⇧ [VIDÉO] You might also like this partner content
Next year, IBM plans to set a new record by developing the most powerful quantum computer ever designed, combining the computing power of several quantum processors connected in parallel. This announcement follows the successful demonstration of linking two quantum processors, a strategy that should enable a record number of qubits processed by 2025 — and more than 4,000 qubits by 2026.
Quantum processors, or quantum processing units (QPUs), constitute the basic elements of quantum computers, allowing the manipulation of quantum bits (qubits). They are in a way the quantum equivalent of the central processing unit (CPU) of classical computers. About six years ago, the first QPU developed by IBM supported 20 qubits. Currently, Condor, the company’s largest QPU, has 1,121 qubits, while that of startup Atom Computing has 1,180.
Starting next year, IBM plans to develop even larger processors that can support up to several thousand qubits. However, the development of QPUs faces a major challenge: error rates increase as systems become larger. In order to overcome this obstacle, reliable error correction systems are essential.
To do this, IBM offers a modular approach consisting of combining different components. “Our goal is to build quantum-centric supercomputers», Explained in a blog post on the company’s 2025 roadmap, Jay Gambetta, vice president in charge of quantum computer research at IBM. “The quantum-centric supercomputer will integrate quantum processors, classical processors, quantum communications networks, and classical networks, all working together to transform the way we compute“. This strategy would ultimately make it possible to exploit the computing power of several QPUs connected in parallel and operating as one.
Up to 4,158 qubits processed by 2026
Last November, the IBM team performed the first experimental demonstration of two QPUs connected in parallel and operating as a single device. Each unit (an Eagle QPU) can process up to 127 qubits and connects to the other in real time through a traditional backhaul system. This makes it possible to control quantum gates (the building blocks of quantum circuits) in a classical way within a dynamic circuit. The assembly also has an integrated error mitigation device and a quantum state induction system requiring periodic connectivity.
The company plans to extend this strategy to its new generation of Flamingo QPUs that can process 462 qubits. At least three of these would be connected in parallel to be integrated into a system that could support up to 1,386 qubits, far surpassing the current record.
«Our scaling learning will bring all these advances together to exploit their full potential», explains Gambetta. This step would therefore allow us to move on to the next, which will consist of developing the Kookaburra multichip processor, processing 1,386 qubits alone. In 2026, the company plans to connect three Kookaburra QPUs via a quantum communications link and that would support an impressive number of qubits: 4,158 in total.
See also
With their communications link supporting quantum parallelization, three Kookaburra QPUs could be connected to form a 4,158-qubit system. © IBM
«The combination of these technologies (classical parallelization, multi-chip quantum processors and quantum parallelization) gives us all the ingredients we need to take our computers to where our roadmap takes us», Says Gambetta.
However, a few more years will be necessary to demonstrate the proper functioning of these modular QPUs, specifies the expert. Indeed, their connectivity and synchronization require hundreds of components called “couplers”, and the IBM team has so far only developed two types. Additionally, an entirely different type of coupler (under development, according to the company) will be essential to fully modular quantum computers. These components differ depending on whether they connect adjacent or distant QPUs.
Nevertheless, “by 2025, we will have removed the main barriers to scaling quantum processors through the modular quantum hardware, control electronics, and cryogenic infrastructure that accompany it », concludes Gambetta.
