Google’s Quantum Computer | Vol. 3 / No. 6.4

Wafer of D-Wave quantum chips | Photo: Steve Jurvetson, CC BY 2.0
Wafer of D-Wave quantum processors (detail) | Photo: Steve Jurvetson, CC BY 2.0

In which I largely fail to describe what’s going on.


In the news yesterday, it came out that the “D-Wave X2 Quantum Computing System” that Google teamed with NASA to buy in 2013 does seem to be much, much quicker at solving certain kinds of problems than a traditional computer. This is pretty impressive news, but it’s important to insert a caveat.

This isn’t exactly a quantum computer; it’s more like a quantum ASIC (or maybe a quantum ASSP?). But I’m going to need to back up a little to explain.

So a normal processor, like you have in your laptop, can do a lot of things. It can process instructions in the form of software to help you with everything from doing your taxes to fighting imaginary aliens. But being good at everything comes at a cost. There are other parts of your computer that do more specific things much better, for example your computer’s GPU. The GPU or Graphics Processing Unit focuses on doing one thing and doing it well, in this case, rendering graphics. They’re optimized to do the same calculation over and over again, and since they’re only doing that one thing, they can do it really efficiently.

The next level of focus is something called an ASIC, or an Application-Specific Integrated Circuit. These have become really popular in the Bitcoin world, because they allow users to perform a single function — in this case performing the single function that is mining Bitcoin — very, very well. They do it so well, in fact, that unless you’re using an ASIC setup, you’re not going to be doing much mining these days. These devices take the GPU specialization idea even further — they’re whole devices that only do one thing, but do it very very well.

Enter the D-Wave.

The D-Wave system is something called a quantum annealer, which is pretty hard to understand and maybe even harder to explain. Annealing uses some of the more unusual physical properties of the universe to solve a special kind of question. If you think of Schroedinger’s Cat (the one that’s alive, dead, or both all at the same time) a quantum annealer puts its bits into a state of superposition (on, off, and both), applies a field to it that bounces them around until they settle in their most comfortable state, and that low-energy state is the answer to the question. Yeah… that didn’t explain it very well at all.

Anyhow, it’s very good at answering questions that can be posed just that way, but this is different from a “traditional” or “universal” quantum computer (an odd name given that none exist as yet), which in theory would be able to run something called Shor’s Algorithm which allows for all the crazy encryption-busting things you’ve heard about (among others).

So what Google and NASA have done is to show that (a) quantum effects are definitely taking place in this machine, something that not everybody was convinced of, and (b) that those quantum effects make it do certain things fast. They did this by comparing it with something called a simulated (or thermal) annealer, which uses temperature measurements to mimic the function of the quantum tunneling effects. Basically, it’s a non-quantum annealer built to model quantum annealer properties. Something like that SNES emulator you’ve been running on your phone (you know who you are).

And when they tested it against the simulated annealer with very specifically designed questions — ones designed to prove that the quantum tunneling effects were taking place and that they had a positive effect — they found a very big difference: “For instances with 945 variables this results in a time-to-99\%-success-probability that is ~108 times faster than SA [simulated annealing] running on a single processor core.” Just so we’re clear, that “~108 unpacks as “around one hundred million times faster.”

Now, these are pretty darn specific questions they’re asking it. And they aren’t applicable to a whole lot just yet. But according to Google’s team, they “are optimistic that the significant runtime gains we have found will carry over to commercially relevant problems as they occur in tasks relevant to machine intelligence.” And they’re using what they’ve learned to help steer the direction these specific-function quantum-effect-utilizing computing devices will go in.

So while it’s not quite what you’re thinking of when you say “quantum computer,” it’s still a kind of “quantum computing,” and that makes it pretty darn cool, especially because they’re gearing it toward future advances in machine learning. Who knows, maybe this is another step toward AI.

Check out the coverage from Ars Technica and Engadget, as well as the press release from Google, and the prepress paper for more.


Richard Ford Burley is a writer and doctoral candidate at Boston College, as well as an editor at Ledger, the first academic journal devoted to Bitcoin and other cryptocurrencies. In his spare time he writes about science, skepticism, feminism, and futurism here at This Week In Tomorrow.