Enabling Practical Quantum Computation: Compiler and Architecture Techniques for Bridging the Algorithms-to-Devices Resource Gap

Abstract

Quantum computing (QC) is an emerging computational paradigm poised to fundamentally change what is computable in material science, machine learning, optimization, and other domains. From the first 1- and 2-qubit systems in the early 2000s, today's QC hardware landscape includes cloud-accessible systems with 10-50 qubits and multiple qubit technology candidates for large-scale QC. In spite of rapid hardware progress, the first practically useful QC applications have not been demonstrated yet, even though hundreds of QC algorithms have been developed in the last three decades. This is fundamentally because of a large gap between the resource requirements of QC applications and the capabilities of quantum hardware that is buildable in the near-term; qubit counts and operational noise constraints of applications exceed hardware capabilities by 5-6 orders of magnitude. This dissertation seeks to close the resource gap between quantum algorithms and hardware. The resource gap in QC is similar to the resource gap that existed in classical computing in the 1950s. Taking inspiration from the vital role played by computer architecture in scaling up classical computers, this dissertation develops quantum compilation and architectural techniques. Unlike prior research efforts which largely focused on designing individual layers in the QC execution stack in isolation, this dissertation develops a cross-cutting design approach to optimize the QC stack. Using this approach, Part I of this dissertation develops techniques to bridge the applications-to-devices resource gap from the top of the stack and Part II develops techniques to bridge the gap from the bottom of the stack. Part I includes noise-adaptive compilation techniques that adapt program executions to the large spatial and temporal noise variations that occur in near-term QC systems. Part I also includes the first software technique to mitigate the impact of crosstalk noise on applications. The techniques presented in Part I offer one to two orders of magnitude improvements in application fidelity compared to vendor compiler toolflows and related works. Part II includes an extensive study of architectural designs of real QC systems, a study on designing trapped ion systems based on application requirements and a study on instruction set design to balance application and hardware needs. Through hardware-software co-design, Part II offers up to four orders of magnitude (i.e., 10000X) improvement in reliability for near-term QC devices. This dissertation has already influenced several industry toolflows and architectures. Noise-adaptivity is now a standard optimization in industry compilers, including IBM's Qiskit, Rigetti's Quilc, Cambridge Quantum Computing's TKET and Oak Ridge National Laboratory's XACC and QCOR toolflows. IBM's Qiskit compiler also incorporated this dissertation's crosstalk mitigation techniques. This dissertation's instruction set design recommendations have also been adopted by IBM and these recommendations were instrumental in achieving a quantum volume of 64 on their hardware. Further, the architectural study in Part II was instrumental in driving the community towards application-level benchmarking instead of relying only on low-level benchmarking and metrics like qubit counts. In summary, this dissertation shows that cross-cutting design offers several orders of magnitude improvement in reliability and performance for QC systems, compared to existing approaches. We expect that this approach will be beneficial both for near-term NISQ hardware and the longer-term systems that follow. The research contributions and directions laid out in this dissertation have the potential to accelerate the progress towards practically viable QC by several years, rather than relying solely on hardware or application improvements.