Intro and Motivation:Closing the Gap Between Quantum Algorithms

and Hardware through Software-Enabled Vertical Integration and Co-Design

Fred ChongSeymour Goodman ProfessorDepartment of Computer ScienceUniversity of Chicago

Lead PI, the EPiQC Project, an NSF Expedition in Computing

With Margaret Martonosi, Ken Brown, Peter Shor, Eddie Farhi, Aram Harrow, Diana Franklin, David Schuster, John Reppy, and Danielle Harlow (UChicago, MIT, Princeton, Duke, UCSB)

Tutorial Software Installationn The USB drives with Docker tools and images are provided for you to

complete the tutorial software installation.n The instructions are in the USB drives.

n After you run the tutorial_starter script, your web browser will open the jupyter notebook automatically.

n If the jupyter notebook doesn’t run automatically, please copy/paste the URL, displayed in your terminal, to your web browser manually.

n Skip the step 4, “Edit IBM APIToken”, in the instruction since we don’t need to use IBM token for this tutorial.

n Disk space requirement: 12GB.

Raise your hand if you need any support.

Why Quantum Computing?n Fundamentally change what is computable

q The only means to potentially scale computation exponentially with the number of devices

n Solve currently intractable problems in chemistry, simulation, and optimizationq Could lead to new nanoscale materials, better photovoltaics, better nitrogen

fixation, and more

n A new industry and scaling curve to accelerate key applicationsq Not a full replacement for Moore’s Law, but perhaps helps in key domains

n Lead to more insights in classical computingq Previous insights in chemistry, physics and cryptographyq Challenge classical algorithms to compete w/ quantum algorithms

3

NISQNow is a privileged time in the history of science and technology, as we are witnessing the opening of the NISQ era (where NISQ = noisy intermediate-scale quantum). – John Preskill, Caltech

413:29

IBM 50 superconductor qubits

Google72 supercond qubits

Innsbruck 20 atomic ion qubits

The Algorithms to Machines Gap

5

1

10

100

1000

10000

100000

1000000

1995 2000 2005 2010 2015 2020 2025

Grovers Algorithm (Database search)

Shor’s Factoring Alg. (Crypto)

Gap!

Year

#Qubits

#Qubits Needed

#Qubits Buildable

The Algorithms to Machines Gap

6

1

10

100

1000

10000

100000

1000000

1995 2000 2005 2010 2015 2020 2025

Grovers Algorithm (Database search)

Shor’s Factoring Alg. (Crypto)

Gap!

Year

#Qubits

#Qubits Needed

#Qubits Buildable

Quantum Sim, Q Chem, QAOA

The Algorithms to Machines Gap

7

1

10

100

1000

10000

100000

1000000

1995 2000 2005 2010 2015 2020 2025

Grovers Algorithm (Database search)

Shor’s Factoring Alg. (Crypto)

Gap!

Year

#Qubits

#Qubits Needed

#Qubits Buildable

Quantum Sim, Q Chem, QAOA

Co-Design

Closing the Gap: Software-Enabled Vertical Integration and Co-Design

1

10

100

1000

10000

100000

1000000

1995 2000 2005 2010 2015 2020 2025

Grovers Algorithm (Database search)

Shor’s Factoring Alg. (Crypto)

Gap!

Year

Quantum Sim, Q Chem, QAOA

Modeling

Architecture

Compiler

Prog Lang

Algorithms

Devices

Co-Design

Goal

1

10

100

1000

10000

100000

1000000

1995 2000 2005 2010 2015 2020 2025

Grovers Algorithm (Database search)

Shor’s Factoring Alg. (Crypto)

Gap!

Year

Quantum Sim, Q Chem, QAOA

Result: Crossover by 2023!

Modeling

Architecture

Compiler

Prog Lang

Algorithms

Devices

Develop co-designed algorithms, SW, and HW to close the gap between algorithms and devices by 100-1000X, accelerating QC by 10-20 years.

Co-D

esign

Space-Time Product Limits

13:29 10

Qubits

Gates

1024x1

1x1024

Gate Error ~ 10-3

32x32

Space-Time Product Limits

13:29 11

Qubits

Gates

Gate Error ~ 10-5

128x1024

“Good” Quantum Applications

n Compact problem representationq Functions, small molecules, small graphs

n High complexity computationn Compact solution n Easily-verifiable solutionn Co-processing with classical supercomputersn Can exploit a small number of quantum

kernels

12

Quantum Compiler Optimizations

n Similar to circuit synthesis for classical ASICsn Program inputs often known at compile timen Manage errors and precisionn Scarce resources

q Every qubit and gate is important

13:29 13

Tool Flow

13:29 14

LLVM Infrastructure

Mod

ified

Cla

ng P

arse

r

Con

vers

ion

to R

ever

sibl

e C

ircui

t

Cla

ssic

al C

ontr

ol R

esol

utio

n

LLVM

Inte

rmed

iate

Rep

rese

ntat

ion

Res

ourc

e E

stim

atio

n

Mod

ule

Inlin

inin

g an

d Fl

atte

ning

No-

Clo

ning

V

erifi

catio

n

Qub

it R

edun

danc

y

Compilation Program Checks

Faul

t-To

lera

nt

Gat

e C

onve

rsio

n

Scaf

fold

Qua

ntum

Pro

gram

Dec

ompo

sing

to

Sta

ndar

d G

ates

Zero

Sta

tes

for E

CC

Fault-Tolerant Redundancies

Mag

ic S

tate

s fo

r T G

ates

EP

R s

tate

s fo

r Te

lepo

rtat

ion

Architectural Simulator

Logi

cal

QA

SM

Gen

erat

ion

LPFS

Sch

edul

er

Logical Backend

Logi

cal S

ched

ule

Phys

ical

Sch

edul

e

Phy

sica

l QA

SM

Gen

erat

ion

Physical Backend

Phy

sica

l Sch

edul

er

Com

mun

icat

ion

Opt

imiz

er

https://github.com/epiqc/ScaffCC

Scaffold tools, 41K lines of code, open sourceepiqc.cs.uchicago.edu

Increasing Parallelism

13:29 15

n Compiler Optimizations:q Loop unrolling, constant propagation, inlining,

function cloning, DAG scheduling[Heckey+ ASPLOS 2015]

Microarchitecture

16

[Fu+ Micro 2017 Best Paper]

Breaking ISA Abstraction

n Multi-Qubit Operators for QAOA q Direct translation from compiler to control pulses

17

[Joint work with David Schuster]

Static vs Dynamic: Mapping Datan Static spectral and graph

partitionersn Map for clustering

q Probably necessary to get to 1000 qubits

n Map for irregular physical constraintsq Qubit couplings, hardware defects

n Granularity of mappingsn Interaction with qubit reuse

18

Spectral communities for 2-level Bravyi-Haah magic-state factory

How do I know if my QC program is correct?

n Check implementation against a formal specification

n Check general quantum propertiesq No-cloning, entanglement,

uncomputationn Checks based on programmer

assertions (quantum simulation)n Heuristic bug-finding systems

[Altadmri SIGCSE15]n Can we check useful properties in

polynomial time for programs with quantum supremacy?

What are the right abstractions?

n Specification Languages q Coq, Hamiltonians

n Programming Languages q Scaffold, Quipper, Q#, Quil …

n Instruction-Set Architectures q OpenQASM

n Physical Control q OpenPulse

20

Modeling

Architecture

Compiler

Prog Lang

Algorithms

Devices

Specialization vs Abstraction

13:29 21

Short-term SW Long-term SW

100 1000 10000 100000

qubits

Gap?