$IONQ ran real-time error correction on a single off-the-shelf CPU.
Qubits pick up errors from their surroundings, constantly. A classical computer has to find those errors and fix them while the quantum machine runs. It is called the decoder, and when it falls behind, the quantum computer stops and waits for it.
IonQ announced today that its decoder kept up, from end to end, on one processor. The paper went up on arXiv on August 25, signed Min Ye, Andrii Maksymov and Nicolas Delfosse. What they measured:
➡️ 408 logical qubits, in 68 memory blocks and 20 magic state factories, 11,680 physical qubits in all
➡️ 555,130 T gates and 1.3 million logical measurements, across 31.5 million rounds of error checking
➡️ 0.02% "stretch", the extra time the decoding cost that computation
➡️ under 0.3% on all three test workloads at the error rate they assume, under 12% if that rate is five times worse
➡️ the hardware: a 2024 Apple M4 Max, the chip in a MacBook Pro, 12 of its 16 cores
IonQ splits the job between two decoders on the same chip. One runs continuously over the error data. A lighter one wakes up for each logical measurement, where the answer is needed fastest. Both are built on the beam search decoder Ye, Delfosse and Dave Wecker published in December, now out in PRX Quantum.
Nicolas Delfosse, paper co-author and quantum research lead at IonQ, says a decoder that runs on a single CPU "provides a practical path to commercial-scale fault-tolerant quantum computing".
The quantum computer here is simulated, noise included. The decoding is real work, timed on real silicon.
IonQ reads the result as validation of a pillar of its Walking Cat architecture: the classical side does not have to grow exponentially as the machine gets wider in logical qubits or deeper in operations.
John Gamble, VP of Architecture at IonQ, says the numbers IonQ steers by are time to solution, cost to solution and energy to solution.
The release calls the result a foundation for the roadmap beyond 256 physical qubits, toward thousands.
Congratulations to Min Ye, Andrii Maksymov and Nicolas Delfosse, and to the people they thank in the paper: Felix Tripier, Nolan Coble, Jacob Young, Finn Buessen, John Gamble and the whole IonQ team.
🔗
#IonQ #Quantum #QEC
