Insider Brief:
- IBM released a full-stack roadmap to build a fault-tolerant quantum computer by 2029, starting with modular processors and culminating in the Quantum Starling system, capable of executing over 100 million quantum operations using 200 logical qubits.
- The company is shifting from surface codes to quantum LDPC codes, reducing physical qubit overhead by up to 90% and enabling scalable architectures through non-local connectivity.
- IBM introduced a decoder design optimized for qLDPC codes that runs in real time on classical hardware, eliminating the need for co-located HPC systems and addressing a bottleneck in fault-tolerant architectures.
- The roadmap emphasizes iteration over linear progress, bringing in insights from past processors and focusing on modularity, decoding, and hybrid quantum–classical integration to reach practical utility.
In quantum computing, progress is rarely linear. Inflection points emerge from long feedback loops, iterations between experiment and theory, dead ends that inform new directions, and prototypes that may not make it to market but inform what does. IBM’s latest announcement reflects this truth. Rather than fixate on a single processor or product, the company has announced a full-stack plan to build the world’s first large-scale, fault-tolerant quantum computer that scales in both size and utility.
The IBM Quantum Starling, slated for deployment in 2029 at IBM’s Quantum Data Center in Poughkeepsie, New York, is expected to be capable of executing over 100 million quantum operations using 200 logical qubits. It will embody the company’s transition from quantum systems that demonstrate advantage in restricted domains to those that have the potential to run industrially relevant workloads. In tandem with the announcement, IBM released two new technical papers (here and here) and a refined quantum development roadmap, detailing the architecture and supporting infrastructure required to support fault-tolerant quantum computing.
From Logical Concepts to Physical Systems
The challenges of building a fault-tolerant quantum computer go beyond adding more qubits. Fault tolerance requires encoding quantum information in such a way that it persists despite the challenge of errors introduced by decoherence, control noise, and gate imperfections. To do this, quantum error correction aggregates many physical qubits into fewer logical qubits. However, doing so traditionally comes with extensive hardware overhead.
The dominant approach for years has been the surface code, which uses local qubit interactions to suppress error rates exponentially as the system grows. While effective in principle, surface code implementations require thousands of physical qubits per logical qubit—potentially reaching tens of millions of physical qubits for algorithms like Shor’s—at error rates sufficient for meaningful computation.
IBM is carving another path. In a 2024 Nature paper, the company introduced quantum low-density parity check (qLDPC) codes, a new class of quantum error correction that reduces the physical qubit overhead by up to 90% compared to surface codes. This efficiency gain emerges from qLDPC’s allowance of non-local interactions, connections between distant qubits within a code block, which are infeasible in surface code layouts but essential for scalable architectures.
While qLDPC codes reduce overhead, they introduce new engineering challenges, especially around decoding. Real-time decoding, the ability to identify and correct errors as they occur, is essential. IBM’s new technical papers, released alongside the roadmap, propose a practical decoding solution that operates with conventional computing resources, circumventing the need for co-located high-performance computing infrastructure.
The Roadmap: Loon to Starling to Blue Jay
IBM has structured its plan as a sequence of modular milestones. Each processor in the roadmap plays a specific role in realizing fault-tolerant quantum computing.
- Loon (2025): Demonstrates architectural elements needed for qLDPC, such as high-connectivity layouts and c-couplers that link distant qubits on a chip.
- Kookaburra (2026): Integrates logic and memory into the first fault-tolerant module, using logical processing units to perform encoded operations.
- Cockatoo (2027): Establishes entanglement between modules using l-couplers, paving the way for distributed computation across chips.
- Starling (2028–2029): Demonstrates magic state injection across multiple modules in 2028, then scales in 2029 to a full system capable of executing 100 million quantum gates on 200 logical qubits.
The modular design reflects an essential learning that scaling quantum computing may not only be found by increasing the quantity of qubits on a chip, but also in enabling chips to communicate and function cohesively as parts of a larger system. IBM’s choice to pursue modularity also reflects a commitment to physical and engineering feasibility.
Engineering Tradeoffs and the Role of Decoders
One tradeoff of adopting qLDPC codes is the need for more complex qubit connectivity, specifically, six connections per physical qubit, including two that are non-local. IBM’s roadmap addresses this through hardware innovations such as tunable couplers, c-couplers, and l-couplers. These technologies extend the range of gate operations without degrading performance, enabling scalable layouts that avoid the congestion and heat load of monolithic chips.
Central to making the architecture work is the ability to decode error syndromes in real time. There has been industry-wide concern about whether qLDPC codes can be decoded fast enough for real-time fault-tolerant operation. IBM’s forthcoming decoder design, outlined in a new paper, is intended to run in real time using conventional computing hardware, which avoids the need for HPC or GPU clusters. IBM’s dedicated decoder architecture runs efficiently on classical processors, removing this bottleneck and reinforcing the feasibility of the approach.
Starling and Beyond: What Defines “Fault-Tolerant” at Scale?
IBM defines a large-scale fault-tolerant system as one capable of executing more than 100 million quantum operations across hundreds of logical qubits. At this scale, computations are not only protected from noise but can outperform classical simulations in practical domains such as materials design, quantum chemistry, and combinatorial optimization.
Starling, IBM’s 2029 system, will be the first to meet that threshold. Starling is expected to execute fully encoded quantum programs and interface with classical high-performance computing systems, a model IBM anticipates will be necessary for future quantum-classical hybrid workloads. It will also test the full pipeline of quantum compilation, decoding, and modular execution at scale.
Eventually, IBM plans to build on Starling with a system named Blue Jay, targeted for 2033 and beyond. Blue Jay is expected to scale operations to the billion-gate level with over 2,000 logical qubits, which will complete the arc from early experiments to fault-tolerant utility.
The Non-Linear Path to Quantum Advantage
There is the possibility that not everything in IBM’s innovation roadmap will end up in productized systems. That’s by design. The roadmap is explicitly iterative, built on an understanding that dead ends can be just as informative. Technologies like Condor and Flamingo helped IBM refine packaging, coupling, and chip layout, even if they never became commercial platforms.
This iterative approach contrasts with linear narratives of quantum progress. It reflects a maturing field in which architectural pivots, engineering constraints, and decoding theory are part of a feedback loop, not detours. IBM’s roadmap stands out for its specificity, this acknowledgment that practical fault tolerance requires both innovation and infrastructure discipline.
0 Comments