Infleqtion 与 Nvidia 降低量子计算机物理量子比特与逻辑量子比特之比

COMPUTE

A lot of advancements are being made in quantum computing that move beyond the basics of quantum processors (QPUs) and hardware fundamentals. Companies are expanding out into the architectural realm, subsystems for these machines are being pulled together, and quantum software and algorithms are being developed, while efforts to make quantum computing scalable and integrated into HPC environments are being made.

Still, a key to getting quantum computing from the research phase to commercial viability is solving the error correction challenge. Qubits are notoriously fragile and can split apart – and lose their quantum properties – due to various “noises” like light, sound, temperature, or the activity of other qubits, and a quantum system that makes too many errors isn’t commercially viable. To improve error correction, multiple entangled physical qubits are grouped together to create logical qubits.

However, that comes at a price, from the technical challenges of needing thousands of physical qubits to make a single logical one to cooling and manufacturing costs to adding more noise. Given that, reducing the number of physical qubits per logical qubit is important.

“An unofficial law of physics is ‘there’s no such thing as a free lunch,’” Quantinuum executives wrote earlier this year. “Creating high quality, low error-rate logical qubits often costs many physical qubits, thus reducing the size of calculations you can run, despite your new, lower-than-ever error rates.”

Work is being done researchers and standards groups to reduce the logical-to-physical qubit ratio. Vendors also are making gains. Sheir Yarkoni, director of quantum software for Qarakal Quantum, told The Next Platform that the startup’s new Pangaea architecture will reduce the physical qubit requirements by 10 times. D-Wave has said the same about its developing dual-rail superconducting systems, and Quantinuum in a paper published in March detailed its technology that creates 48 error-corrected and 64 error-detected logical qubits out of 98 physical qubits.

At the IEEE Quantum Week event this week in Toronto, Infleqtion executives said that, thanks to the integration of their open source quantum low-density parity-check (qLDPC) software library and Nvidia’s new CUDA-Q Logical – a logical orchestration and compiler layer also unveiled at the show – they built a validated high-rate quantum error correction code that uses 98 physical data qubits for 18 logical qubits, or about 5.4 physical qubits per logical qubit and driving an 18.4 percent encoding rate.

Infleqtion, which uses neutral atoms in its Sqale systems (below) to create its qubits, used an early-access version of Nvidia’s CUDA-Q Logical software to build a hypergraph-product simplex (HGPS) code block, create stabilizers and logical operators, and put in place the necessary underlying algebra, Victory Omole, senior quantum research engineer for the vendor, wrote in a report. The code also contains 49 X checks and 49 Z checks, used to detect specific types of physical errors and each of weight six.

The canonical logical representatives returned by qLDPC software have weight four.

“Many familiar QEC constructions devote a large number of data qubits to each logical qubit,” Omole wrote. “High-rate qLDPC codes change that arithmetic by encoding multiple logical qubits in one block while keeping each stabilizer check sparse. The appeal is obvious: if the rest of the architecture cooperates, more of the machine can carry useful logical information.”

Doing the work in that sentence is “if the rest of the architecture cooperates.” As Omele wrote, while such numbers as the physical-to-logical qubit ratio are good, other factors in a quantum computing architecture – syndrome ancilla qubits (used to perform error syndrome measurements), scheduling, readout, control, and classical decoding, among others – need to ensure those numbers are preserved throughout the system stack.

That’s where CUDA-Q Logical comes in. Infleqtion’s gLDPC software provides the underlying mathematics for the code.

“CUDA-Q Logical supplies the logical ownership, block allocation, and compilation model,” he wrote. “The code interface in CUDA-Q Logical takes sparse row supports. A small adapter converts the dense NumPy arrays, after which the qLDPC software-generated checks and logical operators can be declared as an 18-port CUDA-Q Logical code.”

Using gLDPC and CUDA-Q Logical together, Infleqtion engineers were able to build the code and validate the structure, which then lead to the creation of the compiler pipeline. The block coming out of the pipeline is confirmed to hold multiple logical qubits inside a single encoded structure, the engineers wrote, adding that the pipeline builds on the architecture research done by Fred Chong, Infleqtion’s chief scientist for quantum software.

The integration of Infleqtion’s and Nvidia’s technologies not only led to the narrowed physical-to-logical qubit ratio, but also that the stabilizer’s matrices and the X and Z checks commute, and that the CUDA-Q code has 18 logical ports and 49 each X and Z checks. In addition, the workflow that was compiled ensure two distinct logical qubit owners within a single encoded block.

“This integration focuses on code construction, logical ownership, and compilation,” Omele wrote. “In coming work, we plan to extend this workflow through physical syndrome extraction, neutral-atom noise modeling, detector generation, and decoder benchmarking.”

The work in Chong’s laboratory focuses on reconfigurable neutral-atom arrays that are good for high-rate qLDPC codes due to their parallel operations and flexible connectivity.

“For our team, the opportunity is to trace that promise through a real machine model,” he wrote. “That means accounting for where ancillas live, how atoms move, which checks can run in parallel, how leakage and loss appear in readout, and whether the classical decoder can keep up with the syndrome stream. Logical performance emerges from the coordination of hardware, control, compilation, and classical compute.”

Infleqtion was founded in 2007, spinning out from atomics physics research at the University of Colorado initially under the name ColdQuanta before rebranding at Infleqtion in 2022. The company’s plans include an aggressive roadmap that sees it reaching more than 1,000 logical qubits by 2030.

The work with Nvidia and its CUDA-Q Logical software is only the latest step in what has been a busy year for the company. Recent events include a $20 million follow-on contract with NASA to continue developing the Quantum Gravity Gradiometer Pathfinder, a Jet Propulsion Laboratory mission to fly the first space-based quantum gravity sensor, the opening of the Colorado Quantum Innovation Center, and the awarding of three projects in the Department of Energy’s Genesis Mission, which aims to accelerate research and innovation using AI and includes a quantum arm with the goal of deploying the world’s first fault-tolerant quantum computer.

Infleqtion executives last week also announced plans to bring together the company’s neutral-atom quantum systems and sensors with Cisco Systems’ quantum networking portfolio in a joint R&D effort to scale quantum computing by networking systems together. Cisco last year began laying the groundwork for a quantum network architecture, and in April unveiled its Universal Quantum Switch, a working prototype for connecting quantum systems from different vendors that use different modalities and quantum sensors to create a single network.

Infleqtion engineers noted that neutral atoms work naturally with photons, which are particles of light used to carry quantum information across networks, making their systems well suited for connecting into such networks and to work with within Cisco’s vision of a modality-agnostic network architecture.

SPONSORED FEATURE: HPE and NVIDIA help organizations apply accelerated computing, software, enterprise infrastructure, networking, control plane, and services to the workloads and business outcomes that matter most.

October 1, 2026

September 30, 2026

AMD Buys World Labs For $8.2 Billion To Get AI Model Cred