A mirror that reflects light is unremarkable. A mirror that reflects exactly one photon at a time, and lets the second photon through, is a different object — and it is the object photonic quantum computing has been waiting for. A paper published in Nature Communications on 11 November 2026, from Tarun Patel, Matteo Pennacchietti and colleagues at the University of Waterloo, the National Research Council Canada, and Queen’s University, reports the first clean demonstration of it: coherent reflection of single photons from a semiconductor quantum dot strongly coupled to the fundamental mode of a nanowire waveguide, with a raw second-order correlation of g²(0) of about 0.02. The device is a waveguide-QED quantum mirror — a single InAsP quantum dot on the axis of an InP nanowire.
This is a component experiment. It does not demonstrate a photonic gate, a network node, or a quantum computer, and it does not establish deterministic two-photon gates. What it demonstrates is the interface those things would be built on: a single emitter so strongly coupled to a single guided mode that essentially every photon interacts with it, in a geometry that separates reflected photons from light that never touched the dot.
The one-dimensional atom
Waveguide quantum electrodynamics — waveguide-QED — is the study of an emitter coupled to light that can only travel in one dimension. In free space, a photon aimed at an atom usually misses: the atom’s scattering cross-section is small, and a focused laser spot is a poor match to a dipole’s radiation pattern. More than 25 years ago, van Enk and Kimble showed that most of the light in a tightly focused Gaussian spot never interacts with the atom at all. The fix, proposed by Domokos and co-workers, was to put the emitter inside a waveguide whose guided mode has a cross-section comparable to the emitter’s scattering cross-section, so the photon has nowhere to go except through the atom. In that regime the emitter becomes a one-dimensional atom: all of its spontaneous emission goes into the guided mode, and all of the incident light meets it. Classically, such an atom is a perfect mirror. Quantum-mechanically, it has a catch — a two-level system can only scatter one photon per radiative lifetime. A single photon arriving alone is reflected; two arriving within the same lifetime find the atom already busy with the first, and the second is transmitted. The device sorts light by photon number — the nonlinearity photonic quantum computing needs, because photons otherwise pass through each other without interacting at all.
The problem was not the dot; it was the light that never interacted with it. Any stray field — light scattered off anything but the dot — backscatters into the same collection path as the reflected signal, and because that stray light is ordinary laser light, its Poissonian statistics wash out the antibunching. Previous attempts stalled at g²(0) of about 0.6. The Waterloo result is, at its core, a solution to that stray-light problem.
The device and the mode match
The system is a single InAsP quantum dot on the axis of a cylindrical InP nanowire waveguide, grown site-selectively by chemical beam epitaxy. The dot sits where the fundamental HE₁₁ mode is strongest, and the nanowire’s diameter-to-wavelength ratio is about 0.29, a regime in which roughly 90 percent of the dot’s spontaneous emission couples into that mode. That β-factor means nine out of ten photons the dot emits enter the one guided mode.
Getting light in is the second number. The nanowire tapers adiabatically toward its tip, at about one degree, so the HE₁₁ mode expands gradually into a free-space Gaussian beam with minimal back-reflection. Simulations put the ideal in-coupling at 97 percent, and the dot’s own emission confirms it: measured with a 4f imaging setup, the profile at about 895 nanometres overlaps a two-dimensional Gaussian by 99 percent.
Achieving it is a matter of focus position: the Gaussian spot must sit about four micrometres from the growth substrate. The team’s solution is to watch the extinction pattern in k-space — the angular distribution of the reflected light. Light that enters the mode leaves through the growth substrate, so it shows up as reduced reflection against the bare substrate’s Fresnel reflection: the extinction, E = 1 − I_NW/I_substrate, with a modelled ideal of 90 percent. The measurement agrees: across ten nanowires the team observes an average extinction of 86 percent, and the main device shows 85 percent. From that agreement the authors conclude that about 95 percent of the input Gaussian beam is coupled into the HE₁₁ mode. With 95 percent of the incoming light in the mode, and 90 percent of the dot’s emission returning to it, essentially all the light in the system has interacted with the dot.
The setup is deliberately spare: the input laser comes from a single-mode fibre, reflects off a 93:7 non-polarising beamsplitter into an objective inside a 4 kelvin cryostat, and is focused onto the nanowire; the reflected light returns through the beamsplitter into a second single-mode fibre. Because the collection fibre only accepts a Gaussian mode, the high-numerical-aperture back-scatter — the stray light — is simply not collected.
The quantum mirror, measured
The measurements are performed in the Heitler regime: weak, resonant, continuous driving, where the scattered light is dominated by elastic, coherent scattering — photons that keep the frequency and phase of the drive. The team scans the laser across the exciton resonance at input powers corresponding to a mean photon number ⟨n⟩ per dot lifetime as low as 0.001.
The reflected intensity, normalised to the off-resonant background — the light interpreted as not having interacted with the dot — is constant over seven decades of input power, from ⟨n⟩ = 10⁻⁶ to 10⁻¹. That flat plateau is the mirror: in the single-photon regime, the one-dimensional atom reflects essentially perfectly. Above ⟨n⟩ ≈ 0.1 the probability of multiphoton components in the drive grows, the dot saturates, and the normalised reflected intensity falls. The mirror stops being a mirror when, on average, more than one photon arrives per lifetime.
The photon statistics confirm it. Hanbury Brown–Twiss measurements give a raw g²(0) of about 0.02 — 0.025 at ⟨n⟩ = 0.001 in the preprint text, with no background corrections applied — against the classical value of 1 and the single-photon threshold of 0.5. The antibunching does not degrade as the drive is increased to ⟨n⟩ = 1, and even at ⟨n⟩ = 10 — ten photons per lifetime on average — g²(0) is only about 0.21. The reflected field stays non-classical far into the regime where the dot is saturated, because the multiphoton components are transmitted rather than reflected.
A Fabry–Pérot etalon measurement completes the picture: the emission is dominated by coherent scattering, and at the lowest drive the measured linewidth is indistinguishable from the etalon’s 66 MHz instrument response — the true coherent emission is on the order of the 10 kHz laser linewidth. More than 95 percent of the reflected signal is emission from the one-dimensional atom. At resonance the dot also imposes a π phase shift on the drive, the same phase flip a mirror gives a reflected wave — but saturation limits the reflection to one photon per radiative lifetime, so multiphoton components interfere destructively in the detection channel.
The numbers, and the boundary
Four figures carry the result, each with its condition attached. The coupling of the dot’s emission into the HE₁₁ mode is about 90 percent — a β-factor, not a device efficiency. The in-coupling of the input beam into the waveguide mode is about 95 percent, inferred from the extinction measurement rather than measured directly. The extinction itself is 85 percent for the main device, 86 percent averaged over ten nanowires, against a modelled ideal of 90 percent. And the reflected light’s raw g²(0) is about 0.02 at ⟨n⟩ = 0.001, rising to about 0.21 at ⟨n⟩ = 10. None of these is a total single-photon reflection efficiency, and they should not be combined into one.
One further figure sets the honest boundary of the device. The reflected resonance, fitted with a Voigt profile, has an inhomogeneous linewidth of 4.4 ± 0.6 GHz — more than twenty times the Fourier-transform limit of 177 MHz set by the dot’s radiative lifetime. The broadening is Gaussian, consistent with charge noise, and a weak above-band laser was needed to stabilise the charge environment. The dot is a good mirror and a noisy emitter at once: its reflection is coherent with the drive, but its resonance wanders by gigahertz. The applications the paper points to need Fourier-limited linewidths, and the authors say reaching them will require electrostatic control over the charge environment.
What this does not show
Three qualifications apply, all visible in the paper itself.
First, this is a single device, measured at 4 kelvin, in a vertical free-space geometry. The nanowire is addressed from above with a focused beam; it is not yet part of an integrated photonic circuit, and the transmission port — the light that passes through the dot and exits via the growth substrate — was not measured in this work. The transmission port can be accessed through the transparent substrate. Until that measurement exists, the mirror has been characterised from one side only.
Second, the mirror is not a gate. The nonlinearity demonstrated here — one photon reflected, two photons split — is the physical ingredient a deterministic photon–photon gate would be built from, and the paper’s framing is careful: such devices are “necessary for developing deterministic gates,” and the platform “opens up applications” in few-photon nonlinear optics. The step from a number-sorting mirror to a gate requires, at minimum, two such mirrors interacting, or one mirror in a configuration where the reflected photon’s phase can be conditioned on a second photon. None of that is in this paper.
Third, the wavelength. The dot emits at about 895 nanometres — near-infrared, but not a telecom wavelength, and not matched to the fibre infrastructure a deployed quantum network would use. That is not a flaw; the physics of waveguide-QED does not depend on the wavelength. But it is a constraint on any deployment story, in the same way the 4 kelvin operating temperature is.
None of these qualifications diminishes the result; they locate it. What Patel, Pennacchietti and colleagues have done is solve the measurement problem that had kept the reflection port unreadable, and then show the physics the theory predicted. The previous best in the reflection port, g²(0) of about 0.6, was a measurement limited by stray light. This one, at about 0.02, is a measurement limited by the dot.
What to watch next
The follow-ups that would matter most are easy to name, because the paper names them. The first is the transmission port: reading the light that passes through the dot via the transparent substrate, and showing the complementary photon statistics — bunching in transmission where reflection shows antibunching. The second is electrostatic control of the charge environment, to push the 4.4 GHz linewidth toward the 177 MHz transform limit; the authors flag it as critical. The third is the gate itself: the paper envisions single-photon waveplates and photon–photon gates built on this platform, following Kimble’s original proposal.
There is also a quieter lesson in the method: the extinction technique gives any group working with nanowire emitters a way to know, quantitatively, whether their light is actually entering the guided mode — and it is the part of the paper most likely to be reused first.
A single quantum dot in a nanowire, reflecting one photon at a time, is not a quantum computer. It is the mirror a quantum computer made of light would need — and for the first time, the mirror can be seen clearly, with the stray light removed and the photon statistics to prove it.
Sources and further reading
- Tarun Patel, Matteo Pennacchietti, Sayan Gangopadhyay, Dan Dalacu, Philip J. Poole, Stephen Hughes & Michael E. Reimer, “Single photon reflection from a quantum dot in a waveguide,” Nature Communications (2026). DOI: 10.1038/s41467-026-78984-x. Published 11 November 2026; received 30 July 2026; accepted 6 October 2026. Primary source for all figures in this article. Preprint: Research Square, posted 2 April 2026, DOI: 10.21203/rs.3.rs-8979997/v1.
- Prior waveguide-QED results cited in the paper: Javadi et al., Nature Communications 6, 8655 (2015); Le Jeannic et al., Physical Review Letters 126 (2021) — including the prior reflection-port result limited to g²(0) of about 0.6.
- Theoretical background cited in the paper: Duan & Kimble, Physical Review Letters 92, 127902 (2004); Hacker et al., Nature 536, 193–196 (2016); Pichler et al., Proceedings of the National Academy of Sciences 114, 11362–11367 (2017).
- Note on verification: all quantitative values here are taken from the Nature Communications abstract and the Research Square preprint text, which match the published abstract; the published full text was not accessible in full at writing, so the preprint’s values are quoted where they differ in precision (notably the raw g²(0) of 0.025 at ⟨n⟩ = 0.001, against the published headline of about 0.02). No market-size figure, commercial claim, or investment judgement appears in this article; the site’s Terms carry the not-investment-advice disclaimer.