← SCPN-Quantum-Control

Quantum neuromorphic

A spiking neuron and a qubit turn out to fit together. Take the leaky integrate-and-fire neuron of neuromorphic computing — a membrane that charges toward a threshold and fires — and put its decision on a qubit: encode the membrane voltage as a rotation angle, and let the spike be a measurement. The integration stays classical and smooth; the firing becomes genuinely probabilistic, a quantum coin weighted by how close the membrane is to threshold. Run one below.

A membrane, a rotation, a measurement

The membrane charges and leaks by the usual Euler step. Its distance to threshold is mapped to a rotation angle $\theta$ from $0$ to $\pi$, applied as an $R_y$ gate to a qubit that starts in $\lvert 0\rangle$. Measuring in the computational basis gives a spike with probability $\sin^2(\theta/2)$ — zero at rest, one at threshold, and a smooth quantum ramp in between. Fire, and the membrane resets. It is the classical stochastic LIF neuron, re-expressed as a single-qubit circuit.

$$v \leftarrow v - \tfrac{\Delta t}{\tau}\,(v-v_{\rm rest}) + R\,I\,\Delta t,\qquad \theta=\pi\,\mathrm{clip}\!\Big(\tfrac{v-v_{\rm rest}}{v_{\rm th}-v_{\rm rest}},0,1\Big),\qquad P(\text{spike})=\sin^2\!\tfrac{\theta}{2}$$
Live — the library's quantum LIF neuron
membrane v(t) threshold spike
membrane v: — rotation θ: — P(spike): — rate: —

The left panel is the membrane charging toward threshold and resetting on each spike; the raster ticks along the top. The right panel is the firing law itself: the quantum probability $\sin^2(\theta/2)$ rises smoothly from rest to threshold, where a classical neuron would have a hard step. In deterministic mode the neuron fires once that probability passes a half; switch to stochastic and each step is an actual weighted coin flip — the same neuron, now genuinely noisy near threshold, which is what a real measured qubit would give.

Deeper: why put a neuron on a qubit, and what is claimed
The point is not speed — a single classical neuron is trivial — but a shared substrate: once the neuron is a parametrised circuit, a whole spiking layer is a circuit, its synapses are gate parameters, and it can be trained by the same parameter-shift gradient the variational algorithms use, or interfaced with the quantum dynamics the rest of the portal runs. The library's quantum-neuromorphic bridge carries an explicit claim boundary: it maps the classical membrane dynamics faithfully — this panel's membrane trace and firing probability match the library's quantum LIF to machine precision — but a genuine quantum advantage for spiking networks is an open research question, not a delivered result. The stochastic firing shown here is the honest consequence of measurement, not a claim of speedup.
Where it fits

This is the bridge between two of the institute's substrates — the neuromorphic stack, where the same leaky neuron is a stochastic bitstream, and the quantum-control stack here, where it is a qubit circuit. The membrane, threshold and reset are identical; only the firing mechanism differs, a measurement in one and a bitstream comparison in the other. Learning stays local through a quantum spike-timing rule, and training reuses the parameter-shift gradient from the variational side.

PartClassical neuromorphicQuantum here
Membraneleaky Euler integrationidentical leaky Euler
Firingthreshold / bitstream compare$R_y$ rotation + measurement, $P=\sin^2(\theta/2)$
LearningSTDPquantum STDP (local)
Trainingsurrogate gradientparameter-shift gradient
Parameter-shift training Applications Neuromorphic neuron models Back to overview

Evidence boundary: the panel runs the library's quantum LIF neuron exactly — the membrane Euler step, the $R_y$ angle map and the $\sin^2(\theta/2)$ firing probability — live in your browser, its deterministic trajectory checked against the library to machine precision. It demonstrates the neuromorphic-to-qubit mapping, not a quantum speedup for spiking networks, which the library flags as an open question.