Quantum Computing Modalities: Photonic QC

(For other quantum computing modalities and architectures, see Taxonomy of Quantum Computing: Modalities & Architectures)

What It Is

Photonic quantum computing encodes quantum information in particles of light. A qubit can be the polarization of a single photon (horizontal = |0⟩, vertical = |1⟩), its path through an interferometer (left arm vs. right arm), its time of arrival (early vs. late, called time-bin encoding), or the presence or absence of a photon in a given optical mode (dual-rail encoding, where |0⟩ = photon in rail A, |1⟩ = photon in rail B). The dual-rail approach has become dominant in leading commercial systems because it converts photon loss (the most common error) into a detectable erasure rather than an undetectable bit flip.

Photons have properties that no other qubit carrier shares. They travel at the speed of light, interact minimally with their environment, and operate at room temperature. These properties make photons the natural carrier for quantum communication and quantum networking. But they also create the central challenge of photonic quantum computing: photons do not naturally interact with each other. Two photons can pass through each other without effect. This means two-qubit gates cannot be implemented deterministically by simply crossing two photonic beams. Instead, photonic quantum computers use probabilistic entangling operations (fusion gates), where two photons interfere on a beam splitter and a measurement on auxiliary photons either confirms that entanglement succeeded or flags a failure. The computational architectures that handle this probabilistic gate model, measurement-based quantum computing (MBQC) and fusion-based quantum computing (FBQC), are covered in my separate MBQC modality article.

The practical consequence: photonic quantum computers consume vastly more physical resources per logical operation than matter-qubit systems. A single logical gate on a superconducting or trapped-ion processor involves two physical qubits. A single logical gate in a photonic architecture may require hundreds or thousands of photons, most of which are consumed by the probabilistic gate process and the error correction overhead. The photonic bet is that semiconductor fabrication can produce those photons cheaply enough that the resource overhead is offset by manufacturing throughput.

From a CRQC perspective, photonic quantum computing occupies a unique position. It is the only modality where the core computational elements (integrated photonic circuits) are fabricated at existing semiconductor foundries (GlobalFoundries for PsiQuantum, dedicated facilities for Xanadu). If the architecture works at scale, fab-produced photonic chips could make CRQCs cheaper and more widely accessible than any cryogenic matter-qubit approach. That “if” is the largest open question in the modality.

How It Works

Single-Photon Sources

The first requirement is producing individual photons on demand, in identical quantum states, at high rates. Three source technologies compete:

Quantum-dot single-photon sources. Semiconductor quantum dots (InAs/GaAs, typically cooled to ~4 K) emit single photons when electrically or optically pumped. Each photon is nearly identical to the next (high indistinguishability, >99% demonstrated). Quandela (France) is the leading commercial supplier. Aegiq (UK) also develops quantum-dot sources. The advantage: high brightness and near-deterministic emission. The challenge: each quantum dot is slightly different (frequency tuning required), and the sources must be cryogenically cooled.

Spontaneous parametric down-conversion (SPDC) and spontaneous four-wave mixing (SFWM). A pump laser passes through a nonlinear crystal (SPDC) or a waveguide (SFWM), probabilistically producing pairs of entangled photons. One photon is detected to “herald” the existence of the other. These sources are simpler and operate at room temperature, but the emission is probabilistic: you cannot guarantee a photon will be produced on each clock cycle. Multiplexing (running many sources in parallel and routing the first successful emission) improves the effective rate but adds optical switching overhead. This is the most common source type in research labs.

GKP (Gottesman-Kitaev-Preskill) states. GKP states are not a single-photon source technology. They are an error-correctable bosonic encoding that represents a qubit in the continuous-variable properties of light (amplitude and phase quadratures of a squeezed vacuum state). This is the encoding used by Xanadu‘s Aurora platform and by Nord Quantique‘s microwave-photonic approach. GKP states have built-in error-correction properties: small displacements in phase space can be detected and corrected without consuming the qubit. The challenge is producing high-quality GKP states reliably. China’s Jiuzhang 4.0 (USTC, 2025) used 108 squeezed-state inputs for Gaussian boson sampling.

Integrated Photonic Circuits

The photonic processor is an integrated optical circuit, fabricated on a chip using semiconductor lithography. Waveguides carry photons. Beam splitters (directional couplers) mix photonic modes. Phase shifters (thermo-optic or electro-optic) apply controlled phase rotations. Together, these components implement arbitrary linear-optical transformations on the photonic modes.

The materials platforms:

Silicon photonics. Standard CMOS-compatible fabrication on silicon-on-insulator (SOI) wafers. Compact waveguides (~500 nm wide), high density, compatible with existing foundry processes. PsiQuantum fabricates its Omega chipset at GlobalFoundries’ 300 mm fab in Malta, NY using silicon photonics with barium titanate (BTO) electro-optic switches.

Thin-film lithium niobate (TFLN). Strong electro-optic effect enables ultra-fast switching (GHz-speed modulators) with low optical loss. QuiX Quantum (Netherlands) builds programmable photonic processors on TFLN. The platform achieved the first below-threshold error mitigation in photonic quantum computing in April 2026.

III-V semiconductors (InP, GaAs). These host quantum-dot single-photon sources and can integrate sources with waveguides and detectors on the same chip. Quandela‘s MosaiQ platform integrates quantum-dot sources with reconfigurable photonic circuits.

Entangling Gates: The Probabilistic Challenge

The KLM theorem (Knill, Laflamme, Milburn, 2001) proved that universal quantum computing is possible with linear optics (beam splitters and phase shifters) and single-photon detection, without any nonlinear photon-photon interaction. The mechanism: two photons enter a beam splitter simultaneously. Quantum interference (the Hong-Ou-Mandel effect, where identical photons arriving simultaneously always exit together rather than separately) creates correlations between the output modes. Detecting auxiliary photons in specific patterns confirms that an entangling operation succeeded.

The catch: these operations are probabilistic. A basic KLM CNOT gate succeeds with probability ~1/16. The rest of the time, it fails in a detectable way (an “erasure”). This means photonic quantum computers must either repeat failed operations or use error-correction codes designed for erasure-dominated noise.

PsiQuantum‘s fusion-based quantum computing (FBQC) architecture addresses this by generating small entangled photonic “resource states” and then performing fusion measurements (Type-II fusions) that probabilistically connect them into a larger entangled structure. Failed fusions are detected and handled by the error-correction code, which is specifically designed to tolerate high erasure rates. PsiQuantum’s Omega chipset (February 2025, Nature) demonstrated 99.98% state preparation and measurement, 99.5% Hong-Ou-Mandel visibility, 99.72% chip-to-chip fidelity, and 99.22% two-qubit fusion gate fidelity, all fabricated at GlobalFoundries.

Single-Photon Detectors

Photonic quantum computing requires counting individual photons with high efficiency (>90%), low dark counts (false detections), and precise timing resolution (<100 ps). The technology of choice is the superconducting nanowire single-photon detector (SNSPD): a thin superconducting wire cooled to 1–4 K that absorbs a photon, momentarily loses superconductivity at the absorption point, and produces a measurable voltage pulse.

SNSPD suppliers include Photon Spot (US), Single Quantum (Netherlands), Quantum Opus (US), and ID Quantique (Switzerland). Detection efficiency exceeds 95% at telecom wavelengths (1550 nm). The cryogenic requirement for SNSPDs is the only part of a photonic quantum computer that needs cooling, and it operates at 1–4 K (simpler than the 10 mK dilution refrigerators required for superconducting qubits), typically using compact cryogen-free Cryomech or Bluefors systems. PsiQuantum uses cuboid cryogenic racks fed by industrial cryoplants, qualitatively different from the dilution-fridge model.

Transition-edge sensors (TES) are an alternative detector technology that can resolve the exact number of photons in a pulse (photon-number resolution), which is useful for certain error-correction and calibration protocols. TES detectors operate at ~50–100 mK and are used in some research setups but are less common in commercial systems.

Key Academic Papers

KLM (Knill, Laflamme, Milburn, 2001). “A scheme for efficient quantum computation with linear optics.” Proved that universal quantum computing is possible with only linear optics and single-photon detection, using probabilistic gates with ancilla photons and post-selection. Published in Nature.

Raussendorf & Briegel (2001). Introduced measurement-based quantum computing (one-way quantum computing) using cluster states. This architecture became the dominant paradigm for photonic QC because it separates entanglement generation (preparing the cluster state) from computation (measurements), fitting naturally with photonic platforms where entanglement is probabilistic but measurement is deterministic.

O’Brien et al. (2003). First experimental demonstration of a photonic CNOT gate using two photons and linear optics. ~81% fidelity. Published in Nature.

Jiuzhang (USTC, December 2020). Gaussian boson sampling with 76 detected photons from 100 squeezed-state inputs, claiming photonic quantum advantage over classical supercomputers. Published in Science. Jiuzhang 3.0 (2023) scaled to 255 detected photons. Jiuzhang 4.0 (2025) used 108 squeezed-state inputs.

PsiQuantum Omega (February 2025, Nature). Demonstrated the key metrics for a fault-tolerant photonic architecture: 99.98% SPAM, 99.72% chip-to-chip fidelity, 99.22% fusion gate fidelity, fabricated at GlobalFoundries 300 mm fab. My analysis.

QuiX Quantum below-threshold (April 2026). First below-threshold error mitigation in photonic quantum computing on a thin-film lithium niobate processor. My analysis.

The Vendor Landscape (May 2026)

Major Vendors

PsiQuantum (Palo Alto, USA). The most heavily funded photonic quantum computing company. PsiQuantum’s approach: fusion-based quantum computing using dual-rail photonic qubits on silicon photonic chips with BTO electro-optic switches, fabricated at GlobalFoundries’ 300 mm fab. The Omega chipset (Nature, February 2025) demonstrated the component-level fidelities needed for fault tolerance. $1 billion Series E (September 2025) led by BlackRock, Temasek, Baillie Gifford, NVIDIA NVentures, Macquarie, and Qatar. PsiQuantum is building million-qubit photonic compute centers in Brisbane (Australia, with AUD $940M Australian and Queensland government funding) and Chicago (USA). The company operates at 2–4 K (for SNSPDs), not millikelvin.

PsiQuantum’s thesis is uniquely aggressive: skip the incremental NISQ era entirely and build fault-tolerant machines from the start, using semiconductor manufacturing to produce the enormous photonic resource overhead that fusion-based architectures require. If the thesis holds, PsiQuantum could be the first to reach a million physical qubits, but the per-qubit utility (after probabilistic gate and error-correction overhead) may be far lower than a million matter qubits. Whether PsiQuantum reaches fault-tolerant operation before superconducting or trapped-ion approaches is the defining question for the modality.

Xanadu (Toronto, Canada). Uses squeezed-state photonic qubits and GKP encoding on a silicon photonics platform. Xanadu’s Aurora processor is operational, and its Borealis system (accessible on cloud) demonstrated quantum advantage in Gaussian boson sampling. Xanadu develops PennyLane, a widely used quantum ML framework that is hardware-agnostic. The GKP encoding approach has built-in error-correction properties that could reduce overhead compared to dual-rail encoding, but generating high-fidelity GKP states at scale remains technically challenging.

Quandela (Paris, France). The leading quantum-dot single-photon source company. Quandela’s MosaiQ platform integrates quantum-dot photon sources with reconfigurable photonic circuits, producing near-deterministic, highly indistinguishable photons. NVQLink partner. The Belenos system targets 30+ photonic qubits. Quandela has shipped systems to research institutions and is available on OVHcloud.

ORCA Computing (London, UK). Uses time-bin encoding in quantum memories to store and synchronize photons. ORCA’s PT-1 and PT-2 systems are installed at the UK NQCC and at Israel’s IQCC quantum data center. ORCA targets near-term applications in machine learning and optimization using boson sampling and photonic interference.

QuiX Quantum (Enschede, Netherlands). Builds programmable photonic processors on thin-film lithium niobate (TFLN). Achieved the first below-threshold error mitigation in photonic quantum computing (April 2026), a significant milestone. QuiX positions itself as a photonic QPU component supplier, the closest analogue to an open-architecture photonic vendor.

Other Vendors

Photonic Inc. (Toronto, Canada). Silicon T-center color-center photonic networking. T-centers emit telecom-band photons for fiber-based distributed quantum computing. Focused on quantum networking and multi-node architectures rather than standalone compute.

QCI (Quantum Computing Inc.) (USA). Dual-rail bosonic qubits; NVQLink partner.

USTC / Jiuzhang (China). University of Science and Technology of China’s Jiuzhang series demonstrated photonic quantum advantage in boson sampling: 76 photons in 2020, 255 photons in 2023, and 108 squeezed-state inputs in 2025. These are sampling demonstrations, not gate-model quantum computers, but they proved that photonic systems can outperform classical supercomputers on specific tasks.

The Fabrication Advantage

The defining strategic feature of photonic quantum computing is that the core computational components (integrated photonic circuits) can be manufactured at existing semiconductor foundries. PsiQuantum at GlobalFoundries. Xanadu at its own silicon photonics facility. QuiX on TFLN. This is a property shared only with silicon spin qubits among the major modalities. Superconducting circuits require specialized quantum fabs (QuantWare’s KiloFab, IBM’s Fishkill/Poughkeepsie). Trapped-ion and neutral-atom systems require precision vacuum and optics engineering that does not leverage foundry infrastructure.

PsiQuantum’s argument is that this fabrication advantage is decisive at scale: once you need millions of physical qubits (which photonic architectures do, due to probabilistic gate overhead), only a semiconductor fab can produce them at acceptable cost and yield. The counter-argument is that if other modalities achieve fault tolerance with thousands rather than millions of physical qubits, the fabrication advantage becomes irrelevant because the total resource requirement is smaller.

Comparison to Other Modalities

Photonic vs. Superconducting

The starkest contrast among quantum computing modalities. Superconducting qubits are matter-based, deterministic, and cryogenic (10 mK). Photonic qubits are light-based, probabilistic, and mostly room-temperature (with 1–4 K detectors). Superconducting systems have demonstrated below-threshold QEC (Google Willow) and operate at MHz cycle rates. Photonic systems have demonstrated component-level fidelities (PsiQuantum Omega) but have not yet run an error-corrected logical circuit.

The fundamental tradeoff: superconducting gates are deterministic but the qubits are expensive (specialized fabrication, dilution refrigerators, helium-3). Photonic gates are probabilistic but the qubits are cheap (semiconductor-fab photonic chips, room-temperature operation). At small scale, deterministic gates win. At million-qubit scale (if photonic architectures require it), fab economics may dominate. The question is whether superconducting systems reach fault tolerance with ~10,000 physical qubits (via qLDPC codes) before photonic systems reach it with ~1,000,000 photonic qubits (via FBQC). As I discussed in the superconducting modality article, the error-corrected operations per second is the metric that will resolve this.

Photonic vs. Trapped Ion

Trapped ions achieve the highest gate fidelities in quantum computing (99.921% on Quantinuum Helios) with deterministic, all-to-all connectivity. Photonic systems cannot match these per-gate metrics. But trapped-ion systems scale by shuttling ions through complex trap architectures or connecting modules via photonic links, and both paths introduce overhead. The irony is that trapped-ion scaling may end up depending on photonic interconnects, making the two modalities complementary rather than competitive at the architectural level. IonQ‘s photonic-link quantum networking program and Quantinuum’s Lightsynq efforts reflect this convergence.

Photonic vs. Neutral Atom

Both operate at room temperature (for the core processor) and both benefit from erasure-dominated error models. Neutral atoms perform deterministic Rydberg gates with near-unity success probability per attempt; photonic fusion gates succeed probabilistically (~75% for optimized Type-II fusions, lower for basic approaches). Neutral atoms have demonstrated 96 logical qubits on 448 physical atoms; photonic systems have not yet demonstrated a logical qubit. The comparison favors neutral atoms for near-term fault tolerance, but photonic fabrication throughput could dominate at very large scale.

Photonic vs. Silicon Spin

As I discussed in the silicon spin article, both claim semiconductor-fab compatibility but through different physics. Silicon spin produces deterministic gates on cryogenic electron-spin qubits; photonic produces probabilistic gates on room-temperature photons. Silicon spin favors compute density per chip area; photonic favors total throughput at scale. Both modalities are betting that foundry economics will ultimately determine the winner in quantum computing, but they arrive at different architectures from that same premise.

Advantages

Room-temperature core processor. The integrated photonic circuits that perform computation operate at room temperature. Only the single-photon detectors (SNSPDs) require cooling to 1–4 K, far simpler than the 10 mK dilution refrigerators needed for superconducting or silicon spin qubits. No helium-3 required for the core system.

Semiconductor fabrication. Photonic circuits can be manufactured at existing 300 mm semiconductor foundries (GlobalFoundries for PsiQuantum). This provides access to the most advanced lithography, process control, and yield engineering infrastructure on Earth. If photonic quantum computing works at scale, the cost per photonic qubit could follow semiconductor cost curves.

Native quantum networking. Photons travel through optical fiber at the speed of light. Photonic quantum computers are natively compatible with quantum communication networks. Connecting two photonic processors requires fiber, not the microwave-to-optical transducers that superconducting systems would need. This makes photonic systems the natural platform for distributed quantum computing.

Erasure-dominated error model. Photon loss (the dominant error) is detectable: a missing photon is noticed by the detector. This converts the primary error channel into a known erasure, which is cheaper to correct than unknown errors. Dual-rail encoding and erasure-based QEC codes exploit this structure.

Clock speed. Photonic gates operate at GHz rates (limited by the speed of electro-optic switches, ~10 GHz for BTO modulators). This is comparable to or faster than superconducting gate speeds and far faster than trapped-ion or neutral-atom gates.

Disadvantages

Probabilistic gates. The fundamental limitation. Two-qubit entangling operations succeed with probability <1 (even optimized Type-II fusions have ~75% success rate). Failed attempts consume photons and time. This means photonic architectures require orders of magnitude more physical resources per logical gate than deterministic-gate modalities. The entire FBQC/MBQC architectural stack exists to manage this overhead.

No quantum memory. Photons travel at the speed of light and cannot be stored indefinitely. Once a photon is generated, it must be used within nanoseconds or it is lost. This means photonic quantum computers cannot “park” qubits while waiting for other operations, the way trapped ions (seconds of coherence) or neutral atoms (40 seconds on Atom Computing) can. Quantum memories (optical delay lines, fiber loops, atomic ensembles) partially address this but add complexity and loss.

No demonstrated logical qubit. As of May 2026, no photonic system has demonstrated an error-corrected logical qubit. PsiQuantum’s Omega demonstrated component-level fidelities sufficient for fault tolerance in principle, and QuiX demonstrated below-threshold error mitigation, but neither has assembled a complete logical qubit from photonic resources. Superconducting, trapped-ion, and neutral-atom systems have all demonstrated logical qubits.

Photon loss. Every optical component (waveguide, coupler, phase shifter, fiber connection) introduces some photon loss. Cumulative loss through a deep circuit degrades computation quality. While erasure detection mitigates this, high loss rates still consume error-correction budget. Reducing per-component loss is a continuous materials and fabrication challenge.

Vertically integrated supply chain. PsiQuantum fabricates at GlobalFoundries but does not sell chips to external integrators. Xanadu operates its own facility. Quandela sells photon sources but not complete quantum computers as open-architecture components. The photonic QOA ecosystem is less mature than the superconducting component market. QuiX Quantum’s programmable photonic processor is the closest thing to a standalone photonic QPU component, but the broader integration stack (sources, detectors, control electronics, error correction) remains vendor-specific. I discuss what exists for integrators in my deep dive on building a photonic quantum computer.

Impact on Cybersecurity

Photonic quantum computing introduces a distinctive CRQC risk profile. If PsiQuantum or a successor builds a million-qubit photonic quantum computer at a semiconductor foundry, the manufacturing economics change the proliferation picture. A photonic CRQC chip produced at GlobalFoundries could, in theory, be manufactured in volume and sold, rather than existing as a one-of-a-kind installation in a national laboratory. This is the same proliferation argument I raised for silicon spin qubits, but photonic systems arrive at it from a different direction: silicon spin reduces cost through qubit density; photonic reduces cost through fab-compatible manufacturing.

The resource overhead is the key unknown. The Gidney 2025 estimate of ~1,400 logical qubits for RSA-2048 translates to a much larger number of physical photonic qubits than physical matter qubits, because of the probabilistic gate overhead. Exact resource estimates for photonic FBQC architectures breaking RSA-2048 are not publicly available at the level of detail that exists for superconducting surface-code architectures. PsiQuantum’s internal estimates presumably exist but are not published. The total photonic resource count is likely in the tens of millions of photonic modes, but the critical variable is whether the fab can produce them.

The practical response is unchanged: PQC migration now. The photonic path to a CRQC is longer and less certain than the superconducting or trapped-ion paths, but its manufacturing scalability means that if it works, CRQCs could proliferate faster than in a world where every machine requires a custom dilution-refrigerator installation. Regulators, insurers, and clients are setting deadlines that don’t wait for any modality to resolve its engineering challenges.

Future Outlook

2026–2027. PsiQuantum’s Brisbane and Chicago compute centers begin commissioning. The first multi-chip photonic systems with integrated sources, circuits, and detectors are assembled and tested at scale. QuiX Quantum pushes toward a logical qubit demonstration on TFLN. Xanadu advances GKP-encoded fault-tolerant protocols. Quandela scales MosaiQ toward 30+ photonic qubits. The central question for this period: can PsiQuantum demonstrate a fault-tolerant logical qubit using FBQC? If yes, the photonic architecture validates at the most fundamental level.

2028–2030. PsiQuantum targets operational fault-tolerant quantum computing at its compute centers. If the FBQC architecture works, the company begins demonstrating useful computations that justify the $1B+ investment. Xanadu‘s GKP approach matures as an alternative photonic encoding. Photonic-to-matter-qubit hybrid architectures (photonic interconnects connecting trapped-ion or superconducting modules) begin to blur the modality boundaries.

Beyond 2030. If photonic fault tolerance is achieved, semiconductor manufacturing throughput could push photonic quantum computers to scales (millions of photonic modes) that no cryogenic matter-qubit approach can match economically. The photonic CRQC becomes a plausible manufacturing proposition rather than a laboratory achievement. If photonic fault tolerance proves too resource-intensive, the modality may settle into a complementary role: photonic interconnects for quantum networking and distributed quantum computing, with matter qubits doing the local computation.

The photonic bet is the most binary in quantum computing. Either fab-scale manufacturing overcomes the probabilistic-gate overhead and photonics produces the largest quantum computers on Earth, or the overhead proves too high and the modality remains confined to quantum communication and sensing. There is less middle ground here than in any other modality. The next three years will determine which outcome prevails.