Creating entangled logical qubits in the heavy-hex lattice with topological codes

TL;DR

This paper demonstrates how to implement entangled logical qubits using topological codes on a heavy-hex lattice in IBM Quantum devices, enabling fault-tolerant entanglement and measurement of logical observables.

Contribution

It introduces a method to utilize unused qubits in sparse connectivity architectures to implement multiple quantum error correcting codes simultaneously.

Findings

Realized surface and Bacon-Shor codes on 133 qubits.

Achieved entanglement with code distance up to 4.

Verified Bell inequality violation with high fidelity.

Abstract

Designs for quantum error correction depend strongly on the connectivity of the qubits. For solid state qubits, the most straightforward approach is to have connectivity constrained to a planar graph. Practical considerations may also further restrict the connectivity, resulting in a relatively sparse graph such as the heavy-hex architecture of current IBM Quantum devices. In such cases it is hard to use all qubits to their full potential. Instead, in order to emulate the denser connectivity required to implement well-known quantum error correcting codes, many qubits remain effectively unused. In this work we show how this bug can be turned into a feature. By using the unused qubits of one code to execute another, two codes can be implemented on top of each other, allowing easy application of fault-tolerant entangling gates and measurements. We demonstrate this by realizing a surface code and a Bacon-Shor code on a 133 qubit IBM Quantum device. Using transversal CX gates and lattice surgery we demonstrate entanglement between these logical qubits with code distance up to $d = 4$ and five rounds of stabilizer measurement cycles. The nonplanar coupling between the qubits allows us to simultaneously measure the logical $XX$, $YY$, and $ZZ$ observables. With this we verify the violation of Bell's inequality for both the $d=2$ case with post selection featuring a fidelity of $94\%$, and the $d=3$ instance using only quantum error correction.