Tuesday, August 29, 2006

Rose on Analog

Gordie Rose, of D-Wave Systems, has a new blog, and in a recent post talks about hating the gate model, and his strong preference for adiabatic quantum computation. Interestingly, he refers to the gate model as a useful theoretical construct but impractical to implement, whereas cluster state and adiabatic are practical.

My opinion, admittedly less informed than Rose's, is exactly the opposite. This comes, no doubt, from my background as a classical digital computer guy. I know how to create languages and compilers for them, and how to program them, how to build them, how to make them fault tolerant. Despite having taken a class from Carver Mead, I understand almost none of those critical topics for analog (whether quantum or classical).

Moreover, many quantum algorithms are essentially digital in nature (though the interference in the QFT can be described as an analog phenomenon). The best use of cluster-state computing we know of at the moment is as a substrate for the circuit model. And, at least for, say, digital arithmetic (something I know well), the cluster-state model uses one hundred times the resources of a more straightforward implementation. Michael Nielsen is working on some intriguing things, combining cluster-state with error correction in ways that may be more robust or more efficient, but that 100x penalty is a big one to overcome as we struggle to get even a few gates working properly.

Don't get me wrong -- I'm a big fan of analog computing. But Rose's opinion I find a little startling, and it will prod me into some more reading on adiabatic quantum computation, which has been on my list of things to do anyway...


At 1:00 PM, Blogger superperro1 said...

Rose says that his definition of analog does not mention continuous variables. Well, Ignoring the truth does not make it go away. The earliest classical computers were also analog; people quickly found out that they were too noisy. I predict D-Wave will learn the same hard truth very soon. If they don't have a plan B, I don't think they will last very long

At 9:42 PM, Blogger superperro1 said...

This reminds me of slide rules: analog computers, easy to implement, and they do work. But with E-3 accuracy, whereas PCs give you E-15 accuracy


Post a Comment

<< Home