Neural-network Decoders for Surface Codes

A guided research project in Quantum Computing.

← Quantum Computing projects

Experience Research Quantum Computing

Question

Can a neural network infer the right correction from a surface-code syndrome?

The decoder sees only the syndrome and must choose a correction that preserves the logical state.
The decoder sees only the syndrome and must choose a correction that preserves the logical state.
Logical error-rate curves for surface codes of several distances.
Decoder performance curves from the surface-code research materials.

Goals

  • Generate error–syndrome data for a surface code.
  • Train a decoder and measure its logical error rate.

Method

Learn the pattern from stabilizer measurements to an effective correction class.

  • Simulate physical errors and compute the resulting syndromes.
  • Train a suitable neural network on syndrome and correction pairs.
  • Compare its accuracy, speed, and logical failure rate with a standard decoder.