In epitaxial quantum-dot growth, the recipe is usually written in terms of what you deposit: how much material, at what rate, at what temperature, until strain forces the flat layer to break into islands. A new preprint from a team at Leibniz University Hannover, posted to arXiv on 31 October 2026, turns that logic around. In their route, nothing is deposited to form the dots at all. The dots appear — or fail to appear — during cooling, and the cooling path is the control knob.

The paper, “Cooling-path-selected self-assembly of telecom InAsP quantum dots” by Yiteng Zhang and colleagues, with Fei Ding and Michael Zopf as corresponding authors, reports a growth method the authors call cooling-path-selected self-assembly, or CPSA. The dots are indium arsenide phosphide (InAsP) on indium phosphide, InP(001), and the capped emitters span 1.2 to 1.6 micrometres — the full telecom range, from the O-band through the L-band. Single-photon emission is observed up to 100 K. This is a preprint, not yet peer-reviewed, and it is a growth paper: its claims are about how dots form and how their wavelengths can be targeted, not about a finished device. Read on those terms, it is a careful and unusually well-controlled piece of epitaxy.

Why InP dots have been hard to grow reproducibly

The context matters, because the difficulty this paper addresses is specific to InP. Epitaxial quantum dots are the leading solid-state sources of on-demand quantum light, and the standard route — Stranski–Krastanov growth, in which a strained deposited layer breaks into three-dimensional islands past a critical thickness — works beautifully for InAs on gallium arsenide. On InP, the same recipe misbehaves. The lattice mismatch between InAs and InP is about 3.2 percent, roughly half the roughly 7 percent mismatch of InAs on GaAs, so there is less strain energy available to drive islanding. InAs growth on InP often produces elongated quantum dashes or rods instead of compact dot-like structures, and the final morphology is highly sensitive to temperature, group-V conditions, and substrate orientation.

Other routes exist. Droplet epitaxy forms group-III droplets and crystallizes them under a group-V flux; local droplet etching uses droplets to etch nanoholes that are then filled with dot material. The local-droplet-etching route has recently achieved C-band single-photon emission, but it requires separate control of nanohole formation, filling, and overgrowth. What has been missing, in the authors’ framing, is a reproducible InP route that gives both reliable dot formation and predictable emission wavelengths without separate droplet, etching, or filling steps. CPSA is their candidate answer.

The experiment: the same surface, four different cool-downs

The core of the paper is a control series, and it is worth describing in some detail because the controls are the argument. The growth is molecular-beam epitaxy. The sequence starts with high-temperature arsenic exposure of the InP surface. Arsenic replaces phosphorus in the top few surface bilayers — As–P exchange — producing a shallow near-surface InAsP layer. Crucially, no group-III dot material is intentionally deposited at any point. The dots, if they form, must come from material already present in that exchange layer.

The team then interrupted the process at different stages and cooled the surface in different ways, watching the surface evolve in situ with reflection high-energy electron diffraction (RHEED) and examining the result afterwards with atomic force microscopy. The RHEED pattern evolves from a streaky pattern, through the appearance of arc-like features, to an arc-stabilized stage at high temperature. The four labelled cooling paths diverge from there.

Rapid cooling right after the streaky stage (sample S3) leaves a nearly flat surface with only sparse protrusions. Rapid cooling after the arc-stabilized stage (S5) produces shallow protrusions but not large three-dimensional islands. Cooling slowly from the same arc-stabilized stage with the arsenic shutter closed (S7) suppresses island formation. But slow cooling with the arsenic flux maintained, followed by a low-temperature hold (S9), produces large three-dimensional islands — and strong telecom emission. The corresponding capped samples confirm the optical consequence: S5 shows weak telecom-range emission, S7 shows little apart from occasional residual peaks, and S9 shows strong telecom emission.

The logic is clean. Every path shares the same high-temperature treatment, yet the outcomes differ. The high-temperature sequence therefore does not fix the final morphology; the cooling path does. Within the tested paths, large islands develop only during slow arsenic-on cooling and the subsequent low-temperature hold. Close the shutter and the islands do not form; cool quickly and they do not have time to.

There is a subtlety the authors are careful about, and it is worth repeating. The RHEED arc pattern at high temperature does not by itself identify large islands — similar curved diffraction features appear on reconstructed surfaces without any three-dimensional islanding. The evidence that the islands form during cooling comes from the combination of the growth-interruption controls, the arsenic-on/arsenic-off comparison, and the in situ sequence: spotty and chevron-like RHEED features, the signatures of three-dimensional roughening and faceting, appear only during cooling. Atomic force microscopy records the surface only after cooling, so the rapid-cooling samples are path-interruption endpoints rather than instantaneous snapshots — a limitation the paper states plainly.

What the islands actually are

Composition measurements support the exchange picture. Angle-resolved X-ray photoelectron spectroscopy shows near-surface arsenic enrichment that decreases toward the InP substrate across the control samples — consistent with an arsenic-rich layer at the surface, though the authors note that oxide contributions differ between samples, so the absolute ratios cannot be compared directly. Cross-sectional scanning transmission electron microscopy of the island-forming sample shows continuous lattice planes across the island–substrate interface, with no obvious extended defects in the field of view. Electron energy-loss spectroscopy identifies an arsenic-rich, phosphorus-depleted island region with a compositionally graded transition into the surrounding InP matrix. The islands, in other words, are graded, arsenic-rich InAsP — formed from the exchange layer, not from deposited dot material.

The paper offers a qualitative thermodynamic framing rather than a kinetic model, and says so. As–P exchange changes local composition and lattice parameter, so the free-energy balance includes exchange-related chemical, surface and interface, and elastic contributions. Cooling changes the relative free energies of the accessible surface states and the rates of exchange, surface diffusion, and group-V desorption. For representative island geometries, the authors’ estimates suggest the exchange-related chemical contribution is the largest term — which is consistent with the observation that the arsenic-on/arsenic-off condition is the decisive variable — but the estimates compare only characteristic energy scales. They do not define a unique atomistic pathway, and the authors do not pretend otherwise. Whether the same principle applies to other III–V systems where group-V exchange occurs remains to be tested.

The emitters: single photons to 100 K

For the optical measurements, the team fabricated arrays of truncated-cone structures on capped island-forming samples to enhance light extraction, and studied individual emitters by micro-photoluminescence. The headline emitter, labelled QD2, shows two lines near 1278 and 1287 nanometres that remain spectrally resolved from 1.6 to 100 K. Excitation-power and polarization measurements support assigning them to the neutral exciton and the charged exciton, respectively: over the fitted low-power range the neutral-exciton intensity is approximately linear in excitation power while the charged-exciton intensity is superlinear, and the neutral line shows a stronger apparent polarization modulation.

The single-photon figures are the numbers to quote, with their conditions attached. The fitted second-order correlation at zero delay, g(2)(0), for the charged-exciton line is 0.07 at 1.6 K and remains below 0.25 at 80 and 100 K, with no background correction applied. Two additional emitters give fitted g(2)(0) values of 0.16 and 0.073. Under pulsed excitation, the charged-exciton decay is described by a biexponential fit giving an effective lifetime of about 1.6 nanoseconds at 1.6 K. In a magnetic field, the charged-exciton line shows resolvable Zeeman splitting, with an effective optical g-factor of 0.88 ± 0.03. Photoluminescence-excitation measurements show the signal peaking near 770 nanometres in the InAlAs-related spectral region and falling sharply toward the InP-related region around 875 nanometres.

Two things are notable here, and one is missing. The notable points: single-photon emission persisting to 100 K is genuinely useful, because it relaxes the cryogenic burden — liquid nitrogen sits at 77 K, and an emitter that still antibunches at 100 K is within reach of far simpler cooling than the few-kelvin regime most quantum-dot sources require. And the g(2)(0) values are measured without background correction, which is the stricter way to report them. The missing piece is indistinguishability: the paper reports no Hong–Ou–Mandel measurement, so nothing here establishes whether photons from these emitters interfere. Purity is one of three metrics a source must satisfy, alongside indistinguishability and brightness, and this paper addresses the first. Any claim that these are ready sources for interference-based protocols would be premature.

The calibration: one number that predicts the wavelength

The most practically consequential part of the paper is not any single emitter but the wafer-level statistics. Across 20 wafers grown with arsenic-on cooling, the median emission wavelength shifts monotonically to longer wavelengths with increasing surface-effective exchange dose, D — a single empirical growth variable that combines the applied arsenic flux, recorded as beam-equivalent pressure, with the logged durations of the high-temperature, cooling, and low-temperature windows, weighting the lower-temperature contribution by a factor derived from the logged temperatures. The red-shift saturates: further increases in dose produce smaller wavelength shifts, described by a saturating exponential fit with a saturation wavelength of about 1538 nanometres, a shift amplitude of about 738 nanometres, and a characteristic dose of about 15.1 × 10⁻⁵ Torr·min. Representative spectra from the four dose groups collectively span 1.2 to 1.6 micrometres.

The individual wafer medians show scatter, and the pooled emitter distributions retain both within-wafer and wafer-to-wafer spread — the paper does not hide that. But the grouped means shift monotonically, which is what makes D useful as a calibration: a grower can pick a target median wavelength and read off the dose, rather than hunting for it empirically wafer by wafer. That is the difference between a demonstration and a process.

The morphology statistics tell the same story from the structural side. Comparing a lower-dose sample (S8) with a higher-dose one (S9), the islands grow larger and sparser with dose: about 25 islands per square micrometre in S8 versus about 7.5 per square micrometre in S9, from height and lateral-size distributions measured over central 5 × 5 micrometre regions (626 islands for S8, 187 for S9). Effective-mass calculations for lens-shaped InAsP islands embedded in InP show that increasing either island height or arsenic fraction shifts the calculated ground-state transition to longer wavelengths, consistent with the observed red-shift — but because size and composition can change together, the authors explicitly do not use the calculation to separate their contributions or to infer an arsenic fraction from the optical data. That restraint is characteristic of the paper.

What would change the picture

CPSA is a growth result, and its significance should be judged as one. It does not by itself deliver an electrically driven device, a cavity-integrated source, or a measured indistinguishability. The authors’ own stated next step is to combine CPSA with charge stabilization and nanophotonic structures toward electrically controlled, integrated telecom quantum-light sources. That is the right next step, and it is also where the hard work remains: the history of quantum-dot sources is full of growth routes that produced good emitters in isolation and struggled at the integration stage.

What to watch, in order. First, independent reproduction: a second group growing InAsP dots by cooling-path selection on InP, with the same dose calibration, would turn a single-lab process into a route. Second, indistinguishability: a Hong–Ou–Mandel measurement on a CPSA emitter, reported with its excitation conditions, would establish whether these dots are candidates for interference-based protocols or only for applications like quantum key distribution that tolerate moderate indistinguishability. Third, brightness and extraction: the paper reports no collection efficiency, and the truncated-cone structures used here are measurement aids, not a device architecture. Fourth, yield at a target wavelength: the dose calibration predicts the median, but the spread around it — how many emitters on a wafer actually land within, say, 10 nanometres of a target — will decide whether this is useful for wavelength-matched applications.

For now, the contribution is clear and bounded: cooling, long treated as the passive end of a growth recipe, is an active epitaxial step. On InP, where the standard routes have struggled with reproducibility and wavelength control, a team has shown that the cool-down itself — its rate, its duration, and whether the arsenic stays on — decides whether dots form and where they emit. That is a result about process control, and process control is what turns quantum light from a laboratory demonstration into something that can be manufactured.

Sources and further reading

  • Y. Zhang, D. Abdelbarey, M. Etzkorn, Z. Jiang, A. Choudhary, A. Schaate, T. Fandrich, X. Cao, C. Ma, A. Chakraborty, E. P. Rugeramigabo, M. Zopf, and F. Ding, “Cooling-path-selected self-assembly of telecom InAsP quantum dots,” arXiv:2611.01214 (submitted 31 October 2026). Preprint, not yet peer-reviewed. All growth, spectroscopy, and statistics figures in this article are taken from this preprint.
  • Review and experimental literature on telecom-band quantum-dot technologies for long-distance quantum networks, in Nature Nanotechnology.
  • Literature on droplet epitaxy and local droplet etching routes to telecom quantum dots, in Nature Materials, Nanophotonics, and ACS Nano.
  • Literature on Stranski–Krastanov growth, strain-reducing capping, and the indium-flush technique for wavelength control of InAs quantum dots, in Applied Physics Letters and Journal of Applied Physics.
  • Standard quantum optics references on the second-order correlation function g(2), photon antibunching, and Hong–Ou–Mandel interference.
  • Note on verification: all quantitative values quoted above (g(2)(0), lifetimes, g-factors, island densities, and the dose–wavelength calibration parameters) are taken from the preprint as cited, with measurement conditions stated alongside them. No indistinguishability or brightness figures are reported in the preprint, and none are claimed here.