Photonic Quantum Memristors

• Photonic quantum memristors are optical devices that couple quantum photon states with memory feedback to emulate hysteresis behavior in photonic circuits.
• They use measurement-based feedback to modulate beam splitter reflectivity, enabling non-Markovian dynamics and advanced neuromorphic architectures.
• Implemented in diverse platforms, they preserve coherence while offering scalable potential for quantum reservoir computing and neural network applications.

Photonic quantum memristors (PQMs) are optical circuit elements that bring the history-dependent nonlinearity of classical memristors into the quantum photonic regime. By coupling the coherent quantum evolution of photons to feedback-modulated dissipation, they enable explicitly non-Markovian quantum information processing. PQMs implement a tunable beam splitter or interferometer whose transmissivity or reflectivity is determined in real time by the measurement record of photon numbers in an ancillary output, inducing device dynamics that depend on the full input history. This configuration preserves quantum coherence and allows the direct emulation of memristive (hysteresis-bearing) behavior with quantum states of light, making PQMs central to architectures for neuromorphic quantum computing, photonic quantum reservoirs, and quantum neural networks (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Ferrara et al., 2024, Lim et al., 30 Jan 2026).

1. Theoretical Foundations and Operational Principles

The PQM is motivated by the memristor paradigm, in which output current and voltage depend not only on the instantaneous input but also on an internal state variable updated by the input's history: $$V(t) = R[s(t)]\,I(t),\qquad \dot{s}(t) = h\bigl(s(t), I(t), t\bigr)$$ where $s(t)$ is a state variable integrating the stimulus over time (Micco et al., 4 Mar 2025).

In PQMs, the electrical quantities are replaced with photonic observables. The system begins with an optical mode in a pure or mixed quantum state, typically a vacuum–one-photon superposition (single-rail qubit) $$\ket{\psi_\mathrm{in}} = \alpha(t)\ket{0} + \beta(t)\ket{1}$$, with mean photon number $\langle n_\mathrm{in}(t)\rangle = |\beta(t)|^2$ (Micco et al., 4 Mar 2025, Ferrara et al., 2024). This state passes through a beam splitter with reflectivity $R(t)$ set by a control parameter. The output photon numbers in the various arms are measured; specifically, the number of photons in a designated "feedback" arm is used to update $R(t)$ via

$$\dot{R}(t) = \langle n_\mathrm{in}(t)\rangle - \frac{1}{2}$$

or, equivalently,

$$R(t) = \frac{1}{2} + \frac{1}{T}\int_{t-T}^t \bigl(\langle n_\mathrm{in}(t')\rangle - \frac{1}{2}\bigr)dt'$$

where $T$ is a memory integration time, establishing non-Markovian, history-dependent response (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Ferrara et al., 2024). The device thus realizes direct quantum-analogs of memristive state equations.

Generalizations to dual-rail and frequency-bin encodings are possible, as are implementations where the "resistance" update is controlled by either coherent, squeezed, or entangled photonic input states (Gonzalez-Raya et al., 2019, Sanz et al., 2017). The memristive feedback loop in all cases produces characteristic pinched hysteresis in plots of output versus input observables.

2. Experimental Implementations and Architectures

Realized PQMs have been reported in multiple photonic platforms. One approach utilizes bulk or integrated interferometers with tunable beam splitters or Mach–Zehnder interferometers, in which the phase (and hence reflectivity) is continuously modulated by a liquid-crystal, thermo-optic, or electro-optic phase shifter. Feedback is provided by photon-number (or equivalently, count-rate) measurements in a designated port, using fast photodetectors or superconducting nanowire single-photon detectors (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Lim et al., 30 Jan 2026).

The system is initialized by preparing single-rail Fock-basis superpositions using a quantum-dot single-photon source, with conditional purity $\mathcal{P} \approx 0.95$ (Micco et al., 4 Mar 2025). The adjustable phase is controlled so the beam-splitter reflectivity adapts according to the measured photon-number over an integration window, as described by the feedback law above.

More elaborate circuits use multiple cascaded PQMs, with each element possessing its own distinct memory kernel and feedback loop. In networked architectures, memory sharing is achieved by coupling each PQM node's update not only to its own measurement history but also to that of adjacent nodes, implementing a photonic quantum "memtransistor" network (Lim et al., 30 Jan 2026). Frequency-bin PQMs exploit electro-optic pulse shapers and mixers to construct tunable frequency-domain beam splitters with similar real-time feedback (Gonzalez-Raya et al., 2019).

Table 1: Architectures of Experimentally Realized PQMs

Platform	Feedback Mechanism	Key Components
Bulk optics (liquid-crystal, QDs)	Photon counting + LC	Single-rail qubits, tunable BS, feedback
Integrated photonics (MZI, GLASS)	Thermal/electro-optic	Dual-rail encoding, thermal phase shifters
On-chip MZI network	Multiplexed detectors	Multiple PQMs, memory-sharing, photonic RC
Frequency-bin encoder	Spectral detection	EOM–pulse shaper–EOM sequence

3. Nonlinear, Memory-Dependent Quantum Dynamics

The core operational hallmark is the emergence of memristive hysteresis at the quantum level. When driven with periodic modulation of the photon number, the PQM output maintains a history-dependent relationship with the input. For single-rail qubits with input $$\langle n_\mathrm{in}(t)\rangle = \sin^2\bigl(\pi t / T_\mathrm{osc}\bigr)$$, the hysteresis loop observed in $\langle n_\mathrm{out}\rangle$ versus $\langle n_\mathrm{in}\rangle$ is quadratic for short memory ($T \ll T_\mathrm{osc}$), and collapses to linear for long memory ($T \gtrsim T_\mathrm{osc}$), explicitly verifying Chua's memristive criteria.

For dual-rail or frequency-bin encodings, analogous loop areas quantify the classical and quantum nonlinear response. PQMs display "pinched" loops characteristic of true memristors. The area of these loops serves as a metric for memory depth and non-Markovianity (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Ferrara et al., 2024, Gonzalez-Raya et al., 2019).

Quantum coherence and entanglement are preserved by PQMs under these dynamics. Experimentally, conditional purity of the output state after one or two PQM stages remains near unity ($\mathcal{P}_\mathrm{single} = 0.99 \pm 0.01$), confirming that the feedback-induced loss is incoherent and does not degrade the superposition (Micco et al., 4 Mar 2025, Spagnolo et al., 2021). Networks of independent PQMs allow tracking of memory-dependent coherence and entanglement, both analytically and experimentally, as quantified by the $l_1$-norm of coherence and Wootters concurrence (Ferrara et al., 2024).

4. Large-Scale Networks: Reservoir and Neuromorphic Architectures

PQMs are key elements for photonic quantum reservoir computing, extreme learning machines, and neuromorphic architectures. Cascading multiple PQMs yields nontrivial emergent phenomena; for example, a two-PQM cascade with distinct time constants enables richer, asymmetric hysteresis and effective memory separation in the network response (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Lim et al., 30 Jan 2026).

Distributed memory sharing in PQM networks is realized by setting each node's internal state update as a function of both its own and neighboring nodes’ measurement histories, for example via a smooth "gating" function of photon-count histories. This architecture, where each PQM functions as a memtransistor (i.e., a three-port device with both local and nonlocal memory dependence), increases both the classical and quantum hysteresis and enhances the nonlinear separability of reservoir states (Lim et al., 30 Jan 2026).

Concrete benchmarking on image-classification tasks shows that PQM-based photonic reservoirs attain higher accuracy ($82\%$ for three-class Fashion-MNIST) and confidence than uncoupled or memoryless photonic reservoirs. Pinched hysteresis areas and coherence-loop areas serve as operational witnesses of scalable, network-level non-Markovianity. PQMs remain passive linear-optical devices with all nonlinearity and memory implemented at the quantum level through measurement-based feedback.

5. Alternative Photonic Memristive Devices and Non-Volatile Memory

Variants of PQMs include devices where electrical or optoelectronic memristive effects are combined with photonics. The heterogeneously integrated Al$_2$O$_3$ memristive quantum-dot microring laser realizes nonvolatile optical memory by using oxygen-vacancy filaments to modulate refractive index states, resulting in persistent wavelength-encoded logic. SET/RESET voltage pulses switch the device between high- and low-resistance, which correspond to distinct emission wavelengths. Features include nanosecond switching, endurance $>10^7$ cycles, and compatibility with photonic integrated circuits (Tossoun et al., 2024). While not functionally equivalent to measurement-based PQMs in terms of input–output memristive quantum maps, such devices advance nonvolatile optical memory and hybrid opto-electronic synapses for neuromorphic photonic circuits.

6. Limitations, Performance Metrics, and Future Directions

Current limitations for experimental PQMs include bulk-optics implementation (slow phase shifters, large footprint), limited feedback bandwidth ($\tau_\mathrm{latency}\sim 200$ ms to 4 s per step for certain bulk setups), and sensitivity to optical loss, especially for single-rail quantum encoding. Integrated photonic platforms promise scalable, low-loss, high-speed versions using electro-optic or MEMS phase shifters (Micco et al., 4 Mar 2025, Spagnolo et al., 2021). Current insertion loss for basic units is $\lesssim2$ dB, and coherence loss is negligible.

Prospective research directions focus on (i) integrating PQMs on lithium-niobate or silicon-photonics for GHz-rate feedback, (ii) exploring active networks with genuine quantum memory sharing and entanglement between memristors (Lim et al., 30 Jan 2026, Ferrara et al., 2024), (iii) interfacing PQMs with standard linear quantum circuits or variational quantum algorithms, and (iv) developing hybrid quantum-classical architectures that leverage PQM dynamics for enhanced quantum machine learning tasks.

7. Summary and Outlook

Photonic quantum memristors realize nonlinear, memory-dependent, and non-Markovian transformations of quantum optical states via tunable measurement-based feedback. They preserve quantum coherence, exhibit hysteresis both in classical observables and quantum coherence/entanglement, and scale efficiently to large reconfigurable networks suitable for reservoir computing and neuromorphic processing. Their theoretical structure, experimental feasibility, and operational performance are now firmly established across several groups and platforms (Micco et al., 4 Mar 2025, Spagnolo et al., 2021, Lim et al., 30 Jan 2026, Ferrara et al., 2024, Gonzalez-Raya et al., 2019, Sanz et al., 2017, Tossoun et al., 2024). Challenges remain in integration, speed, and controlling network-level dynamics, but PQMs are emerging as foundational components for future quantum-aware photonic information processing systems.