Surface Code Error Correction on a Defective Lattice

TL;DR

This paper evaluates the performance of surface code error correction on defective qubit lattices with varying yields, demonstrating that high-yield chips can effectively support large-scale quantum computing despite static faults.

Contribution

It extends surface code error correction to handle static qubit faults and analyzes the impact of different defect locations and yields on error suppression.

Findings

95% yield is sufficient for large-scale quantum computing.

A local gate error threshold of ~0.3% is achievable with a code distance of seven.

Discarding badly fabricated chips improves overall error suppression at 90% yield.

Abstract

The yield of physical qubits fabricated in the laboratory is much lower than that of classical transistors in production semiconductor fabrication. Actual implementations of quantum computers will be susceptible to loss in the form of physically faulty qubits. Though these physical faults must negatively affect the computation, we can deal with them by adapting error correction schemes. In this paper We have simulated statically placed single-fault lattices and lattices with randomly placed faults at functional qubit yields of 80%, 90% and 95%, showing practical performance of a defective surface code by employing actual circuit constructions and realistic errors on every gate, including identity gates. We extend Stace et al.'s superplaquettes solution against dynamic losses for the surface code to handle static losses such as physically faulty qubits. The single-fault analysis shows that a static loss at the periphery of the lattice has less negative effect than a static loss at the center. The randomly-faulty analysis shows that 95% yield is good enough to build a large scale quantum computer. The local gate error rate threshold is $\sim 0.3\%$, and a code distance of seven suppresses the residual error rate below the original error rate at $p=0.1\%$. 90% yield is also good enough when we discard badly fabricated quantum computation chips, while 80% yield does not show enough error suppression even when discarding 90% of the chips. We evaluated several metrics for predicting chip performance, and found that the average of the product of the number of data qubits and the cycle time of a stabilizer measurement of stabilizers gave the strongest correlation with post-correction residual error rates. Our analysis will help with selecting usable quantum computation chips from among the pool of all fabricated chips.