Quantum Computing Becoming Real

Quantum computing will begin rolling out in increasingly useful ways over the next few years, setting the stage for what ultimately could lead to a shakeup in high-performance computing and eventually in the cloud.

Quantum computing has long been viewed as some futuristic research project with possible commercial applications. It typically needs to run at temperatures close to absolute zero, which means most people never actually will see this technology in action, which is probably a good thing because quantum computers today are still a sea of crudely connected cables. And so far, it has proven difficult to create enough qubits for a long enough period of time to be useful. But the tide appears to be turning, both for how to extend the lifetime of quantum bits, also known as qubits, as well as the number of qubits that are available.

A qubit is the unit of quantum information that is equivalent to the binary bit in classical computing. What makes qubits so interesting is that the 1s and 0s can be superimposed, which means that it can perform many calculations at the same time in parallel. So unlike a binary computer, where a bit is either 0 or 1, a quantum computer has three basic states—0, 1, and 0 or 1. In greatly oversimplified terms, that allows operations can be carried out using different values at the same time. Coupled with well-constructed algorithms, quantum computers will be at least as powerful as today’s supercomputers, and in the future they are expected to be orders of magnitude more powerful.

“The first applications will probably be in things like quantum chemistry or quantum simulations,” said Jeff Welser, vice president and lab director at IBM Research Almaden. “People are looking for new materials, simulating molecules such as drug molecules, and to do that you probably only need to be at around 100 qubits. We’re at 50 qubits today. So we’re not that far off. It’s going to happen within the next year or two. The example I give is the caffeine molecule, because it’s a molecule we all love. It’s a fairly small molecule that has 95 electrons. To simulate the molecule, you simulate the electron states. But if you were to exactly simulate the 95 electrons on that to actually figure out the energy state configuration, it would take 10⁴⁸ classical bits. There are 10⁵⁰ atoms in the planet Earth, so there’s no way you’re ever going to build a system with 10⁴⁸ classical bits. It’s nuts. It would only require 160 qubits to do those all exactly, because the qubits can take on exactly all the quantum states and have all the right entanglements.”

Fig. 1: IBM’s 50Q system. Source: IBM

Exact numbers and timing tend to get a bit fuzzy here. While almost everyone agrees on the potential of quantum computing, the rollout schedule and required number of qubits is far from clear. Not all qubits are the same, and not all algorithms are created equal.

“There are two elements driving this technology,” said Jean-Eric Michellet, senior director of innovation and technology at Leti. “One is the quality of the qubit. The other is the number of qubits. You need to achieve both quality and quantity, and that is a race that is going on right now.”

These two factors are closely intertwined, because there also are two different types of qubits, logical and physical. The logical qubit can be used for programming, while the physical qubit is an actual implementation of a qubit. Depending on the quality of the qubit, which is measured in accuracy and coherency time (how long a qubit lasts), the ratio of logical and physical qubits will change. The lower the quality, the more physical qubits are required.

Another piece is the quality of the quantum algorithms, and there is much work to be done here. “We’re still at the beginning of the software,” said Michellet. “This is a new way of doing algorithms.”

Yet even with the relatively crude algorithms and qubit technology today, quantum computing is beginning to show significant progress over classical computing methods.

“The performance (for quantum computing) is exponential in behavior,” said James Clarke, director of quantum hardware at Intel Labs. “For very small problems, you probably have quite a bit of overhead. When we measure traditional algorithms, there is some crossover point. A quantum algorithm is going to be exponentially faster. There are discussions in the community that this crossover point would be 50 qubits. We actually think it’s more like a thousand or so for certain types of optimization algorithms.”

Intel is working on two qubit technologies—superconducting qubits and spin qubits in CMOS. The company recently demonstrated a 49-qubit computer, based on superconducting technology it has code-named Tangle Lake.

Fig. 2: Intel’s 49-qubit Tangle Lake processor, including 108 RF gold connectors for microwave signals. Source: Intel

Faster sums of all fears
One of the key drivers behind quantum computing is a concern that it can be used to break ciphers that would take too long using conventional computers. The general consensus among security experts is that all ciphers can be broken with enough time and effort, but in the most secure operations that could take years or even decades. With a powerful quantum computer, the time could be reduced to minutes, if not seconds.

This has spawned massive investment by governments, universities, industry, and groups composed of all of those entities.

“Cryptography has really driven research at the government level around the world,” said Clarke. “The thought that security would be compromised is perhaps a worry but perhaps a goal for something like a quantum computer. Within this space, it actually requires a very powerful quantum computer that’s probably many years off.”

The more pressing question involves security measures that are in place today.

“Some, not all, cryptography algorithms will break with quantum computing,” said Paul Kocher, an independent cryptography and computer security expert. “We’re probably not at risk of things being broken over the next five years. But if you record something now, what happens in 30 years? With undercover operations you expect them to be secret for a long time. Something like AES 56 is not breakable any faster with a quantum computer than a conventional computer. The same is true for long keys. But with public keys like RSA and Diffie-Hellman key exchange, those could be broken. We’re still far away from building a quantum computer that could do that, but this could up-end the whole game on the PKI (public key infrastructure) front. Those things are suddenly at risk.”

Challenges remain
Today’s qubits are far from perfect. Unlike classical bits, they don’t exist for very long, and they aren’t completely accurate.

“That’s the major focus for quantum computing right now,” said IBM’s Welser. “It’s not only how to increase the number of qubits, which we know how to do just by continuing to build more of them. But how do you build them and get the error rates down, and increase coherency time, so that you can actually have time to manipulate those qubits and have them interact together? If you have 100 qubits, but the coherency time is only 100 microseconds, you can’t get them all to interact efficiently to do an actual algorithm before they all have an error. In order to move forward, we talk about something called quantum volume, which then takes into account the number of qubits, the coherency time, the length of time they stay stable, and the number that can be entangled together. Those factors provide what we believe is the best way to compare quantum computers to each other.”

IBM is focused on quantum volume as the best path forward, but that point is the subject of debate in academic circles today. “Clearly, getting that coherency time to go longer is important,” Welser said. “But even at the level we’re at right now, we simulated three-atom molecules on our 7-qubit machine and showed that it works. They are error prone, but they get the right answer. You just run the simulation of thousand times—it takes very little time to do that—and then you take the probabilities that come out of that and you map out your answer.”

One of the key metrics here is the classic traveling salesman problem. If a salesman has a certain route to cover that involves multiple cities that are not in a straight line, what is the most efficient way to manage travel? There is no simple answer to this problem, and it has been vexing mathematicians since it was first posed in 1930. There have even been biological comparisons based upon how bees pollinate plants, because bees are known to be extremely efficient. But the bee studies conclude that complete accuracy isn’t critical, as long as it’s good enough.

And that raises some interesting questions about how computing should be done. Rather than exact answers to computational problems, the focus shifts to distributions. That requires less power, improves performance, and it works well enough for big problems such as financial valuation modeling.

Welser said banks already have begun exploring the quantum computing space. “They want to get started working on it now just to figure out what the right algorithms are, and understand how these systems will run and how to integrate them in with the rest of all of their simulations. But they’ll continue to use HPC systems, as well.”

Economies of efficiency
With the power/performance benefits of scaling classical computing chips diminishing at each node after 28nm, and the costs rising for developing chips at the latest process nodes, quantum computing opens up a whole new opportunity. That fact hasn’t been lost on companies such as IBM, Intel, Microsoft, Google and D-Wave Systems, all of which are looking to commercialize the technology.

Fig. 3: D-Wave’s quantum chips. Source: D-Wave

The big question is whether this can all be done using economies of scale in silicon manufacturing, which is already in place and well proven.

“These are larger chips,” said Intel’s Clarke. “For a small chip with 2 qubits, those we can wirebond to our package. Any larger than that, we are starting to do flip-chip bonding. Any larger than about 17, we are adding superconducting TSVs.”

That’s one way of approaching qubit manufacturing, but certainly not the only way. “We are also studying spin qubits in silicon,” Clarke said. “Basically, what we are doing is creating a silicon electron transistor. Instead of having a current through your channels, we trap a single electron in our channel. We put a magnet in the refrigerator. So a single electron in a magnetic field spins up or spins down. Those are the two states of the qubit. Why is this appealing? One of these qubits is a million times smaller than a superconducting qubit in terms of area. The idea that you can scale this to large numbers is perhaps more feasible.”

There are different business models emerging around the hardware. Intel’s is one approach, where it develops and sells quantum chips the way it does with today’s computer chips. IBM and D-Wave are building full systems. And Microsoft and Google are developing the algorithms.

Quality control
One of the big remaining challenges is to figure out what works, what doesn’t, and why. That sounds obvious enough, but with quantum computing it’s not so simple. Because results typically are distributions rather than exact answers, a 5% margin of error may produce good-enough results in one case, but flawed results in another.

In addition, quantum computing is highly sensitive to environmental changes. These systems need to be kept at a constant temperature near absolute zero, and noise of any kind can cause disruptions. One of the reasons these systems is so large is they provide insulation and isolation, which makes it hard to do comparisons between different machines.

On top of that, quality of qubits varies. “Because you use a qubit once in a certain configuration, will it behave the same in another? That can affect what is true and what is false,” said Leti’s Michellet. “And even if you have exactly the same state, are you running the same algorithms the next time? We need some tools here.”

And finally, the algorithms being used today are so basic that they will undoubtedly change. While this is generally a straightforward process with machine learning and AI to prune and weight those algorithms more accurately, when it comes to quantum computing the algorithms can leverage the superimposed capabilities of the qubits, adding multiple more dimensions. It’s not entirely clear at this point how those algorithms will evolve, but everyone involved in this space agrees there are big changes ahead.

Conclusion
Quantum computing is coming. How quickly isn’t clear, although the first versions of this technology are expected to begin showing up over the next few years, with the rollout across more markets and applications expected by the middle of the next decade.

While quantum computing is unlikely to ever show up in a portable device, it is a disruptive technology that could have broad implications both for the cloud and for edge devices that connect to these computers. For the chip industry in particular, it provides another path forward beyond device scaling where massive performance gains are not based on cramming more transistors onto a piece of silicon. And given the technical challenges chipmakers are facing at 3nm, the timing couldn’t be better.

—Mark LaPedus contributed to this report.

Related Stories
System Bits: June 5, 2018
Quantum computing light squeezing
Quantum Computing Breakthrough
Aluminum on sapphire qubits provide proof of concept for surface codes and digitized adiabatic evolution.
System Bits: Aug. 1, 2017
Quantum computing steps forward
What’s Next With Computing?
IBM discusses AI, neural nets and quantum computing.
Quantum Madness (blog)
Multiple companies focus on qubits as next computing wave, but problems remain.