IBM roadmap to fault-tolerant quantum computers by 2029: transcript

Key points

• Quantum computers are expected to tackle molecule simulation, structured data analysis, optimization and search problems.
• IBM plans Starling in 2029 with 200 logical qubits running 100 million gates, then BlueJay in 2033 with 2,000 logical qubits and 1 billion gates.
• Quantum utility arrives at around 100 qubits and a few thousand gates, but errors limit complexity, so fault-tolerant machines are needed.
• IBM's high-rate QLDPC gross code stores 12 logical qubits in a block and uses about 10 times fewer qubits than the surface code.
• A modular architecture links code blocks through logical processing units and universal adapters so systems can scale.
• The processor roadmap runs from Loon in 2025 through Kookaburra in 2026 and Cockatoo in 2027 to Starling, with IBM promising a fault-tolerant computer by 2029.

Chapters

1. 0:02Quantum computing promise and IBM's history
2. 0:37Starling and BlueJay fault-tolerant targets
3. 2:06Gate errors and quantum utility
4. 3:14How error-correcting codes and logical qubits work
5. 3:48Surface code versus IBM's QLDPC gross code
6. 4:52Modular architecture and couplers
7. 5:57Processor roadmap from Loon to Starling
8. 6:46Fabrication pace and the 2029 commitment

Full transcript

0:02Quantum computers are widely expected to solve problems that we can't solve with classical computers. Simulations of molecules, analyzing structured data, optimization and search problems. IBM has a long and rich history of successfully taking many, many ideas to a full-on product. Quantum computers will be no different. It's been a really incredible journey. I joined here almost 19 years ago, we were basically trying to work on just single qubits and showing signs of life, trying to get them entangled.

0:37Did we think it was likely back then to get to where we are today? No, but it was our goal to keep making progress, solve one problem at a time, and now looking back this is where we got us today, building the world's most powerful quantum computers. We believe we have the only credible plan to build fault-tolerant quantum computers. We're so confident, we even put a date on it. Our first fault-tolerant quantum computer, Starling, arrives in 2029 with 200 logical qubits capable of accurately running 100 million quantum logic gates.

1:16Then, in 2033, we get BlueJay, a fault-tolerant quantum computer with 2,000 logical qubits capable of accurately running a staggering 1 billion gates. We're building the data center to house the system right now in Poughkeepsie, New York. I'm a senior research scientist at IBM, and I do research on gate errors in quantum hardware. Quantum computers apply quantum logic gates to quantum bits or qubits to perform complex calculations. If we can run more gates on more qubits, we can make our calculations even more complex. My focus is on L-couplers, so devising the technology that we need, having these couplers that go between chips. From a practical perspective, errors limit the complexity we can achieve.

2:06That's why the research community has developed sophisticated tools and techniques to manage them. Better qubits, like the trans bond, which is much less sensitive to charge nodes. Improved materials and processing for qubits. Better ways of shaping the electromagnetic pulses we use to control qubits. Dynamical decoupling error suppression. Error detecting and correcting codes with post-selection. The list goes on. These techniques have brought us to a point we call quantum utility. Quantum utility means that these devices can no longer to be simulated with brute force classical techniques.

2:41This happens at around 100 qubits or so and a few thousand gates. And we are within reach of getting to 15,000 gates. But we're still a far cry from what we need for the most demanding computations. In order to implement hundreds of millions or even billions of gates, we need a fault-tolerant quantum computer. My goal is to get useful quantum algorithms to actually run in the gates inside of this error-corrected quantum computer. Currently, the challenge of getting anything to happen on a quantum computer is errors.

3:14First, we need an error-correcting code, where the computational overhead grows only mildly as complexity increases. An error-correcting code takes many error-prone physical qubits and entangles them to encode high-quality logical qubits, which have less error. These logical qubits have error rates that decrease exponentially, with a polynomial increase in the number of physical qubits. Next, we need the error-correcting code to have a syndrome extraction circuit and an efficient decoder that is fault-tolerant. The code acts as a fault-tolerant memory that stores quantum information and protects it against errors.

3:48Finally, we need the ability to perform arbitrary quantum circuits using a universal gate set, which requires preparing something we call magic states. There are a lot of places that want to build a fault-tolerant quantum computer, And the upfront choice is an error-correcting code called the surface code. This code has a high tolerance to errors with a threshold of approximately 6.7 times 10 to the minus 3, only nearest neighbor qubit interactions, and simple syndrome extraction circuits.

4:19However, you need so many physical qubits that that particular code becomes impractical at large scale. What's different about IBM's QLDPC codes is that they are high-rate QLDPC codes. It requires more connectivity between qubits, but uses 10 times fewer qubits than the surface code with the same high tolerance to errors. The QLDPC code stores 12 logical qubits in one block, using just two sets of 144 physical qubits to represent the encoded state and syndrome extraction, hence we named it the gross code.

4:52So a huge part of it is being able to scale, but at some point we just run out of real estate on a piece of silicon. And so the approach that we've taken is we've designed a modular architecture made up of code blocks representing 12 logical qubits, each connected to a logical processing unit and universal adapter to define a single module. The logical processing unit implements simple gates. The universal adapter allows entanglement between modules. We can connect the modules together and scale the system size in a straightforward way.

5:22We haven't solved all the experimental challenges to realize this architecture yet, but... Everything we're doing today is geared towards making sure that we can achieve them. We need C-couplers for long-range coupling within chips, L-couplers for long range coupling between chips, and more. But we've already demonstrated prototypes of these technologies, and combining this with our modular architecture gives us confidence we have a realistic path to practical, fault-tolerant quantum computing. We've baked this new architecture into the next four quantum processors on our innovation roadmap.

5:57We start in 2025 with the IBM Quantum Loon, an important test vehicle and first demonstration of the technology required to run our error correcting code. In 2026, KUKABOR will be the first quantum processor that will use our logical processing unit and QLDPC codes to demonstrate a quantum memory and simple gates in 2027 we'll introduce cockatoo a quantum processor that will allow us to test the universal adapter connecting multiple modules together and in 2028 we have starling our first full demonstration of the architecture capable of running our qld pc codes with quantum gates magic states and an integrated real-time hardware decoder Everything is coming together, and we really have a plan.

6:46We have a strong quantum research team and advanced semiconductor technology expertise, so at any moment we have over 20 device types being built across multiple fabrication facilities and new quantum devices in fabrication every 15 to 20 days. This means we're moving much faster than anyone else in this space. We stand by what we said. IBM will deliver a fault-tolerant quantum computer to our clients by 2029.

Source and licence

Recording: File:IBM Quantum Industry Webinar Series： Quantum Computing for the Automotive Industry.webm by IBM Research (2025-04-21), via Wikimedia Commons. Licence: CC BY 3.0.

The transcript is machine speech-to-text by the TidyTools Audio & video transcriber; the summary, key points and chapters were written by Claude from that transcript. Both may contain errors: check the recording before quoting it.