Quantum light sources live or die on a number most papers quote as an average: the biexciton emission efficiency, the probability that a nanocrystal holding two electron-hole pairs emits a photon from that state instead of losing the energy to Auger recombination. It decides whether a material can lase at low power, serve as an entangled-photon source, or be trusted as a cascade pair source. Until now, that number for colloidal quantum shells has mostly been an ensemble property, averaged over every shell in the batch. A result published online on November 2, 2026, in Nano Letters changes the unit of analysis: Tjom Arens and colleagues at Utrecht University and Bowling Green State University measured biexciton emission shell by shell — more than a thousand individual particles — with a single-photon avalanche diode (SPAD) array camera, and found that the average conceals a distribution wide enough to matter.
This is a metrology result, and it should be read as one: no new quantum network, entangled-photon source, or laser. What it does is give the field a way to measure, at scale, the quantity that determines whether any of those devices can be engineered from this material — and it revises an earlier interpretation of the material’s biexciton energetics. For a desk that has followed the biexciton cascade from the epitaxial-dot side, where cavity engineering recently lifted photon indistinguishability from about 60 percent to about 90 percent, this is the complementary half of the story: before engineering a cascade, you must know what each emitter in your batch actually does.
Why the biexciton, and why shells
A biexciton is two excitons — two bound electron-hole pairs — occupying the same nanocrystal, and it decays in a cascade: one pair recombines, emitting a photon and leaving a single exciton, which then recombines to emit a second photon. That pair structure makes biexcitons interesting for quantum optics: the photons are correlated in time, and under the right conditions in polarization. The problem is competition. A biexciton can also decay nonradiatively through Auger recombination, in which one pair recombines and hands its energy to a third carrier instead of to a photon. In strongly confined colloidal quantum dots, Auger recombination is fast and quenches multiexciton states almost completely — which is why ordinary colloidal dots make such pure single-photon sources: the biexciton rarely survives to emit, so g²(0) stays low.
Quantum shells invert that trade. In a CdS/CdSe/CdS quantum shell, the emitting CdSe layer is a thin spherical shell sandwiched between CdS core and outer shell material, so carriers spread over a larger volume than in a compact dot. A larger volume means weaker carrier-carrier interaction, slower Auger recombination, and a biexciton that survives long enough to emit. Thick-shell dots and dot-in-rods exploit the same volume logic, but quantum shells push it further, and the material has already performed excellently as a laser gain medium in prior work, including a 2022 ACS Nano report and later continuously tunable red lasing demonstrations. At and around the gain threshold, the biexciton is arguably the most important state — yet gain is a single-particle property wearing an ensemble disguise. The paper’s opening calculation: two ensembles with the same mean efficiency, 0.5, one homogeneous and one split between 1 and 0, have identical averages but different gain thresholds and lifetimes. An ensemble measurement cannot tell them apart. A single-particle measurement with real statistics can.
The measurement problem: crosstalk that looks like physics
Measuring biexciton efficiency on one particle at a time is a photon-counting problem, and the standard tool is the Hanbury-Brown–Twiss experiment: split the emission onto two detectors and count coincidences. Under pulsed excitation, same-cycle photons produce the zero-delay peak; different-cycle photons produce reference side peaks. Under low excitation and assuming unit exciton efficiency, the ratio of the zero-delay peak to the side peaks — g²(0) — equals the biexciton emission efficiency, ηBX. Done one particle at a time with conventional time-tagging detectors, this yields beautiful data on a handful of particles. Done on a thousand, it runs out of patience.
SPAD arrays promise to parallelize the measurement. An area SPAD array is a camera with thousands of single-photon-sensitive pixels; the PI Imaging SPAD512 used here has 512 × 512 pixels and runs at up to 100 kHz, so one widefield image excites tens to hundreds of particles simultaneously. The catch: SPAD arrays are not innocent photon counters. When one pixel fires, its avalanche can emit a photon that a neighboring pixel collects, registering a spurious detection. For correlation measurements, which rely on genuine photon pairs, that is a counterfeit signal: crosstalk manufactures exactly the coincidences being counted. Previous approaches corrected for crosstalk in post-processing, but those corrections discard real data or lean on statistical assumptions.
The team’s fix is simple, and it is the paper’s central technical advance. Instead of correcting crosstalk, they prevent it. The sample emission is split by a 50:50 beamsplitter and projected twice onto the detector, onto two regions separated by more than 200 pixels, so each emitter appears twice on the camera. The team first mapped crosstalk versus pixel separation; it falls rapidly with distance, and at more than 200 pixels is effectively zero. Correlations are then computed only between matched images of the same emitter across the two regions. A genuine photon pair lands one photon in each region. A crosstalk artifact, which by nature strikes neighboring pixels, cannot. The price, stated plainly by the authors: the scheme discards half of all photon coincidences, those arriving on the same detector region, and it uses more pixels and therefore accumulates more dark counts.
One more piece of engineering was needed. The SPAD512 does not time-tag individual photons; each frame records, per pixel, whether a photon arrived during a defined integration window. The team synchronized the experiment so that each camera frame corresponds to exactly one laser pulse, running the laser at the camera’s maximum frame rate of 80 kHz, with each pixel recording zero or one photon per frame at 1-bit depth. The integration window then becomes a tunable gate: with a fluorescence lifetime near 30 nanoseconds, a 10-microsecond window drowns the signal in dark counts and a 20-nanosecond window clips the emission, while 250 nanoseconds captures most of the fluorescence and strongly suppresses dark-count coincidences. Validation against the old method came first: for the same emitter, the frame-based measurement gave g²(0) = 0.51, against 0.54 from conventional time tagging.
What a thousand shells actually do
The high-throughput dataset is built from 17 measurements on different regions of the substrate, each consisting of 20 million frames, yielding 1,162 matched emitting spots with enough coincidences to analyze. The raw g²(0) distribution across those spots is broad and asymmetric, mean 0.56, standard deviation 0.14; with a median per-spot error of only 0.03, the width is not measurement noise. But it is not yet the answer: clusters of two or three shells hiding inside one diffraction-limited spot, blinking between charged and uncharged states, and the charged states’ own correlation contribution also broaden it.
Two steps peel these apart. First, clusters: shifting the detection window’s start later exploits the fact that biexciton photons arrive faster than exciton photons. Delaying the window suppresses the biexciton contribution to the zero-delay peak for a genuine single emitter, driving its time-gated correlation toward zero, while a cluster keeps a finite value from independent excitons in different particles. A threshold of 0.3 on this time-gated correlation separates likely individual emitters from clusters, narrowing their distribution to an approximately Gaussian shape with mean g²(0) of 0.51 and standard deviation 0.10. Second, charging: shells blink, and unlike many nanocrystals their charged states are relatively bright — about 65 percent emission efficiency — so charged-state emission biases the inferred efficiency. The team isolates uncharged periods by dividing each trajectory into adaptive bins averaging 100 detected photons and keeping bins above 75 percent of the particle’s maximum binned intensity, automated across all thousand-plus particles. Using only those bright, uncharged periods gives mean biexciton efficiency ⟨ηBX⟩ = 0.46 with standard deviation 0.10; subtracting the median per-particle error of 0.06 leaves intrinsic shell-to-shell heterogeneity of 0.08. An independent mixture fit of the unfiltered distribution gives a nearly identical ⟨ηBX⟩ = 0.47 with the same heterogeneity. Mean efficiency also rises with brightness, used as a proxy for size, consistent with volume scaling of Auger recombination.
A blue shift that was not what it seemed
The paper’s second act revisits a puzzle from the ensemble literature, and it is the part most likely to be misreported. Earlier power-dependent studies had assigned a strongly blue-shifted feature in transient-absorption and stimulated-emission spectra to exciton–exciton repulsion in the biexciton state, with reported energies of roughly 50–70 meV. If the biexciton really repelled that strongly, gain and cascade energetics would look quite different from this paper’s efficiency picture.
To test that assignment, the team applied the same double-projection trick to spectroscopy, dispersing emission with a grating onto two separated regions of an event-based linear SPAD array that time-tags photons with sub-nanosecond resolution. Selecting only time-ordered cascade pairs across the two regions reconstructs separate exciton and biexciton spectra, free of the crosstalk peak that otherwise contaminates the zero-delay correlation. Both spectra are bimodal: the exciton spectrum has a main peak at 1.817 eV and a shoulder around 1.905 eV, a splitting of 88 meV. The biexciton’s main peak sits 6 meV to the red of the exciton’s — not tens of meV to the blue. Across 16 particles, the fitted S-level biexciton shift averages +2.8 meV, slightly repulsive, with a median error of 1.6 meV — nowhere near 50–70 meV.
The blue-shifted emission is real; the interpretation was not. The shoulder is stronger in the biexciton spectrum, consistent with thermal excitation of carriers to a higher P level rather than strong exciton–exciton repulsion. The observed main-peak-to-shoulder ratio in the exciton spectrum, 0.12, matches the Boltzmann prediction of 0.098 for a P/S degeneracy ratio of 3 at the measured 88 meV splitting, and the biexciton’s brighter shoulder follows from Fermi–Dirac level filling. Conditional spectra confirm it: the exciton spectrum does not depend on whether the biexciton photon came from the main peak or the shoulder, and vice versa, indicating thermal redistribution between S and P levels much faster than the exciton lifetime. A small triexciton contribution, estimated at about 8 percent, does not change the conclusion. The earlier spectra were not wrong; only the assignment of their blue feature to a large repulsive interaction is unsupported by state-resolved single-particle spectroscopy.
What this does and does not change
Placed against this desk’s biexciton-cascade arc, the result fills a specific, previously empty slot. The October cascade result from Paderborn, Basel, and Bochum showed that timing jitter in an epitaxial dot’s cascade can be engineered away with a cavity, lifting indistinguishability to about 90 percent — making a good cascade better. This work is about knowing, for a colloidal platform with different strengths — solution processing, room-temperature handling, strong multiexciton emission — what the cascade’s first step delivers, emitter by emitter. A cavity can only fix timing if the biexciton emits in the first place, and a gain medium can only be optimized if the distribution behind its average is known.
What the result does not do is equally clear. It does not report photon indistinguishability for quantum shells, entanglement quality, or any device-level demonstration. A biexciton efficiency of 0.46 is strong for a colloidal material — nearly half of biexcitons emit rather than Auger-decay — but it says nothing about photon indistinguishability, which depends on dephasing, spectral diffusion, and cascade temporal shape, none of which this paper measures. Nor does it solve the heterogeneity it reveals: with a standard deviation of 0.08 around a mean of 0.46, any device built from unscreened shells inherits that spread. The authors are candid about the method’s costs: discarding half the coincidences, more dark counts across more pixels, and a sensitivity floor near efficiencies of about 0.05.
The most informative follow-ups are easy to name: whether the method migrates to other emitters, where single-particle biexciton statistics are thin; whether growers can narrow the efficiency distribution, using the brightness correlation as a guide; whether the corrected energetics — a small shift, a thermally populated P level — enter models of shell gain and cascade emission; and whether a colloidal cascade put through Hong-Ou-Mandel interference shows photon quality matching its efficiency numbers. Efficiency is necessary for a good cascade source. It is not sufficient. Excitonic devices are engineered emitter by emitter but too often characterized as ensembles, and averages hide exactly the variation that decides whether a material scales from a cuvette to a device. A camera that counts photon pairs from a thousand shells at once does not remove the variation. It makes it visible, quantified, and therefore engineerable — a smaller claim than a breakthrough, and a more useful one.
Sources and further reading
- Tjom Arens, Dulanjan Harankahage, Divesh Nazar, Tycho Q. W. Savelkoel, Mikhail Zamkov, and Freddy T. Rabouw, “SPAD-Array Characterization of Biexciton Emission from Single Colloidal Quantum Shells,” Nano Letters (published online November 2, 2026). DOI: 10.1021/acs.nanolett.6c04178. Received August 24, 2026; accepted October 27, 2026. Utrecht University (Debye Institute) and Bowling Green State University (Center for Photochemical Sciences). Supported by NSF Award CHE-2505151 and NWO Vi.Vidi.203.031.
- Prior quantum-shell gain literature as cited in the paper: Cassidy et al., ACS Nano (2022); Harankahage et al., Journal of the American Chemical Society (2023); Tanghe et al., ACS Nano (2024); Waters et al., Journal of the American Chemical Society (2025). Gain framing here is scoped to those prior results, not to claims of the new paper.
- This desk’s earlier coverage: “Nearly Identical Photons on Demand From a Biexciton Cascade” (October 26, 2026) and “Single-Photon Sources From Excitons: Purity, Indistinguishability, Room Temperature” (October 2, 2026).
- Note on verification: all quantitative values — the 1,162 spots, the 0.51/0.54 validation pair, the 0.56 raw mean, the 0.51 individual-emitter mean, the 0.46 uncharged mean efficiency with 0.10 standard deviation and 0.08 intrinsic heterogeneity, the 65 percent charged-state efficiency, the 1.817 eV main peak, the 88 meV splitting, the 6 meV main-peak red shift, the +2.8 meV mean S-level shift, the 50–70 meV prior repulsion range, and the ~8 percent triexciton contribution — are taken from the Nano Letters paper as published on the ACS site. The paper’s supporting information was not separately reviewed; details confined to it are not quoted here. Where the paper reports two related but distinct shift quantities (the main-peak comparison and the fitted S-level shift), both are reported as stated rather than reconciled into a single number.