In celebration of World Quantum Day, NVIDIA today (April 17) announced the world's first open-source AI model series designed specifically for building quantum processors (QPUs) – "NVIDIA Ising". NVIDIA emphasized that the biggest challenge for quantum computing to reach the practical application stage of millions of qubits lies in solving the quantum noise problem, and quantum workloads are essentially "AI workloads", thus relying on AI to drive decoding, control, and error correction.
Breaking through the bottleneck of manual calibration: Ising Calibration (Quantum Hardware Calibration)
Modern qubits are extremely susceptible to noise interference and are physically unstable. Traditionally, even calibrating just 50 to 100 qubits requires physics experts to spend several days manually fine-tuning them, which is completely impractical when scaling up to thousands or millions of qubits.
To address this issue, NVIDIA released an open-source model called "Ising Calibration".
• Visual language model architecture:This is a visual language model (VLM) with 350 billion parameters, which can directly read the measurement data of the QPU and automatically perform calibration.
• Lightweight and efficient:Compared to existing alternatives, its model is 15 times smaller, yet it achieves world-class performance in calibration benchmark tests that include six metrics.
• Significantly reduced time:It can significantly reduce the calibration time, which originally required human experts to spend "several days," to "several hours."
Enhancing error correction efficiency: Ising Decoding (Quantum Decoding)
In quantum error correction (QEC), the system needs to process massive amounts of data, up to terabytes in a very short time, which places stringent requirements on computational speed and accuracy.
"Ising Decoding" employs a convolutional neural network (CNN) architecture and is specifically designed for quantum error correction. NVIDIA provides two variant models for different application scenarios:
• Pursuing ultimate speed:The speed-optimized version operates 2.5 times faster than the current industry standard (Pine-marten).
• Pursuing ultimate precision:The version optimized for accuracy is three times more accurate than the existing standard.
In addition, the amount of data required for training the model has been reduced by 10 times, which is a great boon for resource-intensive quantum research environments.
Embrace the open-source ecosystem and combine it with NVIDIA's existing quantum platform.
NVIDIA Ising is not a closed system, but a complete "family of open source models". NVIDIA not only releases the models themselves, but also provides cookbooks for fine-tuning, quantization, and inference workflows, as well as related open source research papers and benchmark data, allowing ecosystem partners and researchers to customize and fine-tune them for their specific hardware.
At the same time, NVIDIA Ising has also deeply integrated existing resources from the NVIDIA Open Quantum Platform, including the Quantum-GPU platform CUDA-Q, cuQuantum which provides GPU-accelerated computing, and the NVQLink reference architecture for low-latency integration. Through cuQuantum, even developers who have not yet obtained expensive physical quantum hardware can train and develop synthetic data in a GPU simulation environment, further democratizing quantum computing.
Currently, NVIDIA Ising has been widely adopted by the quantum computing ecosystem, including many top research institutions and companies such as Lawrence Berkeley National Laboratory, Harvard University, IonQ, IQM, and Atom Computing, which have incorporated the Ising model into their calibration and decoding R&D workflows.
In summary, NVIDIA is leveraging its AI advantages to lay the foundation for the software infrastructure of future hybrid supercomputing.
Faced with the uncertainties of quantum computing, NVIDIA is leveraging its absolute dominance in AI to address the most challenging issues in quantum hardware development: control and debugging. By open-sourcing the Ising model, NVIDIA is essentially establishing an unshakeable foundational software standard and ecosystem for future quantum-GPU hybrid supercomputing.
