← back to jason-chadwick.com

Verity: a resilient kernel for magic state distillation

in preparation

Verity: a resilient kernel for magic state distillation Verity: a resilient kernel for magic state distillation Verity: a resilient kernel for magic state distillation
Jason D. Chadwick*
University of Chicago
Christopher Kang
University of Chicago
Sophia Fuhui Lin
University of Chicago
Frederic T. Chong
University of Chicago
Jason D. Chadwick*
University of Chicago
Christopher Kang
University of Chicago
Sophia Fuhui Lin
University of Chicago
Frederic T. Chong
University of Chicago
Jason D. Chadwick*
University of Chicago
Christopher Kang
University of Chicago
Sophia Fuhui Lin
University of Chicago
Frederic T. Chong
University of Chicago

Abstract

Magic state distillation is a key component of fault-tolerant quantum computing. However, we find that it is susceptible to variety of physical noise sources, thereby degrading quantum programs. The Verity kernel dynamically estimates T infidelities using already available measurement syndromes from the 15-1 T distillery. These estimates can be used to diagnose faulty devices, triggering dynamic recalibrations as noise events occur, and can also provide accurate estimates of mean fidelity over the entire program duration. Our long-time evaluations show that Verity seamlessly adapts to different noise environments, without requiring device characterization: in stable noise regimes, Verity operates at competitive cost, and in challenging noise regimes, Verity maintains infidelity guarantees where existing baselines degrade or completely fail.

[slides]   [code]   [data]  

My Contributions

  • Designed noise model and wrote open-source evaluation code.
  • Designed figures and co-wrote manuscript.

    • Things I Learned

  • Code design & organization: I was much more careful in the design of the code than I have been in the past. Knowing from the start that this would be a code-heavy project encouraged me to carefully plan out the structure of the codebase. This increased attention to design also led me to make a much more modular, understandable, and easy-to-modify codebase than I might have otherwise made, which are design goals that I will continue to apply to my new projects from now on.
  • Performance optimization: To run our simulations of ~10 million distillations, we needed to do some very heavy performance optimizations on the code to achieve runtimes of a few hours. The first iterations of the simulation code would have taken days or months to do the same thing. I learned a lot about profiling, parallelization, smart use of memory, batching, and large-scale memoization during this project.
  • Knowing when to pivot: ...TODO
  • Presentation: ...TODO

    • ← back to jason-chadwick.com