTP), -A₋ exp(Δt/τ₋) for Δt<0 (LTD)
Memristor: V = M(q)·i; HP 2008 model dx/dt = μ_V·R_on·i/D², 0≤x≤1
LIF firing rate: r(I) = 1/(τ_ref + τ_m·ln((R·I)/(R·I − (V_th-V_rest)))) for suprathreshold
STDP integral: ∫₀^∞ A₊ exp(-Δt/τ₊) dΔt = A₊·τ₊; ∫_{-∞}^0 A₋ exp(Δt/τ₋) dΔt = A₋·τ₋
HP memristor I-V loop: V(t) = (R_on x + R_off(1-x))·i(t), pinched at origin
Hopfield storage capacity via Hoeffding concentration on pattern overlaps
Echo state property: spectral-radius rho(W) < 1 via Perron-Frobenius on |W|
Oja rule as PCA power iteration: w(t+1) = w(t) + eta*(y*x - y^2*w); w_inf = first eigenvector
Hopfield pattern-overlap concentration: P(|m-E[m]|>=1/2) <= 2*exp(-N/2)
ESN spectral-radius verification: W = [[0, 1/2], [1/2, 0]] has rho = 1/2 < 1 (ESP holds)
Oja convergence: C = diag(2, 1) -> w_inf = (+-1, 0); power-iteration fixed-point
Memristor (Chua 1971)
SNN (Maass 1997)
PCM (Ovshinsky 1968)
Crossbar VMM
Loihi (Intel 2018)
DVS (Lichtsteiner 2008)
Memristor (Chua 1971)
STDP (Bi-Poo 1998)
Hodgkin-Huxley (1952)
TrueNorth (Merolla 2014)
Mead (1989)
LSM (Maass 2002)
Explore the neuromorphic-computing subtree on the interactive graph →