Hadamard Get the app

Device-Independent Quantum Key Distribution at 100 km, Explained

Device-Independent Quantum Key Distribution at 100 km, Explained — Hadamard

Device-independent quantum key distribution (DI-QKD) is a way to share a secret key without trusting how the hardware works. Its security comes from a measured violation of a Bell inequality. In February 2026, a team at USTC reported in Science that it had run DI-QKD between two single rubidium atoms connected by up to 100 km of fibre. The full finite-key security proof, though, covers only 11 km, and all the equipment sat in one laboratory.

That combination is why the headlines are both accurate and easy to misread. This article covers what “device-independent” means, how it differs from BB84 and MDI-QKD, what the USTC paper (arXiv 2602.09596, published as Science 391, 592–597) actually measured, and how far the result still is from real use.

What is device-independent quantum key distribution?

Practical attacks on QKD rarely break quantum physics. They go after the equipment: detectors that can be blinded, sources that leak side information, components that do not behave the way the security proof assumes. Every protocol that relies on a model of its devices inherits that model’s blind spots.

DI-QKD removes the model. Alice’s and Bob’s devices are treated as black boxes. Each box gets an input (a measurement setting) and returns an output. Nothing is assumed about the source, the measurement apparatus or Alice’s and Bob’s own detectors. If the statistics of the inputs and outputs violate a Bell inequality strongly enough, a secret key can be extracted, whatever is inside the boxes.

“Device-independent” does not mean nothing is trusted. A few assumptions remain:

  • Alice’s and Bob’s labs must not leak information to the outside.
  • The randomness used to choose measurement settings must come from a trusted source.
  • The classical post-processing that turns raw outputs into a final key must be trusted.

How a Bell violation becomes a security meter

The idea of using entanglement and Bell tests for key distribution goes back to Ekert’s protocol, covered in our explainer on E91 and how Bell tests secure quantum keys. Turning that intuition into a device-independent security proof took much longer.

A key step was the 2007 paper by Acín, Brunner, Gisin, Massar, Pironio and Scarani, “Device-independent security of quantum cryptography against collective attacks” (Phys. Rev. Lett. 98, 230501). It gives a tight bound on how much an eavesdropper can know as a function of how strongly a Bell-type inequality is violated. It makes no assumptions about how the devices work.

In practice, the meter is the CHSH value, S. Any classical, pre-programmed explanation of the results gives S of at most 2. The further S rises above 2, and the lower the error rate between Alice’s and Bob’s outputs (the QBER), the less an attacker can know and the more secret key survives. The finite-key result in the 2026 paper is proven against general attacks, which is a stronger guarantee than the collective-attack setting of the 2007 bound.

DI-QKD vs MDI-QKD vs BB84: what each one trusts

Each generation of QKD removes another thing you have to trust. The table below sets out the differences.

Protocol Source / state preparation Measurement devices Where security comes from
BB84 (prepare-and-measure) Trusted and characterised Trusted and characterised Quantum physics plus accurate device models
MDI-QKD Trusted (Alice and Bob prepare states) Untrusted relay in the middle Quantum physics plus trusted state preparation
DI-QKD Untrusted Untrusted, including Alice’s and Bob’s own Observed Bell (CHSH) violation

MDI-QKD closes the detector side-channels that caused most real-world QKD hacks, as covered in our MDI-QKD explainer. DI-QKD goes further by also removing trust in the source. The cost is steep. The protocol needs high-quality entanglement between the two ends and a Bell violation strong enough to survive noise and loss. That is why every DI-QKD experiment so far has been slow and short-range.

The 2022 firsts: 2 metres, 400 metres and 220 metres

Three groups published the first experimental DI-QKD results in 2022, each with a different platform and level of proof:

  • Oxford, trapped ions. Nadlinger et al., Nature 607, 682–686 used 88Sr+ ions in two nodes about 2 m apart. They produced 1.5 million Bell pairs over about 8 hours (7.9 h of acquisition) and ended with 95,628 net secret key bits. The characterisation run showed S = 2.677 and a QBER of 1.44%. It was the first complete DI-QKD protocol with finite-statistics security. The 2026 USTC paper describes it as finite-key security over 3.5 m of fibre.
  • Munich (LMU), single atoms. Zhang et al., Nature 609, 687 heralded entanglement between two independently trapped 87Rb atoms, with users 400 m apart. Security was asymptotic, with an expected key rate of 0.07 per entanglement event.
  • USTC, photons. Liu et al., PRL 129, 050502 used a polarization-entangled photon source with about 87.5% heralded detection efficiency. They showed a positive key rate for fibre up to 220 m. This was a proof-of-principle verification against collective attacks.

As the Chinese Academy of Sciences release put it, DI-QKD had been limited to a few hundred metres.

How USTC took DI-QKD to 100 km with single atoms

The 2026 paper, first-authored by Bo-Wei Lu, includes Ernest Y.-Z. Tan, Feihu Xu, Qiang Zhang, Xiao-Hui Bao and Jian-Wei Pan among its authors. It ran DI-QKD between two single-atom nodes over total fibre lengths of 11, 20, 50, 70 and 100 km. Three engineering choices made the longer distances possible.

Telecom-band photons

The 87Rb atoms emit photons at 780 nm, where fibre loses about 3 dB per kilometre. Quantum frequency conversion shifts them to 1315 nm in the telecom O-band, where loss is about 0.32 dB/km. That is almost a tenfold drop per kilometre, and loss compounds over distance.

Single-photon heralding

The atoms’ entanglement is heralded by single-photon interference: a detection event signals that the two distant atoms are now entangled. According to the full text, this makes the entangling rate “orders of magnitude higher” than the two-photon-interference schemes used in the earlier ion and atom experiments.

Low-noise emission

A Rydberg-based emission scheme suppresses the recoil that photon emission imparts to the atom, without adding noise. This matters because noise lowers fidelity, and fidelity drives the Bell violation.

Reading the numbers: finite-key at 11 km, asymptotic at 100 km

Two kinds of key rate appear in the paper. An asymptotic rate assumes an unlimited number of rounds. A finite-key rate accounts for the statistical uncertainty of the data actually collected, and only a finite-key result gives a usable security guarantee for a real key.

Total fibre Atom–atom fidelity QBER Security analysis shown
11 km 0.947 2.85% Finite-key, general attacks
20 km 0.933 — Asymptotic only
50 km 0.931 — Asymptotic only
70 km 0.921 — Asymptotic only
100 km 0.911 7.16% Asymptotic only

At 11 km the team measured S = 2.612 ± 0.031, well above the classical bound of 2. They collected 1.208 million heralded events over 624 hours. With security parameter ε = 10−5, the finite-size secure key rate was 0.112 bits per event. That works out to about 135.3 kbits of key in total, or an average of 0.06 bit/s. The asymptotic rate at the same distance was 0.275 bits per event. With a much stricter ε = 10−15, the rate was still positive at 0.075 bits/event.

From 20 km to 100 km, the paper shows only positive asymptotic key rates. There is no finite-key analysis at those distances. So “100 km” means a demonstrated positive asymptotic rate, not a finite-key secret key distributed over 100 km.

The long run also shows how fragile the setup is. The mean event rate at 11 km was 0.53 per second, against 0.72 per second in an earlier measurement. The authors attribute the drop to long-term efficiency drift.

What the 100 km result does not prove

The paper is open about its limits. The main ones:

  • One laboratory. “All three nodes are housed within the same laboratory,” and the distance comes from spooled fibre, not separated sites.
  • Locality loophole open. “The present implementation does not close the locality loophole,” because the measurement events are not space-like separated.
  • Low duty cycle. Reading out the atom destroys it, so it has to be reloaded. That takes hundreds of milliseconds and limits the duty cycle to about 0.15.
  • Finite-key proof only at 11 km, at an average of 0.06 bit/s.

Outside experts reacted along the same lines. Speaking to the Science Media Centre Spain, Antonio Acín (ICREA, ICFO) called it “Excellent. It is a major achievement and a very significant improvement over the state of the art.” He also noted that the nodes “are located in the same laboratory, but connected by a fiber whose length is tens of kilometers,” and that “a complete implementation has not yet been achieved.” Carlos Sabín (Universidad Autónoma de Madrid) said “we are still far from being able to achieve a completely secure, error-free quantum key distribution.”

What comes next

The authors project, as an estimate rather than a result, that their proposed improvements could push the entangling rate above 10 kHz over 10 km. Beyond that, a real deployment needs genuinely separated nodes, space-like separated measurements and a way to bridge longer distances. That last point brings in quantum repeaters, which we cover in our quantum internet explainer.

For systems that need quantum-resistant security now, the practical route is still post-quantum cryptography running on ordinary hardware. That is the approach Hadamard takes with BB84, a hash-based, ECDSA-free Layer 2 rollup on Ethereum whose accounts use one-time addresses based on hash commitments instead of ECDSA keys.

FAQ

What is the difference between device-independent QKD and measurement-device-independent QKD?

MDI-QKD removes trust in the measurement devices by moving measurement to an untrusted relay, but Alice and Bob still have to trust their own state preparation. DI-QKD trusts neither the source nor any measurement device, Alice’s and Bob’s included. Its security rests on an observed Bell violation.

Did China really send a device-independent quantum key over 100 km?

Not quite in the sense the headlines suggest. The 100 km was spooled fibre inside one laboratory, and beyond 11 km the paper shows only positive asymptotic key rates. The finite-key secure key, about 135.3 kbits against general attacks, was produced at 11 km.

Why is DI-QKD so slow?

Every key bit depends on a heralded atom–atom entanglement event. These arrived at about 0.53 per second at 11 km. Each atom also has to be reloaded after readout, which keeps the duty cycle near 0.15, and finite-key security needs a very large number of events. The result was 624 hours of data and an average of 0.06 bit/s.

Does DI-QKD make QKD unhackable, and what loopholes remain?

It removes the need to trust the devices, which closes the device flaws that most QKD attacks exploit. But the 2026 experiment leaves the locality loophole open. Any DI-QKD system still assumes that the labs do not leak information, that the setting choices come from a trusted random source, and that the classical post-processing is sound.

Sources