20130912_mit _geordie_rose

The D-Wave Two

Dr. Geordie RoseFounder and CTO, D-WaveHigh Noon September 12th 2013 @ MIT

Image from http://www.nas.nasa.gov/quantum/quantumcomp.html

What I’m going to talk about

1. What we are doing (& why) 2. The machines 3. What the machines do

2 © Copyright 2013 D-Wave Systems Inc.

Quantum computation … will be the first technology that allows useful tasks to be

performed in collaboration between parallel universes. David Deutsch @ TED 2005

Image from http://conferences.ted.com/TEDGlobal2010/program/3 © Copyright 2013 D-Wave Systems Inc.

Mungo Park … sailed from Portsmouth on the 30th of January, 1805, resolved to trace the Niger (River) to its source or perish in the attempt. He perished. Introduction to Travels in the Interior Districts of Africa 1795-7Mungo Park (1799)http://ebooks.adelaide.edu.au/p/park/mungo/travels/complete.html


… quantum computers … can solve problems whose solution will never be feasible on a conventional computer.

Quantum computing for everyoneMichael Nielsen (2008)http://michaelnielsen.org/blog/quantum-computing-for-everyone/

Image from http://www.longexposure.ca/2011/03/tedx-waterloo-the-uncharted/5 © Copyright 2013 D-Wave Systems Inc.

© Copyright 2011 D-Wave Systems Inc.

Someday, perhaps soon, we will build a machine that will be able to perform the functions of a human mind, a thinking machine. The Connection MachineDanny Hillis (1985)

6 © Copyright 2012 D-Wave Systems Inc. Image from http://compass-summit.com/danny-hillis/

II. The machines



Footprint

• ~ 200 square feet

• Closed cycle fridge

• Consumes ~ 15 kW


Processor environment

• 168 lines from room temperature to processor

• 10 kg of metal at 20 milliKelvin

• 1 nanoTesla in 3D across processor; 50,000x less than earth’s magnetic field

2.725 K 10 © Copyright 2013 D-Wave Systems Inc.


What is a quantum computer

•dfd


Fabrication cross-section

© Copyright 2011 D-Wave Systems Inc.

BASE

WIRA

WIRB

WIRC

Resistor

VIA

SiO2

SiO2WIRD

WIRE

Junction


One of very few processes in the world capable of fabricating VLSI superconducting processors

QubitsCompound compound Josephson junction RF-SQUID niobium flux qubits

Control the amount of superposition – ‘sigma X’

Control the bias towards one or the other state – ‘sigma Z’

Processor architecture overview available at http://www.youtube.com/watch?v=PqSgmCg1kew

Qubits are stretched into long thin loops and coupled together


Unit cells tile the plane


=

A physical ‘computing fabric’


The evolution of an idea

Processor evolution overview available at http://www.youtube.com/watch?v=6VIAL8gQRTI18 © Copyright 2013 D-Wave Systems Inc.

442 qubits

509 qubits

III. What the machines do



Program real numbers on a graph, draw samples



sk [-1, +1]



sk [-1, +1]

�⃗�=[+𝟏 ,−𝟏 ,+𝟏 ,+𝟏 ,−𝟏 ,+𝟏 ,−𝟏 ,−𝟏 ]


Program real numbers on a graph

h1

J15



N

jji

EjiijjjN ssJshss

1 ,1 ,,

h1

J15

sk [-1, +1]



N

jji

EjiijjjN ssJshss

1 ,1 ,,

h1

J15

sk [-1, +1]

Ising modelAKA

QUBOAKA

Weighted Max2SAT


A simple example

3

1321 ,,

jji

jiijjj ssJshsss

h Jh1= +1.0 J12= - 1.0

h2= - 1.0 J13= - 0.5

h3= 0 J23= +0.3

h1

h2h3

J12

J13

J23

s1

s2 s3


s1 s2 s3

-1 -1 -1 -1.2

-1 -1 +1 -0.8

-1 +1 -1 -1.8

-1 +1 +1 -0.2

+1 -1 -1 3.8

+1 -1 +1 2.2

+1 +1 -1 -0.8

+1 +1 +1 -1.2

Notation: +1 = , -1 =

3

1321 ,,

jji

jiijjj ssJshsss



N

jji

EjiijjjN ssJshss

1 ,1 ,,

h1

J15

sk [-1, +1]

TsZ

sP /exp1 ⃗⃗


TsZ

sP /exp1 ⃗⃗


Programmable processor devicesTwo types: qubits and couplers

s1 s5

h1 h5

J15


Machine language programming

N

jji

EjiijjjN ssJshss

1 ,1 ,,

Optimization HardwareVariables sj = ±1 Qubits j

z

Linear terms hj Qubit biases hj

Quadratic terms Jij Coupler biases Jij

Allowed edge set E Locations of couplers between qubits

Objective function values (s1,…,sN) Allowed energy eigenvalues E

1 ,

Nz z z

P j j ij i jj i j E

H h J

Quantum annealingStart with all qubits in superposition


s3s2 s4s1

s7s6 s8s5

All 28 states equally likely


Quantum annealingTurn down superposition, turn up {h, J}


Vesuvius SR10-V6500 qubits, 1,472 couplers


Duty cycle for Vesuvius

Step 1. Program {h, J} into hardware• About 50msStep 2. Quantum anneal• About 20 microsecondsStep 3. Measure qubits (draw a sample)• About 100 microsecondsStep 4. Repeat 2-3 to gather N samples

~ 50 ms for 1 sample, ~ 1.2 s for 10,000 samples


A quantum Boltzmann-like machineProgram real numbers on a graph, draw samples

N

jji

EjiijjjN ssJshss

1 ,1 ,,

h1

J15

sk [-1, +1]

TsZ

sP /exp1 ⃗⃗


There’s a fascinating hypothesis that a lot of human perception … can be explained by a single learning algorithm. Unsupervised Feature Learning and Deep LearningAndrew Ng (2011)http://www.youtube.com/watch?v=I56UugZ_8DI

Image from http://coursera.com

[email protected]


Technology

20130912_mit _geordie_rose