Nvidia, the chipmaker whose processors power the vast majority of artificial intelligence systems worldwide, has released a set of free, open-source AI tools designed to help researchers and companies build quantum computers that actually work reliably.
The tools, collectively called Ising, target the two engineering problems that most urgently need solving before quantum computers can move from laboratory curiosities to machines capable of tackling problems in drug discovery, materials science and cryptography.
The first problem is calibration, the painstaking process of tuning a quantum processor so that its components behave as precisely as possible.
Quantum computers use qubits, the quantum equivalent of the bits that store information in ordinary computers, but qubits are extraordinarily sensitive to their surroundings and drift out of alignment constantly, requiring engineers to recalibrate their machines repeatedly.
Nvidia's calibration tool is a large AI model with 35 billion internal settings that can read the measurements coming off a quantum processor, interpret what has gone wrong, and automatically adjust the machine's tuning, a process the company says can reduce calibration time from days to hours.
The second problem is error correction.
Even after careful calibration, today's best quantum processors still make an error roughly once in every thousand operations, a rate that would need to fall to around one in a trillion before the machines become useful for serious work.
Quantum error correction works by spreading information across many physical qubits and constantly checking for mistakes, but the checking process itself demands enormous computing speed because errors must be caught faster than they accumulate.
Nvidia's error correction tool uses a type of AI called a three-dimensional convolutional neural network, a system designed to recognise patterns across multiple dimensions of data simultaneously, and comes in two versions: one optimised for speed and one for accuracy.
The company says its decoder is up to two and a half times faster and three times more accurate than the current open-source standard used by most quantum research groups, while requiring ten times less data to train.
Jensen Huang, Nvidia's chief executive, described AI as becoming the control plane for quantum hardware, effectively the operating system that manages the noise inherent in quantum processors and turns fragile qubits into something scalable and reliable.
The models have already been adopted by a range of institutions including Harvard University, Fermi National Accelerator Laboratory, IQM Quantum Computers and the UK's National Physical Laboratory.
Related reading
- Nvidia quantum AI launch triggers buying frenzy across Asian and US tech stocks
- Nvidia launches Ising AI models to speed quantum processor calibration and error correction
- Meta and Arm to co-develop data centre chips built for AI workloads
By making the tools freely available on platforms including GitHub and Hugging Face, Nvidia is positioning itself at the centre of the quantum computing ecosystem without building quantum hardware itself, a strategy that mirrors how it came to dominate AI by providing the chips and software that others use to build their systems.
The quantum computing market is expected to surpass $11 billion by 2030, according to analyst firm Resonance.
The recap
- NVIDIA launches Ising, an open family of quantum AI models
- Ising Decoding delivers up to 2.5x faster and 3x more accurate
- Models, data and tools available on GitHub, Hugging Face, build.nvidia.com
