Is Circuit Depth Accurate for Comparing Quantum Circuit Runtimes?
Authors:
Matthew Tremba,
Ji Liu,
Paul Hovland
Abstract:
Although quantum circuit depth is commonly used to estimate differences in circuit runtimes, it overlooks a prevailing trait of current hardware implementation: different gates have different execution times. Consequently, the use of depth may lead to inaccurate comparisons of circuit runtimes, especially for circuits of similar scale. In this paper, we introduce an alternative metric, gate-aware…
▽ More
Although quantum circuit depth is commonly used to estimate differences in circuit runtimes, it overlooks a prevailing trait of current hardware implementation: different gates have different execution times. Consequently, the use of depth may lead to inaccurate comparisons of circuit runtimes, especially for circuits of similar scale. In this paper, we introduce an alternative metric, gate-aware depth, that uses unique gate weights, and investigate how its accuracy in comparing circuit runtimes compares to the existing metrics of traditional and multi-qubit circuit depth. To do so, we compiled a suite of 15 practical circuits using different algorithms and compared depths and runtimes between the compiled versions to determine how accurately the size of the change in depth approximated the size of the change in runtime, and how accurately the order of circuits by depth matched their order by runtime. When approximating the size of runtime changes, gate-aware depth decreased the approximation error by an average of 412 times relative to traditional depth and 124 times relative to multi-qubit depth. When matching the order of true runtimes, gate-aware depth achieved the highest accuracy on all devices and a perfect accuracy of 100% on five out of six devices. Additionally, we show that the optimal weights needed to achieve these accuracy improvements can be easily calculated using device gate times, and provide good general weight values for the IBM Eagle and Heron architectures.
△ Less
Submitted 22 May, 2025;
originally announced May 2025.
