Oak Ridge National Laboratory has been investing heavily in quantum computing across the board. From testing new devices, programming models, and figuring out workflows that combine classical bits with qubits, this is where the DoE quantum investment seems to be centered.
Teams at Oak Ridge have access to the range of available quantum hardware devices—something that is now possible without having to own the difficult-to-manage quantum computer on site. IBM’s Q processor is available through a web interface, as is D-Wave’s technology, which means researchers at ORNL can test their quantum applications on actual hardware. As we just learned this week at SC17, the lab just added a new tool to their quantum experimentation kit with an interesting quantum simulation appliance from French cloud and IT company, Atos (which most of us in HPC think of in relation to supercomputer maker, Bull, which Atos acquired a few years ago).
According to Dr. Travis Humble who directs the Quantum Computing Institute at Oak Ridge National Lab, the reason why his teams are spread across so many emerging areas in the quantum computing space is because there is no sense which approach (for example, gate-based models, annealing) will win out in the end. Further, he says, there might not be a clear winner in that technology area because the different devices and models are primed for different, specific application areas. Development to augment post-exascale supercomputers needs to happen now with quantum and other models that require a fundamental rethinking of problem casting–and simulating possibilities is a key to this early exploration process.
The Atos quantum simulation is an unexpected offering from a company that is otherwise quiet on the quantum front (meaning they do not appear to be pursuing a path to quantum system production). Atos wants to be at the bleeding edge of compute, which explains their decision to step ahead of the curve into this arena with something that, if executed properly (and adopted at scale) could be a device-agnostic standard-setting tool for the burgeoning quantum device ecosystem.
As Atos describes their QLM-30, which they say can process up to 30 qubits in memory via the appliance, it represents the first major quantum industry program in Europe. They say the “QLM-30 combines an ultra-compact machine with a universal programming language” to let researchers test their applications and algorithms. To put this into real-world perspective, we asked Humble how Oak Ridge is deploying it—and what it might lend to future quantum development.
“Right now with the quantum hardware vendors, if I write a program and run it on the hardware, it will return a result. If it’s a simple enough program I can figure out what the results will be. But as those programs get more complex, it will be harder to debug. This is because quantum computing doesn’t give a way to checkpoint or copy out the quantum state, so I can’t verify it is what I expect it to be.” This, he explains is where numerical simulators come in. It lets the ORNL quantum teams write their programs, run it on the Atos simulator to see what the answer should be so it can be compared to how it runs on the hardware.
But what exactly is this thing beyond what it does?
The simple answer is that this is “a lot like a Linux server running on some custom hardware and with a funky custom memory management system. It is running a specific application for simulating quantum circuits—one that comes down to a cocktail of linear algebra based matrix vector multiplication and data structures of that limited type. “The way we use it is the brute force way,” Humble says. “We write a program which for simplicity’s sake you can think of as a TXT file where each line has some instruction statement and it says what instruction needs to be applied to a qubit in my simulated environment. This looks a lot like assembly for traditional computers but it’s targeting the logic that we would have in quantum computing environment.” Ultimately, the file is written, compiled to a representation that the Atos simulator understands, and is processed. A printout at the end gives a final answer and can even tell the quantum state for different points of the simulation. There are other methods, he says, but they are too approximate—his team wants more exactness.
The end game here is that eventually, in a world with more commercially available quantum devices, the same program gets submitted to the Atos appliance and processed the same way. But instead of a numerical simulation crunching in, the machine issues instructions to the device itself. “Where the land of pure numerical simulation is and where that gets replaced with a quantum device no one would know. And that’s the holy grail,” Humble says.”
Even though it is still early days for quantum computing as a realistic part of HPC environments, Humble’s team is exploring some promising areas for this approach. We hear a lot already about quantum chemistry and materials science and machine learning as potential targets, but he says teams are exploring application potential in a couple of other areas, including applied engineering and data science. The key to developing these applications he says is continued work at his lab up and down the quantum stack; from device research to the middleware nitty gritty and programming. Being able to interface with these devices in a standard way is the key—and Oak Ridge is at the cutting edge of new tools to help create those standards.