Short answer: Device-independent quantum key distribution (DIQKD) aims to generate secret keys even when the users' devices are uncharacterised and possibly untrusted, by checking a Bell inequality violation. It is very demanding because it needs high-quality entanglement between remote sites with high detection efficiency. A 2022 Nature paper reports a DIQKD experiment between two buildings 400 metres apart, with a secret key rate of 0.07 bits per entanglement event under asymptotic assumptions. This is a laboratory demonstration, not a commercial system.
What device-independent means
The 2022 Nature paper explains that DIQKD generates secret keys over an untrusted channel with uncharacterised, potentially untrusted devices. Their proper functioning is certified by a statistical test using a Bell inequality. That test comes from the foundations of quantum physics, and it also ensures robustness against implementation loopholes, leaving only the integrity of the users' locations to be assured by other means. A 2021 arXiv version of the experiment says the idea was first introduced by Mayers and Yao, and that users need only analyse their input-output measurement statistics to bound what an eavesdropper could know, relying on a loophole-free Bell test.
Why it is hard
Both papers describe realisation as extremely challenging, because it needs a loophole-free Bell test across two remote locations with very high quality entanglement, and that is hard because high detection efficiency is needed at both ends. A 2013 Physical Review X paper makes the same point from a protocol angle: DIQKD in its original design needs a Bell inequality violation across a quantum channel, but real channels are lossy, so such protocols are vulnerable to attacks exploiting the detection loophole. It adds that DIQKD requires an almost lossless channel, which drastically limits the possible distance between the users.
A local Bell test proposal
The 2013 authors propose a different protocol in which the Bell test runs entirely on two causally independent devices inside the sender's laboratory. Because the test does not cross the channel, the detection loophole caused by channel loss is avoided.
The 2022 experiment
The experiment uses event-ready entanglement between two independently trapped single rubidium-87 atoms in buildings 400 metres apart. The authors report an entanglement fidelity of at least 0.892, and using a DIQKD protocol with random key basis they observe a Bell inequality violation of S = 2.578, above the classical limit of 2, with a quantum bit error rate of 0.078. For the protocol this gives a secret key rate of 0.07 bits per entanglement generation event in the asymptotic limit, which the authors say demonstrates the system's capability to generate secret keys. The 2021 arXiv version states the key rate is positive with a probability error of 3 percent and analyses finite-size security against collective attacks.
What this does not tell you
The key rate is an asymptotic estimate, and finite-size security was analysed only for collective attacks. The distance is 400 metres in a laboratory arrangement. The papers do not describe a deployed network.
FAQ
What does device-independent mean?
Security is checked from the measurement statistics of a Bell test, without trusting the internal workings of the devices.
Why is it hard to build?
It needs remote entanglement of very high quality and detection efficiency, per both experimental papers.
What distance was reached?
400 metres between two trapped-atom setups in separate buildings.
What was the key rate?
0.07 bits per entanglement generation event in the asymptotic limit.
Is it ready for use?
These papers describe a laboratory experiment.
All field notes