Date post: | 18-May-2015 |
Category: |
Technology |
Upload: | insidehpc |
View: | 1,364 times |
Download: | 2 times |
Quantum Computing:A New Resource for HPC
Colin P. WilliamsD-Wave Systems Inc.
© 2013 D-Wave Systems Inc. All Rights Reserved 2
What I’ll Talk About
• D-Wave Systems Inc.• What are quantum computers?• D-Wave’s approach & rationale for it• Our current product: The D-Wave TwoTM Quantum Computer
• Integrating QC into HPC
© 2013 D-Wave Systems Inc. All Rights Reserved 3
D-Wave Systems Inc.
• Privately owned based in Vancouver, Palo Alto & Washington D.C.– Investors: DFJ, Goldman Sachs, In-Q-Tel, Jeff Bezos, et al
• Offering quantum computing systems– Built from superconducting processors
• ~100 employees (27Ph.D., 18 B.Eng. 11 M.Sc.)
• 100+ patents / 60+ peer reviewed papers
• Unique infrastructure (design, fab, test, systems, software)
© 2013 D-Wave Systems Inc. All Rights Reserved 4
Ranked 4th in Patent Power for Computer Systems
© 2013 D-Wave Systems Inc. All Rights Reserved 5
Working with Leaders
Defense Intelligence National Labs
Universities
Web
Finance Energy
© 2013 D-Wave Systems Inc. All Rights Reserved 6
What are QuantumComputers?
© 2013 D-Wave Systems Inc. All Rights Reserved 7
What are Quantum Computers?
• Computers that harness quantum physical phenomena not available to conventional computers, e.g.,– Superposition– Entanglement– Co-tunneling– Non-determinism
© 2013 D-Wave Systems Inc. All Rights Reserved 8
Why are they Interesting?
• Pragmatically– Allows us to solve problems in new ways to beat the best
we can do classically in many cases
– Exponential speedups• Factoring• Simulating quantum systems• Quantum chemistry
– Polynomial speedups• Unstructured search• Structured search (NP-complete & NP-hard problems)
• Philosophically– As Prof. David Deutsch of Oxford University says …
© 2013 D-Wave Systems Inc. All Rights Reserved 9© 2013 D-Wave Systems Inc. All Rights Reserved PAGE
“Quantum computation … will be the first technology that allows useful tasks to be
performed in collaboration between parallel universes”
Prof. David Deutsch – The Fabric of Reality
© 2013 D-Wave Systems Inc. All Rights Reserved 10
Significance to HPCof harnessing quantum mechanics
Metric High Performance Computer Quantum Computer
Concurrency 108 cores & 1010 threads One 512-qubit core has ~10154 “virtual threads” in superposition, but quantum mechanics limits our ability to read them
Robustness Reduced operating voltages & channel widths, will make devices less reliable. Need new programming style that is intrinsically probabilistic and tolerant to errors
Naturally probabilistic programming.Quantum annealing degrades gracefully to errors
Power Expect 25-100MW systems. Few locations can support this demand. Fewer data centers can afford it. Power demand dominated by data movement
15kW for cooling & ~0kW for computation. Cooling power will stay constant up to thousands of qubits! Almost no energy to compute. No data movement needed
Storage Needs to be 100PB capacity but will be constrained by physical & economic limits (density, power, cost)
Memory exploits parallel universes. Create & process superposition of all 2N configurations at once. N > 300 qubits provide more storage than there are particles in the known Universe
Speed 1018 FLOPS Potential to be fast but runs at 0 FLOPS
© 2013 D-Wave Systems Inc. All Rights Reserved 11
Will QCs Make HPCs Obsolete?
• Probably not . . .
• They’re suited to different tasks
– HPCs: Computational fluid dynamics, molecular simulation, weather forecasting, nuclear weapons modeling, etc.
– QCs: discrete combinatorial optimization, artificial intelligence, machine learning, sampling
• But together they can enhance each other
© 2013 D-Wave Systems Inc. All Rights Reserved 12
D-Wave’s ApproachQuantum Annealing
inspired by Adiabatic Quantum Computing
© 2013 D-Wave Systems Inc. All Rights Reserved 13
How Quantum Annealing Works
• Space of solutions defines and energy landscape & best solution is lowest valley
• Classical algorithms can only walk over this landscape
• Quantum annealing can tunnel through the landscape
© 2013 D-Wave Systems Inc. All Rights Reserved 13
© 2013 D-Wave Systems Inc. All Rights Reserved 14
Our Product
© 2013 D-Wave Systems Inc. All Rights Reserved 15
D-Wave TwoTM Quantum Computer
© 2013 D-Wave Systems Inc. All Rights Reserved 15
© 2013 D-Wave Systems Inc. All Rights Reserved 16
Cooling
• Closed cycle dilution refrigerator (“fridge”)
• Fridge + servers consume 15.5kW
• Power demand will remain constant as we scale up to thousands of qubits
© 2013 D-Wave Systems Inc. All Rights Reserved 17
• 192 lines from room temperature to chip
• 10kg cooled to 0.02K• 150x colder than
interstellar space• Shielded room excludes
external RF• Magnetic field < 10−9 Tesla
across chip• 50,000x weaker than
Earth’s field• Isolated from vibrations
Processor Environment
© 2013 D-Wave Systems Inc. All Rights Reserved 18
Product Roadmap
D-Wave Two
D-Wave One
2,048-qubits Q4 2015
… and still g
oing!
Talk of “the” D-Wave chip is a misnomer – the architecture evolves 6-8 times per year
© 2013 D-Wave Systems Inc. All Rights Reserved 19
512-Qubit Vesuvius Processor
© 2013 D-Wave Systems Inc. All Rights Reserved 19
© 2013 D-Wave Systems Inc. All Rights Reserved 20
Processor Architecture
© 2013 D-Wave Systems Inc. All Rights Reserved 20
Wiring layout• 8×8 array of 8-qubit unit cells
• Within each unit cell each vertical qubit is coupled to each horizontal qubit
• Vertical (horizontal) qubit coupled to corresponding qubit in vertical (horizontal) neighboring cells
• Non-planarity of interconnect network makes the problem of finding the lowest energy state of the qubits NP-hard
• NP-Hardness guarantees you can map many practical problems to the architecture
Physical Layout
Logical Layout
Physical Unit Cell
Logical Unit Cell
q2
q1
q3q4
q6
q5
q7q8
Topology of interconnect network
© 2013 D-Wave Systems Inc. All Rights Reserved 21
Programming Languages
• User does not need to know anything about quantum physics– Just pass a matrix of hi’s and Jij’s to the machine
• Currently have interfaces for:– Python
– Matlab
– C/C++
© 2013 D-Wave Systems Inc. All Rights Reserved 22
Integrating QC into HPC
© 2013 D-Wave Systems Inc. All Rights Reserved 23
A Strategy for Integrating QC with HPC• HPCs excel at large scale numerical simulations
• QCs excel at discrete combinatorial optimization
• Can we use a HPC + QC for engineering design optimization?
• Problem setup …– Suppose engineering design is specified by a bit string
– Various designs can be “scored” by running some HPC simulation
– Goal is to find the bit string whose score meets design criteria
© 2013 D-Wave Systems Inc. All Rights Reserved 24
Division of Labor
• HPC’s job – Map bit string to design; simulate design; score the result; send score to QC
• QC’s job– use sequence of bit string/score pairs to tweak the h’s and J’s in the QMI so
that it will yield samples that correctly mimic the ordering of energies of solutions in the neighborhood of the highest scoring bit string
– Yields a new QMI
– Run new QMI many times to yield new candidates (~104 solutions/sec)
– HPC scores the candidate solutions and returns scores to QC
– Iterate until design meets desired criteria
© 2013 D-Wave Systems Inc. All Rights Reserved 25
• Using the QC + HPC together allows faster convergence on optimal design than is attainable by using HPC alone
• Avoids unnecessary HPC cycles & power consumption– Increases availability of HPC for running other computations– Works best when cost of running the HPC simulator is high
• Supercomputer + quantum computer
Quantum-Accelerated HPC
Better “guesses”
“Scores” for the guesses
Quantum-Accelerated HPC
© 2013 D-Wave Systems Inc. All Rights Reserved 26
Why Does it Work?
• Imagine you’ve reached an intermediate point in design space and want to pick the next bit string to try
• Classical methods only sense the local neighborhood• Quantum methods have potential for greater horizon• Make a better next move possibly leading in different direction
Classical
Dis
crep
ancy
Design Parameter
Quantum
Dis
crep
ancy
Design Parameter
© 2013 D-Wave Systems Inc. All Rights Reserved 27
ExampleRadiotherapy Optimization
© 2013 D-Wave Systems Inc. All Rights Reserved 28
Another Example: Radiotherapy OptimizationPROBLEM:
Deliver lethal dose to tumor whilst minimizing damage to healthy tissues
APPROACH:
Hybrid: QC + Conventional Computer• Design = bit string• Quality = result of running extensive
radiation transport simulation• D-Wave system learns from
simulations to predict better designs
IMPACT:• Hybrid quantum-classical design found a radiation
therapy treatment that minimized the objective function to 70.7 c.f. 120.0 for tabu, and ran in 1/3 the time making fewer calls to radiation transport simulation
© 2013 D-Wave Systems Inc. All Rights Reserved 29
Conclusions
• We see quantum computing as a new resource for HPC
• Technology scaling faster than Moore’s Law
• 1,024-qubit quantum computer by mid-2014
• Performance is encouraging
• Many potential uses in combinatorial optimization, engineering design optimization, A.I., machine learning & sampling
• Seeking early adopters to explore QC-HPC synergy
Email: [email protected]