1
Architectural Challenges
for mm-scale Sensor Nodes
2013. 9. 29.
Yoonmyung Lee, Ye-Sheng Kuo, Pat Pannuto,
Ron Dreslinski, Prabal Dutta, David Blaauw
Department of Electrical Engineering and Computer Science
The University of Michigan, Ann Arbor
mm-Scale Sensor Nodes
2
[1]
[2]
[3]
[4]
[5]
[6]
§  Emergence of mm-scale systems
§  Smaller form-factor
§  Cheaper price
[1] G. Chen et al. ISSCC 2010
[2] G. Chen et al. ISSCC 2011
[3] Y. Lee etal. ISSCC 2012
[4] P-J. Chen et al. JMEMS 2010
[5] R. Haque et al. MEMS 2011
[6] R. Casanaov ISSCC 2009
2
mm-Scale Sensor Nodes
3
§  Key enabler: Low power circuit technologies
Energy optimal operation
- f / V scaling
- Super-pipelining
Standby mode operation
Low power antenna
Low power f-ref
Radio duty-cycling
Low power (<nW)
Small uncertainty
Temperature
sensitivity
Low power (<nW)
Efficient mode
conversion
Radio
Control
Data
Memory
Instruction
Memory
Timer
CPU
Energy
Harvesting
Wake Up
Control
Sensor
Power
Mgmt
Low leakage (<pW/b)
Small cell area
Robustness
Harvesting from < µW
< mm3 harvester
Low voltage
harvesting
Sporadic availability
Wide range Iload
Efficient mode
conversion
Brown-out detection
Low voltage ADC
Low power sensors / interfaces
3
Challenges in mm-Scale Sensors
4
§  Building a “SYSTEM” out of low power building blocks
§  Integration requirements
§  Volume limit
§  Energy limit
§  Controllability
§  Extremely wide application space
§  Medical / Environmental
Structural Integrity / Surveillance
Building Management / and so on …
4
Layer Based Modular Architecture
5
§  Layer approach
§  Maximizing Si area in limited volume
§  Modularization
§  Technology optimization
Generic
Layers
r 1
ye
La 2
r ye
La 3
r ye
La 4
r ye
La
r 5
ye
La
Application-specific
Layers
Processor
Imager
Motion Detection
Flash
Memory
Pressure
Sensing (C to D)
Temperature
Sensing
Decap
ECG monitor
Strain Sensing
(R to D)
Battery
Light Sensing
Analog Voltage
(ADC)
Power Mng.
E-Harvesting
Near Field
Radio
Far Field
Radio
5
Inter-Layer Interconnect
6
§  How are the layers connected?
§  Delivery of Power
§  Control
§  Data exchange
§  Traditional solutions
1
er
y
La 2
r ye
La 3
r ye
a
L
r 4
ye
a
L
r 5
ye
a
L
§  I2C
§  Bidirectional – pull up resistor power
§  Distributed clock control – clock generation on each device
§  SPI
§  >3 wires - limited expansion
§  No slave TX initiation / slave ACK
§  No slave to slave
§  Need low power solution for control & data exchange
6
MBus: Low Power Inter-Layer Interconnect
7
§  MBus
§  1 Master, multiple Slave layers
§  2 wires: CLK, DATA
§  Unidirectional: Daisy-chained
Layer 1
Layer 2
CLK
DATA
CLK
La
bi tch
t 2 Dr
bi ive
t 2
La
bi tch
t 1
Dr
bi ive
t 1 La
bi tch
t 0
Dr
bi ive
t 0 §  2 phase clocking: Improve setup/hold tolerance
§  for arbitrary layer combination
§  for irregular loading
Set up Margin
Hold Margin
DATA
7
MBus: Low Power Inter-Layer Interconnect
DATA
(idle)
PMU
(in Master) Standby
Slave
§  Interrupt by clock blocking
è Prioritize messages
è Reset bus
Clock blocking
= INT request
Wait for
an event
PMU mode
switch
Active
(send clock)
TX request
Send Data
Data toggle wo CLK
= INT grant
IN
bi T Ct
t 1 rl
(idle)
IN
bi T Ct
t 0 rl
è No clock generation on other layers
è Slave request TX by DATA pull down
è Request wake up by DATA pull down
CLK
St
a
IN rt o
T f §  CLK driven by Master layer
8
CLK_IN
CLK_OUT
DATA
§  Termination sequence
è Utilize interrupt as termination sequence
è Enable slave ACK (control bit) & arbitrary length message
8
1st System
9
§  Motion detection surveillance sensor
§  Application specific: Motion detector layer
§  Motion detection in standby mode (200nW)
§  Data transmission in active mode (40µW)
CPU
(180nm)
SRAM(3kB: retentive)
PMU
DECAP
Motion detection
pixel array
(180nm)
Thin-film Li Battery
TX Radio
(180nm)
Motion Detection
& Imager
(130nm)
9
2nd System
10
§  Implantable medical sensor (tumor pressure monitoring)
§  Application specific:
MEMS pressure sensor
CDC conversion
CPU
(180nm)
SRAM(3kB: retentive)
PMU
DECAP
(180nm)
Thin-film Li Battery
TX Radio
CDC Converter
(180nm)
[Pressure Readout from mouse]
MEMS Pressure Sensor
10
3rd System
11
§  Imager and Temperature monitoring
§  Application specific:
Imager
Temperature sensor
Imager
(130nm)
Timer
Temperature Sensor
Thin-film Li Battery
DSP CPU
(65nm)
SRAM(16kB:non-retentive)
CPU
(180nm)
SRAM(3kB: retentive)
PMU
DECAP
(180nm)
11
12
Thank You
12