Fault-Tolerant System IBM Aims to Realize A Scalable Quantum Computer by 2029

Related Vendors

IBM Deutschland GmbH

FAULHABER Drive Systems

Wuxi InfiMotion Technology Co., Ltd.

IBM has unveiled a new quantum architecture that aims to enable the construction of a fault-tolerant quantum computer by 2029, with significantly reduced hardware requirements.

IBM has unveiled a new architecture for quantum computers that has the potential to overcome one of the central bottlenecks of the technology: error correction. The company plans to deliver a market-ready, fault-tolerant system based on this by 2029.

At the center is the construction of a modular quantum computer called Starling, which is to use 200 logically error-corrected qubits. With this, IBM aims to realize a device that can perform up to 100 million error-corrected quantum operations—a magnitude that goes significantly beyond previous "gadget experiments."

Redundancy Against Instability

Quantum computers are susceptible to disturbances, such as stray radiation or thermal influences. To stably store or process usable information, many physical qubits must be combined into a "logical qubit." This redundancy enables the detection and correction of errors—albeit at the cost of significant hardware complexity.

Previous methods, such as the surface code, require several hundred to a thousand physical qubits per logical qubit. IBM is now pursuing a new approach: so-called quantum Low-Density Parity Check (qLDPC) codes. These are intended to manage with about one-tenth of the previous number of qubits, allowing for a more compact system configuration.

Another innovation is the capability for real-time error diagnosis. Using special FPGAs (Field Programmable Gate Arrays), errors in reading out qubit states can be immediately detected and classified—a crucial step toward stable and scalable error correction.

Modular Design as the Key to Scaling

The first building block of the new system architecture is the chip Loon, which is to be introduced as early as 2025. It allows for non-local coupling of qubits on the same chip—a prerequisite for the implementation of qLDPC codes.

In 2026, Kookaburra will follow as the first module containing both computing and storage units. A year later, IBM plans to network two of these modules with Cockatoo into a functional composite system. The modular design is intended to allow flexible scaling in the future, comparable to data centers of classic architecture.

The completion of Starling is planned for 2028, with cloud-based availability a year later. The system will be operated in a specially built quantum data center in Poughkeepsie, New York.

For IBM, Starling is a milestone on the way to Blue Jay, a system with 2,000 logical qubits and a target of one billion error-corrected quantum operations. Even though Starling is unlikely to implement economically relevant algorithms, it represents an important interim step.

Technical hurdles and competitive pressure

Experts like Wolfgang Pfaff (University of Illinois) consider IBM's plan ambitious but realistic. The modular strategy fits the current development of the industry. Nevertheless, implementation remains a technological challenge.

Particularly critical is the reduction of error rates in quantum operations. IBM views the improvement of coherence times as a central lever—that is, the duration over which a qubit can reliably maintain its state. Advances with the Heron chips indicate initial successes in this direction.

IBM thus positions itself in a close race with other players like Google, AWS, QuEra, or PsiQuantum. The outcome is open—but IBM sees itself at an advantage due to its vertically integrated development of hardware, algorithms, and system architecture.

The focus is now on engineering work. "The foundations have been laid— now it's about building the machine," says Jay Gambetta, head of IBM Quantum. The next few years will show whether the step from research lab to industrial quantum computer can actually be realized. (sg)

Machine Authentication

Machine Operator Authentication - Why knowing "who" matters