ARM TechCon Software and Systems Keynote: Why Ultra-Low Power Computing Will Change Everything Simon Segars, speaking of the importance of continuing low power initiatives, introduced Dr. Jonathan Koomey, Consulting Professor at Stanford. (First impression, our kind of guy: He wears engineer shoes, not sales shoes!)
Koomey spoke for “Revolutionary change, and that’s not just loose talk….” in energy efficiency for the future. He cited Proteus Digital Health in biomedical; testing a new device embedded in a pill with no battery! It uses digestive juices to power the device! You take the pill, it communicates to report compliance…(Orwellian?) Just one example of how low power devices will transform medicine: delivering more info about the individual. Imagine sensors which can report pH, temperatures, etc.
Koomey summarized results of a 2011 study on computing energy efficiency he was part of. He stated:
- Energy efficiency of computation has doubled every year and a half (actually 1.6 years) from 1946 (ENIAC) to the present, which is a 100x improvement every decade…..….think Moores Law!
- Definition of energy efficiency: Computations/kWh, based on full load, using measured power data.
My personal note: A real weakness in the usefulness of this study is that it does not reflect idle modes, which are of critical importance in really understanding the issues.
Toomey delivered several truisms this audience well understands, such as: The things we did to improve performance also benefitted power efficiency. Laptops are taking over desktop market share. Revolution is driven by confluence of trends: computing, communications, sensors, controls. Importance of idle modes.
He presented interesting data on cell phone efficiency (minutes of talk time/Wh) vs year of introduction; on this graph, the efficiency increase he is seeing is 8% per year.
Are we burning up our planet? Well, we are using more devices worldwwide, but they are more efficient. Cloud is more efficient than data centers. Data centers in the US use 2% of our power, use around 1.5% in the world. Overall electronics is maybe 5% of our worldwide energy usage.
Power management trends: from process technology, to multicore, to software optimization matters, to “approximate computing”…its not just a silicon problem, it’s a system challenge.
The most fascinating topic: Energy harvesting, from ambient sources….and the future....
• Intelligent trash cans, self-solar-powered, compact trash, send a text when full, for fewer truck trips! Economic and environmental home run!
• Wireless no-battery sensors from Univ of Washington scavenge power from radio and tv signals in metro areas….
• MicroMote, generic sensing platform with ARM core, at University of Michigan. Ultra low power: 40uWatts in active mode, an astounding 11nW in sleep mode. Devices like this could someday monitor growth of a tumour.
• Streetline Networks, promoting sensors in parking spaces, and a mesh grid, in LA. Motes use only 400uW, with Mote Technology from Dust Networks. Think about variable messages signs or subscriptions to find parking spaces.
Deeper implications Toomey summarized included:
• Move bits not atoms (things)
• Customized data collection (nanodata not big data)
• Ever more precise control of processes
• Real time analysis
• Enabling “the internet of things”
How far can these energy efficiency trends continue?
- One physicist calculated a theoretical limit…this graph takes us to 2041, three decades!!!
- That is if we are clever...our horizon includes exciting applications…
- Vision of a world with low power, cheap devices distributed everywhere..for medical, for transportation, for services, for manufacturing…..
And as Toomey commented, "Our community has a high “do to talk” ratio….."
(Yes, the shoes don't lie.....)
Shhh.....Researchers at Purdue and the University of New South Wales recently invented a 1 electron computer…