Scalable Quantum Circuit Optimization Using Automated Synthesis

We describe a scalable technique for circuit optimization, with emphasis on depth reduction. Our Quantum Global Optimizer (QGo) uses a combination of partitioning and synthesis: 1) partition the circuit into blocks; 2) each block is re-generated and optimized using synthesis; and 3) re-compose the circuit by stitching all the blocks together. For NISQ benchmarks, we show that QGo can reduce the number of CNOT gates by 29.8% on average (up to 38.8%), when compared with industrial compilers such as IBM Qiskit and T|ket. We demonstrate the QGo scalability to optimize circuits of 60+ qubits.
Approaches similar to QGo should be incorporated into any quantum compiler. Traditional compilers are scalable and rely on a combination of local peephole optimization and global mapping. Recent synthesis results show global optimization on circuits with few qubits. As unitaries scale exponentially, this imposes hard limits on scalability for global optimization. Intuitively QGo provides a framework for peephole optimization using customizable size blocks, each block being subject to global optimization.