IBM proposes unified architecture for hybrid quantum-classical computing

Quantum computing has moved past the “science curiosity” stage, but, ultimately, hasn’t reached the general-purpose, fault-tolerant phase.

This, researchers say, requires the use of quantum computers and classic high performance computing (HPC) in tandem. When combined, they can process emerging algorithms and workflows, and truly scale real-world applications.

To this end, IBM has released what it calls the industry’s first published quantum‑centric supercomputing (QCSC) reference architecture. The new blueprint outlines how quantum and classical systems can work in one computing environment to tackle scientific challenges that “no single computing approach can solve on its own,” the company contends.

This hybrid approach is promising, but as with any emerging technology, there are still challenges. “There’s a big difference between quantum hardware and classical hardware, and you have to somehow match those two up,” noted Paul Smith-Goodson, VP and principal analyst at Moor Insights & Strategy. “It’s kind of like trying to run your Tesla off a gasoline engine.”

Combining the technologies

IBM’s architecture integrates quantum processors (QPUs) with modern supercomputing environments comprising GPU and CPU clusters, high‑speed networking, and shared storage across on‑premises systems, the cloud, and research centers. This coordinated workflow can support computationally intensive workloads and algorithm research, the company says.

The infrastructure also integrates orchestration and open-source frameworks such as Qiskit, aPython-based software development kit (SDK) for programming quantum computers. This means developers and scientists can access quantum through tools and workflows they are familiar with.

Quantum computers and classical HPC are traditionally “disparate systems [that] operate in isolation,” IBM researchers explain in a new paper. This can be “cumbersome,” because users have to manually orchestrate workflows, coordinate scheduling, and transfer data between systems, thus hindering productivity and “severely” limiting algorithmic exploration.

But a hybrid approach can simplify the process of applying quantum computing to problems in areas like chemistry, materials science, and optimization, and “solve problems that were previously out of reach,” IBM says.

The researchers describe quantum-centric supercomputing (QCSC) as evolving through three distinct phases: quantum systems as specialized compute offload engines within HPC environments; quantum and classic HPC systems coupled through advanced middleware; and fully co-designed HPC and quantum systems for hybrid workflows.

The first phase focuses on establishing “foundational integration across multiple dimensions,” the researchers explain. The second phase focuses on reducing latency, creating multiple sophisticated feedback mechanisms, and supporting complex hybrid algorithms. The third phase represents the “culmination” of the integration through “fully co-designed heterogeneous systems where quantum and classical resources are architected as unified platforms from the ground up.”

The latter mirrors the trajectory of GPUs in HPC systems, the researchers note; early GPUs typically functioned as external accelerators attached to host processors. But then interconnects were established between GPUs and CPUs, and from GPU to GPU, to provide much higher bandwidth and lower latency.

“Similarly, quantum systems will transition from standalone units to fully integrated components within co-designed quantum-HPC platforms,” the researchers contend.

IBM says scientists are already using its quantum-centric architecture to deliver “accurate results for real experiments.” This includes the simulation of one of the largest molecular models by Cleveland Clinic; the creation and verification of a first‑of‑its‑kind half‑Möbius molecule with an unusual electronic structure; and one of the largest simulations of iron‑sulfur clusters, a fundamental molecule in biology and chemistry.

Challenges of classical plus quantum

IBM has been aggressively pushing into quantum, and seems to have the “most definite plan so far,” noted Smith-Goodson. Other tech giants including IonQ, Google, Microsoft, and Amazon are also rolling out quantum roadmaps, while more specialized companies like Quantinuum, QuEra Computing, and Xanadu, are exploring novel techniques.

IBM’s type of quantum-centric supercomputer will likely be the standard for some time, noted Smith-Goodson. “It’s not only a standalone thing, it’s going to be really supercharged between the two.” The quantum computer will need to become more fault tolerant, so it can operate even in the presence of errors, because it will be permanently plugged into the supercomputer, he pointed out.

In algorithms that use both classical and quantum processing, each technology plays its own part in a talkback loop. For instance, the classical component will assume parameters, then send them over to quantum; quantum will run a circuit, measure the result and lob it back; classical will then update it and send back even more parameters for the next round. “They do a back and forth until they finally get a good answer,” Smith-Goodson explained.

Ultimately, most problems aren’t 100% quantum, he pointed out; quantum can perform very sophisticated calculations, but much of the rest of the work stays over on the classical side, which has to do “a lot of heavy lifting” in areas such as error correction, with confidence.

One of the big challenges of these hybrid environments, however, is speed, as quantum is orders of magnitude faster. The cloud isn’t optimal, because network latency can be “thousands of times longer” than quantum’s execution needs, Smith-Goodson pointed out.

Ultimately, though, while there are still a lot of challenges with quantum, the industry is “working through them,” as evidenced by IBM’s new framework, Smith-Goodson noted. 

“I’m glad to see this finally getting on arXiv and laying it out so that we get a good idea of what they intend to do,” he said. “It’s getting closer, and we have to get fault tolerance. And I think this is next on the agenda.”

This article originally appeared on NetworkWorld.