Operating below the Sub-microwatt Barrier - Explorations in Analog by ocv22853

VIEWS: 8 PAGES: 37

									Operating below the Sub-microwatt Barrier –
    Explorations in Analog Computing



                  Shantanu Chakrabartty

        Department of Electrical and Computer Engineering
                    Michigan State University




                    WIMS presentation April 2009
                   Adaptive Integrated Microsystems
    “Interfacing Biology with Analog Circuits”

                              Biology
                                                    Silicon

      Morphing                                                        Monitoring

-Neuronal noise-shaping    Source: wikipedia.org
                                                                - Potentiostats for
for designing speech                                            biosensors (DHS).
processors and ADC.
(NSF,APL).                                                     - Power harvesting and
                                            Synthesis          structural health
- Log-likelihood                                               monitoring (NSF, FHWA).
computing for high-speed         - Protein based analog
decoders (NSF).                  circuit synthesis for
                                 reliable biosensors
                                 (NSF).




                            Adaptive Integrated Microsystems
                           Vision




                                        Finite-element sensor array          Source:GmBH
Source: ECE Networks
                                        equivalent to FEM models.




                                                                Remote
                                             RF-Power         Interrogator




                          Smart Pebble       Event Indicator


                       Adaptive Integrated Microsystems
                       Outline

• Motivation


• Sub-threshold barrier and Structural Health Monitoring

• Sub-microwatt Analog VLSI


• Results


• Other Applications


                   Adaptive Integrated Microsystems
           Motivation: Biomechanical Implants

                  Dental
                  implant                             By 2030, cases of primary hip
  Intra-ocular
  pressure
                                                      arthroplasty in U.S. is expected to
                                                      double to 0.5 million and cases of
                                                      knee arthroplasty will grow by 7
                                                      fold to 3.5 million.
                          Vertebrae
                        replacement
                                 Spinal ligament
                                     strains

Ligament Strain
 Measurement
                        Hip implant




  Knee implant




                                Adaptive Integrated Microsystems
                                             Mechanical Fatigue

• Fatigue and wear are major causes affecting stability of
  biomechanical implants. (10-20% of the artificial joints need to be
  replaced within 15-20 years).




Source: Heaton-Abdigle, Talyor et. al 2006




• Smart sensor that can predict onset of long-term fatigue without
  affecting the functionality of the implant.


                                                Adaptive Integrated Microsystems
 Mechanical Fatigue in Civil Structures
                                                                          Source:Wikipedia




                          0         5       10    15

                                  Minutes




                              0         8    16    24

                                        Hours


On January 15, 2008 the National Transportation Safety Board announced
they had determined that the I-35 bridge's design specified steel gusset plates
that were undersized and inadequate to support the intended load of the
bridge, which had increased over time.

                                  Adaptive Integrated Microsystems
      What is Mechanical Fatigue ?




“the progressive, localized, and permanent structural
damage that occurs when a material is subjected to
cyclic or fluctuating strains at nominal stresses that
have maximum values less than the static yield
strength of the material.”




                 Adaptive Integrated Microsystems
                       Fatigue prediction
1.00       OFHC
           Copper                                                                              6




                                                             strain
                                                                                               5
                                                                                               4
                                                                                               3
0.10                                                                                           2

∆ε
                                                                                               1
                               Hardened
        Annealed
                               4340 steel                                                       t
        4340 steel
0.010
                                                             ni       : Number of times the strain
                                                                      level exceeded level i
                   Fatigue Limit for 4340 steel
0.001                                                        Ni       : Number for strain level i
     100     102        104          106        108                   obtained from S-N curve.
                          Nf
                                                             Onset of fatigue by Miner’s
                                                             formula
                                                                              m
                                                                                    ni
                                                                            ∑i =1   Ni
                                                                                       ≥1

                               Adaptive Integrated Microsystems
                  SHM meets Silicon
                                       (Straser, Lynch, Tanner)




                                                                                                           Crossbow Inc.

                   www.hbm.com




                                 Sukun K. et al. “Health monitoring of civil infrastructures
                                 using wireless sensor networks, University of California at
                                 Berkeley, IPSN 2007                                                     Microstrain Inc.




- Existing solutions use strain gauges,
temperature gauges, Data converters, Micro-
controllers and on-board memory.

- Power for sensing, computation and storage
provided by on-board batteries.
                                   Adaptive Integrated Microsystems                            Univ. of Michigan
Challenges in long-term fatigue monitoring
 Electronic powering is one of the major obstacle !

 Strain-gauges with implanted batteries are impractical




                                                                                             www.hbm.com
solution:
     Batteries self-discharge !
     Batteries life is a tiny fraction of the fatigue life of the
    monitored components
     Batteries replacement is either costly or impractical
    (implants)

 Remote RF powering is impractical for long-term
autonomous monitoring.

                                                               B. D. Beynnon et al. - Microstrain

       Solution: Self-powered sensing using piezoelectric materials
           Harvest computing power from the localized strain.

                            Adaptive Integrated Microsystems
Fundamental Limits of Strain-based Power Harvesting


                                                               bL 
                                                         C = ε         C
                                                               h 
                                         Sensor
                                                             Y E d 31h
                                                                             V
                                                        V=
                                                                ε

                                       - Voltage per µε is independent
                                       of area: For h=0.1mm, a PZT-5H
                                       generates 65V/µε.

                                       - Series capacitance scales
                                       with area.
                                       Maximum harvestable power
                                                     PZ ≈ 2πfCV 2
                  Adaptive Integrated Microsystems
    Challenges for in-vivo harvesting
Typical strain levels in in-vivo biomechanical structures in the
                     order of 100 – 1000 µε.

             Structures                       Typical Strain levels

                nerves                  0.1% - 20% (T. W. Wright et al.)

                                       0.04% - 0.16 % (J. Borodkin et al.)
                bones
                                               (D. T. T. Lie et al.)

              Ligaments                0.1% - 4 % (B. D. Beynnon et al.)


               Muscles                 0.1% - 5 % (R. L. Pradhan et al.)


                                                                      7


                                                   Power Level (uW)
                                                                      6
                                                                      5
                                                                      4
                                                                      3
                                                                      2
                                                                      1
                                                                      0
                                                                          200      400    600     800    1000
                                                                                Strain levels (micro-strain)
        Hip implant with PVDF strain sensor                           PVDF sheet dimensions = (31,16,0.028) mm.
              (US Patent no: 06034296)                                Loading frequency = 0.5 Hz

                            Adaptive Integrated Microsystems
                    Strain power harvesting in Civil Structures
                                                                Elvin et. al (2006,2007)




                                                                                     102
              100
                                5mW                    Earthquake
                                                                                     101
                                                                                                                           Earthquake
                                                       Wind                                                                Wind
                                                      Traffic                        100                                   Traffic
               10
                                                                                                1μW




                                                                       Energy (mJ)
Energy (mJ)




                                                                                     10-1
                    10μW
                1                                                                    10-2

                                                                                     10-3
              0.1
                                                                                     10-4

                                                                                     10-5
                                                                                            Bridges     Tall      Earth-
                      Bridges     Tall      Earth-
                                                                                                      Buildings   quakes
                                Buildings   quakes
                      Maximum harvestable power                                             Maximum transferable power




              - Maximum transferable power less than 1 microwatt.


                                                     Adaptive Integrated Microsystems
           Power Dissipation Scale

  109 W      Hoover Dam

 745W        1 Horse Power

 170W        Intel Itanium Quad-core

 100W        Metabolic rate of Human Body

   80W       Intel Pentium 4

   30W       Power consumption of human brain

 10 −2 W     Laser in DVD ROM

10 −5W       Quartz wristwatch                                10 −6 W
10 −9 W      Air flow at 5m/s per sq mm

10 −12 W     Average power consumption of a human cell

10 −18W      Thermal noise

10 −21W      Power received by deep space antenna from Galileo probe
                  Adaptive Integrated Microsystems
        CMOS: Summary of Constraints


- Maximum power dissipation per loading cycle < 1µW.




- Time between significant loading is long (hours/days) - Can not use
voltage multiplication or trickle charging !



- Energy either needs to be used on the spot and variables stored
on a non-volatile memory !




                          Adaptive Integrated Microsystems
               What can we do with less than 1 µW ?

                            Operation                                   Commercial                Research

                  Microprocessor Operation+                          350µW (TI)         20µW (Amirtharajah)

                       Write 1 bit to NVM                            200nJ (Micron)     25pJ (Cavendish)

          Analog-Digital (A/D) Conversion – 1bit                     2nJ (TI)           50pJ (PicoRadio)

                               Sleep                                 300nW (TI)         5nW (PicoRadio)

                   Digital Signal Processing+                        200µW (LSI)        20µW (Amirtharajah)

                           Pin Leakage                               100nW (TI)         2.2nW (Kao)


+ Operation is scaled to processor speeds of 500 kHz                                   Normal Power           Sleep Mode
TI:- Texas Instrument MSP430 microprocessor
PicoRadio:- University of California – Berkeley PicoRadio project                         (mW)                   (µW)
X-bow: X-Bow Technologies Mica2Dot Wireless sensor Mote
Micron:-Micron Technology NOR Flash Memory                             AD5933                    33              2.3
Cavendish:-Cavendish Kinetics Corp, MEMS memory
LSI: - LSI Logic Corp - LSI403US Digital Signal Processor            ZigBee Tx                  148              33
Telos:- Moteiv Corporation
                                                                    ATmega128L                   31              16

                                                                                Digital techniques used in SHM



                                             Adaptive Integrated Microsystems
          Pico-watt Mechano-receptors in Biology




- Senses short-term and long-term variations in mechanical pressure.
- Integrates sensing, computation, storage within a single device.

                          Adaptive Integrated Microsystems
                            Principle of operation

 - Open circuit voltage generated by piezoelectric transducer for a nominal
 strain can be very high (10-1 V/µε) but with limited driving currents (<1nA).

                                                                                                                        Presynaptic cell
                           Gate           SiO2
                                                                                                Synaptic cleft
   Source                               Drain
                         -
                     -poly Si - -
                       - --                                                                                             Postsynaptic cell
                               - + -+
     p+                                    p+




                                                           Mean postsynaptic membrane current
                          n-well                    -
                                                    -
                                                    -                                                                                (from Shepherd 1979)
                                                    -


    Piezoelectric shell                                                                             Presynaptic membrane potential



-Voltage controlled energy barrier for                  - Pre-synaptic membrane voltage
hole current                                            controlled neurotransmitter release.
                                   Adaptive Integrated Microsystems
               Piezo-floating-gate Injector Model
                                      Chakrabartty, et. al 2007, 2008, (patent pending)




       Sub-threshold Conduction                                           Impact-ionized hot electron
                                                                                  injection (IHEI)
                      −κ
                           V fg       Vs
                                                               +                                    C dVS
         I s = I 0e        UT
                                  e   UT
                                                                             I inj = βI S e sd inj = T
                                                                                           V /V

                                                                                                     κ dt




                                                                                            =
                                  Is                                              Injector response

                                                                         Vs = −
                                                                                    1
                                                                                               (      −K V
                                                                                       log K1 K 2t + e 2 g 0   )
                                  Vs                                                K2
PVDF                                          CT                                      κβ I S       1
                                                                      =K1                 =  , K2
                                                                                          Ct      Vinj
                 Q1
                                        Vg          VB
                                               Adaptive Integrated Microsystems
                                     Log-linear Response
                                          Huang, Chakrabartty, 2008




     Injector response                                             4.55
                                                                                     Linear

Vs = −
         1
         K2
                 (         −K V
            log K1 K 2t + e 2 g 0     )                             4.5
                                                                               4.5




                                              Output Voltage (V)
                                                                              4.45
            κβ I S    1                                            4.45
=K1          =  , K2                                                           4.4
                                                                                                                              Log-linear
             Ct      Vinj                                                     4.35
                                                                    4.4
                                                                               4.3

                                                                              4.25                                              4
                                                                   4.35                 2     4   6            8   10     12 x 10


           t >> e K 2VS 0 / K1 K 2                                                    Measured
                                                                    4.3
                                                                                      Model (K1 = 0.202× 10-22, K2 = 8.873)

                     t 
                                                                          0                                2                          4
               1                                                     10                               10                            10
         ∆Vs =    log 
                     t                                                                                       Time (t)             Single electron per
               K2     0                                                                                                            second injection

                                                                     The log-linear response of the injector
                                                                     is independent of the initial conditions
                                                                            and biasing conditions !
                                          Adaptive Integrated Microsystems
                                                     Characterization of log-linear monitors
                          4.55


                                 4.5
Output Voltage (V)




                          4.45


                                 4.4
                                           I = 17n, K = 10.0604
                                                     2
                                           I = 13n, K = 10.1420
                                                          2
                          4.35             I = 10n, K2 = 8.787
                                           I = 8n, K = 9.2166
                                                          2
                                 4.3 0           1                     2                   3            4            5
                                   10       10                    10                  10        10              10
                                 4.56                                      Time (t)


                                 4.54

                                 4.52
                                                                                                                             - Variation in injection slope with
            Output Voltage (V)




                                  4.5
                                                                                                                             bias current < 15%
                                 4.48
                                            Chip1, K = 9.5493
                                                                                                                             - Variation in injection slope due to
                                                              2
                                 4.46
                                            Chip2, K2 = 10.9104                                                              mismatch < 10%
                                 4.44       Chip3, K = 10.5166
                                                              2
                                            Chip4, K2 = 10.0577
                                 4.42
                                    10
                                       0
                                                     10
                                                          1
                                                                              10
                                                                                 2
                                                                                               10
                                                                                                    3
                                                                                                                 10
                                                                                                                         4   - Temperature coefficient < 0.01 T-1V-1
                                                                            Time (t)
                                                                                                            Adaptive Integrated Microsystems
                                                                Event counting

                      5
                                                           000/256 cycles
                     4.9                                   220/256 cycles                                       IS
                                                           240/256 cycles
                     4.8                                   248/256 cycles
Output Voltage (V)




                                                           256/256 cycles
                                                                                                       Triggering
                     4.7
                                                                                                         Circuit
                     4.6
                                                                                    Piezoelectric               Vout CT
                     4.5                                                            transducer
                                                                                                          Q1
                     4.4
                                                                                                                     Vg   VB
                     4.3

                     4.2
                           10
                                3
                                                           10
                                                                4
                                                                                                                                 6 bit
                                    Loading Cycles (n)
                                                                                                                               counting

                                t0 >> e K 2VS 0 / K1 K 2


                                     1      t + ∆t 
                           ∆Vs =        log 0
                                            t      
                                                    
                                     K2        0   


                                                                    Adaptive Integrated Microsystems
                                                      Level Counting
    Startup          Reference
V
                M1               M5        P1          P3         P5        P7                  P9                  P11                   P13


      S1        M2               M6        P2          P4         P6        P8                  P10                 P12                   P14

                                                 D1         D2         D3                  D4                  D5                    D6

           S2


                M3               M7
                                            O1         O2         O3                  O4                  O5                    O6                    O7
       C0
                                      C1         C2         C3               C4                  C5                       C6                    C7
                M4               M8

                            R
V                                      F1         F2         F3                  F4                  F5                    F6                    F7


                                                  Trigger Circuit

       Injection Thresholds:
        +       −                   Is 
     V − V = Vsat + Vs 0 + MU T ln  
                                    I0 

                                                      Adaptive Integrated Microsystems
                                                                                     Results


                      5                                                                                             5
                                                                              Channel1                                                                                     Channel1
                                                                              Channel2                                                                                     Channel2
                     4.9                                                      Channel3                             4.9                                                     Channel3
Output Voltage (V)




                                                                                              Output Voltage (V)
                     4.8                                                                                           4.8

                              4.7                                                                                           4.7
                     4.7                                                                                           4.7
                             4.65                                                                                          4.65

                              4.6                                                                                           4.6
                     4.6                                                                                           4.6
                             4.55                                                                                          4.55

                              4.5                                                                                           4.5
                     4.5                                                                                           4.5
                             4.45                                                                                          4.45
                                0.5   0.6    0.7       0.8 0.9   1 × 104                                                      0.5     0.6    0.7       0.8 0.9   1 × 104

                     4.4 1                                                                                         4.4 1                         2                     3              4
                                                 2                     3                  4
                       10                   10                10                     10                              10                     10                    10             10
                                                 Loading Cycles (n)                                                                                  Loading Cycles (n)
                                                                                1s                                                                                          1s

                                                                                2s                                                                                          2.5s
                                                                                3s                                                  5.3V             6.1V   6.9V            3s
                                      5.3V           6.1V   6.9V




                                                                           Adaptive Integrated Microsystems
                                          Impact Rate Counting
    Startup          Reference                 Channel 1                Channel 2               Channel 3
V
                M1               M5        P1           P3          P5           P7             P9        P11


      S1        M2               M6        P2           P4          P6           P8         P10           P12

                                      C                       2C                      3C
           S2
                                                        SW1                     SW2                       SW3
                                          N1                       N2                      N3
                M3               M7   R            O1         R            O2         R              O3
       C0
                                                  C1                       C2                        C3
                M4               M8

                           RF
V                                                   F1                      F2                        F3


                                                         Trigger Circuit




                                                  Adaptive Integrated Microsystems
                     Fabricated prototype
                          Lajnef, Chakrabartty, 2008




• Prototype fabricated in a 0.5 µm CMOS process.

• Die size is 1.5 mm x 1.5 mm, QCP package 1cm x 1cm.

• Total load current < 160nA, Startup time < 1ms


                         Adaptive Integrated Microsystems
       Validation




Adaptive Integrated Microsystems
                 level 1 ~ 2100 µε                        level 2 ~ 2500 µε




Piezo-electric
 transducer

                            Vref



                                     level 3 ~ 2800 µε




                                    response of the floating-gate injector array
                           Measured Adaptive Integrated Microsystems
     Impulse Monitoring




Measured response of the floating-gate injector array




                                                        www.pointblankarmor.com
        Adaptive Integrated Microsystems
                    “Smart Pebble” Architecture
  - Long-term computation of fatigue features.
  - Remote Interrogation of accumulated statistics.
  - Remote configuration and reset.

                                     Self-powered                                        Self-powered



                                         Long-term                                              Long-term
                                          Features                                               Features
                                                             Piezo
            Piezo                                            transducer
            transducer

                                               Features                                                 Features



               RFID                                                                             Classifier
               Interface               Classifier
                                                                           Inspect Bit
                           Inspect
RFID coil                  bit




                                        Adaptive Integrated Microsystems
Analog VLSI Inner product Computation
                             (Chakrabartty JSSC 2007)


   Input Block                Memory Block                   - Sub-threshold operation.

  I ref                                         I prog                     −κ (Vg −Vgref ) / U T            I prog
                                                          I out = I in e                           = I in
                                                                                                            I ref
                                                                                  Features
                                                                                                       S-N
                                            I out                                                  parameters
                                                         - Operation with pico-amperes of
                                                         current.




          - Cancels out common mode
          effects (30dB Rejection).
                          Adaptive Integrated Microsystems
 Analog VLSI Classifiers
                   (Chakrabartty JSSC 2007)




                                          • 28000 programmable
                                            parameters.
                                          • 720 classifiers, 14
                                            dimensional templates.
                                          • 840nW @ 40 classifications
720 classifiers
                                            per second.
                                          • Hardware-in-loop calibration
                                            and training.
                                          • >1012 Multiply Accumulate
                                            Operations per Watt.
      2.5mm




                  Adaptive Integrated Microsystems
               “Approximate” Analog VLSI
                                           (Kucher & Chakrabartty 2006)




                            Output Stage
          Template Vector

         Innerproduct
         Computation




                                                     • “Approximate” Gibb’s function.
                                                     • Non-linear operations become
                                                       linear.
                                                     • Linear operators approximated by
                                                       “margin normalization”.
                
z = log ∑ e f i       z : ∑ [ f i − z ]+ = 1       • Bias independent and energy
        i                           i                scalable operation (350nW @ 40
                                                       classification per second).
                                           Adaptive Integrated Microsystems
                                                                                                                 Future
                                                                   Microwatt Barrier
University of
Michigan (2007)

                                                                                                       ~ 0.08 µW


   UC Berkeley
   (2004)
   The Berkley Mote
       ~ 720µW                                                                                   MSU (2009)




             MicroStrain (2007)
                  ~ 100µW                                                               Michigan State
                                                                                       University (2007)
       -3                  -4                      -5                                  -6   ~ 0.8 µW        -7
  10                  10                      10                           10                          10        W
                                Adaptive Integrated Microsystems
 Embedded Structural Health Monitoring

                   Manufacturing

 Diagnostics                                          Packaging


                                Sensors
Prognostics                                        Micro/Nanoelectronics
               Structures


CAD Modeling                                         Power Sources

                     Networking


                Adaptive Integrated Microsystems
            References and Acknowledgement
 1.    S. Kurtz, K. Ong, E. Lau, F. Mowat, M. Halpern, “Projections of Primary and Revision Hip and Knee Arthroplasty in the United
       States from 2005 to 2030” J. of Bone and Joint Surgery, 2007; 89(4):780-785.
 2.    M.A. McGee, et. al., “Implant retrieval studies of the wear and loosening of prosthetic joints: a review,” Wear, vol. 241, No. 2,
       2000, pp. 158-165.
 3.    Wright et al., “Ulnar nerve excursion and strain at the elbow and wrist associated with upper extremity motion”, J. Hand Surg
       [Am], July 2001, 26(4):655-62
 4.    Borodkin J et al (1999), “The influence of mechanical strain magnitude on bone adaptation to porous coated implants” IEEE Eng.
       Medicine Biology Soc. Conf. p 768
 5.    Beynnon BD, Fleming BC., “Anterior cruciate ligament strain in-vivo: a review of previous work”. J Biomech. 1998;31:519–525
 6.    Lie, DTT, Gloria, N, Amis, AA, et al, “Patellar resection during total knee arthroplasty: effect on bone strain and fracture risk”,
       KNEE SURG SPORT TR A, 2005, Vol: 13, Pages: 203 - 208, ISSN: 0942-2056
 7.    B. A. Morris, D. D. D’Lima, J. Slamin, N. Kovacevic, S.W. Arms, C. P. Townsend, and C. W. Colwell, Jr., “E-Knee: Evolution of
       the electronic knee prosthesis” J. Bone Joint Surg. Am., pt. 1, vol. 83-A, no. Suppl 2, pp. 62–66, 2001.
 8.    Colville MR, Marder RA, Boyle JJ, Zarins B., ‘Strain Measurement in Lateral Ankle Ligaments”. American Journal of Sports
       Medicine. 18(2): 196-200, 1990.
 9.    Beynnon et al. “The effect of functional knee braces on strain on the anterior cruciate ligament in vivo”, J. Bone & Joint Surgery
       (American), Vol. 74(9), pages 1298-1312, October 1992
 10.   Elvin, N.G. , Lajnef, N. , Elvin, A.A. (2006), "Feasibility of Structural Monitoring With Vibration Powered Sensors", Smart Mater.
       Struct., 15, p 977-986.
 11.   N. Lajnef, S. Chakrabartty, N. Elvin and A. Elvin, Piezo-Powered Floating Fate Injector for Self-Powered Fatigue Monitoring in
       Biomechanical Implants, IEEE Symposium on Circuits and Systems (ISCAS'2007), New Orleans 2007.
 12.   N. Lajnef, N. Elvin and S. Chakrabartty, “Piezo-powered floating gate injector for self-powered fatigue monitoring in biomechanical
       implants,” Transactions of Biomedical Circuits and Systems (to appear 2008).
 13.   S. Chakrabartty, N.Lajnef, N.Elvin and A. Elvin, “Towards self-powered circuits for biomechanical Implants”, VLSI Circuits for
       Biomedical Applications”, ed. K.Inieweski, Artech House, 2008



 This work is currently being supported by a research grant from the National
Science Foundation (NSF-CMMI 0700632) and Federal Highway Administration


                                            Adaptive Integrated Microsystems

								
To top