Nobody agrees on how to build a quantum computer. Why?
There are a dizzying number of competing technologies that underlie quantum computing — and so far "it's really too early to pick a winner."
Microsoft has bet the farm on something called anyons. IBM, Intel and Google have backed superconductors. And even if those scientific words don't make much sense, at least you've heard of the organizations.
Less well-known outfits, like startups IonQ, QuEra, D-Wave, PsiQuantum and Silicon Quantum Computing, are using a range of other esoteric scientific approaches to build quantum computers.
No one involved doubts that it's worth attempting to build a machine that can process information using the strange laws that govern matter at the quantum level. But there is no consensus about how best to proceed — or who is going to get there first.
"I think it's really too early to pick a winner," says Peter Knight, senior research investigator at Imperial College London.
Quantum computers are created from processing units known as quantum bits, or qubits. As well as superconductors (which are materials with zero electrical resistance below specific low temperatures) and anyons (which are so-called "quasi-particles"; more on both below), researchers are creating qubits from ions, atoms, photons and even individual phosphorus atoms embedded in silicon, among others. Each one, as you might imagine given the lack of consensus, has its advantages and drawbacks.
Leader of the pack
At the moment, superconductors and trapped ions are widely considered the two leading options.
Google's superconducting qubits, which are loops of superconductor constricted by a "pinch" at one point — the pinch, called a Josephson Junction, gives the loop similar quantum properties to an atom — are currently considered by many to be the class leader. Late last year, they performed a computation that would take the world's most powerful supercomputer 10,000 years. Google's quantum computer, called Sycamore, did it in 3 minutes and 20 seconds. (Read more.)
It was a "very significant" milestone, according to Travis Humble, director of Oak Ridge National Laboratory's Quantum Computing Institute. "We finally had a quantum processor that needed a supercomputer to compare whether we were getting the right answer," he says. "As far as I know, that's the first time that has ever happened."
Many startups, such as Rigetti of Berkeley and Finland's IQM, share the optimism about superconducting qubits. As do IBM and Intel. But that enthusiasm and Google's result don't mean other technologies will lose their investors' interest. "My impression is that there are a few different areas generating enthusiasm, so I don't think the field has skewed too much towards Google yet," says Joe Fitzsimons, CEO of Singapore-based startup Horizon Quantum Computing.
There are some doubts about whether superconducting qubits can be built and operated in large enough numbers to make them actually useful in the long run. "The question is, how do you get to hundreds of thousands of qubits?" asks Benjamin Bloom of Berkeley-based Atom Computing.
And that really is a big question. Google's Sycamore uses just 54 qubits. But no one knows how big a truly useful quantum computer will have to be; some experts claim that 1 million qubits might be required. And because superconducting qubits need to be cooled to around -270 Celsius (-454 Fahrenheit), cooling even thousands of them could prove to be an almost insurmountable headache.
The other main contender
A strong contender to leapfrog superconductors is so-called ion traps, which are used by companies including Alpine Quantum Technologies, based in Innsbruck, Austria. In ion traps, atoms forming the basis of the qubit have an electron removed, and the ion is held in position using magnetic fields. They have a big plus in terms of practical implementation: "Ion trap systems operate at room temperature and don't require special cooling," says Thomas Monz, a co-founder of Alpine.
Monz certainly doesn't characterize Alpine as "behind" in the quantum race. He points out that it and its main competitor on this technology — IonQ, a startup spun out of the University of Maryland — are operating with more than 100 qubits that induce fewer errors than superconducting qubits. They also have established interfaces for operation and modular designs that will make scaling up feasible.
That Alpine has a device with more quibits than Google but hasn't demonstrated supremacy isn't necessarily a sign of inferiority: Performance isn't just about raw qubit numbers, and algorithms aren't easily transferred between quantum computers anyway. Instead, think about it as a different but competing technology that is maturing at a different rate, with different priorities. As Monz puts it: "What's the best computer? Do you value the small and mobile one? Or the one with the fast CPU?"
Some competitors remain unconvinced about superconductors and ion traps, despite their "frontrunner" status. "Their systems are currently almost unusable — you can only put toy problems on them," says Alan Baratz, CEO of D-Wave Systems, alluding to the currently fragile nature of the world's leading quantum computers. (D-Wave has its own unique — and controversial — approach to quantum computing hardware; see sidebar.)
Besides D-Wave, plenty of other technologies are up for the challenge. In Australia, Silicon Quantum Computing hopes that its ambitious scheme to piggyback on the well-established fabrication routines of the semiconductor industry means it can assemble qubits that are easy to manufacture at very large scales. SQC's qubits are the electrons hosted on a single phosphorus atom embedded in a silicon chip, but the company doesn't expect to have a useful general purpose quantum computer until the 2030s.
Microsoft might take even longer. Its technology, called topological quantum computing, centers on a particle called — wait for it — a non-abelian anyon. This doesn't occur naturally. It will only pop into existence in very particular circumstances, such as when strong magnetic fields are applied to ultrathin sheets of semiconductors. Not many people outside the company are sure that it can ever be created in a reliable enough way to support a technology infrastructure.
"The Microsoft bet is really interesting: It's been extraordinarily hard to make these qubits," Knight says. But, he adds, their properties could allow Microsoft's quantum computations to run without generating errors. That would mean a significant reduction in the number of qubits required because, for most of the technologies, the overwhelming majority of the qubits in a working quantum computer — at least 5 in 6 — are required exclusively for error correction. "Their approach demonstrates that when you've got serious resources, you can look at alternative platforms."
That also seems to be true of the most recent entrant to the race. In November 2019, news leaked that Palo Alto-based PsiQuantum raised $230 million to develop its photon-based qubit technology, originally developed at the University of Bristol, into a fully fledged quantum processor.
So, there are stronger contenders and underdogs, for sure, but the field is crowded — and there is no outright winner right now. But then, there may never be: The reality is that quantum computers could end up never even usurping classical machines. "If we ever use quantum computers," Humble says, "they are probably going to be integrated into our supercomputing systems."
When is a quantum computer not a quantum computer?
D-Wave Systems has been seen as a kind of bad boy of quantum computing for a few years. Its approach, known as annealing, doesn't involve operating quantum versions of logic gates. Instead, its machines allow thousands of superconducting qubits to interact in loosely defined ways. "What that means is that we can scale much more rapidly," says Alan Baratz, D-Wave's CEO.
To use a D-Wave quantum computer, you first formulate your question in a way that mirrors looking for the lowest point in a landscape. Asking the right question kicks the qubits into a high-energy state where they occupy many quantum states at once. If the question is correctly formulated, it is akin to simultaneously occupying all the points in the landscape of answers. As the qubits settle toward their lowest energy state, they reveal the lowest point in the landscape — the required answer.
Criticisms have arisen because D-Wave's is not a general purpose machine: Relatively few problems can be efficiently solved this way. Still, it has plenty of fans, including users at NASA, Volkswagen, Lockheed Martin and Oak Ridge National Laboratory. "We're seeing interesting results," says Travis Humble, director of ORNL's Quantum Computing Institute. "I can test much larger problem sizes than I could on devices that only have 50 qubits."