Photonic quantum computing builds qubits from light rather than matter, using single photons or squeezed light in integrated optical circuits; component-level progress through 2026 is rapid, but a mid-2026 analysis shows the leading all-linear-optical architecture cannot reach useful logical error rates at low overhead without adding matter qubits.
Summary
Photonic quantum computing encodes quantum information in light: either in discrete single photons (dual-rail or similar encodings) or in continuous-variable modes such as squeezed states and Gottesman-Kitaev-Preskill (GKP) grid states. Because photons barely interact, entangling operations are not deterministic gates but measurements. The dominant design pattern is fusion-based (or measurement-based) computing: small entangled resource states are generated, then stitched together by interferometric “fusion” measurements, with the computation defined by the pattern of measurement outcomes and adaptive feedforward. Fusion-based architectures are described in the current literature as the leading approach to photonic quantum computing.
The parameters that decide the platform are therefore not qubit counts but component-level numbers: photon source efficiency and indistinguishability, waveguide and modulator insertion loss, detector quantum efficiency, and above all fusion success probability. Without ancillary resources, type-I and type-II fusion gates succeed only half the time; existing 3/4-efficient schemes need Bell-pair ancillas that are themselves probabilistic to prepare. Every failed fusion is a lost link in the entangled lattice, so the whole architecture is an exercise in beating down loss and failure probability with redundancy.
The most consequential recent result is negative. In the sub-threshold regime, where logical error rates must reach levels useful applications demand, fusion failure imposes a noise floor on logical error that prevents all-linear-optics architectures from reaching the required rates at low overhead; introducing quantum-emitter spins lowers that floor by orders of magnitude at lower overhead. That pushes the field towards hybrid designs pairing stationary matter qubits with flying photons, which are argued to give an intrinsically networked and resource-efficient route to error-corrected computation.
Evidence base
- 30 June 2026: fusion failure imposes a noise floor on logical error rate that prevents all-linear-optics fusion-based architectures from reaching application-required rates at low overhead; emitter-spin variants lower the floor by orders of magnitude at lower overhead.
- 5 May 2026: unboosted type-I and type-II fusion gates are limited to 1/2 success; a new scheme reaches 3/4 total (5/8 direct plus distillation) using four single-photon ancillas and passive linear optics.
- 3 June 2026: FPGA feedforward system with fibre homodyne detection demonstrates 15 dB clearance at 1 GHz with a 4 mW local oscillator, greater than 95% quantum efficiency and 196 ns total system latency.
- 17 August 2026: on a 13-factor equity library across a 164-month test window with 48 hyperparameter configurations, a photonic quantum annealer found superior risk-return topologies only in a narrow operating range while classical mixed-integer programming remained superior for risk-constrained mandates.
- 4 May 2026: cryogenic operation of dual-layer graphene phase modulators on silicon nitride is being pursued precisely because fully cryogenic integrated photonic platforms need low loss and compactness, undercutting the no-fridge framing of the platform.
- 15 May 2026: an open-source PyTorch platform provides Fock, Gaussian and Bosonic photonic backends and claims first closed-loop integration of gate-model, photonic and measurement-based paradigms.
Open questions
- Can the fusion-failure noise floor identified for all-linear-optics architectures be circumvented by better boosted gates alone, given that the best passive-optics scheme reaches only 3/4 success with four single-photon ancillas?
- Do the cryogenic dual-layer graphene modulator designs survive fabrication, and what measured insertion loss and footprint do they achieve against the simulated optima?
- How far do the claimed industry-ready spin-photon interface metrics (photon efficiency, indistinguishability, spin coherence) sit from the values the sub-threshold analysis assumes for emitter-assisted fusion?
- Is there any workload where photonic hardware beats a commercial classical solver outside a narrow operating range, given the constrained portfolio result?
Assessment drafted 2026-08-31 from up to 15 KB sources using the technology-scorecard framework; scores are a draft read pending review.