Case Study: The Photonic Bet — Trading Determinism for Room Temperature
Executive Summary
Photonic quantum computing makes a distinctive wager: accept that two-qubit gates are probabilistic in exchange for room-temperature operation, natural networking, and manufacturing on existing silicon-photonics lines.
The wager is either brilliant or fatal depending on one number — the loss per component — and this case study works out why. It follows the argument from why photons make poor interacting qubits, through the multiplexing strategy that rescues probabilistic gates, to the loss threshold that determines whether the whole architecture closes.
Skills applied
- Understanding why linear optics gives probabilistic gates (§28.9).
- Analyzing multiplexing as a route to near-determinism.
- Computing loss budgets in a fusion-based architecture.
- Comparing photonic and matter-based scaling arguments.
Phase 1: Photons as qubits — the good and the fatal
Excellent qubits in most respects. A photon's polarization or path is essentially immune to thermal noise: at room temperature there are no thermal optical photons, so a photonic qubit does not decohere the way a matter qubit does. Single-qubit gates are wave plates and beam splitters — passive, fast, near-perfect.
Fatal in one respect. Photons do not interact. Two photons pass through one another with no effect. A CNOT requires the state of one qubit to affect another, and there is no direct mechanism.
The KLM workaround (Knill, Laflamme, Milburn, 2001) showed that measurement can induce effective interaction: interfere the photons on beam splitters, measure ancilla modes, and — when the ancilla detection pattern is right — the surviving photons have undergone an entangling gate.
But only sometimes. The basic KLM CNOT succeeds with probability 1/16. When it fails, it fails destructively.
Phase 2: Why probabilistic gates are a crisis, not an inconvenience
A circuit of $g$ two-qubit gates, each succeeding with probability $p$, succeeds overall with probability $p^g$:
| Gates | $p = 1/16$ | $p = 0.75$ |
|---|---|---|
| 10 | $10^{-12}$ | 0.056 |
| 100 | $10^{-121}$ | $3\times10^{-13}$ |
| 1,000 | — | — |
Exponential decay in circuit size. Without a fix, photonic quantum computing is impossible for any non-trivial circuit.
Phase 3: Multiplexing — the fix
The strategy: run many attempts in parallel and route the successes forward.
If a gate succeeds with probability $p$ and you attempt it $M$ times simultaneously, at least one succeeds with probability $1 - (1-p)^M$:
| $M$ attempts | $p=0.25$ | $p=0.5$ |
|---|---|---|
| 1 | 0.250 | 0.500 |
| 5 | 0.763 | 0.969 |
| 20 | 0.997 | $1 - 10^{-6}$ |
| 50 | $1 - 10^{-6}$ | $1 - 10^{-15}$ |
With ~20–50 parallel attempts, a probabilistic gate becomes effectively deterministic.
The cost is component count. Every logical gate needs tens of physical attempt-units, plus fast optical switches to route the successful outcome onward. A machine needing $10^6$ gates needs on the order of $10^7$–$10^8$ optical components — which is precisely why photonics is pursued by groups with silicon-photonics foundry access. Integrated photonics can print millions of components on a wafer; bulk optics cannot.
Phase 4: Fusion-based architecture
Modern photonic proposals (PsiQuantum, Xanadu and others) do not build circuits gate by gate. They:
- Generate many small entangled resource states (a few photons each) offline.
- Perform probabilistic fusion measurements joining resource states into a large cluster state.
- Compute by adaptive single-qubit measurements on the cluster (measurement-based computing).
Advantages: resource-state generation is repeat-until-success and can be done in parallel ahead of time; fusion failures are handled by the cluster's redundancy rather than by aborting; and the computation itself is measurement, which is fast and deterministic.
Phase 5: The number that decides everything — loss
Photon loss is the binding constraint. Every component loses photons:
| Component | Typical loss |
|---|---|
| Waveguide (per cm) | 0.1–1% |
| Beam splitter | 0.1–1% |
| Optical switch | 1–5% |
| Detector inefficiency | 1–10% |
| Fibre coupling | 1–10% |
A photon traversing 50 components at 1% loss each survives with probability $0.99^{50} \approx 0.61$. At 3% each, $0.97^{50} \approx 0.22$.
Fusion-based architectures have a loss threshold — roughly a few percent per component for the best schemes. Above it, no amount of multiplexing helps: the cluster state cannot be built faster than loss destroys it.
| Loss per component | Architecture viable? |
|---|---|
| 0.1% | Comfortably |
| 1% | Yes, with overhead |
| 3% | Marginal — near threshold |
| 10% | No |
The whole bet in one line. Photonics wins if integrated-photonic loss can be pushed below roughly 1% per component at scale, and loses if it cannot. Every other advantage — room temperature, networking, manufacturability — is contingent on that number.
Phase 6: Comparing the bets
Each platform bets on a different engineering problem being solvable:
| Platform | The bet | What must be solved |
|---|---|---|
| Superconducting | Wiring and cryogenics scale | Cryo-CMOS at μW/qubit, modular interconnects |
| Trapped ion | Shuttling and photonic links scale | Faster transport, better ion-photon interfaces |
| Neutral atom | Loss and measurement improve | Erasure conversion, faster non-destructive readout |
| Photonic | Component loss drops below ~1% | Low-loss integrated photonics at wafer scale |
Photonics is the highest-variance bet: if loss targets are met, the manufacturing story is by far the strongest — millions of components per wafer on existing fabrication lines, room-temperature operation, and networking built in. If not, no amount of multiplexing rescues it.
It is also the platform whose progress is hardest to read from qubit-count announcements, because the relevant metrics are loss per component, resource-state generation rate, and detector efficiency — none of which are qubit counts.
Discussion Questions
- Multiplexing converts probabilistic gates to near-deterministic ones. Why does it not similarly rescue a lossy system?
- Photonics needs no cryogenics for photons but detectors typically run at 1–4 K. How much does that undercut the room-temperature claim?
- Each platform bets on a different problem. Rank the four bets by your estimate of tractability and defend the ordering.
- Photonic progress is not measured in qubits. What three metrics would you track instead?
Your Turn: Extensions
- Compute the survival probability for a photon through 100 components at 0.5%, 1%, and 2% loss.
- Calculate the multiplexing depth for effective determinism at $p = 0.11$ and 0.999 target success.
- Research current state-of-the-art integrated-photonic waveguide loss and compare against the threshold.
- Estimate the component count for a fusion-based machine with $10^6$ logical operations.
Key Takeaways
- Photons make excellent qubits except that they do not interact, so linear-optical two-qubit gates are inherently probabilistic.
- Success probability decays exponentially in circuit size, making multiplexing — tens of parallel attempts plus fast switching — mandatory rather than optional.
- Fusion-based architectures build cluster states from offline-generated resource states, tolerating fusion failure through redundancy.
- Photon loss, not gate probability, is the binding constraint; the threshold is roughly a few percent per component.
- Photonics is the highest-variance bet: the strongest manufacturing story in the field, contingent entirely on achieving sub-1% component loss at wafer scale.