Scientists have invented a new computing technology that enables multiple people to run programs on a quantum computer for the first time.
Dubbed “HyperQ,” the new system is a type of virtualization technology that balances workloads by dividing a quantum computer’s physical hardware into multiple isolated quantum virtual machines (qVMs) that are then tasked by an intelligent scheduler.
This scheduler operates “like a master Tetris player” that packs multiple qVMs together to run simultaneously on different parts of a single machine, Columbia representatives said in a statement. The end result is a single quantum computer capable of supporting multiple users running different applications. The scientists published their findings in July, in a new study which featured in the 19th USENIX Symposium on Operating System Design and Implementation (OSDI ’25).
“HyperQ brings cloud-style virtualization to quantum computing,” study coauthor Jason Nieh, professor of computer science at Columbia Engineering, said in the statement. “It lets a single machine run multiple programs at once — no interference, no waiting in line.”
Efficiency through virtualization
Typical gate-based quantum computers are expensive compared to their binary counterparts. According to data from Quantum Zeitgeist, the research and development costs for a small-scale quantum computing system range from $10 million to $15 million. That’s before wecount the costs of upkeep, which are estimated at more than a million dollars per year, with software and programming development on top.
Despite the high development and operational costs, most quantum computers are usually only capable of supporting single-user operations due to the intrinsically interconnected nature of qubits — the quantum equivalent of classical computer bits — they comprise.
The researchers took inspiration from the virtualization technology that powers modern cloud computing services such as Amazon Web Services (AWS) and Microsoft Azure. In a classical computing virtual machine (VM) environment, a software layer called a hypervisor or Virtual Machine Monitor allocates unused resources to individual VMs that run entirely independent of one another.
In a quantum environment, however, computer scientists have to take into account “noise” in the quantum signal that could propagate throughout the system. HyperQ gets around this problem by isolating each qVM with a “buffer” of qubits that remain inactive, thus negating the potential for noisy “crosstalk.”
“Previous efforts required specialized compilers and needed to know exactly which programs would run together ahead of time,” said lead author of the paper Runzhou Tao, former doctoral student at Columbia’s Software Systems Laboratory. “Our approach works dynamically with existing quantum programming tools, which is far more flexible and practical for real-world use.”
Dynamic multiprogramming
Quantum programs typically execute via a predictable pattern of qubits. HyperQ determines the optimum time slots for each user request and allocates resources across both time and space by determining which qubits will be necessary for each request and how long they’ll be active, the researchers said in the study.
This might sound like a simple concurrent scheduling task, but previous machine management systems required users to queue up so the system could precompile their requests for execution. HyperQ introduces a concept called “dynamic multiprogramming”, in which usage is streamlined, with programs allowed to be compiled independently for different-sized qVMs.
The team tested its HyperQ software layer on IBM’s Brisbane quantum computer, a 127-qubit gate-based system built on the Eagle chipset. According to the research, HyperQ reduced average user wait times by up to 40 times, lowering project turnaround times from “days to mere hours.” It also enabled up to a tenfold increase in the number of quantum programs executed.
Going forward, the team intends to expand HyperQ to function across the gamut of quantum computing architectures, including machines made by manufacturers other than IBM.
