Dead Time Correlation Attack
The Dead Time Correlation Attack lets a threat actor (Eve) read part of a QKD key without touching the quantum channel at all. Every single-photon detector is blind for a short recovery period, its dead time, after each click. When a receiver (Bob) is run faster than its own detectors can count, two detections in the same basis within one recovery period must come from that basis's two detectors in turn, because the first is still recovering while the second fires, so the two key bits they produce are opposite rather than independent [Rogers2007].
Nothing is injected, measured or altered on the quantum channel. The threat actor only listens to the ordinary public exchange in which the transmitter (Alice) and Bob agree which detections to keep, which tells her when each detection happened and in which basis. From that she reconstructs a whole alternating run of key bits and is left guessing only which detector started it, one bit however long the run is [Rogers2007]. Because she does nothing to the link, the error rate the two parties watch does not move and there is no equipment for them to find.
What triggers the attack is the transmission rate rather than the brightness of the light: a link clocked below its detectors' counting limit is not exposed, and raising the channel loss weakens the effect without removing it [Rogers2007]. How strong it gets depends on the protocol, and of those studied the two-state B92 protocol is the worst exposed [Xu2006].
| [Burenkov2010] | V. Burenkov, B. Qi, B. Fortescue, H.-K. Lo. "Security of high speed quantum key distribution with finite detector dead time" (2014) 10.26421/QIC14.3-4-2. arXiv:1005.0272. |
| [Dixon2009] | A. Dixon, J. Dynes, Z. Yuan, A. Sharpe, A. Bennett, A. Shields. "Ultrashort dead time of photon-counting InGaAs avalanche photodiodes" (2009) 10.1063/1.3151864. |
| [Jiang2024b] | Shuangfeng Jiang, Majid Safari. "High-Speed Quantum Key Distribution Using Dead-Time Compensated Detector Arrays" (2024) 10.1109/JLT.2024.3363145. |
| [Krause2025] | Jan Krause, Nino Walenta. "Exponential-recovery model for free-running SPADs with capacity-induced dead-time imperfections" In: arXiv:2507.10361 [quant-ph]. (2025) 10.48550/arXiv.2507.10361. arXiv:2507.10361. |
| [Kuniyil2026] | Hashir Kuniyil, Asad Ali, Syed M. Arslan, Muhammad Talha Rahim, Artur Czerwinski, Saif Al Kuwari. "Recovery-Induced Erasure Attack on QKD Systems" (2026) 10.48550/arXiv.2603.03217. arXiv:2603.03217. https://arxiv.org/abs/2603.03217v1. |
| [Mengler2025] | M. Mengler, M. Tippmann, T. Walther. "Increasing the secure key rate of a quantum-key-distribution system via synchronized detector dead times" (2025) 10.1063/5.0253835. |
| [Rogers2007] | D. Rogers, J. Bienfang, A. Nakassis, H. Xu, and C. Clark. "Detector dead-time effects and paralyzability in high-speed quantum key distribution" In: New J. Phys. 9, 19. (2007) 10.1088/1367-2630/9/9/319. |
| [Souza2026] | G. C. Souza, et al.. "Efficiency-Resolved Recovery Dynamics of a Free-Running InGaAs/InP Single Photon Avalanche Detector Operated in Gated Mode" (2026) 10.48550/arXiv.2609.17694. |
| [Xu2006] | H. Xu, L. Ma, J. Bienfang, and X. Tang. "Infuence of avalanche-photodiode dead time on the security of high-speed quantum-key distribution systems" In: Conference on Lasers and Electro-Optics and 2006 Quantum Electronics and Laser Science Conference, Long Beach, CA, pp. 1-2. (2006) 10.1109/CLEO.2006.4628713. |
17366ced-f5f6-4b2a-a4ca-bae5d9ed6538
- Quantum Communication
Technique properties
Operational properties
| Description | Countermeasure |
|---|---|
|
This is the one control both of A-0085's own primaries proposed. It has two strategies. Simultaneous hold-off: "one can simultaneously hold off all APDs when any of them is in the dead time", which in simulation holds the adjacent-bit transition probability at exactly one half, since "the probability of neighboring keys being different remains 0.5 independent of the repetition rate" [Xu2006]. The 2010 sequel names the same scheme AlgDeactivate, where "all detectors are actively disabled (or pulses actively blocked) every time any one of the four detectors is hit", and confirms it is secure against an active threat actor as well as a passive one [Burenkov2010]. Self-disabling: the second primary's own alternative time-multiplexes both states of a basis onto one physical detector, which makes the receiver a non-paralyzable counter, at the cost that "the maximum transmission rate as determined by the detector timing resolution is reduced by one half", which its authors say leaves it for cases where the algorithmic route is impractical [Rogers2007]. The rate effect is set by the regime and the protocol, and both signs are in the literature. For BB84-style sifting driven far above the inverse dead time, hold-off costs rate: "One trade-off of this de-correlation is the decrease of the ultimately available sifted-key rate" [Xu2006], with a ceiling of against the per-basis rule's [Rogers2007,Burenkov2010]. For a coincidence-based time-bin entanglement system at realistic fibre loss it gains rate, because blocking a partner detector's unfavourable click avoids spending a second dead time on an event that could never have contributed to the key: a 2025 measurement [Mengler2025] raised the sifted key rate by up to 78% with no significant change in the error rate, with the detectors kept free-running, which is this technique's own precondition architecture. The two results are not in conflict; the sign depends on whether the hold-off removes availability or removes waste. |
CM-0061 Detector's Dead Time Management |
| Description | Category | Impact Type |
|---|---|---|
|
Physical Layer (Quantum and Optical) is the primary layer. The quantity the whole technique rests on is the detector's optical recovery period, "that is typically of the order of 100 ns" for a free-running silicon single-photon avalanche diode, and the forced alternation follows from which detector is physically able to respond to the next photon [Rogers2007]. Nothing above this layer creates the pattern. |
L1 — Physical Layer (Quantum & Optical) | Primary Target |
|
Post-Processing and Cryptographic, secondary, and this is the mapping that makes the technique actionable. The sifted bit stream and the public exchange that produces it are the surface the threat actor reads, and the layer is also where the primary puts the remedy: the security condition "must be imposed on the receiver either by some means of gating the detectors or by the sifting algorithm", with the software route judged "both more practical and more efficient" [Rogers2007]. A reader who sees only the physical layer will look for a faster detector when the published fix is a sifting rule. |
L5 — Post-Processing & Cryptographic | Secondary Effect |
|
Electronics and Analog Components, secondary. The length of the recovery window is set by the detector's quenching electronics rather than by the optics, which is why the figure differs across the device family and why the entity's own text singles out passively-quenched devices, whose slow charge and discharge cycles lengthen the response to consecutive detections. The threat actor does not act on this layer, but it is the layer that decides the transmission rate at which the technique becomes possible. |
L2 — Electronics & Analog Components | Secondary Effect |
| Relation type | Description | Technique |
|---|---|---|
| Alternative to |
The two techniques exploit the same device property and reach a comparable outcome by opposite routes, and the contrast is the clearest way to say what this one is. A-0076's threat actor injects dim correctly-polarised pulses to drive chosen detectors into dead time, so she creates the asymmetry; here she waits for the link's own traffic to create it and only listens, because "she does not interfere with the quantum signals" [Burenkov2010]. That is why A-0076 needs optical access to the channel, a pulsed source and timing electronics while this technique needs none of them. Neither generalises the other. |
A-0076 Blinding Attack Using Dead Time |
| Alternative to |
The closest structural relative in the catalogue, and the pairing a defender needs. Both techniques recover detector identity from publicly announced detection timing with no action on the quantum channel, no equipment beyond a classical-channel tap and no error signature, since in both cases "she knows when bits are detected and in which basis they are sifted" is the whole mechanism [Rogers2007]. They differ in which timing property carries the identity: A-0086 uses the fixed per-detector click-latency difference, so it works at any repetition rate and leaks on every detection, while this one uses the dead-time-forced alternation, so it needs the link clocked above the inverse dead time but then leaks a whole run from one bit. |
A-0086 Timing Attack Exploiting Detector Response Times |
|