Quantum computing’s also-rans and their fatal flaws

[Karlston]

Quantum computing’s also-rans and their fatal flaws 

When it comes to performance, engineering matters more than physics.

Enlarge / IBM's 16-qubit quantum computer from 2017.

IBM quantum experience

Last month, Google claimed to have achieved quantum supremacy—the overblown name given to the step of proving quantum computers can deliver something that a classical computer can't. That claim is still a bit controversial, so it may yet turn out that we need a better demonstration.

 

Independently of the claim, it's notable that both Google and its critics at IBM have chosen the same type of hardware as the basis of their quantum computing efforts. So has a smaller competitor called Rigetti. All of which indicates that the quantum-computing landscape has sort of stabilized over the last decade. We are now in the position where we can pick some likely winners and some definite losers.

Why are you a loser?

But why did the winners win and the losers lose?

 

In the end, the story comes down to engineering. A practical quantum computer requires that we can create many quantum bits (qubits). Those qubits have to stay in a quantum state for multiple gate operations. Gate operations require that we are able to manipulate qubits on both an individual basis and in groups (or at least pairs). And, of course, you have to be able to read out the result of a computation.

 

Many of these individual features have been demonstrated to work using qubits in liquids, in Rydberg atoms, in Bose Einstein condensates (BECs), in solid-state systems, nitrogen vacancies in diamond, defects in silicon, trapped ions, light, and of course, superconducting rings. That list is incomplete, but, most of those options are pretty much dead in the water, and for very good reasons. While qubit behavior is dictated by physics at the individual qubit level, once you think about scaling, engineering really matters, and a lot of these options aren't very amenable to scaling.

Randomness is bad

Early in the decade, nitrogen-vacancy centers, silicon vacancies (which took a lot longer), and solid-state materials were among the front runners. These materials all operate on similar principles: a small percentage of a contaminant material is introduced to a crystal. Nitrogen is put in diamond, phosphorous in silicon, and ytterbium in yttrium-aluminum-garnet crystals.

 

The qubits in each material are formed by similar physics. The contaminant material can't match the bonding requirements of its neighboring atoms, which will leave an isolated electron or positively charged nucleus (an ion). The states of these isolated objects can be used as a qubit, and its states can last for a very long time—often longer than their more successful counterparts.

 

But there are fundamental disadvantages to these technologies as well. A good example of many of these can be seen in nitrogen-vacancy centers in diamond. Each qubit consists of a single electron left hanging by nitrogen's inability to bond with a fourth carbon atom. This electron is addressed (set and read) optically. Hence, the first problem is to search through a crystal for the few isolated vacancies that can be individually addressed. Optically addressing the qubits means that the vacancies are too far apart to directly couple, so qubit operations and entanglement between qubits has to be done via optical and microwave photons. Unfortunately, microwave emission will couple to all qubits, reducing the precision with which the qubit scan be controlled.

 

Even worse, each vacancy is different. The quantum properties of the vacancy are determined by the precise arrangement and type of atoms that surround it. For instance, in diamond, the two common isotopes of carbon provide enough difference that the presence of carbon-13 changes the performance of nearby qubits. To make the qubits identical, local magnetic fields need to be applied, which shifts the energy levels of the qubit states. That needs to be done by running relatively strong currents through nearby wires but simultaneously isolating the effects so that they don't influence other qubits.

 

Essentially, every chip of diamond will produce a different computer, with a different arrangement of qubits that have different properties. The wire routing to ensure that the local magnetic fields are truly local to the target qubit seems insanely difficult. Then you have to engineer tiny lens arrays (milled directly onto the diamond surface) to couple all the qubits to the outside world.

 

The tiny, suppressed part of my brain that understands engineering screams at the very thought.

 

These issues apply to almost all vacancy-based qubit systems, which is why we're hearing less and less about them.

Solids are memorable

The case for ions in crystal, like ytterbium in yttrium-aluminum-garnet, is a bit different. Here, the quantum state is not generally stored in a single ytterbium ion. Instead, the state is spread over a population of ions, which makes it incredibly robust—these are some of the longest-lived quantum states. However, it also makes defining the location of the qubit a bit difficult. Indeed, the location is defined by the optics that focus the light being used to set and read quantum states.

Enlarge / Look, you never forget a solid.

Getty Images / mikroman6

Essentially, the qubit state is set by pulses of light that interact with many ions within a volume of the crystal. To generate more than a few of these qubits requires a rather sophisticated optical setup. And that level of sophistication doesn't even take into account the requirement to be able to entangle qubits and generate gate operations. The engineering is, again, not favorable for a full quantum computer. On the other hand, these crystal make awesome quantum memories and may well still find application in that limited role.

Neutrality is indifference

Moving even further away from the practical, we come to the more outsider chances; examples of these include Rydberg atoms and Bose-Einstein Condensates (BECs).

 

Rydberg atoms are created by exciting the outer-most electron in an atom to a highly energetic state. In that state, the electron orbits much more like a planet around a star. A qubit can be created by managing the transitions between different Rydberg states. The state can be set and read by optical pulses and optical emissions. Cold Rydberg atoms can be trapped optically, holding them in a location, allowing them to be addressed with an optical system.

 

Unfortunately, their very nature prevents them from directly interacting with each other, so qubit operations have to be performed by the exchange of photons. This, like the case for ions in crystals, makes the optical system and the computational procedure much more difficult than it is for a more successful system. But there's also a difficulty involved in creating these qubits. Getting large numbers of Rydberg atoms in an identical starting state is not at all trivial.

 

A BEC provides a wonderful quantum state that can be manipulated and maintained with very high precision. And they're relatively easy to create. But, like Rydberg atoms, that quantum state doesn't really influence the quantum state of a neighboring BEC directly, making the arrangement of gates much more difficult.

That very definable winning quality

Now compare that with ion-trap quantum computers and superconducting qubit computers. In the case of ion traps, the quantum state is stored in and read from an individual trapped ion. Qubits can directly interact with each other via their motion in the trap and also via the emission and absorption of light and microwaves. That optical address system is still complicated, but it is much simplified by the use of microwaves and trap motion to take care of some operations. This is enough to make the engineering feasible.

 

Superconducting qubits are manufactured. They probably have the worst quantum properties of any qubit contender. However, the fact that they are manufactured also gives us a great deal of control. Gate operations, setting and reading qubit states, and storing operations can all be designed to keep the computer functional for as long as possible. It is this sense of control that gave engineers confidence to start scaling up the number of qubits.

Enlarge / Obviously winning.

Getty Images / Flashpop

Photonic qubits are the oddball of the three that seems to be succeeding. Photonic qubits don't stand still, so gate operations require excellent timing as two or more qubits have to overlap in space and time. This requirement makes designing photonic circuits challenging. But, given a desired computer program, a photonic circuit can be designed.

 

The problem is making a photonic circuit programmable. It's difficult but not the sort of challenge that makes an engineer run away screaming. So, in that sense, photonic qubits still have a good chance of remaining in the mix.

Cost is king

Will we end up with one technology to rule them all? I think that, for the most part, yes, a single technology will dominate. And, I think photonic quantum computers will win out, even though superconducting qubits rule the roost at the moment. Essentially, it comes down to cost: superconducting qubit boards are much cheaper to produce than either ion-trap computers or photonic circuits. But photonic circuits are like integrated circuits, where costs come down with scale. So at volume, the price difference will be small.

 

Then we get to operating costs. Ion-trap computers require vacuum systems with expensive pumping systems, while superconducting qubits operate below liquid helium temperatures. Not only is helium expensive, but dilution fridges are not cheap either. Photonic circuits have none of those costs.

 

Yes, there are design challenges to getting photonic circuits on a par with others, but once overcome, then, the costs favor photonics massively. At the risk of sounding like a futurist (excuse me while I go vomit at the thought), the first two to three generations will be a mix of superconducting qubits and ion computers, then photonic quantum computers will hit their stride. By the fourth generation, no one will know what a transmon qubit is anymore.

 

So, there you have it: I am thankful for the wafer-steppers that will give me access to an awesome light-based quantum computer.

 

 

Source: Quantum computing’s also-rans and their fatal flaws (Ars Technica)