For years, quantum computers have been the subject of dreams. They are said to have the ability to simulate molecules to design new drugs, optimize gigantic logistical chains, or break supposedly unbreakable encryption systems. However, in reality, the technology faces a persistent obstacle: qubits, the basic building blocks of quantum computing, are extremely unstable. Sensitive to the slightest environmental noise, they accumulate errors as soon as one tries to make them work together on a large scale. As long as this problem persists, quantum computers remain a laboratory promise.
This is precisely the barrier that Nvidia aims to overcome. Introduced on April 14, World Quantum Day, Ising is a family of open-source AI models designed to help researchers and industry professionals build quantum processors capable of running practical applications. The name pays tribute to the Lenz-Ising model, a statistical mechanics formalism that significantly simplified the understanding of complex physical systems in the last century.
Two Models to Tame Quantum Qubit Noise
Ising comes in two complementary tools, each targeting one of the two major challenges of quantum computing.
The first, Ising Calibration, is a vision-language model with 35 billion parameters. Essentially, it reads the measurements produced by a quantum processor, interprets them, and automatically triggers the necessary adjustments. For non-experts, it’s a bit like a robotic mechanic constantly monitoring an ultra-sensitive engine and making real-time adjustments. The result: a calibration that used to take several days can now be completed in a few hours.
The second, Ising Decoding, focuses on error correction itself. It is based on a 3D convolutional neural network, offered in two variations: one emphasizing speed, the other precision. Nvidia claims performance improvements of up to 2.5 times faster and 3 times more precise than pyMatching, the open-source standard reference in the field.
Labs and Industries Already Onboard
Ising is not just a demonstration confined to conference slides. Atom Computing, IonQ, Infleqtion, and the Lawrence Berkeley National Laboratory are already utilizing the calibration aspect. The University of Chicago, Sandia National Laboratories, and IQM Quantum Computers have deployed the decoding aspect. Nvidia also provides training datasets, ready-to-use workflows, and NIM microservices to allow developers to adapt the models to their own hardware architectures, all executable locally to protect sensitive data.
The global quantum market is estimated to exceed 11 billion dollars by 2030. With Ising, Nvidia adds a layer of artificial intelligence to the core of quantum infrastructure and bets on the GPU-quantum convergence to have a lasting impact in this race.
The models are available on GitHub, Hugging Face, and build.nvidia.com.
