Nvidia unveils Ising AI models for quantum error correction and calibration  – News

Technology and computing giant Nvidia Corp. today announced the release of Ising, the world’s first open artificial intelligence model family aimed at quantum computing calibration and error correction. 

Nvidia, whose main business is the graphics processing units that power AI, said these AI models will allow researchers and enterprise companies to build better quantum computers capable of running useful applications at scale. 

To build and run useful applications, quantum computers must handle millions of qubits — the atomic computational units of quantum information. The essential problem is that qubits are fragile, error-prone and susceptible to noise at scale. As quantum computers grow, they must be error-corrected and calibrated in real time to account for environmental factors and remain useful. 

“AI is essential to making quantum computing practical,” founder and Chief Executive Jensen Huang said. “With Ising, AI becomes the control plane — the operating system of quantum machines — transforming fragile qubits into scalable and reliable quantum-GPU systems.” 

Ising is named after the landmark mathematical model that helped simplify the understanding of complex physical systems by describing how interacting particles, or spins, influence one another. Nvidia is providing two models: one for real-time error correction and one for calibration. 

The need for error correction is obvious: It turns noisy systems into coherent outputs. That is where Ising Decoding comes in. Decoding comes in two variants of a 3D convolutional neural network model, one optimized for speed and the other for accuracy, that perform real-time decoding for quantum error correction. Nvidia said the models provide up to 2.5 times more speed and three times more accuracy than pyMatching, the current open-source industry standard. 

Ising Calibration allows physicists to prepare systems by tuning, measuring and optimizing physical control signals, such as microwaves or lasers. This calibration is necessary to ensure high-fidelity outputs by correcting for noise, hardware instability and parameter drift over time. It’s a vision-language model that can rapidly interpret and react to measurements from quantum processors, driving AI agents that automate continuous calibration. 

The path ahead to better quantum computers 

Speaking at a briefing, Sam Stanwyck, Nvidia’s director of quantum product, said the company chose decoding and calibration first because they address the most immediate obstacles to scaling quantum systems.  

He described both as “AI-shaped workloads,” where models can make an immediate impact today, but said Nvidia’s longer-term vision goes further. Over time, the company expects AI to help build and optimize quantum circuits as well, making decoding and calibration the first milestones on a broader path toward scalable quantum-GPU-based supercomputers. 

Ising Decoding and Ising Calibration are already being adopted by enterprise and research organizations. Decoding is being deployed by Cornell University, Sandia National Laboratories, the University of California at San Diego and UC Santa Barbara, among others. Calibration is already in use by Atom Computing, Academia Sinica, EeroQ, IonQ, IQM Quantum Computers, Q-CTRL and others. 

Additionally, Nvidia released a cookbook of guides, including quantum computing workflows and training data, along with an Nvidia NIM microservice. This will allow developers to customize, train, fine-tune and build models for different hardware setups, and run them locally on researchers’ systems to protect sensitive data. 

Image: News/Microsoft Designer