Koomey on why ultra-low power computing will change everything

Copyright Jonathan Koomey 2012

SOFTWARE & SYSTEMS DESIGN

Why ultra-‐low power compu1ng will change everything

Jonathan Koomey Research Fellow, Steyer-‐Taylor Center for Energy

Policy and Finance, Stanford University [email protected] hSp://www.koomey.com

ARM Tech Con, San Jose, CA October 31, 2012


RevoluXon

2

Copyright Jonathan Koomey 2012 3

Cheap Smart Small

Connected Low power + Self powered


CONSIDER THE POSSIBILITIES

hSp://proteusdigitalhealth.com/technology/


Research quesXon: How has the energy efficiency of compuXng changed over Xme?

5


MOORE’S LAW

•  Not a “law” but an empirical observaXon about components/chip – 1965: doubling every year – 1975: doubling every 2 years

•  Characterizes economics of chip producXon, not physical limits

•  Ofen imprecisely cited, interpretaXons changed over Xme (Mollick 2006)


MOORE’S ORIGINAL GRAPH

7


TRANSISTORS/CHIP (000S)

The doubling Xme from 1971 to 2006 is about 1.8 years. Data source: James Larus, Microsof CorporaXon.

8


How to measure the energy efficiency of computaXon?

9


METHOD

•  ComputaXons per kWh =


DATA

•  Performance from Nordhaus (2007) or normalized to that source using benchmarks for more recent computers

•  Used measured power data, either published (e.g. Weik 1955, 1961, 1964) or from archival or recent computers – with computer fully uXlized – with screen power subtracted for portables


Doubling Xme for performance per computer = 1.5 years in the PC era

12


PERFORMANCE TRENDS (2): COMPUTATIONS/S/COMPUTER

Source: Nordhaus (2007) with addiXonal data added by Koomey (2009b)


REAL POWER MEASUREMENTS OF REAL COMPUTERS


COMPUTING EFFICIENCY

• Doubling about every year and a half since the 1940s

• 100x improvement every decade

• Enabled the existence of laptops and smart phones


GOOD CORRELATION, CLEAR RESULTS

•  R2 for computaXons/kWh – 0.983 for all computers, 1946-‐2009

– 0.970 for PCs, 1975-‐2009 •  Doubling Xme for computaXons/kWh

– All computers: 1.6 years – PCs: 1.5 years – Vacuum tubes: 1.35 years

•  Big jump from tubes to transistors


IMPLICATIONS

•  AcXons taken to improve performance also improve computaXons per kWh –  Transistors: Smaller, shorter distance source to drain, fewer electrons

–  Tubes: Smaller, less capacitance, lower currents

•  Trends make mobile and distributed compuXng ever more feasible (baSery life up 100x per decade at constant compuXng power)


LAPTOPS GROWING FAST (WORLD INSTALLED BASE, MILLIONS)

Sources—1985: Arstechnica + Koomey calcs 1996-‐2008: IDC


IT’S NOT JUST ABOUT COMPUTING EFFICIENCY

•  Low power > high efficiency •  RevoluXon is being driven by the confluence of trends allowing low-‐power – compuXng – communicaXons – sensors – controls

•  Energy harvesXng and storage also criXcal •  Idle modes more important than acXve


REDUCE AREA UNDER THE CURVE


A MICROCONTROLLER “RACE”

Source: Eduardo Montanez, Freescale Semiconductor (microcontroller is ARM ® Cortex™-‐m0+ core)


EFFICIENCY OF SIMPLE CELL PHONES OVER TIME


AN EXAMPLE OF MOBILE COMPUTING + COMMUNICATIONS ENABLED BY EFFICIENCY

http://www.bigbellysolar.com

• Compacts trash 5 x • Sends text message when full • PV panel generates power from sunlight • An economic and environmental home run


HOW ABOUT REALLY LOW POWER SENSORS?

•  Consider the wireless no-‐baSery sensors created by Joshua R. Smith of the University of Washington hSp://www.nyXmes.com/2010/07/18/business/18novel.html, hSp://www.economist.com/node/16295708 –  sensors use 60 microwaSs on

average (60x10-‐6 waSs)

–  scavenge power from radio and TV signals

•  Other possible power sources for similar devices: light, heat, moXon, blood sugar, digesXve fluids

Images courtesy of Josh Smith, U of WA


UNIVERSITY OF MICHIGAN MICRO-‐MOTE

Slide courtesy of David Blaauw and Dennis Sylvester, U of MI

P=11nW sleep, 40 μW acXve, ARM® M0 core


TUMOR PRESSURE MONITORING

Slide courtesy of David Blaauw and Dennis Sylvester, U of MI


STREETLINE NETWORKS: SMART PARKING, SMART CITIES

Slide courtesy of Mark Noworolski, Streetline Networks

Motes use <400μW on average. For LA, With 40,000 parking spots, that implies total mote power of about 15W. Mote technology is from Dust Networks


STREETLINE NETWORKS: SMART PARKING, SMART CITIES



STREETLINE NETWORKS: VARIABLE PARKING SIGNS



DEEPER IMPLICATIONS

• Move bits, not atoms • Customized data collecXon (focus on nanodata, not big data) • Ever more precise control of processes • Real-‐Xme analysis • Enabling “the internet of things”

• Bo#om line: beSer matching of energy services demanded with those supplied, beSer real-‐Xme control, and beSer analysis.


THESE TRENDS STILL HAVE A LONG WAY TO RUN

Psssst: Researchers at Purdue and the University of New South Wales recently created a reliable one atom transistor…

2041


BIG UNANSWERED QUESTIONS

•  Could we do beSer than historical trends? •  Might we do worse? If so, why?

•  What’s next as we approach theoreXcal limits?


CONCLUSIONS

•  QuanXtaXve results –  In the PC era (1976-‐2009) performance per computer and computaXons per kWh doubled every 1.5 years

–  From ENIAC to the present, computaXons per kWh doubled every 1.6 years

•  Performance and efficiency improvements inextricably linked

•  SXll far from theoreXcal limits •  Big implicaXons for mobile technologies •  The future belongs to low power systems!


Viva la Revolución!

34


THANKS!

•  Rob Bernard, Microsof (funder)

•  Lorie Wigle, Intel (funder) •  Stephen Berard of Microsof (coauthor) •  Marla Sanchez of LBNL and CMU (coauthor), and

•  Henry Wong of Intel (coauthor) •  The Computer History Museum •  IEEE Annals of the History of Compu8ng


FURTHER READING

Koomey, Jonathan G., Stephen Berard, Marla Sanchez, and Henry Wong. 2011. "ImplicaXons of Historical Trends in the Electrical Efficiency of CompuXng." IEEE Annals of the History of Compu8ng. vol. 33, no. 3. July-‐September. pp. 46-‐54. [hSp://doi.ieeecomputersociety.org/10.1109/MAHC.2010.28]

Koomey, Jonathan. 2012. "The CompuXng Trend that Will Change Everything." In Technology Review. April 2. [hSp://www.technologyreview.com/news/427444/the-‐compuXng-‐trend-‐that-‐will-‐change-‐everything/]

Greene, Kate. 2011. "A New and Improved Moore's Law." In Technology Review. September 12. [hSp://www.technologyreview.com/compuXng/38548/?p1=A1]

Eisenberg, Anne. 2010. "Bye-‐Bye BaSeries: Radio Waves as a Low-‐Power Source." The New York Times. New York, NY. July 18. p. BU3. [hSp://www.nyXmes.com/2010/07/18/business/18novel.html]


REFERENCES

•  Feynman, Richard P. 2001. The Pleasure of Finding Things Out: The Best Short Works of Richard P. Feynman. London, UK: Penguin Books.

•  Hilbert, MarXn, and Priscila López. 2011. "The World's Technological Capacity to Store, Communicate, and Compute InformaXon." Science. vol. 332, no. 6025. April 1. pp. 60-‐65.

•  Koomey, Jonathan. 2008. "Worldwide electricity used in data centers." Environmental Research Le#ers. vol. 3, no. 034008. September 23. <h#p://stacks.iop.org/1748-‐9326/3/034008>.

•  Koomey, Jonathan G., ChrisXan Belady, Michael PaSerson, Anthony Santos, and Klaus-‐Dieter Lange. 2009a. Assessing trends over 8me in performance, costs, and energy use for servers. Oakland, CA: AnalyXcs Press. August 17. <hSp://www.intel.com/pressroom/kits/ecotech>.

•  Koomey, Jonathan G., Stephen Berard, Marla Sanchez, and Henry Wong. 2011. "ImplicaXons of Historical Trends in The Electrical Efficiency of CompuXng." IEEE Annals of the History of Compu8ng. vol. 33, no. 3. July-‐September. pp. 2-‐10. <hSps://files.me.com/jgkoomey/u0zi7l>

•  Koomey, Jonathan. 2011. Growth in data center electricity use 2005 to 2010. Oakland, CA: AnalyXcs Press. August 1. <hSp://www.analyXcspress.com/datacenters.html>

•  Mollick, Ethan. 2006. "Establishing Moore’s Law." IEEE Annals of the History of Compu8ng (Published by the IEEE Computer Society). July-‐September. pp. 62-‐75.


REFERENCES (2)

•  Yoonmyung, Lee, Kim Gyouho, Bang Suyoung, Kim Yejoong, Lee Inhee, P. DuSa, D. Sylvester, and D. Blaauw. 2012. A modular 1mm3 die-‐stacked sensing pla]orm with op8cal communica8on and mul8-‐modal energy harves8ng. Proceedings of the Solid-‐State Circuits Conference Digest of Technical Papers (ISSCC), 2012 IEEE InternaXonal. 19-‐23 Feb. 2012. [hSp://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=6171933]

•  Moore, Gordon E. 1965. "Cramming more components onto integrated circuits." In Electronics. April 19.

•  Moore, Gordon E. 1975. "Progress in Digital Integrated Electronics." IEEE, IEDM Tech Digest. pp. 11-‐13. <hSp://www.ieee.org/>

•  Nordhaus, William D. 2007. "Two Centuries of ProducXvity Growth in CompuXng." The Journal of Economic History. vol. 67, no. 1. March. pp. 128-‐159. <hSp://nordhaus.econ.yale.edu/recent_stuff.html>

•  Weik, MarXn H. 1955. A Survey of Domes8c Electronic Digital Compu8ng Systems. Aberdeen Proving Ground, Maryland: BallisXc Research Laboratories. Report No. 971. December. <hSp://ed-‐thelen.org/comp-‐hist/BRL.html>

•  Weik, MarXn H. 1961. A Third Survey of Domes8c Electronic Digital Compu8ng Systems. Aberdeen Proving Ground, Maryland: BallisXc Research Laboratories. Report No. 1115. March. <hSp://ed-‐thelen.org/comp-‐hist/BRL61.html>

•  Weik, MarXn H. 1964. A Fourth Survey of Domes8c Electronic Digital Compu8ng Systems (Supplement to the Third Survey). Aberdeen Proving Ground, Maryland: BallisXc Research Laboratories. Report No. 1227. January. <hSp://ed-‐thelen.org/comp-‐hist/BRL64.html>

Koomey on why ultra-low power computing will change everything

Technology

Transcript of Koomey on why ultra-low power computing will change everything