By Katie Elyce Jones, Editor, PillarQ
As the High Performance Computing (HPC) community pushes beyond Moore’s Law for solutions to accelerate future systems, one technology at the forefront is quantum computing, amassing billions of dollars in global R&D funds each year.
Perhaps it’s no surprise that HPC centers — including the Oak Ridge Leadership Computing Facility (OLCF), home of the world’s first exascale supercomputer Frontier — are finding ways to harness and advance quantum systems.
The OLCF’s Quantum Computing User Program (QCUP) is located at Oak Ridge National Laboratory (ORNL) in Tennessee and is funded by the US Department of Energy (DOE). It provides users in the sciences with remote access to key commercial quantum computing systems. Currently, the program provides access to various superconducting architectures from IBM Quantum Services and Rigetti Quantum Cloud Services, as well as Quantinuum Trapped Ion computers and emulators. The program also prepares access to an IonQ Trapped Ion system.
In a new initiative this year, OLCF and QCUP are bridging quantum and HPC through a hybrid assignment program that offers dual access to QCUP’s quantum providers and OLCF’s supercomputers.
“The purpose of QCUP is to help us understand how the [quantum] Technology is evolving and it helps us predict when we may want to incorporate this technology into the next HPC system,” said Travis Humble, QCUP director.
Humble is also director of ORNL’s Quantum Science Center, which is funded through another DOE program — the National Quantum Information Science Research Centers — but shares overlapping interests in quantum research and development. He will be a panelist for “Quantum Computing: A Future for HPC Acceleration?” at SC22 (The International Conference for High Performance Computing, Networking, Storage, and Analysis) on Friday November 18th.
According to Humble, QCUP offers a range of quantum computing systems to explore what works best for specific problems, and that classical computing is part of that exploration. “We don’t yet know the best hardware and how the applications fit together. Quantum computing as a theory gives us a whole new playground where we can try computation, inform scientific discoveries, change the types of problems we can actually compute. A supercomputer is powerful – but also limited. Hybrid combines the best of both worlds.”
However, he cautioned that not many applications currently make good use of both devices, and the intent of QCUP’s new hybrid quantum-classical assignments is to find applications that run well on both.
QCUP has about 250 users and since 2016 has evolved from an internal laboratory program to the current user program. Sponsored by the DOE’s Advanced Scientific Computing Research (ASCR) program, the Quantum User Program adopted the same HPC user model as ASCR’s leading computing facilities, which review scientific proposals for potential impact and merit to allocate time for computing systems.
“We’re looking for feasibility — try to solve a problem that even fits on a quantum computer — and for technical readiness and application,” Humble said.
QCUP’s user support includes a Science Engagement Team that helps researchers port their code, though many users have historically been “expert quantum users,” he said. “You’ve written programs and you’re ready to go.”
Many users come from scientific programs related to quantum research, such as high-energy and nuclear physics and fusion energy. For example, a team led by the Lawrence Berkeley National Laboratory used QCUP resources to simulate part of two protons colliding and broke down physical calculations into those best suited to classical versus quantum computing to include quantum effects that a classic computer would otherwise approach.
“Physics has by far the greatest presence. Coming second is probably computer science, which involves building tools that allow for better performance of a quantum computer,” Humble said.
In another QCUP project, a team led by researchers from the University of Chicago and Argonne National Laboratory simulated quantum spin defects with applications to encode information in quantum computers. In this case, they used classical calculations to validate and reduce errors in their quantum calculations.
Artificial intelligence (AI) also appears at the interface between classical and quantum computing. Humble said the goal of some computer science projects is to use quantum computing to speed up AI and machine learning operations, or to uncover quantum-specific information in AI-generated data.
Although the Program provides access to quantum computers through an HPC user facility, these computers are not integrated with HPC systems. One of QCUP’s ultimate goals is to connect quantum and HPC systems, but there are short-term obstacles.
“Part of the barrier now is that quantum computing is so early. If you look at what a quantum computer is today, in 6 months it will be replaced by something new,” Humble said.
From a technical point of view, quantum computers still require special maintenance and cannot yet keep up with the performance of HPC. From a user perspective, training hurdles have mostly relegated quantum computing to the quantum experts.
“The training material you need to start using quantum computing is also in its infancy,” Humble said. “For the vast majority of HPC users who want to use quantum, we need to create training resources for them.”
Although many HPC quantum collaborations are still in their infancy, lessons learned from programs such as QCUP and quantum projects at other HPC centers can help set the stage for future HPC quantum integration.
Katie Elyce Jones is the founder and editor of research news publication PillarQ.