At Cambridge Quantum, we develop quantum entropy products, so at first it may come as a surprise that we thoroughly agree with the NCSC’s position and amplify it loudly to our customers. The NCSC got it spot on – there’s no place in a high-security environment for noisy QRNGs.
To mark the (almost) anniversary of the NCSC statement, this article carefully examines the criticisms levelled at existing QRNGs and explains the flawed approach taken so far. We then explain the approach taken by Cambridge Quantum, which tackles the issues head-on; not by slippery debate, but with deliverable and measurable actions. Cambridge Quantum’s approach provides truly perfect quantum random numbers that we then use in our cybersecurity products.
We explain how, if required, we can supply perfect and verifiably-quantum randomness, but that our mission is to solve our customers’ problems with actual products and services that draw on this randomness and work in a zero-trust environment.
The Flawed Approach to QRNGs
Before we get into the specifics of the NCSC statement, it’s helpful to understand how a typical QRNG device works. Later in the article, we’ll discuss the fundamentally different approach Cambridge Quantum has taken.
The goal of any QRNG is to generate a stream of random numbers. To do this, most QRNGs fire photons from lasers down a series of paths within their device. Along the way, photons are directed towards angled mirrors, which deflect about half of the photons down one route, while allowing the remainder to pass through. These photons eventually hit detectors, which can register their presence. Ideally, half the photons hit one detector and half hit the other. By labelling one detector as “1” and the other “0”, these QRNG devices can generate strings of random-looking data.
These QRNG devices rely on their physical construction as proof of randomness. To have any confidence in the randomness of their output, one has to inspect and consider the physical structure of the device. Since the devices cannot be built perfectly (since engineering perfection is impossible when you look close enough), these devices usually produce raw output that demonstrates a small bias in one direction or another. Perhaps there is slightly more 1s than 0s, for example. To try and mitigate this, the raw output is usually post-processed with software functions to try and reduce the bias. In short, the output from the devices isn’t truly random, so software is used to try and mask this and improve the situation.
Are These Devices Suitable for High-Security Use?
The NCSC squarely addresses these flawed QRNGs when it presents the advice in its statement. The NCSC acknowledges that, in theory, quantum technology can provide “truly unpredictable numbers”, but in practice, these QRNGs have fallen short.
One concern the NCSC raises is that classical noise is introduced by all the measurements we discussed earlier. Because these flawed QRNGs rely on fine-grained measurements of photons, they are heavily dependent on the quality of the detectors and influenced by electrical noise from surrounding components. It’s like trying to accurately dictate what someone is saying, in a room where a rock band is playing at full volume. You’re not going to get it right every time and that’s a real issue for security. As the NCSC says: