Quantum Fan-out: Circuit Optimizations and Technology Modeling
Instruction scheduling is a key compiler optimization in quantum computing, just as it is for classical computing. Current schedulers optimize for data parallelism by allowing simultaneous execution of instructions, as long as their qubits do not overlap. However, on many quantum hardware platforms, instructions on overlapping qubits can be executed simultaneously through __global interactions__. For example, while fan-out in traditional quantum circuits can only be implemented sequentially when viewed at the logical level, global interactions at the physical level allow fan-out to be achieved in one step. We leverage this simultaneous fan-out primitive to optimize circuit synthesis for NISQ (Noisy Intermediate-Scale Quantum) workloads. In addition, we introduce novel quantum memory architectures based on fan-out. Our work also addresses hardware implementation of the fan-out primitive. We perform realistic simulations for trapped ion quantum computers. We also demonstrate experimental proof-of-concept of fan-out with superconducting qubits. We perform depth (runtime) and fidelity estimation for NISQ application circuits and quantum memory architectures under realistic noise models. Our simulations indicate promising results with an asymptotic advantage in runtime, as well as 7--24% reduction in error.