What is quantum-centric supercomputing?

A quantum-centric supercomputer is a next-generation connection of a quantum computer with a classical supercomputer—deeply integrated through co-designed infrastructure, middleware and applications. Many experts in the field believe that quantum-centric supercomputing is the future of computing.

In the era of quantum computing, quantum-centric supercomputing is predicted to help researchers make major breakthroughs in the fields of simulation, optimization, machine learning and solving challenging math equations. This approach can provide value for chemistry, materials science, healthcare and life sciences, energy, high-energy physics and more, potentially ahead of large-scale quantum systems.

Quantum-centric supercomputers—including those supercomputers based on the IBM Quantum System Two® architecture, the building block of quantum-centric supercomputing—leverage the complementary strengths of both quantum technology and traditional supercomputers. Quantum processors can provide speedups for computational tasks that involve generating and sampling from structures that require exponential resources to represent classically but arise naturally in quantum circuits.

In 1994, MIT mathematician Peter Shor discovered an algorithm that can divide large numbers into prime factors exponentially faster than the best classical algorithms by using a hypothetical quantum computer.

In 1995, Shor showed that information can be encoded into noisy quantum computers to protect the information from errors. The next year, Lov Grover discovered a quantum algorithm that can search a database faster than a classical search algorithm. These discoveries greatly accelerated interest in quantum computing. 

This work proved, at least in theory, that a useful quantum computer can process certain complex workloads faster than classical methods—hundreds of thousands of years faster. Even the most advanced supercomputers in the world, like those used in high-profile data centers and universities, are simply not able to process large quantum workflows fast enough. It also proved that quantum computers would be feasible to build and run accurate computations.

Today, quantum computers are no longer theoretical and are available on cloud platforms like IBM Quantum® Platform. Quantum processors like IBM Quantum Heron have proven the viability of quantum computing, where an ecosystem of users across startups, enterprises and research institutions have been using it as a tool for scientific discovery as they work to realize valuable use cases with quantum.

However, today’s quantum computers are limited by obstacles such as the number of qubits with which they can process, and errors inherent to quantum hardware.

Quantum-centric supercomputing, sometimes shortened to QCSC, combines the strengths of quantum and classical computing by using the unique properties of qubits to perform calculations that are otherwise infeasible for classical systems. This approach aims to overcome the limitations of classical high-performance computing by introducing quantum computers into existing workflows, thus enhancing the computational efficiency and capability of both types of systems.

The following are some of the main differences between HPC and quantum-centric supercomputing:

Traditional HPC:

• Built on classical computer architecture
• Limited by binary processing and linear scalability

Quantum-centric supercomputing:

• Includes tightly coupled quantum computers to accelerate workloads
• Optimized to orchestrate work across the quantum computers’ and HPC’s compute clusters in the same data center or in the cloud
• Offers potentially exponential speedups and processing power greater than either quantum or classical computing can provide for certain problems

As experimental quantum computing continues to advance rapidly, we predict that quantum-centric supercomputing will be a pivotal bridge to achieving quantum advantage. Quantum advantage is the milestone by which researchers measure whether a quantum machine can outperform classical hardware simulating a quantum system or any other classical methods for solving a practical problem.

However, quantum computing is not expected to fully replace classical computing. Instead, quantum-centric supercomputers combine quantum computers and classical computers, with each type of system working together to run computations beyond what’s possible on either alone.

Globally, multiple supercomputer facilities have already begun to research the incorporation of quantum-computing hardware. These facilities include Jupiter, developed by the Jülich Supercomputing Centre (JSC) in Germany, Fugaku at the Riken Center for Computational Science in Japan, the Poznan Supercomputing and Networking Center in Poland, and AiMOS at RPI in the United States.

As part of the IBM Quantum Roadmap, IBM hopes to build quantum-centric supercomputers with thousands of logical qubits by 2033.

Unlike traditional computers, quantum computers use the fundamental qualities of quantum physics to potentially solve complex problems. Four key principles of quantum computers are as follows:

• Superposition: Superposition is the state in which a quantum object or system can represent not just one possibility, but a combination of multiple possibilities during the calculation.
• Entanglement: Entanglement is the process in which multiple quantum objects or systems become correlated more strongly than regular probability allows.
• Interference: Interference is the phenomenon in which entangled quantum states can interact and produce more and less likely outcomes.
• Decoherence: Decoherence is the process in which quantum objects and systems can decay, collapse or change, converting into single states describable by classical physics.

While classical computers rely on binary bits (zeros and ones) to store and process data, quantum computers can encode even more data at once using quantum bits (qubits) in superposition. 

A qubit can behave like a traditional bit and store a value of either a zero or a one, but its power comes from its ability to calculate with data in superpositions: a weighted combination of zero and one at the same time.

When combined, a set of qubits in superposition can store more information than the same number of bits. However, each qubit can output only a single bit of information at the end of the computation. Quantum algorithms work by storing and manipulating information in a way that is inaccessible to classical computers, which can provide speedups for certain problems.

Controlling qubits requires delicate hardware that is sensitive to interference and must be kept at extremely cold temperatures. Quantum researchers use cryogenic refrigeration to keep qubits at temperatures colder than the void of space. 

Currently, quantum hardware is expensive, large and error-prone. While researchers work daily to address the challenges of building larger quantum computers, quantum computing is not expected to completely replace traditional computing anytime soon (or potentially ever). That’s because quantum computing is best suited for certain complex problems.

In a matter of minutes, a fully realized quantum computer can potentially solve a simulation problem that would take a traditional supercomputer hundreds of thousands of years. This performance speedup has not yet been realized.

However, IBM quantum computers have already demonstrated quantum utility, the ability to solve problems at a scale beyond brute-force classical simulation. Today, researchers are searching for quantum advantage, the ability for information processing tasks on quantum hardware that satisfy two essential criteria:

1. The correctness of the output can be rigorously validated.
2. It demonstrably offers superior efficiency, cost-effectiveness or accuracy over what is attainable with classical computation alone.

Quantum computing is built on the principles of quantum mechanics, which describe how subatomic particles behave differently from macro-level physics. But because quantum mechanics provides the foundational laws for our entire universe, on a subatomic level, every system is a quantum system.

For this reason, we can say that while conventional computers are also built on top of quantum systems, they don’t innately use quantum mechanical mathematics to run their calculations. Quantum computers take better advantage of quantum mechanics to conduct certain calculations that even high-performance computers cannot. 

Classical computation models use strings of binary digits (bits) to reduce all information into binary code composed of zeros and ones. Using a set of simple logic gates like AND, OR, NO and NAND, we can process that information to perform advanced calculations by using logic circuits.

However, each logic gate can act only on one or two bits at a time. We determine the “state” of a classical computer based on the states of all its bits. Classical computers use transistors and semiconductors to store and process binary information. 

Quantum computers use a special type of quantum hardware called a quantum processing unit (QPU) to store and process data differently. Classical computers use transistors to store bits of information, but quantum computers use qubits, typically made of quantum objects (those objects that behave like the smallest known building blocks of the physical universe). Unlike traditional bits, qubits hold more than two states of information. Instead of simple logic gates, quantum computers use quantum logic gates like X, Y, Z, CNOT, the Hadamard gate and the Toffoli gate. They combine these gates into quantum calculations called quantum circuits.

Different types of qubits are better for different use cases and systems. IBM uses superconducting qubits favored for speed and precise control. Qubits made from photons (individual light particles) are commonly used in quantum communication and quantum cryptography. Other types of qubits include trapped ions, neutral atoms and single electrons held by small semiconductors known as quantum dots.  

In order to brute-force simulate a quantum circuit, classical computers must use exponentially more resources than quantum computers. Classical computers must recreate quantum circuits by using large data structures called tensors, which are like multidimensional spreadsheets of numbers. Quantum computers can run quantum circuits innately.

At the heart of a quantum-centric supercomputer is the quantum processing unit (QPU). IBM’s QPU includes the hardware that routes inputs and outputs as well as a multilayer semiconductor chip etched with superconducting wiring. It’s this chip that contains the qubits used to perform calculations and the gates that perform operations on them. The chip is divided into a layer with the qubits, a layer with resonators for readout and multiple layers of wiring for input and output. The QPU also includes the interconnects, amplifiers and signal-filtering components.

The type of physical qubit used by IBM is made of a superconducting capacitor wired to components called Josephson junctions, which together behave like lossless, nonlinear inductors. Because of the superconducting nature of the system, the current flowing across Josephson junctions can assume only specific values. The Josephson junctions also space out those specific values so that only two of those values are accessible for the calculation.

The qubit is encoded in the lowest two energy values of this system, which then become the zero and one for computation (or as a superposition of both zero and one). Programmers change the qubit states and couple qubits together with quantum instructions, commonly known as gates. These are a series of specially crafted microwave waveforms. 

To keep the qubits operating at the required temperature, the quantum chip must be held inside a dilution refrigerator, which keeps them cold using a mixture of two isotopes of liquid-helium. Other QPU components require room-temperature classical computing hardware. Then, the QPU is connected to runtime infrastructure, which also does error mitigation and results processing. This is a quantum computer.

A quantum-centric supercomputer connects a classical high-performance computing (HPC) cluster or supercomputer to a quantum computer that is either colocated or located over the cloud.

The integration of quantum and classical systems is achieved through middleware and hybrid cloud solutions that facilitate seamless interaction between the two. This hybrid approach helps ensure that quantum processing units can be effectively used within quantum computers connected to existing computational frameworks, maximizing their impact without needing a complete overhaul of current infrastructures.

Over time, quantum-centric supercomputers will evolve in three distinct phases.

1. First, quantum computers will function as specialized compute offload engines within existing HPC complexes, similar to how GPUs augment CPU-based systems today. During this phase, foundational integration between quantum and classical computing will be established across hardware, resource management software and workflow managers, enabling quantum resources to be managed within familiar HPC software tools. As these efforts mature, emerging applications use integrated quantum-HPC systems for closed-loop hybrid computations and error mitigation algorithms, demonstrating the viability of offloading computing onto quantum while informing requirements for deeper integration.
2. Second, quantum and classical resources will become tightly coupled through advanced middleware, enabling seamless workload distribution, dynamic resource allocation and sophisticated error mitigation strategies that span both computing domains.
3. Third, fully co-designed quantum-HPC systems will emerge. These systems will feature unified programming models, integrated system software and hardware architectures optimized for quantum-classical synergy from the ground up. These systems will efficiently partition problems between quantum processors and classical processors.

Quantum computers will connect to classical computers as part of a quantum-centric supercomputing workflow. This workflow will be reliant on an architecture that connects quantum computers to classical computers using high-latency links, middleware, familiar HPC software tools and open source quantum software tools. These elements all connect in a quantum-centric supercomputing architecture.

1. Applications: At the highest level, the architecture considers quantum-centric applications, programs that incorporate both quantum and classical libraries for workflows like simulation, optimization or differential equation solving.
2. Application middleware: Going down a layer, these libraries map problems to appropriate data structures, including tensors and quantum circuits (the core units of computation). In turn, the middleware layer prepares these structures to run on the appropriate hardware. Tools like OpenMP, MPI and SHMEM prepare data to process on GPUs that use CUDA, Triton and PyTorch, while quantum SDKs like Qiskit, TKET and CirQ prepare circuits to run on QPUs.
3. System orchestration: Below the middleware are the workflow and resource management tools that perform orchestration and allocate resources across the appropriate hardware. The quantum resource management interface (QRMI) is one such open tool, a vendor-agnostic library for high-performance compute (HPC) systems to access, control and monitor the behavior of quantum computational resources.
4. Compute: At the bottom is the actual processing and post-processing. These system are the workflow and resource management systems that orchestrate the problem across hardware. Use cases guide the orchestration at this lowest layer, incorporating QPUs and interconnects to scale up and scale out systems of CPUs and GPUs. For example, certain algorithms require scale-out and closed loops, yielding temporal and spatial coupling considerations. Meanwhile, error mitigation requires high-throughput CPU and GPU resources, while users would explore error correction by having low-latency classical systems more closely integrated.

While quantum computing matures, classical computing can assist with computing during a workflow. For example, controlling qubits is a major challenge. External noise and crosstalk between control signals destroy the fragile quantum properties of qubits, and controlling these noise sources has been key in furthering the development of useful quantum-centric supercomputers. Meanwhile, GPUs can perform tensor computations to aid in the running of hybrid quantum-classical algorithms.

Alongside hardware improvements, researchers have demonstrated the ability to deal with some noise by using error-mitigation algorithms that analyze how system noise changes program outputs. Researchers use this information to create a noise model. They then use classical computing, including GPUs, to reverse engineer a noise-free result based on the model’s predictions. Quantum error mitigation is part of the continuous path that will take today’s quantum hardware to tomorrow’s fault-tolerant quantum computers.

In the following video, IBM Quantum researchers Andrew Eddins and Youngseok Kim explain the crucial role that error mitigation will play in achieving useful quantum computing in the near term.

Unlike error mitigation, where post-processing fixes noise after a computation, quantum error correction can remove noise in real-time during processing, without the need to create a specific noise model first. Though effective to a point, error mitigation is limited in scale. As quantum circuits increase in complexity, error correction remains effective in large-scale systems.

Quantum error correction requires numerous resources, such as more qubits and more gates in a circuit. Computing with more qubits requires many more qubits for error correction. Better hardware and better error-correcting codes are bringing error correction closer to reality. Earlier this year, IBM published a new type of error-correcting memory that can conceivably be implemented on near-term quantum computers.

Certain algorithms employ techniques that use both quantum circuits and tensors. Quantum circuits run best on QPUs, while tensors run best on graphics processing units (GPUs). Tensors are data structures that store and compute with many numbers at once.

Algorithms like sampling-based quantum diagonalization (SQD) can use GPUs and QPUs together to squeeze the most performance out of existing high-performance computing systems. A QPU can handle quantum circuits that would otherwise require far larger tensors than a GPU can handle. Meanwhile, CPUs and GPUs can take over at smaller scales for parts of a problem requiring many parallel operations of simpler tensors or traditional processing and orchestrating tasks.

Common conceptions of “quantum computers” often envision a single QPU using millions of physical qubits to run programs independently. “Instead,” writes VP of Quantum and IBM Fellow Jay Gambetta, “we envision computers incorporating multiple QPUs, running quantum circuits in parallel with distributed classical computers.”

Quantum computers excel at solving certain complex problems with the potential to speed up the processing of large-scale datasets. From the development of new drugs to supply-chain optimization to material science and climate change challenges, quantum computing might hold the key to breakthroughs in several critical industries.

• Pharmaceuticals: Quantum computers capable of simulating molecular behavior and biochemical reactions can massively speed up the research and development of new, life-saving drugs and medical treatments.
• Chemistry: For the same reasons quantum computers might impact medical research, they might also provide undiscovered solutions for mitigating dangerous or destructive chemical byproducts. Quantum computing can lead to improved catalysts that enable petrochemical alternatives or better processes for the carbon breakdown necessary for combating climate-threatening emissions.
• Machine learning: As interest and investment in artificial intelligence (AI) and related fields like machine learning ramps up, researchers are exploring whether some quantum algorithms might be able to look at datasets in a new way, providing a speedup for some machine learning problems.

Quantum computers, as they exist today, are scientific tools useful for running specific programs beyond the brute-force ability of classical-only simulations—at least when simulating certain quantum systems. However, for the foreseeable future, quantum computing will work in tandem with modern and future classical supercomputing to be useful. In response, quantum researchers are preparing for a world where classical supercomputers can use quantum circuits to help solve problems.

The key challenges facing quantum-centric supercomputing include maturing the middleware and software that orchestrate workloads across classical and quantum computers, as well as general challenges facing the scale-up of quantum computers themselves. As they work toward achieving quantum advantage, developers are working to overcome the following key obstacles.

A fully realized large-scale quantum computer requires millions of physical qubits. However, practical hardware constraints make scaling single chips to these levels prohibitively challenging.

As a solution, IBM is developing next-generation interconnects capable of encoding quantum information across multiple chips. This solution provides a modular scalability to reach the required qubits needed to perform error correction.

IBM has demonstrated these new interconnects—called l-couplers and m-couplers—with proof-of-concept chips called IBM Quantum Flamingo and IBM Quantum Crossbill. These couplers are responsible for scaling chips. IBM has also demonstrated c-couplers with a chip called IBM Quantum Loon. These are on-chip couplers that connect qubits beyond their nearest neighbors, responsible for assisting with error correction.

While quantum processors relying on qubits used in quantum computing have the potential to massively outperform bit-based processors for certain problems, current quantum processors are limited in the number of qubits and in the number of gates they can run in a single circuit.

As research progresses, IBM plans to introduce a quantum system with 200 logical qubits capable of running 100 million quantum gates by 2029. The goal is 2,000 logical qubits capable of running 1 billion gates by 2033.  

Qubits require large cooling systems capable of creating temperatures lower than outer space in order to operate properly. Researchers are developing ways to scale cryogenics, electronics, infrastructure and software to reduce footprint, cost and energy usage.

Qubit coherence is brief, but integral, for generating accurate quantum data. Decoherence, the process in which qubits fail to function properly and produce inaccurate results, is a major hurdle for any quantum system.

Quantum error correction requires that we encode quantum information into more qubits than we would otherwise need. In 2024, IBM announced a landmark new error-correcting code about 10 times more efficient than prior methods. While error correction is not a solved problem, this new code marks a clear path toward running quantum circuits with a billion logic gates or more. 

Quantum advantage requires two components. The first is quantum circuits that today’s hardware can verifiably run more accurately than any classical hardware or method. The second is a means to demonstrate that those quantum circuits are the best way to solve a quantum problem over any other state-of-the-art method.

Quantum algorithm discovery is what will take quantum advantages and make them useful.

Quantum-centric supercomputing will require a highly performant and stable software stack to write, optimize and execute programs across classical and quantum resources. IBM’s Qiskit® is by far the most widely used quantum software in the world. It is Python-based and composed of open source SDK and supporting tools and services. It is useful for executions both on IBM’s fleet of superconducting quantum computers and on systems that use alternative technologies, such as ions or neutral atoms.