Solving the Information Technology Energy Challenge Beyond ...cxro.lbl.gov/PDF/DOE_and_meeting_the_future_IT_energy_challenge.pdfx Petabit cm-2 Densities x Replaces DRAM & flash x

Solving the Information Technology Energy Challenge Beyond Moore’s Law

2x transistor/2yrs at same power & cost Started slowing ~2007

0

2000

4000

6000

8000

10000

2000 2005 2010 2015 2020 2025 2030

TWH

r

Year www.alliancetrustinvestments.com/sri-hub/posts/Energy-efficient-data-centres www.iea.org/publications/freepublications/publication/gigawatts2009.pdf

Projection based on consumer electronics + data centers

The problem: IT projected to challenge future electricity supply

~40% of 2014 world electricity

0

2000

4000

6000

8000

10000

2000 2005 2010 2015 2020 2025 2030

TWH

r

Year www.alliancetrustinvestments.com/sri-hub/posts/Energy-efficient-data-centres www.iea.org/publications/freepublications/publication/gigawatts2009.pdf

Projection based on consumer electronics + data centers

The problem: IT projected to challenge future electricity supply

Even worse if Moore’s Law ends!!!

Photonic

ICs

More efficient architecture in first

10 years

Need more efficient materials/devices for

next 10-20+ years

The path: near and long term

Revolutionary Heterogeneous HPC

architectures & software

Industry Academia

SRC

DARPA

NSF IARPA

Complex Landscape

DOE/National Labs • Support • Collaborate • Partnerships • Early Adoption

The Opportunity Fundamental

Science • Move revolutionary

low energy devices and architectures from Lab to Fab

• Accelerate

development and manufacturing base for semiconductor devices

• Create public-private partnerships with electronics industry

• Advanced computing

• Multiscale computing benchmarking and modeling

• Next-generation heterogeneous architectures

• New programming

models and application software

• Leverage unique capabilities in material science

• Scientific user facilities

• Computational approaches to materials design

• Grand challenges driven research

Computing Science

Manufacturing Science

Multiscale Multi-Lab Effort

MESAFab

APS

+++

+-

--

--

-

++

VTE

+

++

+

+

--

+

++

++

+

+

--

--

-

PtTaOxTa

On Chip Universal Memory:

Stacked ReRAM

Petabit cm-2

Densities

Replaces DRAM & flash

<1 pJ per write/read

Silicon

On Chip Memristor Accelerator:

Vector or matrix operations

fJs per operationOn Chip Photonics

Chip to chip communication

<1 pJ per bit transfer

High Performance Logic:

TFET, NcgFET

To next

node

x2

x2

x2

x2

w1,1

w2,1

w3,1

w4,1

w1,2

w2,2

w3,2

w4,2

w2,x

w2,x

w3,x

w4,x

...

...

...

...

Mu

ltis

cale

Ev

alu

atio

n

Fra

me

wo

rk

Advanced Materials

Advanced Devices and Manufacturing

Architecture & Algorithms

Multi-Lab Engagement

(F.B.) Rick McCormick Bruce Hendrickson

Dave Sandison Rob Leland

Ramamoorthy Ramesh Patrick Naulleau

John Shalf Horst Simon

Supratik Guha Rick Stevens

Lou Terminello Nathan Baker Malin Young

Trish Damkroger

Gary Grider John Sarrao Toni Taylor

Jeff Nichols

Building the Public-Private Partnership

0

2000

4000

6000

8000

10000

2000 2005 2010 2015 2020 2025 2030

TWH

r

Year

SUMMARY

100’s fewer power plants

• Looming end of Moore’s Law creating rapidly growing energy gap

• Coordinated public-private partnership will drive breakthroughs by leveraging multiple DOE Office leadership and Lab capabilities

These breakthroughs will enable continued affordability of our information economy, saving construction of 100’s of power plants