Pasqal’s Quantum Processor Reveals Physics Theory Didn’t Predict
Quantum computers are often used to reproduce physics we already understand. French quantum company Pasqal has now gone further. In a new study released as a preprint, its 256-qubit processor revealed quantum magnet behaviour that current theory did not predict (Pasqal).
What the team did
The team arranged rubidium atoms in a 16 × 16 grid to simulate a two-dimensional quantum magnet. They suddenly pushed the system out of balance and watched how it settled back down. Depending on the magnetic field applied, the system behaved in one of three ways:
- It settled quickly, as expected.
- It held a stable in-between state for a long time, also as expected.
- It slowed down far more than theory predicts, which was unexpected.
Where classical computers disagree
This third behaviour appears where three of the best classical simulation methods stop agreeing with one another. The quantum processor’s results, by contrast, stayed consistent as the system grew. The team checked the findings against noise models and exact calculations, but the cause of the slowdown is still unexplained.
“There is still much we don’t understand about quantum materials, and quantum processors can help us explore these new frontiers,” said Loïc Henriet, CEO of Pasqal. Earlier this year, the same processor matched laboratory measurements on a real magnetic crystal. This result goes a step further by uncovering something new.
A European research effort
Pasqal carried out the work with classical simulation experts at the Flatiron Institute (Simons Foundation), together with the University of Augsburg, Forschungszentrum Jülich, the University of Regensburg, and the Institut d’Optique Graduate School and CNRS. Joseph Tindall of the Flatiron Institute stressed the value of comparing quantum and classical results: “This is much more beneficial than just trying to prove quantum advantage as it leads to interesting physics results like we have here that will stand the test of time.”
Read the press release, the technical blog post or the preprint on arXiv.
ALICE & BOB
Alice & Bob: Using AI and GPUs to Build Quantum Computers Faster
What happens when AI enters the quantum lab, with GPUs doing the heavy lifting? At NVIDIA GTC in Berlin, Théau Peronnin, CEO and co-founder of Paris-based Alice & Bob, showed what that looks like in practice. The company now uses AI and GPU computing to speed up every step of building its quantum computers, from chip design to error correction.
Faster from design to experiment
Alice & Bob builds quantum computers based on “cat qubits”, a type of qubit designed to resist certain errors by design. The company showed how AI and GPUs help at each stage:
- Chip design: Dynamiqs, its quantum simulation library, works with NVIDIA’s CUDA-Q software and runs on NVIDIA GPUs. It simulates cat qubit physics up to 75 times faster, so the team can test chip designs before building them.
- Measurements: AI analyses lab measurements using ten times less computing power, with results checked against a digital twin of the chip.
- Experiments: 155 AI agents planned a research study, the team reviewed the plan, and a single agent then ran it on Alice & Bob’s chips by itself. A full study took one day instead of a week.
A step towards error correction at scale
The biggest challenge in quantum computing is correcting errors fast enough to keep up with the machine. On 13 October, Alice & Bob announced a first proof of concept using NVIDIA NVQLink, which connects quantum processors to GPU supercomputers. The company decoded simulated error data from its error-correction codes on GPUs for the first time. Decoding became more efficient as several error signals were processed in parallel, which suggests the approach can scale. The work is not yet in real time, and that is the team’s next goal (Alice & Bob).
“Decoding is one of its most demanding classical tasks,” said Juliette Peyronnet, US General Manager at Alice & Bob. The next step, she said, is making it fast enough to keep pace with the quantum computer. Sam Stanwyck, Director of Quantum Product at NVIDIA, said the work “shows how NVQLink can in fact accelerate the real-time decoding needed to turn cat qubits into fault tolerant quantum-GPU supercomputers.”
A growing European quantum player
Founded in 2020, Alice & Bob has raised €180 million and employs more than 250 people. It says its cat qubit approach could cut the hardware needed for a useful large-scale quantum computer by up to 200 times compared with competing approaches. The company’s stand at GTC Berlin also drew a notable visitor: NVIDIA’s Chief Marketing Officer, Alison Wagonfeld.
