AlphaQubit

AlphaQubit is a neural-network decoder for quantum error correction developed jointly by Google DeepMind and Google Quantum AI. Its task is to identify errors arising inside quantum processors, a critical step toward building reliable, large-scale quantum computers.

How it works

The system is based on the Transformer architecture. It takes repeated consistency-check measurements of a logical qubit as input and predicts whether, when measured at the end of the experiment, the logical qubit has flipped from how it was prepared. The model was first trained on hundreds of millions of examples from a quantum simulator, then fine-tuned on thousands of real samples from a Sycamore processor.

Results

On Sycamore data (surface code at distance 3 = 17 physical qubits and distance 5 = 49 qubits), AlphaQubit reached higher accuracy than previous leading decoders: it made 6% fewer errors than tensor-network methods and 30% fewer errors than the correlated matching decoder. In simulation it scaled to systems of up to 241 qubits (distance 11).

A remaining limitation is speed: in its current form AlphaQubit is too slow to correct errors in real time on a superconducting quantum processor. The results were published in Nature on 20 November 2024.