1. SMART DUST Hardware Limits to Wireless Sensor Networks
Kris Pister
Berkeley Sensor & Actuator Center
Electrical Engineering & Computer Sciences
UC Berkeley –
(on leave to start Dust Inc –

2. Ken Wise, U. Michigan

3. Bill Kaiser, UCLA

4. Wireless dawn sensor

5. Computation
Difference Engine
Charles Babbage, 1822
Steve Smith, UCB

6. Multi-hop message passing

7. Lots of exponentials Digital circuits Communication circuits
Speed, memory
Size, power, cost
Communication circuits
Range, data rate
MEMS Sensors
Measurands, sensitivity
Communication
Computation
Sensing

8. Smart Dust Goal

9. COTS Dust - RF Motes Simple computer Cordless phone radio
Up to 2 year battery life N S E W
2 Axis Magnetic Sensor
2 Axis Accelerometer
Light Intensity Sensor
Humidity Sensor
Pressure Sensor
Temperature Sensor

10. Open Experimental Platform to Catalyze a Community
Services
David Culler, UCB
Networking
TinyOS
WeC 99
“Smart Rock”
Rene 00
Designed for experimentation
sensor boards
power boards
Dot 01
Demonstrate scale
Mica 02
NEST open exp. platform
128 KB code, 4 KB data
50 KB radio
512 KB Flash
comm accelerators
Small microcontroller
- 8 kb code, 512 B data
Simple, low-power radio
- 10 kb
EEPROM storage (32 KB)
Simple sensors
Before we jump in to the technical material, I would like to give you one more bit of context.

11. 800 node demo at Intel Developers Forum
4 sensors
$70,000 / 1000
Concept to demo in 30 days!

12. Structural performance due to multi-directional ground motions (Glaser & CalTech).
Mote Layout
Mote infrastructure 14 13 5 ` 15 15 6 12 11 9 8
Comparison of Results
Wiring for traditional structural instrumentation
+ truckload of equipment

13. Cory Energy Monitoring/Mgmt System
50 nodes on 4th floor
5 level ad hoc net
30 sec sampling
250K samples to database over 6 weeks

14. 29 Palms Sensorweb Experiment
Goals
Deploy a sensor network onto a road from an unmanned aerial vehicle (UAV)
Detect and track vehicles passing through the network
Transfer vehicle track information from the ground network to the UAV
Transfer vehicle track information from the UAV to an observer at the base camp.

15. Last 2 of 6 motes are dropped from UAV
8 packaged motes loaded on plane
Last 2 of six being dropped

16. Available Sensors Demonstrated w/ COTS Dust Available
Temperature, light, humidity, pressure, air flow
Acceleration, vibration, tilt, rotation
Sound
GPS
Gases (CO, CO2)
Passive Infra-red
Contact/touch
Available
Images, low-res video
Gases (VOCs, Organophosphates, NOx…)
Neutrons
Demonstrated Actuators
Motor controllers
110 VAC relays
Audio speaker
RS232: LCD, …

17. Blue Mote Hardware Chipcon cc1000 radio TI MSP430 Processor
RX Power: mA (-102 -> -105 dBm)
TX Power: mA, (-5 to 4 dBm) range ~50m indoors
Bit rate up to 76,800 kbps
TI MSP430 Processor
4MHz
Operating Voltage V
Sleep mode = 3 mA
Same damn 51 pin connector
$50-$100

18. Basic Operations Sleep Listen for activity on radio Sample sensors
Synchronize clocks
Scheduled chat with neighbor
Message via multihop
Data
“Warning!”
“We’re all fine down here”

19. Cost of Basic Operations
Current
[A]
Time
[s]
Charge
[A*s]
Sleep 3u
Sample 1m
20u
20n
Talk to neighbor
15 byte payload
25m 5m
125m
Listen to neighbor
10m 8m
80m
Sound an alarm
1s?
25,000m?
Listen for alarm 2m 4m
QAAbattery = 2000mAh = 7,200,000,000 mA*s

20. Typical Topologies
Star
Linear
Tree

21. Application Energy Breakdown
Collect Data from 3 Children every 15 seconds (RX cost include synchronization)
Send 4 data packets every 15 seconds
Alarm check once per second
Send 10 alarms per day
Expected Lifetime: 4.1 Years
Cost Each (mJ)
Number Per Day
mJ per Day
% of Total RX
0.24
17280
4147.2
28% TX
0.375
23040
8640
59%
Alarm Check
0.012
86400
1036.8 7%
Alarm Send 75 10
750 5%
Total:
14574
Battery Life (years):
4.1

22. HDK Implementation
Report Interval (user controlled, 0 to 256 seconds)
Reporting Slots 32 ms
Collect data from children
Send to parent
Periodic Alarm Message Checks
Alarm Msg Forward
Sensor Sampling
Alarm Msg !

23. Time period definitions
Report Interval (user controlled, 0 to 256 seconds)
Reporting Slots 32 ms
Tepoch
tslot
Periodic Alarm Message Checks
talarm
Sensor Sampling
tsample

24. HDK Extrema Max data rate through a single mote: 1kB/s
Max data rate via linear multihop: 300B/s
Latency in multihop communication: n*tslot
Alarm lifetime = talarm*Qbat/Qcheck
Alarm latency < n*talarm
E.g. talarm = 0.1s; n=20; N=1,000,000
Lifetime = 6 years
Latency < 2 s
“We’re all fine” lifetime = (Qbat / (Qmsg )* (Tepoch /(1+nkids))
E.g. Tepoch = 20 min; nkids = 1000
 Lifetime = 3 years

25. Application Energy Breakdown
Collect Data from 3 Children every 15 seconds (RX cost include synchronization)
Send 4 data packets every 15 seconds
Alarm check once per second
Send 10 alarms per day
Expected Lifetime: 4.1 Years
Cost Each (mJ)
Number Per Day
mJ per Day
% of Total RX
0.24
17280
4147.2
28% TX
0.375
23040
8640
59%
Alarm Check
0.012
86400
1036.8 7%
Alarm Send 75 10
750 5%
Total:
14574
Battery Life (years):
4.1

26. One Chip, Four Dissertations
CMOS ASIC
8 bit microcontroller
Custom interface circuits
External components
antenna
~2 mm^2 ASIC uP
SRAM
Radio
ADC
Temp
Amp
~$1
inductor
crystal
battery

27. Working silicon 8 bit uP 3k RAM OS accelerators
World record low power 8 bit ADC (100kS/s, 2uA)
HW Encryption support
900 MHz transmitter
Functional, running TinyOS,
sending packets to Blue

28. Working mote, happy grad student
Jason Hill
Jason’s mote

29. Power and Energy Sources Storage Usage
Solar cells ~0.1mW/mm2, ~1J/day/mm2
Combustion/Thermopiles
Storage
Batteries ~1 J/mm3
Capacitors ~0.01 J/mm3
Usage
Digital computation: nJ/instruction
Analog circuitry: nJ/sample
Communication: nJ/bit
10 pJ
20 pJ/sample
11 pJ RX, 2pJ TX (optical)
10 nJ/bit RF

30. Energy and Lifetime 1 mAh ~= 1 micro*Amp*month (mAm)
Lithium coin cell: 220 mAm (CR2032, $0.16)
AA alkaline ~ 2000 mAm
100kS/s sensor acquisition: 2mA
1 MIPS custom processor: 10mA
100 kbps, m radio: 300mA
1 month to 1 year at 100% duty
10 year lifetime w/ coin cell  1% duty
Sample, think, listen, talk, forward…
2 times/second!

31. Energy Considerations
Storage
Batteries today: 700 Wh/kg (Tadiran)
Battery limits: 8,000 Wh/kg (Aluminum/air)
Gasoline: 12,700 Wh/kg (upper heating value)
H2: 50,000 Wh/kg (upper heating value)

32. Energy Considerations
Sensing
10 bits (re: 8 bits)
Power ~ 22N f
Scott, Boser, Pister, An Ultra-Low Power ADC for Distributed Sensor Networks, ESSCIRC 2002.

33. Energy Considerations
Computation
Power ~ CV2 f
C = NgC0
C0 = er e0 A/d ~ 5fF/mm2
For 8 bit ops, Ng ~100
A ~ Ld2
A = 0.020mm2 today (Ld =0.13)  10pJ
A = 0.001mm (Ld =50nm)  0.5pJ

34. RF Sensitivity Pn = kBT Df Nf Sensitivity = Pn * SNRmin
e.g. GSM (European cell phone standard), 115kbps
kBT kHz ~8x SNR
S = -174dBm + 53 dB + 9 dB + 10 dB
= -102 dBm
RX power = ~200mW
TX power = ~4W  50 uJ/bit

35. RF Path Loss Isotropic radiator, l/4 dipole Free space n=2
Pr=Pt / (4p (d/l)n)
Free space n=2
Ground level n=2—7, average 4

36. N=4 -102dBm From Mobile Cellular Telecommunications, W.C.Y. Lee
Pt = 10-50W
-102dBm

37. Path Loss Like to choose longer wavelength But… Loss ~(l/d)n
916MHz, 30m,  92dB power loss
 need –92dBm receiver for 1mW xmitter
 power!
Penetration of structures, foliage, …
But…
Antenna efficiency
Size – 1GHz = 7.5cm

38. Output Power Efficiency
Pout RF
Slope Efficiency
Linear mod. ~10%
GMSK ~50%
Poverhead = 1-100mW
Optical
lasers ~25%
LEDs ~50%
Poverhead = 1uW-100mW
True
Efficiency
Slope
Efficiency
Pin
Poverhead

39. Limits to RF Communication
Cassini
8 GHz (3.5cm)
20 W
1.5x109 km
115 kbps
-130dBm Rx
10-21 J/bit
kT=4x
~ cm photons/bit
Canberra
4m, 70m antennas
The cassini High Gain Antenna has a half angle of 0.14 degree in Ka band. I calculate 0.5 degrees at 8.4 GHz. The Canberra dish has a half angle of 0.5mrad, and would almost fit a football field. 10^-21 J is about 5000 photons/bit.

40. Integrated Microwatt Transceiver, Howe/Rabaey, UCB
Radios need filters
The best filters are electromechanical
Power is related to size

41. Mike Sailor’s Smart Dust
M. Sailor
UCSD Chemistry

42. CMOS Cameras Today Tomorrow Soon 5mm scale 1mJ/image 110,000 pixels
50pJ * #pixels / image ~ 1uJ
16k pixels
Soon
1pJ * # pixels /image ~ 1uJ
1M pixel

43. Single Nanotube Inverter - IBM
Atomic Force Microscope image showing the design of an intra-molecular logic gate. A single carbon nanotube (shaded in blue) is positioned over gold electrodes to produce two p-type carbon nanotube field-effect transistors in series. The device is covered by an insulated layer (called PMMA) and a window is opened by e-beam lithography to expose part of the nanotube. Potassium is then evaporated through this window to convert the exposed p-type nanotube transistor into an n-type nanotube transistor, while the other nanotube transistor remains p-type.
Derycke, Martel, Appenzeller, Avouris; Carbon nanotube inter- and intra-molecular logic gates; Nano Letters, August 26, 2001

44. Carbon Nanotube Circuits - Delft
A. Bachtold, P. Hadley, T. Nakanishi, C. Dekker; Logic circuits with carbon nanotube transistors Science, 294, (2001).

45. Nano Dust? Nanotube sensors Nanotube computation
Nanotube hydrogen storage
Nanomechanical filters for communication!

46. Mobility
Walking
Hopping
Flying

47. Mobility

48. Milli-Millennium Falcon
Increase the thrust and decrease the mass, while controlling thermal losses

49. Thrust Measurements vs. Theory
Predicted altitude: 50 m

50. Rocket in Action

51. Synthetic Insects (Smart Dust with Legs)
Goal: Make silicon walk.
Autonomous
Articulated
Size ~ 1-10 mm
Speed ~ 1mm/s

52. 2 Degree of Freedom Legs
1st Link
Motor
2nd Link
1mm

53. Silicon Inchworm Motors

54. Current Layout for Motor and Legs
Solar Cells
Legs
Linkages
CMOS
Motor
7.6 mm

61. Solar Powered Robot Pushups

62. Big Products from Small Workers

63. The Dark Side

64. Conclusion
Tremendous promise
More new questions than answers