According to this paper, the speed of a quantum processor can be measured in Circuit Layer Operations Per Second (CLOPS). Circuit layers are defined in the context of Quantum Volume (QV) benchmarks, and one circuit layer is defined as one layer of permutation among qubits and one layer of pair-wise random parametrized SU(4) 2-qubit unitary gates where the choice of parameters is deferred until run time. When a QV circuit is compiled to the native gate set of a particular QPU, the circuit depth of the compiled circuit will typically be much larger than the number of QV layers.
In the same paper cited above, the CLOPS figure of merit is shown to be dependent on three factors:
- Circuit execution time (1% of total)
- Circuit delay (roughly 9% of total)
- Run time compilation and data transfer (rest of 90%)
The values cited above were correct in 2021. The most optimistic estimates for CLOPS indicate that by the end of 2022 it will reach 10000 as indicated in this video from IBM.
I want to understand what are the prospects for further improving this number in the next 2-4 years. I guess it is conceivable that run time compilation and data transfer can improve even by orders of magnitude. What about the other two numbers? In page six of the cited paper, there is a comment indicating that circuit delay time will shorten dramatically in the future, but I do not understand why this delay exists and how it can be improved upon. How much these numbers can be improved? Is there a fundamental limit to these numbers due to the way superconducting quantum computing devices are known to work and operate?