PsiQuantum Demonstrates Foundry-Manufactured Photonic Quantum Computing Platform With Record Fidelities

February 28, 2025 — PsiQuantum has demonstrated a complete manufacturable platform for photonic quantum computing, achieving record-breaking fidelities in a commercial semiconductor foundry, according to research published today in Nature.

The company’s integrated silicon photonics platform, built at GlobalFoundries’ 300mm facility, achieved 99.98% ± 0.01% state preparation and measurement fidelity for dual-rail photonic qubits, 99.50% ± 0.25% Hong-Ou-Mandel quantum interference visibility between independent photon sources, 99.22% ± 0.12% two-qubit fusion fidelity, and 99.72% ± 0.04% chip-to-chip qubit interconnect fidelity over 42 meters of fiber. All results are conditional on photon detection and do not account for optical loss.

The platform marks the first fully integrated on-chip heralded single-photon source, combining source, filter, and superconducting nanowire single-photon detector (SNSPD) on the same chip. The research team modified an established silicon photonics manufacturing flow to include NbN superconducting layers for single-photon detection, deep metal-filled trenches for optical noise reduction, and thermal phase shifters operating at approximately 2 Kelvin.

PsiQuantum’s next-generation components address key limitations. Silicon nitride waveguides achieved 0.5 ± 0.3 dB per meter loss in multimode configurations. Photon-number-resolving detectors demonstrated 98.9% median efficiency with four-photon resolution capability. Barium titanate electro-optic phase shifters recorded a loss-voltage product of 0.33 dB·V. Edge couplers showed 52 ± 12 millidecibel loss to ultra-high numerical aperture fiber.

The cascaded resonator source design achieved greater than 99% indistinguishability over ±400 picometer resonance wavelength shifts, addressing fabrication tolerance requirements. This compares to less than ±40 picometer tolerance for single-ring sources.

Testing occurred on 300mm wafers containing single-photon sources, superconducting single-photon detectors, and quantum benchmarking circuits. The research team packaged dies into assemblies with more than 1,000 electrical connections and up to 200 optical inputs/outputs, housed in cryostats with 2 Kelvin base temperature and up to 20 watts cooling capacity.

My Analysis

I’ll say it plainly: these fidelity numbers are genuinely impressive. The 99.50% Hong-Ou-Mandel visibility is the highest reported in any quantum computing platform. The 99.98% SPAM fidelity approaches what you’d want for fault tolerance. These aren’t cherry-picked lab results from a hero device—they’re from chips manufactured in a commercial semiconductor foundry.

But here’s the catch hiding in plain sight: every single result is “conditional on photon detection.” That’s quantum optics speak for “we’re ignoring all the photons we lost along the way.”

And that matters enormously. As I discussed in my analysis of quantum manufacturing challenges, fusion-based quantum computing (FBQC), PsiQuantum’s chosen architecture, can tolerate roughly 10% total optical loss. Current silicon photonics waveguides lose about 0.1 dB per centimeter. Do the math on a chip with thousands of components and hundreds of centimeters of waveguide routing, and you’re way over budget.

The next-generation silicon nitride waveguides at 0.5 dB/m are a significant improvement. Combined with the ultra-low loss splitters (0.5 millidecibel) and crossings (1.2 millidecibel), they’re approaching what’s needed. But they’re not quite there yet.

So why am I still impressed? Because PsiQuantum just proved something many thought impossible: you can manufacture every single component needed for photonic quantum computing in a standard semiconductor fab. No exotic materials. No boutique fabrication processes. No hand-assembled components. Just standard CMOS-compatible manufacturing with some clever modifications.

This fundamentally changes the manufacturability story for photonic quantum computing. In my CRQC Quantum Capability Framework, I identify manufacturing maturity as one of the critical dimensions for assessing quantum computing progress. PsiQuantum just moved photonics from “specialized fabrication required” to “commercial foundry compatible.”

The integrated heralded single-photon source is particularly noteworthy. Previous photonic quantum computing demonstrations relied on off-chip components connected through lossy fiber interfaces. By integrating the source, filter network, and detector on-chip, they eliminate those interfaces and dramatically improve the heralding efficiency – a crucial metric for FBQC.

The fabrication-tolerant cascaded resonator source addresses another key scaling challenge. Maintaining greater than 99% indistinguishability over ±400 picometer wavelength shifts means these sources won’t need individual tuning. That eliminates thousands of power-hungry thermal tuners from a large-scale system. For a machine that needs millions of components, removing active control wherever possible is essential.

PsiQuantum’s approach is different than most in this space. They’re not trying to build the best photonic quantum computer in a lab. They’re building a manufacturing process that could produce millions of adequate components. That’s the right strategy for error-corrected quantum computing, where you need massive redundancy rather than perfect individual qubits.

The photon-number-resolving detectors (PNRDs) deserve special mention. With 98.9% median efficiency and four-photon resolution, these enable herald-and-select protocols that can suppress multi-photon errors. For FBQC, where fusion failures cascade through the computation, identifying and rejecting bad events early is crucial.

The barium titanate (BTO) electro-optic switches, while still in development, address photonics’ Achilles heel: the probabilistic nature of spontaneous photon sources. Without fast switching to implement multiplexing, you’re stuck with low photon generation rates or massive resource overhead. The 0.33 dB·V loss-voltage product suggests reasonable switching losses at practical drive voltages.

In my taxonomy of quantum computing modalities, I categorize approaches by their physical implementation and computational model. Photonic quantum computing sits in an interesting position: the qubits themselves are nearly perfect (photons don’t decohere), but the operations are probabilistic and lossy. PsiQuantum’s results show steady progress on both fronts.

Now, let’s talk timelines. Nothing in this paper changes my assessment of when cryptographically relevant quantum computers might arrive. Even with perfect manufacturing, PsiQuantum still needs to solve the loss problem, demonstrate error correction, and scale to millions of qubits. We’re still looking at the 2030s at the earliest.

But this does change the credibility of their roadmap. When PsiQuantum claimed they’d build a utility-scale quantum computer by leveraging semiconductor manufacturing, many were skeptical. Could you really manufacture superconducting detectors, photon sources, and quantum optical circuits in a commercial fab? This paper answers definitively: yes, you can.

The real test comes next. Can they reduce losses enough to reach the fault-tolerant threshold? Can they implement fast enough switching for deterministic operation? Can they maintain these fidelities when scaling to millions of components? The foundations are solid, but the hardest engineering challenges remain.

For organizations tracking quantum risk, this reinforces what I’ve been saying: the threat is real, the timeline is uncertain, but the trajectory is clear. PsiQuantum just removed one of the biggest question marks around photonic quantum computing. They’ve proven the manufacturing approach works. Now they need to prove the physics allows it to scale.

If I had to bet on which quantum computing approach will first achieve crypto-relevant scale, photonics just moved up my list. Not because these results solve all the problems, but because they show a credible path to solving them at industrial scale. In the race to build a useful quantum computer, manufacturing capability might matter more than hero-device performance. PsiQuantum understands that. This paper proves they can execute on it.