Insider Brief
- Researchers at IonQ have published a detailed end-to-end blueprint for a fault-tolerant quantum computer based on trapped ions that they say could run millions of logical operations on hundreds of error-corrected qubits using as few as 2,514 physical qubits.
- The architecture, called “walking cat,” is built on a modern error-correction approach called quantum LDPC codes and relies entirely on hardware components already demonstrated on small devices.
- The researchers estimate that a version of the machine using 10,000 physical qubits could complete a classically intractable physics simulation within one month.
- Image: An AI representation of my cat, Harper, walking across a quantum computer keyboard after I turn away for just a second, a not unlikely scenario should this technology scale.
A team of researchers at IonQ has published a technical blueprint for a fault-tolerant quantum computer built on trapped ions, describing an architecture they contend could execute millions of quantum operations on hundreds of logical qubits using only a few thousand physical particles. They add it’s a design that’s achievable with hardware already demonstrated in the lab.
The study, posted as an arXiv preprint by Felix Tripier, Nicolas Delfosse, and colleagues at IonQ, lays out the full architecture of what they call the walking cat machine, from the high-level compiler that translates quantum programs into device instructions down to the physical movements of individual ions on a chip. The researchers write that their design emphasizes simplicity at every layer, a deliberate choice intended to make the machine buildable in the near term rather than theoretically optimal in some distant future.
The team writes in a blog post on the work that they have developed: “An end-to-end blueprint for an FTQC architecture based on modern quantum error-correcting codes and designed with realistic engineering constraints in mind is still missing in the literature. In this work, we bridge that gap,” and adding that “This is the blueprint that IonQ will use to build the fault-tolerant era.”
The practical implications, if the design can be physically realized, would be a machine that addresses key vulnerabilities and limitations of quantum computing. Today, quantum computers are classified as noisy intermediate-scale quantum, or NISQ, devices — machines that can run thousands of operations on hundreds of physical qubits but accumulate errors too rapidly to tackle industrially meaningful problems. The transition to a fault-tolerant machine requires more than simply adding qubits, according to the paper. It demands an architectural overhaul, a complete rethinking of how logical operations, error correction and hardware control are organized.
The Error Problem and the Error Solution
At the heart of quantum’s problem is noise, the researchers report. Every operation on a physical qubit, whether a gate, a measurement, or even the passage of time, introduces a small probability of error. In a conventional NISQ machine, those errors compound quickly. Fault-tolerant quantum computers address this by encoding information not in single physical qubits but in logical qubits, each of which is spread across many physical qubits. When errors occur on individual physical qubits, the logical information can be reconstructed and corrected before mistakes accumulate.
The walking cat architecture achieves this using a class of techniques called quantum low-density parity-check codes, or LDPC codes. The term “low-density” refers to the structure of the code’s error-detection checks: each check involves only a small, fixed number of qubits, and each qubit participates in only a small, fixed number of checks. This sparse structure is efficient as it allows more logical information to be packed into fewer physical qubits than older designs based on surface codes, which dominated the field for years and are still used by many of the leading quantum hardware companies.
The backbone of the walking cat architecture is what the researchers call a cat factory, a dedicated component of the machine that continuously produces a special type of quantum state called a cat state. Cat states, named after the Schrödinger’s cat thought experiment, are quantum superpositions that allow multiple logical operations to be performed simultaneously and efficiently. These states are distributed throughout the machine’s memory blocks, where they are consumed to drive computations.
A separate component called a magic factory produces another type of quantum resource called a magic state, which is needed to implement the most computationally powerful — and most error-prone — class of quantum gate. Magic states cannot be produced directly from error-correcting codes, so they must be generated separately and distilled to the required purity before use. The walking cat machine includes two different magic factory designs, based on schemes developed by Meier, Eastin and Knill and by an approach the IonQ team calls cat-based Clifford measurements, respectively.
The logical architecture connects these components through a modular framework, with separate units for memory, magic state production, cat state generation and auxiliary Bell state production. A compiler layer translates quantum programs into sequences of logical instructions, which are then mapped onto physical operations on the chip.
The paper describes three specific configurations of the walking cat architecture, each representing a different tradeoff between simplicity, speed and density.
The simplest version, described as the single-code architecture, uses one LDPC code for both memory and magic state production. It is the most straightforward to build but the least computationally efficient.
The fast architecture is built around a new error-correcting code the researchers introduce in the paper, designated BB7-[[70, 6, 9]]. This means the code uses 70 physical qubits to encode 6 logical qubits, with a minimum distance — a measure of how many physical errors can be tolerated before a logical error occurs — of 9. This configuration is optimized for speed, using a technique called Clifford frame tracking that allows an entire class of six-qubit logical operations to be executed without physically applying them to the hardware, which reduces gate overhead. Using this architecture with 10,000 physical qubits, the researchers estimate that a quantum simulation of the Heisenberg model — a standard benchmark in quantum physics — running on 100 sites could be completed within one month, including all repetitions needed to achieve chemical accuracy. According to the paper, that level of precision would suggest the machine has entered the regime of classically intractable physics simulations.
The dense architecture is built on a new code the researchers call Q102, designated [[102, 22, 9]], which encodes 22 logical qubits per block using 102 physical qubits. Because each block holds more logical information, fewer blocks are needed to reach a target qubit count, reducing total physical qubit requirements. Using the dense architecture, the team calculates that a machine supporting 110 logical qubits and executing approximately one million T gates — the expensive non-Clifford operations that power universal quantum computation — per day requires only 2,514 physical qubits in total. That figure includes every qubit in the system, such as memory, error correction, magic factories, cat factories, reservoirs for fresh ion loading and routing, according to the paper.
Moving Ions, Not Wires
Non-technical readers might want to think of the trapped ion system as a microscopic assembly line, where individual charged atoms play the role of workers who are physically moved from station to station on a chip, briefly paired together to perform an operation, then shuttled back to storage until they’re needed again.
More technically, in trapped ion devices, individual ions — typically atoms stripped of one electron so they carry a slight electric charge — are held in place using electric fields. Quantum information is stored in the internal energy levels of the ions, and operations are performed using precisely controlled microwave or laser pulses. Two-qubit gates, the basic building blocks of quantum circuits, are performed by briefly bringing two ions close together in a dedicated interaction zone on the chip.
The architecture is built on top of what the researchers call the quantum charge-coupled device, or QCCD, framework. In this design, ions can be physically transported across a two-dimensional chip, shuttled between storage zones and interaction zones as needed. The IonQ team simplifies this picture for their error-correction design using what they call a moving-qubit model, which represents the chip as a square grid where ions move between neighboring sites and interact with adjacent neighbors. This simplified model is used to design and simulate error-correction protocols while abstracting away the fine-grained detail of ion transport. A separate micro-architecture layer translates those protocols back to actual device instructions.
The researchers emphasize that all hardware operations their design relies upon have been experimentally demonstrated on small trapped-ion devices. Two-qubit gates using a technique called EQC, or entangling qubit coupling, and physical ion transport have both been validated in laboratory settings. The challenge of scaling to thousands of physical qubits remains, but the team indicates that by using only demonstrated components and explicitly designing for simplicity, the architecture avoids relying on speculative advances.
Limitations and Future Directions
It’s important to note that the paper is a theoretical blueprint, not a hardware demonstration, and the researchers acknowledge that moving from design to physical machine will be a significant undertaking. Scaling trapped-ion systems to thousands of physical qubits is itself a major engineering challenge. Ion traps must be fabricated with precise tolerances, classical control electronics must scale alongside the quantum hardware, and the loss and leakage of ions — events in which a qubit either escapes the trap or falls into an inaccessible internal state — must be managed continuously.
The walking cat architecture explicitly accounts for ion loss and leakage in its error-correction protocols. A dedicated loss-detection scheme monitors for missing ions and triggers a reloading process when one is lost. A leakage-correction subroutine detects and corrects qubits that have fallen into unintended internal states. These additions increase the overhead of the architecture but the researchers argue they are essential for a realistic design.
All performance estimates in the paper are based on assumed hardware parameters, including a physical error rate of one in 10,000 operations, a leakage rate of one in 100,000 operations, and a loss rate of one in 10 million operations.
The researchers say future work could improve on the design at every layer of the architecture, from better error-correcting codes and faster compiler algorithms to co-optimized micro-architectures that exploit the relationship between adjacent design layers. They note that many potentially important quantum applications remain entirely unexplored, and anticipate that a machine capable of running millions of logical operations would open access to a broad range of scientific problems that are currently inaccessible to both classical computers and existing quantum devices.
The IonQ team includes Tripier, Delfosse, and 16 co-authors.
For a deeper, more technical look at the work, the paper is available on arXiv and the IonQ team is creating a series of deep dives on the work. It is important to note that arXiv is a preprint server where researchers can circulate work quickly and receive feedback. It is not a peer-reviewed publication. Peer review remains an important step in validating scientific results.
