In the past two decades, interest in building quantum computers has grown exponentially. Major investments are being made by most industrialised nations, and several times a year, new announcements claim record-breaking milestones.
For example, Google announced a breakthrough in 2019, claiming that its quantum computer was the first in the world to achieve quantum supremacy, solving in 200 seconds a problem that would take a supercomputer 10,000 years to solve. More recently, China made similar announcements, claiming it now has two different quantum computers that have achieved quantum supremacy.
So far, however, the problems that are being solved faster than supercomputers have no practical use. They were carefully selected to demonstrate the immense scaling power of quantum technology for certain kinds of algorithm.
Even though the current generation of quantum computers have no practical application, funds continue to pour in. To be fair, companies such as IBM have managed to make some money, creating a niche by offering access to quantum computers through a cloud interface.
They provide programming tools and simulators that allow researchers to test algorithms and demonstrate how many qubits will be needed to solve a given problem. And IBM continues to push the envelope, announcing recently that it will have a 127-qubit processor by 2025.
“IBM and Google are probably more patient than the venture capitalists who fund startups,” says Per Delsing, programme director of Sweden’s Wallenberg Centre for Quantum Technology (WACQT) and professor of quantum device physics at Chalmers University of Technology. “But our investors are even more patient than IBM and Google. Having investors that aren’t looking for immediate returns allows us to be more realistic in our goals and make the right decisions.”
While adding more qubits attracts a lot of attention in the press, building better qubits is the approach that will yield more results in the long run. WACQT wants to scale up to 100 very high quality qubits, already much more than the 50 to 60 qubits needed to surpass the current generation of supercomputers.
“To make a chip with a lot of qubits is not a difficult task,” says Delsing. “We can do that tomorrow. But to control all of them at the same time in the manner that they need to be controlled – that is the challenge.”
Two things are required to make a good qubit. The first is long coherence – that the qubit lives and behaves quantum mechanically for a long time. The second is that it is easy to scale up to connect many qubits – connections that are required to achieve the quantum superposition and entanglement that create what seems to be magic to those unfamiliar with the underlying science.
The two main technologies currently used for making qubits are ion traps and superconducting circuits. Ions can be held in magnetic or electric traps, where they can be controlled with lasers. Trapped ions typically have longer coherence times than superconducting circuits, but they don’t scale up as easily.
One of the big advantages of superconducting circuits is that they have no intrinsic dissipation and are therefore easier to connect with each other without interference – a superconducting wire carries current without heating up. Also, superconducting circuits can be made with standard nanolithography and nanofabrication processes, which means many circuits can be integrated on a single chip.
“This combination of being dissipation free and easy to integrate makes superconducting circuits very interesting,” says Delsing. “But of course, we might be biased. We’ve been using this approach for a very long time.”
WACQT is in good company – IBM and Google also use superconducting circuits. But Honeywell opted for trapped ions, announcing completion of its first quantum computer in 2020. Several smaller companies are also betting on trapped ions – including IonQ, Universal Quantum and Alpine Quantum Technology.
A high-quality qubit is one that has long coherence, which means it behaves quantum mechanically for a long time. To achieve long coherence, the number of impurities around the qubits must be reduced by choosing the right materials and refining them – and the qubits have to be made as insensitive to the impurities as possible. This requires expertise in both material science and engineering.
But even the highest-quality qubits have a limited lifetime – for superconducting qubits, this is in the order of 100 microseconds. The longer it takes to read a qubit, the higher the probability that it will decay during the readout process, meaning the value may appear as zero when it is really one. Also, a qubit can only be read once, because the process of reading a qubit changes it.
“The term we use is fidelity,” says Delsing. “How true is the measurement? If you read a one, what is the probability that it was one from the start? And if you read a zero, what is the probability that it was zero from the start? We look at this for both single-qubit operations and two-qubit operations. Typically, for a single-qubit, fidelity is 99.9% or a bit better. Two-qubit gates have a lower fidelity, about 99%. We are, of course, working to improve on these numbers.”
Delsing adds: “Because of these high error rates, the current generation of quantum computers are often referred to as noisy intermediate-size quantum (NISQ) computers. Eventually, error correction algorithms will be used to overcome the noise. By reading out in specific arrangements, you can figure out if there was an error and you could correct it.
“When you set up such an error-correction scheme, you need a large number of physical qubits to represent a few logical qubits. It will take some time to master error correction and fault tolerance. But in the meantime, we can still do a lot with 50 to 100 noisy qubits.”
A quantum computer runs differently from a classical computer
A quantum computer is so different from a classical computer that it requires a new set of tools and procedures to program and run. Not surprisingly, both programming tools and algorithm development are currently areas of research.
The process of programming a quantum computer starts by converting the problem to a quantum algorithm. The algorithm is then compiled to produce code that indicates the quantum gates to use and the microwave pulses to send to each gate.
The microwave pulses used to control a qubit are generated at certain frequencies and the pulses are timed to synchronise the control of different qubits, with as little noise as possible. Then the signal is fed down to a quantum computer, which is at -273.14°C (one-hundredth of a degree above absolute zero). Because any cable distorts the pulse, pre-compensation is required to make sure the pulse ends up the way it should when it meets the qubit. Then, at the end of the process, the qubits need to be read.
“Decay time is in the order of 100 microseconds,” says Delsing. “This is also known as decoherence – the time it takes for the qubit to lose its quantum mechanical properties. But a lot of operations can be performed during that time. Each operation is in the order of 10 to 100 nanoseconds, so the number of operations you can do is somewhere between one and 10,000. How many operations you can run depends on how good your qubits are and how fast your gates are. Of course, we expect to improve in this area.
“After decay, you need to let things relax. It takes around 10 milliseconds to get everything back to a known state. Then you can do another run starting with new microwave pulses. Usually, it will take many such cycles to solve a problem on a NISQ computer. But the whole program can be very fast because you can make many runs in a single second.”
Quantum computers can only solve certain kinds of problem
Chalmers University of Technology had already been working on quantum computing way before WACQT was founded in 2018. It was one of the first research groups in the world to build a superconducting qubit in 2003 – and the qubits it produces today are among those with the longest lifespan.
With the know-how from Chalmers and additional support from the Knut and Alice Wallenberg Foundation, the centre has made significant progress since 2018. Using electron beam lithography to pattern the quantum bits and microwave circuits on a silicon chip, the centre produces its own qubits in a 1,000m2 clean room at Chalmers, one of the best in the world.
By 2020, the centre had built a functioning two-qubit quantum computer and used it to solve an optimisation problem involving aircraft scheduling. “While that problem was too small to be practical, it was a worthwhile demonstration,” says Delsing. “By 2022, we expect to scale up to 20 qubits and we will solve a larger version of the same problem. Although this will still be too small to be useful for the airline industry, it will be a nice demonstration of how we can move in the right direction.”
One day, quantum computers will solve useful problems. But even then, they will never be suited to most of the algorithms that are run on classical computers. However, for certain problems, they will find a solution in seconds or minutes, whereas a supercomputer would take thousands or billions of years.
Quantum computers are very good at solving optimisation problems – problems that cause supercomputers to get bogged down as datasets scale up. Even a small number of input parameters creates too many possibilities for a classical computer to analyse to determine the cheapest or fastest solution – or whichever attribute needs to be optimised.
Almost every industry has optimisation problems – and solving them quickly would have a widespread impact. Although WACQT may have more patient investors than most organisations developing quantum technology, it is still aware that one day, its work will have to deliver economic value. To make sure it focuses on what people need, WACQT has been careful to involve future users of quantum computers.
“One of our industry partners is owned by Boeing,” says Delsing. “They do flight logistics, so we are inspired by them to attack this problem in the quantum way. This is, of course, interesting for them in their business, and it could be potentially very fruitful for them once this is working better than on a classical computer.”
Quantum computers will also be good at simulating natural phenomena, such as the behaviour of molecules given a certain structure. Simulating large biological molecules is an essential process that pharmaceutical companies need for drug design. Supercomputers cannot scale up to run such simulations on large molecules, which means some diseases will remain untreated.
“We are also collaborating with the pharmaceutical industry,” says Delsing. “For instance, AstraZeneca is interested in simulating big molecules, which behave quantum mechanically. A quantum computer can simulate bigger and bigger molecules, which means you can screen many more molecules in a much shorter time.”
One day, patience will pay off
Delsing adds: “Eventually, we will have a really good quantum computer, and it will have economic value. We will set up a cloud layer that will allow people to access several different quantum computers and simulators.”
WACQT works on more than just quantum computing. It also works on three other aspects of quantum technology – simulation, communication and sensing. The computing and simulation are coordinated by Chalmers, the communication by the Royal Institute of Technology and the sensing by Lund University.
“If you look at the long term, you can imagine quantum computers in many different places that use quantum communication to talk with one another over long distances,” says Delsing. “This is extremely challenging, and another group at WACQ is in fact working to solve that type of problem.”
