Embracing Randomness A Roadmap to Truly Disappearing Electronics

Transcription

1 Embracing Randomness A Roadmap to Truly Disappearing Electronics I&C Research Days Lausanne July 8, 04 Jan M. Rabaey and the PicoRadio Group Berkeley Wireless Research Center Department of EECS, University of California, Berkeley

2 Bell s Law: A New Computer Class Every 10 Years log (people per computer) Meaning in the Device Meaning in the Connection Meaning in the Collection 1940 s 2000 s Year Source: R. Newton

3 Ambient Intelligence (The Concept) An environment where technology is embedded, hidden in the background An environment that is sensitive, adaptive, and responsive to the presence of people and object An environment that augments activities through smart nonexplicit assistance An environment that preserves security, privacy and trustworthiness while utilizing information when needed and appropriate Fred Boekhorst, Philips, ISSCC02

4 Tackling Societal Scale Problems Disaster Mitigation Smart buildings Traffic management Infrastructure maintenance Energy management Medical

5 The Technology is Not Quite There Yet From 10 s of cm 3 and 10 s to 100 s of mw To 10 s of mm 3 and 10 s of µw

6 The Road to Truly Disappearing Electronics Mesoscale lowcost wireless transceivers for ubiquitous wireless data acquisition that are fully integrated Size smaller than 1 cm 3 are dirt cheap ( the Dutch treat ) At or below 1$ minimize power/energy dissipation Limiting power dissipation to 100 µw enables energy scavenging and form selfconfiguring, robust, adhoc networks containing 100 s to 1000 s of nodes Berkeley PicoRadio Project

7 Energy/Power as the Limiting Factor Power Source P/cm 3 (µw/cm 3 ) E/cm 3 (J/cm 3 ) P/cm 3 /yr (µw/cm 3 /Yr) Primary Battery Secondary Battery MicroFuel Cell Ultracapacitor Heat engine Radioactive( 63 Ni) Solar (outside) Solar (inside) 10 Temperature 40 Human Power 330 Air flow 380 Pressure Variation 17 Vibrations 200 Reasonable Target: 100 µw/cm 3 (/year) Courtesy Shad Roundy (ANU and UCB)

8 Practical Means of Energy Scavenging Piezoelectric bimorphs PVDF PZT 90 µw/cm 3 Photovoltaic µw/cm 2 Capacitive converter using MEMS microvibrator 30 µw/cm 3 (on microwave oven) [Shad Roundy (IML,UCB)]

9 Towards a sub100 µw W Integrated Node RF + Antenna Digital Processor(s) Baseband (mixedsignal) Sensors Clock Generation Power Supply Network Some Overall Guidelines Some Overall Guidelines Keep it simple! Minimize the supply voltage and the ambient currents as much as possible Aggressive use of new technologies (RFMEMS, integrated passives, ) Manufacturability is key

10 Towards a sub100 µw W Integrated Node Base Band Voltage Conv RF + Antenna Baseband (mixedsignal) Clock Generation 64K memory GPIO Serial Interface Interface DW8051 µc Locationing Engine Neighbor List DLL System Supervisor Network Queues Digital Processor(s) Sensors Power Supply Network Simplest possible processor Dedicated accelerators when needed Aggressive power management 1V supply, 16 MHz clock 300 mv standby voltage < 1 mw in full operation; < 1 µw standby

11 Towards a sub100 µw W Integrated Node RF + Antenna Baseband (mixedsignal) Clock Generation Energy Source 1 (solar) Energy Source 2 (vibration, ) Conversion Network 1 Reservoir 1 (capacitor) Reservoir 2 (microbattery) Conversion Network 2 Digital Processor(s) Sensors Energy generation and conversion network Power Supply Network Anchor Spring flexure Comb fingers Electrostatic MEMS vibration converters Microbattery

12 Towards a sub100 µw W Integrated Node RF + Antenna Baseband (mixedsignal) Clock Generation MEMS resonator die flips directly onto CMOS for a compact, integrated clock module. Digital Processor(s) Sensors Power Supply Network 1 µw oscillator Wineglass MEMS resonator

13 LowPower RF: Back to The Future (Courtesy of Brian Otis) 2000 Direct Conversion f c = 2GHz >10000 active devices no offchip components 1949 superregenerative fc= 500MHz 2 active devices high quality offchip passives hand tuning D. Yee, UCB

14 Back to the Future ThinFilm Bulk Acoustic Resonator OSC1 MOD1 OSC2 Preamp PA Matching Network MOD2 FBARbased RF Filter A RF Filter RF Filter Env Det Env Det f clock f clock Minimizes use of active components exploits new technologies Uses simple modulation scheme (OOK) Allows efficient nonlinear PA Downconversion through nonlinearity (Envelope Detector) Tx and Rx in 12 mw range (when on)

15 The Incredibly Shrinking Radio FBAR MOD1 MOD2 OSC1 OSC2 Preamp PA Matching Network TX TX On: On: 4 mw mw Stby: 1 mw mw Off: Off: 0 mw mw 130 nm CMOS Carrier frequency: 1.9 GHz 0 dbm OOK Channel Spacing ~ 50MHz kbps/channel 10 µs startup time Total area < 5 mm 2 RF Filter LNA RF Filter RF Filter Env Det Env Det f clock f clock RX RX On: On: 3 mw mw Off: Off: 0 mw mw

16 PicoBeacon: An EnergyScavenging Radio Regulator Energy Storage Capacitor (10 µf) Antenna (ceramic) Single solar cell An exercise in miniaturization and energy scavenging RF Transmitter Light Level Duty Cycle Low Indoor Light 0.36% Fluorescent Indoor Light 0.53% Partly Cloudy Outdoor Light 5.6% Bright Indoor Lamp 11% High Light Conditions Vibration Level 100% Duty Cycle 2.2m/s2 1.6% 5.7m/s2 2.6%

17 The Return of Superregenerative regenerative Fully Integrated Receiver Frontend 400µA when active (~200µW) with 50% quench duty cycle 1200µm 1500µm 1 0 OOK modulated (80 dbm signal)

18 How to go even further? Trading off accuracy for power! Example: subthreshold RF oscillator using integrated LCs 2.4 ns startup time Measurement bias conditions: Vdd=0.5V, Itail=400µA 2layer inductor 3layer inductor Simulations Oscillation Frequency GHz GHz 1.5GHz Diff. Output Swing 76mV 124mV 150mV

19 The Roadmap to Ultradense Networks Trading off accuracy for power Superregenerative: < 500 µw Untuned Subthreshold < 50 µw Untuned mostly passive < 5 µw Resonant body Finfet (NEMS)

20 The Challenge: Simplicity Threatens Reliability Narrowband radios very sensitive to fast fading effects 20 kbps, +1.5dBm 40 kbps, +3dBm 80 kbps, +4.5dBm Broadcast Success Rate [%] /19 h 11/31 h 12/43 h 13/55 h 15/07 h 16/19 h 17/31 h Time BER PicoRadio Meeting NAMP Meeting effective path loss Factor 10 5 in error rate 6 db

21 A Host of Reliability Issues Channels are unreliable Rapidly changing multipath environment Nodes are inherently unreliable May appear at will May move May temporarily run out of energy May break down Cost and power concerns explicitly decrease reliability Narrowband radios increase sensitivity to fast fading Nodes are duty cycled to preserve energy Further reductions in power affect tuning and calibration Other issues such as sensor calibration

22 Providing Robustness Traditional radio s provide robustness through diversity: Frequency: e.g. wideband solutions (hopping) Time: e.g. spreading Spatial: e.g. multiple antenna s All these approaches either come with complexity, synchronization, or acquisition overhead, or might not even be applicable Data traffic irregular, and in very short bursts A more attractive approach: Exploit redundancy Embrace randomness

23 Exploiting Redundancy and Randomness at the Network Level Multihop sensor network Multihop approach reduces transmission overhead Shortest path algorithm seems to be the optimal choice But Tends to exhaust some paths faster Breaks down in presence of unreliable nodes R. Shah et al, 2003.

24 The Impact of Spatial Diversity nodes Broadcast success rate [%] Deep fade due to multipath distance [cm] 2 nodes Data gathered using PicoNodeI testbed Adding a single node already changes broadcast reliability dramatically spatial diversity is the preferred way to provide robustness in sensor networks

25 Regionbased Opportunistic Routing Onehop neighbors Forwarding region Energy per node Opportunistic routing: Network specifies forwarding region MAC randomly chooses nexthop based on connectivity Improves reliability and energy efficiency Probability of packet success

26 Maximizing SleepTime Average power of node Dominated by TX (+RX) power Parameters: 3 packets/sec 200 bits/packet 20 bit preamble 5 neighbors Range: 10 m Synchronization using cycled receiver with Ton/T = 0.1 Dominated by channel monitoring power *based on analytical model including actual PicoRadio power numbers EnYi Lin et al, ICC 2004

27 Maximizing SleepTime A pseudoasynchronous approach: The Cycled Receiver Src T wait DATA (T p ) Wakeup beacon (T b ) ACK (T b ) Dest T Every node wakes up occasionally and asks for work Allows deep sleep mode of node at the expense of rendezvous overhead (power and time) How to choose the wakeup periodicity?

28 Randomizing the Sleep Discipline SLEEP IF YOU CAN If the node is not necessary, goes to sleep and saves power Maintain sufficient connectivity Create a sense of virtual density Allows nodes that run low in power to back off Incoming traffic Controller Sensors For how long should the node be allowed to sleep? Incoming traffic Given: 1. Loss rate 2. Delay constraint 3. Data generation requirement Choose wakeup time Adaptive Traffic & node density Random Exponentially distributed sleeping times. Avoid phase synchronization.

29 Maximizing the SleepTime Nodes estimate traffic and density based on perhop delay. Adaptively change their mean wakeup time based on estimated wakeup rate and latency requirements Sleep time an exponential random variable Dutycycle as a function of density & channel quality Changing densities Deteriorating channel Under all network conditions, only 1% of the packets fail to meet the latency constraints JanaVan Greunen et al, ICC 2004

30 Going One Step Further: Narrowband Untuned Radios Distribution f tol f LO freq Carrierfrequency variance much larger than the bandwidth Challenge: How to make these unmatched radios communicate?

31 A Statistical Communication Paradigm Strength in Numbers Source Forwarding node Destination H Random frequency multihopping Information packet traverses from source to destination in a multihop fashion. Transmitter broadcasts signal to neighboring block on random channel (as determined by process variations). Receivers randomly select channel to listen to K 2 Rx s 4 Rx s 1 Rx 6 Rx s No Tx Collisions Receptions Legal transmission only when single TX, multiple RX

32 A Statistical Communication Paradigm Some Amazing Properties Reliable communication over this unreliable platform indeed possible. Even more, reliability improves EXPONENTIALLY with a linear increase in network density. Few Tx s clean channels Too many Tx s cause collisions And the process is selfregulating D. Petrovic et al., 2004

33 Summary And Perspectives Scaling of technology leads to ever smaller communication and computation nodes Severe energy (power) constraints can only be met by compromising on complexity (or size). But simple nodes/algorithms tend to be unreliable An appealing solution: exploit the power of the numbers, and avoid brittleness by embracing randomness An opportunity for bold innovation! "Research is what I'm doing when I don't know what I'm doing." W. Von Braun The support of CEC, NSF, DARPA, GSRC Marco, and the BWRC sponsoring companies is greatly appreciated.