Diary/2019-4-13

ASPLOS(1)

ワークショップの日

Practical Aspects of Building a Software Stack for Noisy Intermediate-Scale Quantum Devices

Introduction

• "Quntum Computation and Quntum Information"
• "Quntum Computing for the Determined" - https://www.youtube.com/playlist?list=PL1826E60FD05B44E4
• "A Gentle Introduction"
• http://www.theory.caltech.edu/~preskill/ph219/ph219_2017
• http://www-bcf.usc.edu/~tbrun/Course/
• http://algassert.com/quirk

Qunatum Computing Software Stack: Hardware Technologies

• Three Main Directions
  • 2-Level Quntum Systems (cf. https://en.wikipedia.org/wiki/Two-state_quantum_system)
    • many players
  • Toplogical Qubits (cf. https://en.wikipedia.org/wiki/Topological_quantum_computer)
    • mainly microsft/delft
  • Quantum Qnnealers
    • mainly D-Wave
• Gate Sets in Superconducting Chips
  • Compilers to translate gates sets, do mapping
• Take Away
  • Hardware technologies are far from dtermined
    • no clear front runner
    • no clear path for scaling up
  • near-term(-10 years)
    • noise, noise, noise, ...
    • small width and depth
  • software stack should target actual hardware operations
    • textbook gate set based on theory of the '90s, real hardware differs

Quantum Benchmark - Dar Dahlen

• Why make a Qutnum Proc:
  • Use quantum for it's computational capabilities
    • chemistry, materials science, machine learning, optimization, ...
• The error problem for Quantum
  • digital computing is discrete & predictable:
  • Quantum computing is continuous & error-prone:
• Error Types
  • rotation erros
  • non-rotation errors
    • spontaneous bit flip (T1 Decay most common, 1 to 0)
    • dechoherence (Qubit vector drifting in time, T2)
• Quntifying erros
  • randomized benchmarking
  • gate set tomography
  • interleaved randomized benchmarking
  • cycle benchmarking

Rondomized Compiling (RC)

RC helps to mitigates errors, it is not error correction

Workshop on Datacenter Resource Disaggregation

Keynote: Sharad Singhal, HP -- The Machine―Memory-Driven Computing

• GEN-Z https://genzconsortium.org/
• HPE Superdome Flex
• OpenFAM: programming model for fabric-attached memory
  • cf. https://www.researchgate.net/publication/331836860_The_OpenFAM_API_A_Programming_Model_for_Disaggregated_Persistent_Memory
  • cf. https://github.com/FabricAttachedMemory

• MDC Programming Opportunities
  • data sharing in one large globally-addressable memory
    • pass by reference
    • multi-process - share large pool of data in shared memory
    • use global shared memory for messaging
  • focus on in-memory data formats
    • file system vs database vs direct use of byte-addressable persistent memory
    • opportunity to move away from having multiple data formats in memory and storage; single data format used as both in-memory representation and data storage
    • reduce number of software layers -- simpler to develop and maintain software
• MDC Programming Challenges
  • practicalities of using new tech.
    • accessing memory
    • allocating and managing memory
  • data consistency in face of failures

Practical Aspects of Building a Software Stack for Noisy Intermediate-Scale Quantum Devices

Introduction to IBMQ and Qiskit

https://qiskit.org/

• Qiskit Backend Specifications for OpenQASM and OpenPulse Experiments cf. https://arxiv.org/abs/1809.03452
• Open Quantum Assembly Language cf. https://arxiv.org/abs/1707.03429

Quantum Synthesis aka Quantum Compilation

• Heuristics for Quantum Compiling with a Continuous Gate Set. Davis, Dahlen,Sen,Iancu. On arxiv soon.
  • use A* inspired search to expand solution

Berkeley Hardware

Custom HW + Pulses

• Pulse Description Libraries: OpenPulse
• Not much standardization, not clear we reached maturity
• pulse level compilation nascent research area