D-Wave wooing universities down under
Next wave will be software, quantum computing outfit tells El Reg
While controversy continues over the nature of its machines, quantum computing company D-Wave is wearing out the shoe-leather talking to academic users – and The Register.
In Australia to present to universities and the HPC community at large about D-Wave, the company's director of business development and strategic partnerships, Dr Colin Williams, spoke to The Register about the company's technology, the next generation of its machines, and the controversy that surrounds D-Wave's claim to have a working quantum computer.
While he visited quantum computing researchers in the University of Technology, Sydney, and at Sydney University, Williams said his presentations were aimed at academic HPC users with the kinds of problems that D-Wave's technology addresses.
“We're looking for application expertise in optimisation problems,” he told The Register, citing mining and geophysics as suitable fields, in which Australia has a lot of expertise and experience.
That's quite a long way from the pop-culture view of quantum computing, that pitches it as the perfect all-purpose encryption cracker.
“We're not all that interested in Shor's algorithm,” Williams said [Shor's algorithm describes how to apply quantum computers to encryption schemes], because there “aren't that many customers” for a machine optimised to solve it. Finance, in which D-Wave's ability to solve optimisation problems would be useful in portfolio management, is a much more attractive target market.
By way of illustration, Williams discussed the “travelling salesman” problem with Vulture South. A famous member of the class of NP-complete problems, the travelling salesman problem asks for the shortest path solution in which each of a set of cities is visited once and once only.
With a few thousand variables, it's a problem that defies complete solution with classical computing, but D-Wave says the TSP is in the class of optimisation problems it's addressing. “We would expect that within a few generations, we'll be able to solve the travelling salesman problem very quickly,” Williams told The Register.
“Heuristics can get distracted into exploring the same part of a problem again and again,” he continued. “With a quantum solution, you can look at different approaches - 'show me all possible solutions', or 'give me a count of the possible solutions'.”
Which raises the question of how D-Wave's computers actually function. The hard work, Williams said, is in the programming: to run on the D-Wave quantum annealing device, the user problem has to be expressed as a formula describing the problem's energy state.
“The aim is to minimise the energy function,” he said. The D-Wave chip starts at a low-energy state (which is why it's cryogenically cooled – Williams said today's 10 milli-Kelvin temperatures will seem relatively warm in the future), and a transverse magnetic field is applied which, in theory, gives the qubits an equally-weighted superposition of ones and zeroes.
The act of computing occurs when two things happen simultaneously: the transverse magnetic field is removed, while at the same time, a second field that expressed the problem is increased. This is when the “annealing” in “quantum annealing” takes place – the problem is annealed onto the qubits.
Were things perfect, one computation should yield a result, but that's not the world we live in. First, some problems won't have a single solution; and second, there's still noise. D-Wave doesn't guarantee that an optimal solution will be obtained, Williams said: rather, the user then takes the results of a computation run (the eigenstate), tests the result by calculation in a classical computer, and decides whether or not to re-run the problem on the quantum chip.
Right now, Williams told The Register, the D-Wave environment can run 10,000 quantum computations per second. Over time, that will accelerate, and the company is already working on 1,000-qubit silicon to expand the number of states it can test in each computation.
That quantum-classical-quantum iteration also means the company can offer academic access to its machine, without an institution having to lay out the price of a computer. Williams said one of the things he's discussed with Australian institutions is that they could use D-Wave's machine to run the quantum computations, with remote connections so they can test the outputs on their own machines.
Software the next frontier
To get those tests happening on a wider scale, Williams said, D-Wave is acutely aware that it has to make programming the machines accessible in languages that people actually use, like C, C++, Python and Matlab.
The existence of quantum compilers would relieve users of having to work in what amounts to assembly language, and it's high on the company's agenda. “The biggest growth area in the company is in software development,” Williams said.
Compiler-level accessibility would also relieve a criticism made of quantum computing – that even if there's a quantum speed-up taking place, the time it takes to set up a problem negates it. Making it possible for any arbitrary problem (within, of course, the problem space D-Wave works in) to be handled in familiar languages would eliminate the programming delay.
Of the ongoing controversy playing out in the scientific literature, Williams said the problem with any single benchmark is twofold. Performance benchmarking is a game of leapfrog, with each benchmark superseded by faster machines; and each benchmark only provides a quantum-versus-classical result for a single class of problems.
The other important aspect of the debate is the question “is it a quantum computer?” Williams said the D-Wave chips' behaviour can be explained with quantum mechanical models; and that the company has also posted papers presenting evidence of entanglement on Arxiv.
Vulture South doesn't expect, however, that the controversies will end any time soon. ®