rose.blog

Over the past few years we here at D-Wave have been busy turning blood, sweat, tears and cash into quantum computers. We tend to be tight-lipped about what we’re up to and rarely disclose anything directly related to what we’re building.

One strategy is to continue operating in this mode until we’ve built quantum computers that can solve problems obviously beyond the scope of any conventional computer, go on a road show and build a customer base. Now anyone who has been involved in technology commercialization should immediately realize that this approach has a basic flaw - we would be building the technology without having any input from end users. This is a very common and usually fatal mistake that a lot of techno-centered organizations make. There are several examples just from aborted attempts at commercializing superconducting gadgets. Products - yes, even quantum computers - are all about satisfying some customer need. Build in isolation from end users and usually you die.

Another strategy is to demonstrate versions of the machines we’ve built that are not yet competitive with conventional systems, but demonstrate all of the functionality of the larger scale machines we’re planning on building in the future.

This strategy has the advantage of announcing to potential users of the technology what the capability of the machines is expected to be, how to interact with them, how to program them, etc., with the objective of developing partnerships with eventual customers who will help guide the development of the hardware and applications that will run on it.

There is an unfortunate giant chasm between people who are potential users of this type of technology, and who would benefit alot from it, and the people who are building it (ie. us). This chasm needs to be bridged and I think the best way to do this is to announce what our machines do now.

Now we’ve been working on building real large-scale quantum computers for a long time, and have spend a lot of money. We’ve done things using “best practices”, using a full-time workforce made up of the best and brightest who share our common vision of building these things. We are considerably ahead of the “state of the art” as it’s known in the academic community that works in quantum computation.

Now let’s say we wanted to announce some capability in order to attract co-development partners that will help shape our products to be valuable for these partners. When we announce where we are, the reaction from the expert community is predictable: there will be a large amount of scepticism and disbelief.

“Where are the peer reviewed papers? Where is your proof of (fill in your favorite QC dogma requirement)?” Well heck if we told you everything about what we’re doing, that wouldn’t make much sense now, would it?

But on the other hand the scepticism is of course warranted. If I were working somewhere else on QC, I probably wouldn’t believe it either.

So what I’m going to do is release just enough technical information so that when we do begin to make our announcements, the experts among you should have enough to admit that, just possibly we may have done what we’ve claimed.

The first of these technical releases was the paper on one of the couplers we’ve built. This paper is here. We released this paper to demonstrate that we have the capability to produce world-best, quantum regime superconducting hardware.

The second of these papers, which I’m going to link to in this post in a bit (the first public release ever! Even prior to arxiv! See reading this blog does pay off!) describes the underlying model of quantum computation that our machines use.

See I really hate the gate model. For me, it’s personal, because we tried very hard to build a gate model QC. What I learned is that the gate model is very stupid from the practical standpoint, especially with superconducting systems. Not that I want to belabor the point, but you can actually demonstrate that the interaction of superconducting qubits with the intrinsic nuclear spins in any of the materials that superconducting electronics can actually be built out of (ie. niobium, possibly aluminum) creates errors above the amount required by the threshold theorem. It’s a simple calculation that completely rules out gate model quantum computers built using niobium or aluminum. Try it! Tell your friends!

Anyway once we realized that the gate model was out, we had to either (A) give up and get real jobs or (B) find a different model of computation. So we looked around and found this cool thing called adiabatic quantum computing, which seemed to have an intrinsic resistance to the types of noise found in real niobium electronics.

So we started building components and AQC circuits and fell in love with the model. We started building more and more advanced stuff and found out a huge amount of fascinating things that would make your head fall off if we disclosed it all.

One of the coolest things we found out is that noise actually helps AQC. Yes, you read that right. AQCs can be run in the presence of noise, without quantum error correction, and still provide optimal scaling.

This appears to fly in the face of conventional QC dogma re. decoherence times, threshold theorems etc. but actually it doesn’t. It is entirely consistent with all of the rest of the QC literature, although the details of why this is true are highly non-trivial.

The key finding that we’re disclosing now is that quantum error correction (and fault tolerance) can be achieved PASSIVELY instead of actively, which in practice is the difference between large scale quantum computers today and in 20+ years (or whatever your favorite misty future date is).

Anyway with that as a brief intro, I am pleased and honored to release, for the very first time, an introduction to the model of quantum computation underlying our processors. We’ve disclosed an analysis of the model as applied to the adiabatic Grover search problem, which is of course sort of artificial, but demonstrates the basic principle.

Here it is! Enjoy!