Quantum communication needs photons that are identical, one at a time, and in the right mode. Most of the field’s effort has gone into making the right transition brighter — putting the emitter in a resonator and using the Purcell effect to hurry the wanted photon out the door. A paper published in Nature Communications on 16 July 2026, from Andreas Reiserer’s group at the Technical University of Munich and the Munich Center for Quantum Science and Technology (MCQST), takes the opposite route. Instead of amplifying the wanted emission, the Munich team suppresses everything else: an erbium dopant in silicon sits inside a photonic-crystal waveguide whose bandgap forbids the unwanted decay channels, so the excited state has nowhere to go but the one transition that matters. The share of photons emitted on the desired transition rises from about 23 percent in bulk silicon to an average of 72 percent.

This is not breaking news. The paper is four months old; the TUM press release went out on 18 August 2026. What makes it worth covering now is comparison. In October, this desk covered two other answers to the same bottleneck — a Paderborn–Basel–Bochum cavity result that lifted photon indistinguishability from about 60 to about 90 percent, and a Naval Research Laboratory result on nonvolatile ferroelectric control of tungsten disulfide emitters. Together the three form a natural triptych: each moves a different metric.

The bottleneck in one number

Erbium in silicon is, on paper, the ideal quantum-network emitter. Its lowest crystal-field transition, Y₁→Z₁, sits at about 1538 nanometres, squarely in the telecommunications C band, the lowest-loss window of optical fiber. The same platform offers millisecond ground-state spin coherence and is built on silicon, so it is compatible with foundry-based semiconductor manufacturing.

But in bulk silicon, the Y₁→Z₁ transition carries only 23(±5) percent of the excited state’s decay. The rest of the time it decays to higher crystal-field levels, emitting at wavelengths between 1550 and 1650 nanometres that are useless for the photon source. Three quarters of the photons are born on the wrong transition.

The conventional fix is a resonator: a small, high-quality-factor cavity tuned to Y₁→Z₁, in which the Purcell effect accelerates that one decay until it dominates. But the resonator route carries costs that scale badly. A sufficient Purcell factor demands a small mode volume, which limits how many emitters one device can host — prior resonator experiments in this system addressed only a few dopants — and a high quality factor, which narrows the bandwidth and demands precise tuning. And a large Purcell factor shortens the lifetime, which, as the new paper argues, is not always what a quantum network wants.

Two philosophies of a photon source

The distinction is between enhancing the wanted and inhibiting the unwanted, and it is worth stating plainly because the two are often conflated. A resonator changes the local density of optical states (LDOS) at one frequency, making the desired transition faster. A photonic crystal changes the LDOS across a broad range, making the undesired transitions slower — ideally, impossible. The emitter is not shouted at; it is put in a quiet room with one open window.

The idea is old. Inhibited spontaneous emission goes back to Daniel Kleppner’s 1981 paper and Eli Yablonovitch’s 1987 proposal for photonic crystals. What the Munich team has done is turn it into a working single-photon source with measured numbers, in a platform that matters for networks.

The device: a quiet room with one open window

The device is a W1 photonic-crystal waveguide: a triangular lattice of cylindrical air holes etched into an undercut silicon membrane 220 nanometres thick, with a single row of holes omitted along the Γ–K direction. The slab carries a photonic bandgap 50 terahertz wide for TE-like modes, spanning 190 to 240 terahertz. Between roughly 190 and 195 terahertz, the LDOS falls to less than a tenth of its bulk value. That is where the unwanted Y₁→Z₂…Z₈ transitions live.

The desired Y₁→Z₁ transition, at about 1538 nanometres, sits at the edge of the bandgap, where the guided even mode has a flat dispersion relation — slow light, in which the group velocity tends toward zero and the LDOS rises to more than ten times its bulk value. So the same structure does two things at once: it strangles the unwanted transitions by more than a factor of ten and boosts the wanted one by more than a factor of ten, through the Purcell effect at the band edge. The geometry is tuned so the two effects nearly cancel on the total lifetime.

The proof it is suppression, not filtering

A skeptic’s first objection is that the waveguide might merely be acting as a spectral filter, so the detected light looks purer than the emitted light. The paper answers it with a calculation.

First, the headline measurement: with a tunable narrowband filter in the detection path, the team measured the fraction of detected photons on Y₁→Z₁ for individual emitters. The average is 72 percent, with a standard deviation of 14 percent. Every measured dopant exceeded 40 percent, and some approached unity — against 23(±5) percent in silicon strip waveguides without a tailored LDOS.

Then the filter-exclusion argument. Suppose the waveguide did nothing to the branching ratio, and the excited state still decayed via Y₁→Z₁ with the bulk probability of 23 percent. Even under the most generous assumptions — a lossless waveguide, no non-radiative decay, perfect coupling, and 50 percent excitation efficiency — one would expect a photon to be detected after the filter in less than 1.84(±0.43) percent of excitation repetitions. The measured value is 3.41(±0.23) percent, well above that ceiling. The increase cannot be explained by filtering. The branching fraction itself has changed: the emission has been redirected, not sieved.

The single-photon character is confirmed by autocorrelation. On an isolated line, the second-order correlation at zero delay is g²(0) = 0.07 ± 0.01 after subtracting independently measured detector dark counts — 0.25 ± 0.03 without the correction — comfortably below the 0.5 threshold for single-photon emission. About thirty spectrally separated peaks are resolved in a single waveguide, each a single dopant, individually addressable by frequency. Efficiency is inferred at 20(±2) percent on-chip and 14(±1) percent after fiber coupling — comparable to the state of the art at telecom wavelengths.

Why slower is better

The most counterintuitive claim in the paper is that preserving the lifetime is a feature. The measured Y₁ lifetimes range from about 100 to 600 microseconds across the ensemble, averaging 295 microseconds with a standard deviation of 97 — longer than the 142(±1) microsecond bulk lifetime. A resonator approach would have shortened the lifetime by orders of magnitude. Here it is preserved, or even increased.

The paper’s argument is about the timescale of the network, not the emitter. Long-distance quantum links are limited by the two-way signaling time, which approaches millisecond timescales for typical distances of about 100 kilometres. In that regime, a shorter lifetime does not increase the success rate; it only tightens the demands on control and detection electronics and restricts multiplexing. Slower photons are easier to control. And because the Y₁→Z₁ transition can be Zeeman-tuned with fields up to 3 tesla, one dopant’s lifetime can be tuned to match another’s — the path toward two-photon interference, and toward remote spin–spin entanglement.

The scaling payoff: tens of emitters, one chip

The result that matters most is not a single number but a count: thirty individually addressable emitters in one waveguide, versus a few dopants in prior resonator experiments. The extended mode volume permits lower dopant concentrations, reducing dipolar interactions and improving coherence — concentration-dependent spectral instability has been the scaling limit for erbium in silicon. And because the effect is largely independent of waveguide length, the count is limited by spectral diffusion and the inhomogeneous width of the ensemble, not by the device’s mode volume.

That is the sense in which suppression is a scaling strategy, not merely a clever device: resonators scale by replication — more devices, each tuned, each hosting a few emitters — while a waveguide scales by extension: one device, more emitters, addressed by frequency.

Three ways to a better photon

Paderborn, Basel, and Bochum moved indistinguishability. Their tunable open Fabry–Pérot microcavity shortened the biexciton lifetime in a quantum dot, lifting Hong–Ou–Mandel indistinguishability from about 60 to about 90 percent. Their problem was timing jitter; their tool was selective Purcell enhancement; their ceiling, as they reported it, was phonon-induced cavity feeding, which limits purity.

The Naval Research Laboratory moved control. In work published in ACS Nano in 2024, Sung-Joon Lee and colleagues integrated monolayer WS₂ with the ferroelectric polymer P(VDF-TrFE) and showed that switching the ferroelectric polarization reversibly tunes single-photon purity between the semiclassical and quantum regimes, up to 94 percent — with the state retained without applied power. Their problem was tunability; their tool was a nonvolatile ferroic field; what they did not move is indistinguishability or telecom compatibility.

Munich moved the branching ratio and the multiplexing count. Its problem was wasted emission; its tool was LDOS suppression; its ceiling is spectral diffusion. The three results do not compete: a future device that combined them — a suppressed, tunable, indistinguishable source — would need all three mechanisms, not one.

The honest boundary

Two caveats, stated plainly. First, the physics here is not excitonic: erbium’s 4f transitions are inner-shell transitions of a rare-earth dopant, shaped by the crystal field of the silicon host; no exciton is involved. This belongs on the Quantum desk through the better-photons hardware arc, not through the physics of the emitter.

But the bridge the authors themselves draw is real, and it is the reason this desk is covering the paper at all. Because a photonic-crystal bandgap can span tens of terahertz — wider than the frequencies of localized and propagating phonon modes — the same suppression trick could, in the authors’ words, “fully suppress phonon sideband radiation that has limited the rates in many pioneering quantum networking experiments with solid-state emitters.” Phonon sidebands are the bane of excitonic emitters: quantum dots, color centers, and two-dimensional materials all lose photons to phonon-assisted transitions. If the approach transfers to those emitters, it becomes a design template the excitonic community can steal, even though the demonstration used erbium.

Second, indistinguishability was not demonstrated. The measured linewidths average 27 megahertz with a standard deviation of 12 — about four orders of magnitude broader than the lifetime limit, given a homogeneous linewidth down to 10 kilohertz at 1.2 kelvin. The broadening is spectral diffusion, attributed to charge and spin noise in the devices. Until that diffusion is tamed — the paper suggests strong electric fields to ionize charge traps — no Hong–Ou–Mandel measurement will show high-visibility interference. The per-emitter branching fraction also varies widely, 72 ± 14 percent, because implantation position and dipole orientation are random. None of this diminishes the result. It locates it: Munich moved branching ratio and multiplexing; Paderborn and Basel moved indistinguishability; neither closes the other’s problem.

What to watch next

The most informative follow-ups will be easy to recognize: whether spectral diffusion can be suppressed far enough for a meaningful Hong–Ou–Mandel measurement, and what visibility it returns; whether spatially selective implantation narrows the spread in branching fractions and lifetimes; whether the phonon-sideband suppression argument is demonstrated on an excitonic emitter — a quantum dot or a two-dimensional material — rather than merely asserted as a generality; and whether the Zeeman lifetime-matching trick delivers two-photon interference between two dopants in the same waveguide, and then remote spin–spin entanglement.

The deeper lesson is one this desk has stated before. Solid-state emitters are strongly coupled to light, which makes them fast and optically accessible, and strongly coupled to their environment, which makes them decohere. The art is to engineer the environment so that the coupling that helps dominates the coupling that hurts. The resonator school does this by making the good coupling faster. The Munich school does it by making the bad coupling impossible. Both are legitimate. Only one of them leaves the lifetime alone — and for a network that spans a hundred kilometres, that may turn out to be the more important trick.

Sources and further reading

  • Florian Burger, Stephan Rinner, Andreas Gritsch, Kilian Sandholzer, and Andreas Reiserer, “Inhibited radiative decay enhances single-photon emitters,” Nature Communications 17, 6250 (2026). DOI: 10.1038/s41467-026-75489-5. Published 16 July 2026.
  • TUM press release, “TUM develops single-photon sources for quantum communication,” 18 August 2026 (MCQST version 17 August).
  • Timon L. Baltisberger et al., “Indistinguishable Photons from a Two-Photon Cascade,” Physical Review Letters (2026). DOI: 10.1103/t8sk-b2w4. Reported 1 October 2026.
  • Sung-Joon Lee et al., “Ferroelectric Modulation of Quantum Emitters in Monolayer WS₂,” ACS Nano 18, 25349–25358 (2024). DOI: 10.1021/acsnano.4c10528. Note: a 2024 paper, surfaced in this desk’s October breakthrough scan — not a new 2026 result.
  • Daniel Kleppner, “Inhibited spontaneous emission,” Physical Review Letters 47, 233–236 (1981); Eli Yablonovitch, “Inhibited spontaneous emission in solid-state physics and electronics,” Physical Review Letters 58, 2059–2062 (1987).
  • Prior Reiserer-group work on erbium in silicon: Purcell enhancement in nanophotonic resonators (Optica 10, 783–789, 2023) and optical single-shot readout of spin qubits (Nature Communications 16, 64, 2025).
  • Note on verification: all quantitative values here are taken directly from the Nature Communications paper, cross-checked against the TUM press release and the phys.org recap of 18 August 2026.