If there is one number that decides whether an excitonic quantum device can work, it is the coherence time. Coherence is the property that lets a quantum system maintain a well-defined superposition — a definite phase relationship between its components — long enough to be manipulated and measured. When coherence is lost, the system behaves classically: superpositions become mixtures, interference vanishes, and quantum advantage evaporates. Every proposal for excitonic qubits, simulators, sensors, and sources must therefore answer the same question: how long does coherence last, compared with how long an operation takes?

This article explains what coherence time means, why there are several different coherence times that must not be conflated, what sets them in excitonic systems, and how to read a coherence claim critically.

T1, T2, and T2*: three times, not one

The phrase “coherence time” conceals at least three distinct timescales. T1, the population lifetime or energy relaxation time, is how long an excited state survives before decaying — for an exciton, the time before it recombines and emits a photon. T2, the coherence time or dephasing time, is how long a superposition maintains its phase. T2*, the inhomogeneous dephasing time, is how long coherence appears to last in an ensemble measurement where different members have slightly different energies and therefore dephase relative to each other, even if each individual member remains coherent longer.

The relationship between them is constrained: T2 cannot exceed twice T1, because population decay destroys coherence. In practice, T2 is often much shorter than 2T1, because pure dephasing — phase randomization without energy loss — adds an additional decay channel. Pure dephasing comes from fluctuations in the system’s energy: phonons, charge noise, nuclear spins, and disorder all jitter the transition energy, scrambling the phase. T2* is shorter still in inhomogeneous systems, because ensemble averaging over different energies causes rapid apparent dephasing even when individual emitters are coherent.

Techniques exist to separate these timescales. A Ramsey interference experiment measures T2*, including inhomogeneous broadening. A Hahn echo — a refocusing pulse that reverses inhomogeneous dephasing — measures T2, the intrinsic coherence after removing static inhomogeneity. Dynamical decoupling sequences of many pulses can extend coherence further by filtering low-frequency noise. When a paper reports a “coherence time,” the reader should ask which of these was measured, with what sequence, and on what system. A T2* of a few picoseconds and an echo-protected T2 of microseconds describe very different capabilities, and both numbers may be correct for the same material under different protocols.

What limits excitonic coherence

For excitons in extended semiconductors, the dominant dephasing mechanism at elevated temperatures is phonon scattering. Acoustic and optical phonons collide with the exciton, shifting its energy and momentum and randomizing its phase. The rate grows with temperature, which is why excitonic coherence is longest at cryogenic temperatures and degrades rapidly as temperature rises. At low temperatures, other mechanisms dominate: scattering off defects and disorder, interactions with other excitons and free carriers, and fluctuations in the local charge environment.

For localized excitons in quantum dots, the picture changes. Confinement suppresses momentum scattering, and the discrete level structure reduces phonon coupling, so coherence can be substantially longer than for free excitons. But quantum dots have their own dephasing sources: spectral diffusion from charge fluctuations in the surrounding material, coupling to nuclear spins (in III-V materials, the nuclei have spin and create a fluctuating magnetic field), and phonon-induced transitions between closely spaced levels. Encapsulation, charge stabilization, and resonant excitation all help, but they do not eliminate dephasing entirely.

For valley coherence in monolayer TMDs, the relevant process is intervalley scattering: the electron-hole exchange interaction can flip the valley index, destroying valley superposition. This exchange-driven dephasing is fast — valley coherence times are typically picoseconds in many experiments — though they can be extended in high-quality encapsulated samples, at low temperature, and for specific exciton species (such as interlayer excitons, where the electron and hole sit in different layers and exchange is suppressed). The details depend on material, stacking, and measurement, which is why a single “valley coherence time” is meaningless without context.

For polaritons, coherence is a property of the condensate or the polariton field, and it is shaped by the driven-dissipative balance of pumping and loss, by interactions, and by disorder in the cavity. Polariton coherence can extend over tens to hundreds of micrometers in space and over times set by the condensate’s steady state, but this is collective coherence of a many-particle state, not coherence of an individual qubit. Comparing polariton coherence times directly with qubit T2 values is a category error.

The figure of merit: coherence divided by gate time

Coherence time alone does not determine whether a system can compute. What matters is the ratio of coherence time to operation time — how many gates can be performed before coherence is lost. A system with a coherence time of one microsecond and a gate time of one nanosecond can perform roughly a thousand operations per coherence time. A system with a coherence time of one millisecond and a gate time of one microsecond has the same ratio. Both are, in this crude counting, equally capable of shallow circuits, though their absolute speeds differ enormously.

Excitons have a distinctive position on this tradeoff. Their gate times can be extremely fast — optical manipulation on femtosecond to picosecond timescales — so even a short coherence time can in principle accommodate many operations. If an exciton coherence time is on the order of picoseconds and a gate takes tens of femtoseconds, the ratio is on the order of tens to hundreds. That is enough for proof-of-principle coherent control and for simulation of short-time dynamics, but it is far from the thousands to millions of operations that fault-tolerant computation requires, even before accounting for gate errors.

Spin qubits hosted in quantum dots illustrate a different tradeoff. The spin coherence time can be much longer than the exciton coherence time — microseconds or more under dynamical decoupling — but spin manipulation is slower, typically nanoseconds with microwave or optical Raman control. The ratio can therefore be more favorable for spins than for excitons, which is one reason hybrid schemes that use excitons for fast optical interfacing and spins for storage are attractive. The general lesson is that no single timescale tells the story; the ratio, together with gate fidelity, does.

How coherence is measured, and how claims go wrong

Coherence measurements are technically demanding, and the ways they can be misreported are predictable. The most common error is quoting T2* as T2, or quoting an echo-protected T2 as if it were the coherence available for free evolution. Another is quoting coherence measured on a single, exceptionally good device or a single, exceptionally clean spot as if it were representative of the material or the fabrication process. A third is quoting coherence at cryogenic temperature in a context that implies room-temperature operation. A fourth is comparing coherence times across systems without normalizing for gate time or fidelity.

There is also a subtler issue: coherence measured in one experiment may not transfer to another. Coherence under resonant excitation may be longer than under above-band excitation. Coherence of a neutral exciton may differ from coherence of a charged exciton (trion) or a biexciton. Coherence in photoluminescence may differ from coherence in a transport or interference experiment. A coherence number is only meaningful attached to a specific state, a specific protocol, a specific temperature, and a specific sample. Responsible reporting includes all four; irresponsible reporting includes none.

What would improve excitonic coherence

Improving coherence is a materials, fabrication, and control problem. Cleaner materials with fewer defects and less disorder reduce inhomogeneous broadening and scattering. Encapsulation — for example, sandwiching a TMD monolayer between flakes of hexagonal boron nitride — protects the active layer from environmental fluctuations and has been shown to narrow lines and improve coherence. Charge stabilization, through gates or doping control, suppresses spectral diffusion. Cavity engineering can shorten the radiative lifetime via the Purcell effect, so that emission completes before dephasing — effectively improving the photon’s coherence even when the exciton’s coherence is short. Dynamical decoupling and echo sequences can extend spin coherence by filtering noise. And choosing longer-lived carriers — dark excitons, interlayer excitons, spins — trades brightness for coherence where the application allows it.

Each of these helps, but none changes the fundamental tension: excitons are strongly coupled to light, which makes them fast and optically accessible, and strongly coupled to their environment, which makes them decohere quickly. That coupling is the same physics viewed from two sides. The art of excitonic quantum engineering is to find the point on that tradeoff where the ratio of coherence to operation time, and the fidelity of the operations, are good enough for the task at hand — simulation, sensing, communication, or, eventually, computation.

Coherence in context: a reading guide

When evaluating any excitonic quantum claim, ask for the coherence numbers in this order. First, what is T1 — how long does the state live? Second, what is T2 under free evolution, and what is T2* — how long does phase survive? Third, what is the gate or operation time — how many operations fit in T2? Fourth, what is the gate fidelity — does the operation preserve coherence well enough for error correction? Fifth, at what temperature, on what sample, and with what protocol were these measured? A paper that answers all five gives you a real assessment. A claim that quotes a single large coherence number without the rest gives you a headline.

The field’s honest position is that excitonic coherence is sufficient for the demonstrations achieved so far — coherent control, polariton condensation, entangled photon generation, small-scale simulation — and insufficient, by a wide margin, for fault-tolerant gate-based computation with excitons as the qubits. Closing that gap is not a matter of one breakthrough but of sustained improvement in materials, control, and architecture, measured by the full set of numbers above. Coherence time is the number that decides everything, but only when it is quoted completely.

Sources and further reading

  • Standard quantum optics and quantum information references on T1, T2, T2*, Ramsey interferometry, spin echo, and dynamical decoupling, including standard textbooks on quantum computation and quantum optics.
  • Review and experimental literature on exciton coherence, dephasing, and spectral diffusion in semiconductor quantum dots, in Physical Review Letters, Physical Review B, and Nature Photonics.
  • Review literature on valley coherence and intervalley scattering in monolayer transition metal dichalcogenides, in Nature Materials, Nature Reviews Materials, and Nature Communications.
  • Literature on polariton coherence, driven-dissipative condensates, and spatial coherence measurements, in Nature Physics and Physical Review Letters.
  • Literature on spin coherence in quantum dots, nuclear-spin-induced dephasing, and dynamical decoupling, in Nature, Science, and Physical Review Letters.
  • Note on verification: coherence times are deliberately stated qualitatively or as order-of-magnitude ranges above, because they vary by orders of magnitude across materials, states, temperatures, and protocols. Any specific coherence figure should be checked against the original paper’s state, protocol, temperature, and sample details.