Podcast Summary: Oliver Dial, IBM Quantum CTO on "The Superposition Guy"
Qubit calibration, reducing the energy cost of quantum computation, and modularity as the future for quantum system design
A shorter post today. I was originally going to do a Note, but there were some additional points which merited an actual post. Normally I don’t write about “the news”, because the focus here at The Quantum Stack is on the trends and key ideas, not so much the “sturm und drang” of the day-to-day news cycle.
That said, recently IBM Quantum’s Chief Technology Officer, Oliver Dial, did a podcast with the folks at Global Quantum Intelligence. You can listen to it here. There were a few things discussed in the podcast which I wanted to highlight for you:
Shifting our calibration strategy from "strive for the absolute best performance" towards "ensure the most stable behavior" for the purposes of error mitigation.
Driving down the energy cost per qubit through advances in control electronics.
The importance of modularity for scaling.
Let’s take a quick look at each. (I will again issue my standard disclaimer that the opinions and viewpoints here do not represent those of my employer!)
Calibration Strategy Shift
Calibration refers to the act of tuning up and maintaining your qubits. Previously, there was a big focus on calibrating the qubits to ensure maximal performance (i.e., lowest possible errors). That’s a big advantage when doing any kind of demonstration which requires executing circuits with as little noise as possible (e.g., quantum volume tests).
However, the process of getting those calibrations done can be quite time-intensive. (In the podcast, Oliver says something like 15% of each day is spent calibrating the systems.) If ensuring the absolute best performance was essential, then perhaps that’s an acceptable loss. There are other considerations which have come to change that line of reasoning though – one of them being error mitigation.
Error mitigation “virtualizes” noise, in the sense that it defines via software a simulation of a less-noisy quantum computer. It does so by classically post-processing the outcomes of running an ensemble of circuits. Some error mitigation techniques (such as probabilistic error cancellation) require the noise on the hardware to be learned (or modeled) before the technique can be applied. Learning this noise also takes time, especially if the model has many parameters.
As such, you can see why it would be nice if the systems were stable over time – you could learn a noise model less frequently, and spend more time running useful computations for end-users. Hence, it would be good to change how the systems are calibrated, so an increased level of stability is achieved.
Decreasing the Energy Cost per Qubit
This part of the podcast surprised me a bit. There’s relatively little public information about how much energy a quantum computer consumes (though some estimates exist; see page 13 of this paper). So it was pretty neat for Oliver to go into this a little. In what follows below, I’m using the numbers directly from the podcast.
It turns out that for the first systems we put online (which were 5 qubits), using commercial off-the-shelf electronics, the estimated energy cost per qubit was 300 Watts. So that’s about 1.5kW of energy to run the system.
Later systems (on the order of 20 qubits) also used commercial off-the-shelf electronics, but had a cost of only 70 Watts per qubit. The cost per qubit went down by about a factor of 4. But, at 20 qubits, we’re still talking about 1.4kW of energy for the whole system.
Using the latest generation of electronics, the cost per qubit has been decreased by a factor of 2, and is down to only 35 Watts per qubit. So all in all, from Gen 1 electronics to Gen 3, the energy cost per qubit has gone down by an order of magnitude.
Still, that’s not enough if we’re going to keep scaling. Consider, after all, our recent announcement of a goal to build a 100,000 qubit system by 2033. Blithely scaling the control electronics (and assuming you don’t have any additional overheads) would suggest you’d need 3.5 MW of energy to run such a system. This is about 100 times the amount of energy an average American household uses each day!
So clearly, driving down the energy cost is important. Oliver highlights one path we’re exploring to do so: cryoCMOS, which essentially moves some of the classical control electronics inside the dilution refrigerator. They have the advantage that, being much closer to the qubits, you don’t need to drive as much power to them to do the required control – perhaps on the order of 10 milli-Watts per qubit. That would be several thousand times lower!
System Modularity Enables Scaling
1 million – that’s the common consensus on how many physical qubits we’ll need to build in order to prototype a fault-tolerant quantum computer.
That’s a lot. And building them in a monolithic architecture just really isn’t in the cards. We’re going to need a modular architecture to do so.
This realization isn’t particularly new: for example, with trapped-ion quantum computing, there have been proposals for at least a decade now to construct modules which are networked together to realize a large-scale quantum computer. See here, here, and here.
What is new these days about the topic of modularity is how classical and quantum communication can come together with quantum computation to realize large-scale quantum computing systems.
If you’ll indulge me a little bit of marketing pizzazz, a video we put out about our IBM Quantum System Two last November makes this clear:
The future of quantum is modularity, at all levels of the stack. From clusters of chips inside a fridge, to multiple fridges classically networked together, to whole data centers with quantum interconnects, scaling up quantum requires building well-architected systems which embrace modularity at their core.
So, what do you think? Should I do more of these kinds of posts? Let me know in the comments below! And if you know even just one person who could benefit from reading The Quantum Stack, please share this post with them.