NederlandsNvidia simulates a 104-qubit quantum circuit on a European supercomputer, real errors included
Nvidia and the German research centre Jülich used the JUPITER supercomputer to simulate how a 104-qubit quantum computer makes an important building block, including the errors that real hardware makes. This lets designs be tested before they are built.
In future, quantum computers may do calculations that are too heavy for ordinary computers, for example when designing medicines and new materials. Their computing units are called qubits. They are very sensitive: a little heat or vibration and an error creeps into the calculation. That is why researchers are working on error correction, in which many qubits together form one reliable qubit.
For the hardest calculations, such a protected quantum computer also needs special helper states, called 'magic states'. Making them costs a lot of qubits. According to some estimates, it can take up to 95 per cent of all the qubits in a quantum computer.
A more economical method, tested on a computer first
Researchers at Google came up with a more economical way to make those helper states, 'magic state cultivation'. You can think of it as growing something: you start small, keep checking along the way whether anything has gone wrong, and then let the state grow. According to Nvidia, this takes up to 90 per cent fewer qubits than the older method.
Nvidia and the Jülich Supercomputing Centre in Germany have now simulated that method on an ordinary supercomputer, with 104 qubits. They also included the errors that real quantum chips make, such as small shifts that pile up and qubits that 'leak' into a state where they do not belong. According to Nvidia, it is the largest simulation so far of a quantum computer with error correction.
A big computing job
Simulating qubits quickly becomes very heavy: each extra qubit roughly doubles the memory needed. The researchers therefore used 8,192 AI chips of JUPITER, the first European supercomputer in the highest class. Together they kept track of more than 560 terabytes of data, the memory of thousands of ordinary laptops. Clever tricks helped: qubits that are finished free up their memory for the next ones.
The method was first tested on a smaller Nvidia system and then scaled up step by step. According to Nvidia, such a simulation therefore takes days instead of weeks. Last year, JUPITER already simulated a 50-qubit quantum computer, but then without errors.
The next steps
The simulation is above all a test set-up. Researchers can now follow in it how an error spreads through the whole circuit and which checks catch it. The next step is to calculate how good the resulting helper states ultimately are, and what that means for the calculations a quantum computer has to do. With that, builders of quantum chips can improve their designs before they have to build anything.
Things are moving fast in the lab too. Stanford, for example, recently built a grid of light traps to link quantum computers made of atoms.
What does this mean for the road to the singularity?
Quantum computers and AI supercomputers work together here. The chips made for AI are now helping to design quantum computers that can later help in turn, for example in finding new medicines or materials. Testing in a simulation is much faster and cheaper than building new hardware each time. The better such simulations become, the sooner a reliable quantum computer can arrive. AI can also come to play a role in this itself, by searching thousands of simulated designs for the most economical recipe.
What is the technological singularity? Read the explanation →
