Technology and computing giant Nvidia recently released a family of open source artificial intelligence models called Ising, which it claims is the world's first AI model suite specifically for quantum computing calibration and error correction. NVIDIA said this family of models will help scientific research institutions and companies build more powerful quantum computers, giving them the ability to run truly useful practical applications on a larger scale.
For a quantum computer to run complex applications, the system needs to process quantum information units with a scale of millions of qubits. However, the qubits themselves are extremely fragile, susceptible to noise interference and error-prone. As the scale of quantum computers expands, the system must be able to complete error correction and fine calibration in real time during operation and compensate for environmental fluctuations in order to maintain the validity of calculation results. NVIDIA founder and CEO Jensen Huang said, "AI is the key to making quantum computing practical." In his view, with Ising, "AI will become the control plane of quantum machines - equivalent to the operating system of quantum machines, transforming fragile qubits into scalable and reliable quantum-GPU systems."
The name Ising comes from the Ising model, a classic mathematical model in physics, which is used to describe the interaction between particle spins and characterize complex physical systems in a more concise way. NVIDIA provides two types of models this time: one is used for real-time error correction, and the other focuses on the calibration of quantum systems.
In terms of error correction, Ising Decoding is responsible for "decoding" quantum measurements under noisy conditions into coherent outputs. It is based on a three-dimensional convolutional neural network and offers two variants: one focused on speed and the other on accuracy. Nvidia claims that compared with pyMatching, an error correction tool commonly used in the current open source industry, Ising Decoding can increase decoding speed by up to 2.5 times and improve accuracy by about 3 times.
In terms of calibration, Ising Calibration is mainly aimed at physicists and engineering teams and is used to tune, measure and optimize control signals of quantum hardware. These control signals include microwaves, lasers and other physical means. High-quality quantum output is highly dependent on precise calibration to counteract issues such as noise, hardware instability, and parameter drift over time. According to NVIDIA, Ising Calibration is a visual-language model that can quickly interpret measurement data from quantum processors and drive AI agents to continuously and automatically complete the calibration process.
When talking about the future roadmap, Nvidia Quantum Product Director Sam Stanwyck said at the press conference that the company chose to launch decoding and calibration first because these two links are the most pressing bottlenecks restricting the expansion of quantum systems. He described the two as "AI-shaped workloads" and believed that the introduction of AI in these areas can have immediate and considerable effects. However, he also emphasized that Nvidia’s long-term vision is not limited to this. In the future, it is hoped that AI can also participate in the construction and optimization of quantum circuits, making decoding and calibration the first milestones on the road to quantum-GPU supercomputing platforms.
At present, Ising Decoding and Ising Calibration have begun to be applied in enterprises and scientific research institutions. In terms of error decoding, Cornell University, Sandia National Laboratories, University of California, San Diego, University of California, Santa Barbara and other institutions have begun to deploy relevant models. In terms of calibration, many quantum computing-related companies and research organizations such as Atom Computing, Academia Sinica, EeroQ, IonQ, IQM Quantum Computers, Q‑CTRL, etc. are already using Ising Calibration for system debugging and optimization.
In order to lower the barriers to use, NVIDIA also released a set of "cookbook" guides, which include workflow examples for quantum computing and supporting training data, and provides microservices based on NVIDIA NIM. These resources will help developers customize, train and fine-tune models based on different quantum hardware architectures and run them in local research environments, leveraging AI capabilities while keeping sensitive experimental data within the institution.