There are many different schemes for making quantum computers work (most of them evil). But they pretty much all fall into two categories. In most labs, researchers work on what could be called a digital quantum computer, which has the quantum equivalent of logic gates, and qubits are based on well-defined and well-understood quantum states. The other camp works on analog devices called adiabatic quantum computers. In these devices, qubits do not perform discrete operations, but continuously evolve from some easily understood initial state to a final state that provides the answer to some problem. In general, the analog and digital camps don’t really mix. Until now, that is.

The adiabatic computer is simpler than a quantum computer in many ways, and it is easier to scale. But an adiabatic computer can only be generalized to any type of problem if every qubit is connected to every other qubit. This kind of connectivity is usually impractical, so most people build quantum annealers with reduced connectivity. These are not universal and cannot, even in principle, compute solutions to all problems that might be thrown at it.

The issues with adiabatic quantum computers don’t end there. Adiabatic quantum computers are inherently analog devices: each qubit is driven by how strongly it is coupled to every other qubit. Computation is performed by continuously adjusting these couplings between some starting and final value. Tiny errors in the coupling—due to environmental effects, for instance—tend to build up and throw off the final value.

For annealers with limited connectivity—each qubit is only connected to a few other qubits, rather than all other qubits—this is not such an issue. The coupling between these qubits tends to be strong, so the noise is small compared to the coupling. For a fully interconnected adiabatic quantum computer, however, the weak connections between distant qubits are very sensitive to environmental noise. Thus, errors accumulate—if you are unlucky, pi ends up equal to three.

Digital quantum computing, which uses logic operations and quantum gates, offers the possibility of error correction. By encoding information in multiple qubits, you can detect and correct errors. Unfortunately, digital qubits are delicate things compared to those used in adiabatic quantum computers, and the ability to program and run complex problems with them is out of reach at the moment.

What if the computation was performed by qubits that were operating as an adiabatic quantum computer, but with connections between the qubits controlled via a digital network of qubits?

What about a hybrid approach? That’s the question asked by a international group of researchers in a recently-published paper in *Nature*. They’ve tested a system where the computation is performed by qubits that were operating as an adiabatic quantum computer, but with connections between the adiabatic qubits is controlled via a digital network of qubits. This allows the benefits of scale and flexibility that you get from adiabatic quantum computing, while also making the connections less susceptible to noise.

## Digital vs. analog

Let me make an analogy here. Imagine that I have an instrument that measures the hot air concentration in Congress (dangerously high when in session). The instrument produces an analog voltage that is displayed on an analog meter right at the instrument, and the results are relayed to a second analog meter in my home in the Netherlands. The meter in Washington shows a reasonably accurate value with a strong correlation between the reading of hot air displayed in the Capitol Building and speeches by members of Congress. But the distance to my house is so great that my needle only shows an awful lot of noise.

So, instead of transporting the signal directly, I digitize it, encode it, and send it via a network to my home, where it is re-converted to an analog signal and read by my meter. Now, my meter is *almost* as accurate as the local meter. The only differences between the two are the errors due to the two conversions between digital and analog domains.

In other words, as long as the process that converts between digital and analog domains generates less noise than the transport of the analog signal, you win. And that is exactly the determining factor for a hybrid adiabatic quantum computer as well. The difference is that, instead of measuring methane concentrations, we are varying the coupling between qubits. Instead of a continuous, slow change, the coupling is stepped from value to value in jumps that are determined by the number of qubits in the digital part of the circuit.

Now, as you might have guessed, this is very expensive in terms of quantum gates. With the new hardware, the researches have an adiabatic quantum computer with up to nine *computational* qubits. In the interests of reproducibility, I can say that the digital connection between two qubits involves… wait for it… hmm, well, the researchers don’t say how they are connected to each other.

In fact, the whole paper in *Nature* seems to lack details on how this quantum computer is laid out. All we get is one tiny electron microscope picture of nine qubits in a row, with none of the coupling network shown. But, we get some idea of how the hardware works from various things that the researchers say when describing it.

For a four qubit case, the link between each qubit seems to require a cluster of *48 qubits* for control. The link itself is made up of five entangled qubits, coming to a grand total of 159 qubits. The authors also mention that for nine computation qubits, they need about 1,000 auxiliary qubits for control purposes. Yet despite all of the supporting digital architecture, there are still only 5 steps between the start and end of the computation.

## Controlling communication

The big question is, of course, does it work? And you know that it must have, because otherwise it, and the researchers that worked on it, would be gathering dust in a cupboard somewhere. The researchers were able to compare their implementation to a model of an ideal version and to a noise-free analog model. Now, since the digitization was very coarse, the answers to the toy problems that they got the computer to solve are not terribly accurate compared to the analog case. But they do show that their real digital version performs about as well as can be expected. That is, the model of the digitized adiabatic quantum computer and the real digitized adiabatic quantum computer performed about the same.

That is faint praise, though. My impression is that this is a huge engineering feat. The researchers have implemented not just a multi-qubit adiabatic quantum computer, but also a complex quantum digital network between the qubits. To give you an idea of the complexity: each qubit in the network is, even when you don’t want it to, able to talk to the rest of the network. So, setting the coupling between any two qubits varies the coupling of adjacent qubits too.

To prevent this, the researchers developed an impressive set of decoupling sequences. Instead of actually stating what this means, let me use an analogy. Imagine that you and some friends are in a row of rooms that are joined by a set of vertical windows. If two of you are standing up, you can communicate using hand signs; if you are both lying down, you can communicate with hand signs. But, if one of you is lying down and the other is standing up, you cannot see each other’s hands and no communication is possible. And, if people are in the rooms in between, they might block your view anyway.

So, for you to communicate with one of your friends, you have to do two things: you have to make sure you and your friend have the same orientation. And, since the rooms are all in a row, you have to make sure that all the people in between you are in the opposite orientation.

When I started writing about quantum computing, it was lab stuff… Now, things are starting to get scary.

This is pretty much what decoupling sequences do: they change the state of a qubit so that the coupling between it and the qubit you want to manipulate is at a minimum (ideally, zero, but in practice it is never quite zero). After you’ve performed the desired operation on the target qubit, you reverse the decoupling operation to return the qubit to its original state. This requires exquisite control and timing to get right. And, in this paper, that sort of control was impressively demonstrated on a large scale.

When I started writing about quantum computing, it was lab stuff of mostly academic interest. We celebrated every qubit and every time there was evidence of quantum goodness in our computing. Now, things are starting to get scary. Computations involving many qubits are common. And companies—not just startups, but serious companies that do serious things like setting milestones—are getting involved.

There is still a long way to go before a useful quantum computer emerges. But in the past there were also an awful lot of “if” statements associated with every “when” statement. Those qualifiers are being worked through very quickly, and the “when” is looking a good deal more certain.

*Nature*, 2016: DOI: 10.1038/nature17658