Backflash Attack

System

Avalanche photodiodes emit a faint flash of light every time they register a photon. In a QKD receiver those backflash photons travel back out through the receiver's own input port into the quantum channel, where a threat actor (Eve) can collect them. What she learns is not a setting but an outcome: which of the receiver's (Bob's) detectors fired, and therefore the bit and the basis that detector recorded, so what leaks is key material rather than configuration data [Pinheiro2018,Su2025a].

The attack is passive. The transmitter's (Alice's) photons reach Bob untouched, Eve sends nothing in, and nothing she does changes the error rate, the detection rate or any other quantity the two parties monitor, which is why a link can run normally while part of its key is being read [Su2025a]. How much escapes depends on the detector rather than on the protocol: silicon and InGaAs avalanche devices emit, a photomultiplier tube shows no measurable flash, and a fast-gated avalanche device emits about an order of magnitude less than a slow-gated one [Pinheiro2018,KoehlerSidki2020].

Literature 15
[Arnon2019] S. Arnon. "Quantum technology for optical wireless communication in data-center security and hacking" In: Proceedings Volume 10945, Broadband Access Communication Technologies XIII; 109450H. (2019) 10.1117/12.2505733.
[Dimitropoulou2026a] Maria Dimitropoulou, Constantinos Psomas, Ioannis Krikidis. "Backflash Mitigation via Scheduling in FSO Systems with Quantum Key Distribution" (2026) 10.1109/LWC.2026.3661400. https://zenodo.org/records/18628187.
[ETSI016PP] ETSI. "Quantum Key Distribution (QKD); Common Criteria Protection Profile - Pair of Prepare and Measure Quantum Key Distribution Modules". In: ETSI GS QKD 016. (2024) https://www.etsi.org/deliver/etsi_gs/QKD/001_099/016/02.01.01_60/gs_QKD016v020101p.pdf.
[ISOIEC23837P2] ISO/IEC JTC 1/SC 27. "ISO/IEC DIS 23837-2:2022 Information security - Security requirements, test and evaluation methods for quantum key distribution - Part 2: Evaluation and testing methods". (2022)
[KoehlerSidki2020] "Backflashes from fast-gated avalanche photodiodes in quantum key distribution" (2020) 10.1063/1.5140548. arXiv:2006.04107.
[Kupferman2018] J. Kupferman and S. Arnon. "Zero-error attacks on a quantum key distribution FSO system" In: OSA Continuum 1, 1079-1086. (2018) 10.1364/OSAC.1.001079.
[Kurtsiefer2001] C. Kurtsiefer, P. Zarda, S. Mayer, and H. Weinfurter. "The breakdown flash of silicon avalanche photodiodes-back door for eavesdropper attacks?" In: Journal of Modern Optics, 48(13), 2039–2047. (2001) 10.1080/09500340108240905.
[Meda2017] A. Meda, I. Degiovanni, A. Tosi, Z. Yuan, G. Brida, and M. Genovese. "Quantifying backflash radiation to prevent zero-error attacks in quantum key distribution" In: Light Sci. Appl. 6, e16261. (2017) 10.1038/lsa.2016.261.
[Meda2018] A. Meda, I. Degiovanni, A. Tosi, Z. Yuan, G. Brida, and M. Genovese. "Quantum key distribution security threat: the backflash light case" In: Proceedings Volume 10674, Quantum Technologies 2018; 1067418. (2018) 10.1117/12.2307704.
[Navarrete2026] Á. Navarrete, M. Pereira, G. Currás-Lorenzo, A. Mizutani, M. Curty, K. Tamaki. "Phase-error estimation for quantum key distribution with leaky receivers". (2026) arXiv:2608.09674.
[Pinheiro2018] P. Pinheiro, P. Chaiwongkhot, S. Sajeed, R. Horn, J. Bourgoin, T. Jennewein, N. Ltkenhaus, and V. Makarov. "Eavesdropping and countermeasures for backflash side channel in quantum cryptography" In: Opt. Express 26, 21020-21032. (2018) 10.1364/OE.26.021020.
[Shi2017] Y. Shi, J. Lim, H. Poh, P. Tan, P. Tan, A. Ling, and C. Kurtsiefer. "Breakdown flash at telecom wavelengths in InGaAs avalanche photodiodes" In: Opt. Express 25, 30388–30394. (2017) 10.1364/OE.25.030388.
[Singh2025] "Backflash Attack on Coherent One-Way Quantum Key Distribution" (2025) 10.1109/JPHOT.2025.3557214. arXiv:2502.04081.
[Su2025a] Jialei Su, Junxuan Liu, Zihao Chen, Mingyang Zhong, Qingquan Peng, Jiangfang Ding, Yijun Wang, Anqi Huang, Ying Guo. "Quantifying the Upper Limit of Backflash Attack in Quantum Key Distribution" (2026) 10.1103/xn8h-961f. arXiv:2501.09984. https://arxiv.org/abs/2501.09984.
[Tan2025] Hao Tan, Mikhail Petrov, Weiyang Zhang, Liying Han, Sheng-Kai Liao, Vadim Makarov, Feihu Xu, Jian-Wei Pan. "Wide-Spectrum Security of Quantum Key Distribution" (2025) 10.1103/smdh-dcpm.
QID: A-0080
UUID: 043c643c-4b8c-4f5e-aaa9-f8e761ed0141
Tier: T0
Type: Quantum
Domain(s):
  • Quantum Communication
Tactic: Execution
Created: 2024-04-12
Updated: 2026-11-13
Maturity: Experimental
Technique properties

Attack type: Key extraction, Information obtaining
Attack category: Detector-backlash attack
Target party: Receiver
Target function: Quantum layer
Quantum channel: Fibre optic, Free space
Security layers: L1 — Physical Layer (Quantum & Optical) L2 — Electronics & Analog Components
Maturity: Experimental
Operational properties

Technique → Countermeasures 5
Description Countermeasure

The step that actually removes the leaked information once it has been bounded, and the reason this attack degrades a key rate rather than breaking a link outright. Pinheiro2018 names it as the consumer of its own worst-case quantity: once the receiver has characterised PbP_b and TbT_b, the information leakage to Eve is then PE=PbTbP_E = P_b T_b, and privacy amplification for QKD with tagged signals applies, with the single-photon BB84 key rate becoming l≥A⋅Pdet(1−h(e/A))−leakECl \geq A \cdot P_{det}(1 - h(e/A)) - leak_{EC} where A=(Pdet−PE)/PdetA = (P_{det} - P_E)/P_{det} [Pinheiro2018]. The entropy is taken of e/Ae/A rather than of ee, and the difference is not cosmetic: since AA is below one, the rescaled error rate e/Ae/A exceeds ee, so the charge privacy amplification has to pay for the leak is larger than the unrescaled form implies. Su2025a folds the backflash leak in the same way, as an extra tagged fraction, and reports the cost tied to the detector's gate width rather than as a single number: the maximum transmission distance decreases by 2.87% compared to the no-attack case at a 15 ns gate, where the backflash generation probability was 9.2%, and when the detector gate width was reduced to 4.99 ns, the maximum transmission distance decreased by only 1.16% [Su2025a]. Singh2025 does the same on coherent one-way for about 10 per cent of the key rate [Singh2025]. This link works with CM-0014: the proof supplies the bound, this control spends key to remove what the bound admits.

CM-0001 Privacy amplification

Outbound spectral filtering at the receiver's optical port, and the control with the most quantitative support on this technique. Kurtsiefer2001 proposes it from its own measurement, naming filters blocking the spectral range of 700-1000 mn in which photoemission occurs, and states the condition under which it works: the signal wavelength must lie outside the emission band, or else narrow-band interference filters at the diode emission wavelength are needed instead [Kurtsiefer2001]. Two papers measure the factor. On a silicon module an 808 nm filter of 3 nm bandwidth cuts the escaping-flash probability from at least 0.065 to 4.5×10−34.5 \times 10^{-3}, a reduction by a factor of 14 [Pinheiro2018]. On InGaAs modules a bandpass filter does better, since applying spectral filtering can effectively suppress the rate of breakdown flash by two orders of magnitude [Shi2017]. Scope: this is the fibre and free-space control a deployed receiver can actually retrofit, unlike the detector-side control, and it also reduces the characterisation burden, because with a band-pass filter it is sufficient to characterize the parameters over its spectral pass-band rather than across the whole range in which optics and air are transparent [Pinheiro2018].

CM-0013 Employment of Optical Filters

The control that consumes this attack's leak analytically instead of attenuating it, and the only one on this technique that a system which cannot remove the emission can still rely on. Pinheiro2018 defines the route: the receiver characterises the escaping-flash probability PbP_b and the reverse transmission TbT_b, including all isolators and filters, and the information leakage to Eve is then PE=PbTbP_E = P_b T_b, a fraction of detections tagged without disturbing the quantum state or inducing error, which privacy amplification for tagged signals then removes [Pinheiro2018]. Navarrete2026 is the modern form and the reason this is now practical rather than merely sound: a finite-key phase-error bound for prepare-and-measure QKD with a leaking receiver, motivated by the observation that after Bob's measurement, his receiver may emit signals, needing only a lower bound on the fidelity between the two key-basis leakage states, and loss tolerant, so the tolerable leakage is essentially independent of the channel loss [Navarrete2026]. Su2025a supplies exactly the kind of distinguishability bound such a proof consumes: the extinction ratio she achieves against the four BB84 states, where the measured ER reaches a maximum of 28.3 across multiple trials, with an average value ER of approximately 23.4, and the paper's own model reproduces that mean at a simulated ER=22ER = 22 [Su2025a].

CM-0014 Full Device Imperfections Incorporation in Scientific Security Proof

Time gating reduces this attack's leak because a shorter gate means a smaller avalanche charge, and the emitted photon count scales with that charge. Meda2018 measured the effect directly, finding a drastic drop in information leakage when the signal arrives at the end of the gating window, because the gate's falling edge quenches the avalanche before the slower active-quenching circuit would and so removes the portion of the emission generated during that quench, with a window comparable to the emission's temporal width, about 5 ns or less, as the stated target [Meda2018]. Singh2025 measured the same on a different detector, where smaller gate widths, even in slow-gate detectors lead to reduced number of backflash photons, saturating above about 5 ns [Singh2025]. KoehlerSidki2020 is the limiting case, a 1 GHz self-differencing device leaking about 0.5 per cent against 6.0 to 9.8 per cent for megahertz gating, and states the mechanism: faster gating, resulting in narrower gates and smaller avalanche charges, results in fewer backflashes [KoehlerSidki2020]. Su2025a supplies what none of those three does, the dose-response measured on a single detector rather than compared across device classes: when the SPAD gate width was set to 15 ns, the backflash generation probability was 9.2%, falling to 4.05% at a gate of 4.99 ns, and the matched key-rate cost falls with it, the maximum transmission distance decreases by 2.87% compared to the no-attack case at the wider gate and by only 1.16% at the narrower one [Su2025a]. That turns the control from a direction into a dose a defender can design to.

CM-0026 Time Gating

Reverse optical isolation on the receiver's outbound path, placed so that light leaving the detector towards the channel is attenuated. Meda2018 is the published source for the recommendation: if commercial InGaAs/InP detectors are used for QKD, one has to reduce the amount of light that is coming from a detection event, using e.g. ciculators and/or spectral filtering [Meda2018]. Pinheiro2018 states the same measure in terms of the quantity it acts on, that the information leakage could be limited by decreasing reverse transmission efficiency TbT_b either by adding narrow-band spectral filter or an optical isolator, against a measured TbT_b of about 0.091 through a real receiver [Pinheiro2018]. Arnon2019 names a directional coupler or very narrow filter for the free-space case [Arnon2019]. The cost is the point of contention, and it is detector-dependent: on a GHz-gated self-differencing device KoehlerSidki2020 argues an isolator would not be needed as a countermeasure since even with a very low insertion loss of 0.2 dB, it would have a greater impact on the key rate than the 0.5 per cent leak it removes [KoehlerSidki2020]. The same paper also argues the other side, on a ground that has nothing to do with the key rate. Because the temporal profile of backflashes appears to be unique for different APDs, the emission can provide Eve with information on the detectors used by Bob, allowing her to use a tailored attack, from which it concludes that the use of an isolate may still be desirable as a countermeasure [KoehlerSidki2020]. So the key-rate case against fitting an isolator is about the key material the flash carries, while the fingerprinting case for fitting one is about the second thing it carries, and a receiver whose leak is too small to be worth 0.2 dB still has a reason to fit it.

CM-0008 Employment of Optical Isolation
Technique → Security Layers 2
Description Category Impact Type

Electronics and analog components, as a secondary effect, because the magnitude of the optical leak is set by the detector's analog drive circuit rather than by anything optical. The emission scales with the avalanche charge: with the laser off, the measured rate is linear in the avalanche photodiode's dark current at a fitted slope of 0.98, which strongly points to backflashes originating in the InP multiplication region [KoehlerSidki2020]. The two parameters that move it are electrical. Excess bias raises the leakage because it raises the carrier count; and the gating waveform lowers it, because the gate's falling edge quenches the avalanche before the active-quenching circuit would, so a window of about 5 ns or less removes the portion of the emission generated during that quench [Meda2018]. Gating class is itself a circuit property, a 1 GHz self-differencing front end leaking about 0.5 per cent against 6.0 to 9.8 per cent for a megahertz-gated one [KoehlerSidki2020]. A defender who reads only the L1 mapping would not see where the knob is.

L2 — Electronics & Analog Components Secondary Effect

Physical layer, quantum and optical, and the primary target. The whole mechanism is optical: secondary photons produced by the avalanche in the detector's multiplication region [KoehlerSidki2020] traverse the receiver's own analysing optics, which imprint the detection outcome on them because the photons pass through other security-critical components of Bob's receiver and carry out information about the state of those components [Pinheiro2018], and leave through the legitimate optical port into the channel. Every quantity that bounds the attack is likewise optical: the emission band, the 1.3×10−31.3 \times 10^{-3} photons per detection event coupling into the channel's single spatial mode, the reverse transmission of about 0.091 through a real receiver, and the extinction ratio of about 22 that caps her decoding [Kurtsiefer2001,Pinheiro2018,Su2025a]. Every control that attenuates it, spectral filtering, spatial filtering and reverse isolation, is an optical element in that same path.

L1 — Physical Layer (Quantum & Optical) Primary Target
Related Techniques 1
Relation type Description Technique
Alternative to

A-0049 Quantum Trojan-Horse Attack also reaches information through the legitimate optical port of the QKD equipment. The two rows divide the routes between them: A-0049's own scope section states that it covers the active route only and names this row as the carrier of the passive one. The grouping has a source: Vakhitov2001 treats large-pulse probing and light emission from avalanche photodiodes as two members of one conventional-optical-eavesdropping family, and Navarrete2026 covers receiver leakage through detector backflashes or back-reflected Trojan-horse light in a single clause [Navarrete2026]. What separates them is direction and consequence, and those stay distinct on the two rows: this one is cat04 with channel_action measure, channel_directionality backward alone, leakage_stage detector and outcome_class partial_key, while A-0049 is cat15 with inject, backward and bidirectional, source and full_silent_key. The vulnerability chains are separate too: this technique links V-0030 and V-0065, while A-0049 links V-0002, V-0031 and V-0033.

A-0049 Quantum Trojan-Horse Attack (THA)
See Also
A-0077 Blinding Attack on Self-Differencing APDs
A-0078 Illuminating and Faked States Attack on SNSPDs
A-0081 Spatial Misalignment Exploitation in QKD
A-0082 Wavelength Manipulation Attack in Free-Space
A-0083 Side Channel Exploitation in Transmitter Imperfections