Regarding how their system operates, they write:
EQC operates on the most fundamental principles of quantum physics, especially its measurement postulate, wherein the wavefunction of a quantum system will collapse to a certain eigenstate due to its interaction with a measurement apparatus or, broadly speaking, the surrounding environment.
Well... I'm confident that this sentence is not false. Then again, I'm not sure how else a quantum device would be supposed to operate. By which I mean, they're really not giving very useful info here. They then continue writing:
However, while existing quantum computing architectures must operate on closed quantum systems under extreme requirements to calm the effects of the environment, EQC operates on open quantum systems, carefully coupling a quantum system to an engineered environment, so that its quantum state is collapsed to represent a problem’s desirable solution.
This does actually some clues. Apparently they're claiming to use open system dynamics, in a way that is quite noise-resilient. There are schemes to engineer environments in order to drive a system towards a certain state. A keyword here is quantum reservoir engineering. Although they're writing that the quantum state is "collapsed to represent a problems' desirable solution" confuses me. If they're doing some sort of reservoir engineering, which means, roughly speaking, to create a coupling with an environment such that a certain target state is a fixed point of the evolution, talking about "collapse" seems a bit weird to me. This aside, in this sort of scheme I'd guess that the problem parameters are specified through interactions between reservoir and system, which therefore does not really have an "input" like gate-based models do.
For a scheme like this, a compelling question would be how easy it is to find the parameters to use to tune the dynamics in order to obtain the wanted result. And how to figure out what sorts of dynamical processes both have a good resilience to noise, and allow you to manageably figure out interaction parameters resulting in the solution of a given optimization problem, using only the details defining the optimization problem itself (and hopefully, without having to solve the optimization problem itself in order to find the parameters allowing the device to solve the optimization problem).
The most detailed description of the method is around 5:03 in the presentation, where they say that (emphasis mine):
we use [our entropy quantum computer] as a free source of energy and we engineer a tightly controlled interaction with the environment. When you do that you get a system that allows you to unlock a lot of very powerful quantum operational features. [...] The real secret to how the technology work is to very carefully couple the system, it's a photonic system, to the engineered environment which we'll call the entropy. And that quantum state is allowed to relax or collapse to a desired solution across the many quantum modes in the optical system. We achieve full operation at room temperature and because there are so many photonic modes available to us we can solve very large [variable states?] problems.
They then say
[...] we consume the problem by directly feeding the Hamiltonian into the EQC, no preprocessing or postprocessing software involved, and this controlled feedback, or backaction, from the environment to the system, allows us to produce ground state information for the objective function and capture all of the constraints in the solution which we subsequently analysed. The total runtime was just over 6 minutes.
This is the most interesting part in my opinion. A few comments:
- They say they have a photonic system with many photonic modes. This points towards a continuous variable kind of scheme, maybe akin to what other companies such as Xanadu are doing. Xanadu also claims their scheme to allow for fault-tolerant room-temperature quantum computation, as per https://arxiv.org/abs/2010.02905.
- I'm puzzled by their saying that there is no pre- or post-processing involved, but also that they use a feedback system. I suppose what constitutes "processing" is subjective? Still, the use of a "controlled feedback from the environment" and "produce ground state information" points towards a hybrid optimisation approach: evolve the system a bit, measure to see how it's currently doing, tune the parameters of the interaction accordingly, and iterate.
- Referring to the "engineered environment" as "entropy" is... puzzling. I suppose they couldn't just talk of "reservoir computing" because that name is taken to mean something else?