D-Wave hardware has always occupied a unique place in the computer landscape. It is a general purpose computer that relies on quantum mechanical effects to perform calculations. And while other quantum computer makers have struggled to put more than a few dozen qubits together, D-Wave’s systems have already scaled up to more than 2,000 addressable bits. But the D-Wave systems don’t perform calculations in the same way, and despite all those bits, they haven’t clearly demonstrated performance that can surpass even traditional computer hardware.
But D-Wave has come out with a research paper in Science that suggests that the system can do interesting things even in its current state. The company’s researchers have let it go by modeling a quantum system very similar to the bits used in the hardware itself, allowing them to investigate quantum phase transitions. While this still isn’t an advanced feat, it does give researchers complete control over the physical parameters of a relevant quantum system as it undergoes phase changes.
Spins and spin glass
D-Wave’s systems can be thought of as a large collection of magnets, each of which can change direction. These are not qubits in the same way as the components of IBM or Intel’s quantum processors, but they do rely on quantum behavior to perform calculations. In itself, there is nothing that prefers one orientation over the other. But put a second magnet nearby and the two influence each other; now, if someone reverses their orientation, it changes the energy content of the system. D-Wave’s current system scales this up to 2,048 individual magnets, along with associated control hardware that determines which of these magnets is connected and how strong that connection is.
The hardware can act as a general purpose computer because it is possible to encode the solution to the problem as the minimum power state of the system.
In this case, the system they’re modeling looks suspiciously like the D-Wave computer itself. This is called a ‘transverse field Ising model’, it is a cubic arrangement of magnets that can flip. If these magnets are arranged to change direction as you move in one of the three dimensions, an antiferro magnet is formed. But it is also possible to have configurations in which the orientations are disordered, forming a so-called “spin glass” (magnetic properties arise from the spin of particles). Although spin glasses are disordered, they have well-defined energies, including a low energy state.
Although the individual magnetic bits in a D-wave system are largely in one plane, it is possible to control the interconnections so that the system accurately simulates the behavior of a three-dimensional grid. On current generation systems, the largest grid that fits into the processor is a cube with eight magnets on one side.
Getting an answer isn’t exactly groundbreaking work. “An 8 x 8 x 8 grid is state-of-the-art for 1990 computer models,” said Richard Harris, a D-Wave scientist who was lead author of the research paper. But for the company, the work represents significant progress. “It’s the first time the respectable competition has to be a multicore server, as opposed to something like a pencil,” Harris told Ars.
But because these issues have been previously resolved, they provide important validation of the D-Wave system. Because they provide the same answers, they show that the system can properly set up (and get a result from) the physical system being modeled. “It’s clear evidence that it’s what we said it was,” Harris said. “It contains the necessary physics to represent this model.” Or, as he put it in the paper, D-Wave’s hardware “can be used as a materials physics simulator. The ability to manipulate individual spins and bonds allows one to investigate order parameters in ways not possible in bulk condensed matter systems.”
This is also important because we already know many other issues that can be mapped on these transverse-field Ising models, meaning it’s confirmation that the D-Wave hardware could potentially be used to run a wide range of to solve problems.
The problem, of course, remains one of performance. While the number of addressable bits in D-Wave’s hardware has increased significantly, it still can only compete with multicore hardware, and we routinely throw clusters at physics problems.
But there is also promising news for the performance of these results. Calculations on a D-Wave computer are based on a process called annealing. This means that the computer is placed in a simple configuration and in its energetic ground state and then carefully moved to a configuration where the answer to a question can be read out. If the process is done gently enough, the system will remain in the ground state the entire time – this, of course, points to a wide range of minimization problems.
Problems like this allow researchers to examine the annealing process in more detail in a system where we know what the answer should be. This allows them to examine how the system reacts as they adjust the connections between the components and investigate what happens when the system undergoes phase transitions. “We want to see how the system goes through the phase transitions.” Harris told Ars. “What does it look like, how can it be optimized? If it can’t [be optimized]can it be bypassed?”
And if D-Wave understands the annealing process better, it can allow them to run their hardware faster or detect when something has gone wrong, and the annealing will produce an incorrect note. As he drove home, the company’s CEO, Vern Brownell, broke into the conversation here to say, “This helps design the next generation of D-Wave systems.”
The challenge for D-Wave is to get the next generations ready on time. The company has already scaled its computers to far more bits than IBM or Intel’s quantum processors, and it should maintain its lead for the foreseeable future. But that lead isn’t enough for D-Wave to surpass traditional computers at the moment. However, everything points to the kind of computers IBM and Intel are working on will deliver massive accelerations with far fewer qubits than D-Wave already has. So this has the makings of an intriguing technology race.
Science2018. DOI: 10.1126/science.aat2025 (About DOIs).