Qubit Routing for (Almost) Free
For quantum computing researchers, this provides a theoretical foundation for avoiding costly SWAP-based routing in NISQ devices.
The paper proves tight bounds on CNOT gate counts for phase polynomial synthesis and shows that synthesizing only hardware-allowed gates eliminates routing overhead, reducing the overhead factor from up to O(n log^2 n) to O(1).
In this paper, we give a mathematical proof that bounds the number of CNOT gates required to synthesize an $n$ qubit phase polynomial with $g$ terms to be at least $O(\frac{gn}{\max (\log g, 1)})$ and at most $O(gn)$. However, when targeting restricted hardware, not all CNOTs are allowed. If we were to use SWAP-based methods to route the qubits on the architecture such that the earlier synthesized gates are natively allowed, we increase the number of CNOTs by a routing overhead factor of $O(\log n) \leq α\leq O(n \log^2 n)$. However, if we only synthesize allowed gates, we do not need to route any qubits. Moreover, in that case the routing overhead factor is $1 \leq α\leq 4 \simeq O(1)$. Additionally, since phase polynomials and Hadamard gates together form a universal gate set, we get qubit routing for almost free.