Design Considerations for Next Design Considerations for Next Generation Wireless Power Aware Generation Wireless Power Aware Microsensor Microsensor Nodes Nodes D. Wentzloff, B. Calhoun, R. Min, A. Wang, N. Ickes, A. Chandrakasan Massachusetts Institute of Technology
29
Embed
Design Considerations for Next Generation Wireless Power ... · Design Considerations for Next Generation Wireless Power Aware . Microsensor. Nodes. D. Wentzloff, B. Calhoun, R. Min,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Design Considerations for Next Design Considerations for Next Generation Wireless Power Aware Generation Wireless Power Aware
MicrosensorMicrosensor NodesNodes
D. Wentzloff, B. Calhoun, R. Min, A. Wang, N. Ickes, A. Chandrakasan
““Software” Energy Dissipation is Dominated by Software” Energy Dissipation is Dominated by Overhead and NOT by Useful WorkOverhead and NOT by Useful Work
Leakage Mitigation Using MTCMOSX3 X2 X1 X0
Y3
Y2
Y1
Y0
P0
P1
P2
P3
P4P5P6
P7
pc
pc
VDD
0 50 100 150 200 250 8
10
12
14
16
Dela
y ,n s
Sleep Transistor Width, W/L
A: X=00000000->11111111Y=00000000->10000001
B: X=01111111->11111111Y=10000001->10000001
Vector A
Vector B
Sleep High VT Device
Low VTLogic Virtual
Ground
Device Sizing is a Major Concern in Multiple
Threshold CMOS
Leakage Reduction Using MTCMOSLeakage Reduction Using MTCMOS
A
A
A XOR B
B=1
A=0
0
1
Look at A=0 and B=1.Sneak Leakage!!
B
A
Power Aware Architectures Power Aware Architectures
Single butterfly architecture (4 multipliers, 6 adders)
Data Memory
Twiddle ROM
Dat
a A
ddre
ss
Twiddle Address
R/W AB
XY
W
Control Logic
Butterfly structure
A
BW
X=A+BW
Y=A-BW
FFT Computation
Power Scalable MultiplierPower Scalable Multiplier(modified Baugh(modified Baugh--WooleyWooley))
X{15:0}
32x32
16x3216x32
64x32
128x32
32x32
16x3216x32
64x32
128x32
Control Logic
Y{1
5:8
}
X{15:8}
11
1
Z{15:0}
00
00
00
00
00
00
00
0
Y{15:0}
8-bit feed through
Power Scalable MemoryPower Scalable Memory
Address(write/read)data (write)
Data (read)
gatinginput
0
0
Adder used only in16-bit modeAdder used in 8-bitAnd 16-bit mode
First Generation Power Aware FFT First Generation Power Aware FFT
New Energy Metrics in DSM InterconnectNew Energy Metrics in DSM Interconnect
02468
1012141618
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Normalized Energy, Normalized Energy, EE
0
10
20
30
0 1 2 3
Standard modelStandard model
Normalized Energy, Normalized Energy, EE# of
tran
sitio
ns o
f co
st
# of
tran
sitio
ns o
f co
st EE SubSub--micron modelmicron modelBUS
d2
d1
l2
l1
VDD
CL
CL
CI
3==L
I
CCλ
# of
tran
sitio
ns o
f co
st
# of
tran
sitio
ns o
f co
st EE
Extended Busn+a lines
Recovered DataInput Data (n bits)
... DecoderEncoder
Minimizing Transition Activity is not the Minimizing Transition Activity is not the Right approach to Minimize PowerRight approach to Minimize Power
Computation vs. CommunicationComputation vs. Communication
1E-111E-101E-091E-081E-071E-061E-051E-041E-03
1 10 100 1000 10000
Energy for Electronics + Transmit
R2 Propagation LossLimit (no electronics)Assuming 10pJ/bit/m2
Ener
gy (J
)
Distance (m)
Computation: 1nJ/op (µ-Processor) and Communication (@10m): 150nJ/bit @10 m: ~150 instructions/transmitted bit on a low-power processor@10m: > 1Million instructions/transmitted bit using dedicated hardware
Take advantage of dense sensor networks by using several shorter hops to transmit long distancesPlot of total power used to transmit a given distance for 1, 2, 3, and 4 hops
Large power step in each trace from turning on external PATrace out lowest curve for energy efficiency (i.e. use 3 hops @ 1000 m)
Multi-hop routing is more energy efficient for this particular radio
Adds overhead to the protocolAdds latency to the network
Operational PoliciesRadiated Power Convolutional Code
Ene
rgy
(J)
µAMPS-1 Node1000 bits
Total Communication Energy
Range (m)
Range (m)
higher quality
Energy scales gracefully with communication quality
higher quality
ConclusionsConclusions
Exciting new applications enabled by a network of low-power wireless sensing devicesPower Aware Design Methodology supersedes Energy Efficient DesignSlower is Better – exploit sub-threshold operation as fastest switching speed is not neededCommunication-centric design
Energy per operation (mW/MIPS) will scale with technologyCommunication costs (nJ/bit) will not scale at the same rate
Low Energy Sensor Design Requires a SystemLow Energy Sensor Design Requires a System--level level Approach Approach –– Tight Coupling Between Fabrics, Tight Coupling Between Fabrics,