MEMS Technologies for Timing and Frequency Controlpeople.eecs.berkeley.edu/~ctnguyen/Research/PlenaryTalks/MEMSFr… · Faster GPS acquireFaster GPS acquire Higher jamming margin

C. T.-C. Nguyen, “MEMS Technologies for Timing and Frequency Control,” FCS’05, 8/29/05

DARPADARPA

MEMS Technologies for Timing and Frequency

ControlClark T.-C. Nguyen

Program Manager, MEMS/MPG/CSAC/MX/HERMIT/MGA/RIMS/RFMIP/NGIMG/MCC

Microsystems Technology Office (MTO)Defense Advanced Research Projects Agency

(on leave from the University of Michigan)

FCS/PTTI’05, Aug. 29-31, 2005


DARPADARPA NIST F1 Fountain Atomic Clock

VolVol: ~3.7 m: ~3.7 m33

Power: ~500 WPower: ~500 WAcc: Acc: 11××1010––1515

Stab: 3.3x10Stab: 3.3x10--1515/hr/hr

Physics PackagePhysics Package

After 1 sec Error: 10-15 secAfter 1 sec

Error: 10-15 sec

Loses 1 sec every 30 million years!



DARPADARPA Benefits of Accurate Portable Timing

Secure Communications

Networked Sensors

Faster frequency hop ratesFaster frequency hop rates

Faster acquire of pseudorandom signals


Superior resilience against jamming or

interception


interception

More efficient spectrum utilization


Longer autonomy periodsLonger autonomy periods GPS

Faster GPS acquireFaster GPS acquire

Higher jamming margin


Fewer satellites needed


Larger networks with longer autonomy


Better TimingBetter Timing


DARPADARPA

State-of-Practice

State-of-Research

Datum R2000Vol: 9,050 cm3

Power: 60 WAcc: 5×10–11

Temex RMOVol: 230 cm3

Power: 10 WAcc: 1×10–11

NISTNIST-- F1F1

CSACCSAC

VolVol: 1 cm: 1 cm33

Power: 30 Power: 30 mWmWAcc: 1x10Acc: 1x10--1111

Stab: Stab: 11××1010––1111/hr/hr

Miniaturizing Atomic Clocks

VolVol: ~3.7 m: ~3.7 m33


Stab: 3.3x10Stab: 3.3x10--1515/hr/hr

μs/dayμs/day

ps/dayps/day


DARPADARPA Outline

• Introduction: Miniaturization of Transceiversneed for high-Q

• Micromechanical ResonatorsHF clamped-clamped beamsVHF free-free beamsUHF disks

• Micromechanical Resonator Oscillators• Chip-Scale Atomic Clocks• Micromechanical Circuits• Towards RF Channel-Selection• Conclusions

Chip-Scale Atomic Physics Package

Chip-Scale Atomic Physics Package

1.5-GHz μMechanical Disk Resonator1.5-GHz μMechanical Disk Resonator


DARPADARPA

Wireless Phone

Inductors Capacitors Resistors

Inductors Capacitors Resistors

Transistor Chips

Transistor Chips

RF Filter (ceramic)RF Filter (ceramic) IF Filter

(SAW)IF Filter (SAW)

RF Filter (SAW)

RF Filter (SAW)

Quartz CrystalQuartz Crystal

Problem: high-Qpassives pose a

bottleneck against miniaturization

Problem: high-Qpassives pose a

bottleneck against miniaturization

Motivation: Miniaturization of RF Front Ends


DARPADARPA

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal

Motivation: Need for High Q


DARPADARPA Motivation: Need for High Q

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal

Pre-Select FilteringPre-Select Filtering



Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal




Wireless Phone

FrequencyωRF

DesiredSignal

Mix w/ low noise oscillatorMix w/ low noise oscillatorPre-Select FilteringPre-Select Filtering



Wireless Phone

FrequencyωRF

Mix w/ low noise oscillatorMix w/ low noise oscillator

Baseband filteringBaseband filtering

All require high Q

All require high Q



DARPADARPA

• Fabrication steps compatible with planar IC processing

Surface Micromachining


DARPADARPA

• Completely monolithic, low phase noise, high-Q oscillator (effectively, an integrated crystal oscillator)

• To allow the use of >600oC processing temperatures, tungsten (instead of aluminum) is used for metallization

OscilloscopeOutput

Waveform

Single-Chip Ckt/MEMS Integration

[Nguyen, Howe 1993]

[Nguyen, Howe 1993]


DARPADARPA

Wireless Phone

Raised Inductor Q ~30-70

Raised Inductor Q ~30-70

Vibrating Resonator1.5-GHz, Q~12,000

Vibrating Resonator1.5-GHz, Q~12,000

Vibrating Resonator72-MHz, Q~146,000

Vibrating Resonator72-MHz, Q~146,000

Single-ChipRealization

Benefits of MEMS: High On-Chip Q


DARPADARPA MEMS Replaceable Components

• Next generation handsets need multi-band reconfigurabilityeven larger number of high-Q components needed

• Micromachined versions of off-chip components, including vibrating resonators, switches, capacitors, and inductors, could maintain or shrink the size of future wireless phones


DARPADARPA High-Q Resonator Needs

• High-Q best achieved via vibrating micromechanical resonators

Best if Q >300Best if Q >300

Would likeQ’s >2,000



Would likeQ’s >5,000 Would like

Q’s >10,000Would likeQ’s >10,000

Best when highest Q used

Best when highest Q used


DARPADARPA Thin-Film Bulk Acoustic Resonator (FBAR)

• Piezoelectric membrane sandwiched by metal electrodesextensional mode vibration: 1.8 to 7 GHz, Q ~500-1,500dimensions on the order of 200μm for 1.6 GHzlink individual FBAR’s together in ladders to make filters

Agilent FBAR

• Limitations:Q ~ 500-1,500, TCf ~ 18-35 ppm/oCdifficult to achieve several different freqs. on a single-chip

h

freq ~ thicknessfreq ~ thickness


DARPADARPA Vibrating RF MEMS Wish List

•• MicroMicro--scale waferscale wafer--level fabricationlevel fabricationneed >10,000 parts per wafer (for cost reasons)would like >1,000 parts per die (for performance reasons)need wafer-level packaging

•• SingleSingle--chip integrated circuit or chip integrated circuit or system capabilitysystem capability

discrete parts not interestingmust allow many different frequencies on a single-chipneed on-chip connectivityintegration w/ transistors desiredneed real time reconfigurability

•• QQ’’s >10,000 at RF would allow a s >10,000 at RF would allow a revolution in wireless capabilityrevolution in wireless capability

Frequencies should be determined by lateral

dimensions (e.g., by layout)

Frequencies should be determined by lateral

dimensions (e.g., by layout)

Best if systems can be reconfigured w/o the need

for RF MEMS switches

Best if systems can be reconfigured w/o the need

for RF MEMS switches

900 MHz

1800 MHz

433 MHz200 MHz

70 MHz

1900 MHz


DARPADARPA

Vibrating RF MEMS


DARPADARPA Basic Concept: Scaling Guitar StringsGuitar String

Guitar

Vibrating “A”String (110 Hz)Vibrating “A”

String (110 Hz)

High Q

110 Hz Freq.

Vib.

Am

plitu

de

Low Q

r

ro m

kfπ21

=

Freq. Equation:

Freq.

Stiffness

Mass

fo=8.5MHzQvac =8,000

Qair ~50

μMechanical Resonator

Performance:Lr=40.8μm

mr ~ 10-13 kgWr=8μm, hr=2μmd=1000Å, VP=5VPress.=70mTorr

[Bannon 1996]


DARPADARPA Anchor Losses

Q = 15,000 at 92MHz

Fixed-Fixed Beam Resonator

GapAnchor AnchorElectrode

Problem: direct anchoring to the

substrate anchor radiation into the

substrate lower Q

Problem: direct anchoring to the

substrate anchor radiation into the

substrate lower Q

Solution: support at motionless nodal points

isolate resonator from anchors less

energy loss higher Q

Solution: support at motionless nodal points

isolate resonator from anchors less

energy loss higher Q

Lr

Free-Free Beam

Supporting Beams

λ/4

Anchor

Anchor

Elastic WaveRadiation

Q = 300 at 70MHz

Free-Free Beam Resonator

c5


DARPADARPA

• Constructed in SiC material w/ 30 nm Al metallization for magnetomotive pickup

Lr

Wr

h

Frequency [GHz]

Mag

neto

mot

ive

Res

p. [n

V]

Design/Performance:Lr =1.1 μm, Wr =120 nm, h= 75 nm

fo=1.029 GHz, Q =500 @ 4K, vacuum

Design/Performance:Lr =1.1 μm, Wr =120 nm, h= 75 nm

fo=1.029 GHz, Q =500 @ 4K, vacuum

[Roukes, Zorman 2002]

Nanomechanical Vibrating Resonator


DARPADARPA

Mass Loading Noise

• Differences in rates of adsorption and desorption of contaminant molecules

mass fluctuationsfrequency fluctuations

Temperature Fluctuation Noise

• Absorption/emission of photons

temperature fluctuationsfrequency fluctuations

ContaminantMolecules

mass ~10-13 kg

mk

2π1

of =

Photons

volume ~10-15 m3

• Problem: If dimensions too small phase noise significant!• Solution: operate under optimum pressure and temperature

[J. R. Vig, 1999]

Scaling-Induced Performance Limitations


DARPADARPA Radial-Contour Mode Disk Resonator

VP

vi

Input Electrode

Output Electrode

io ωωο

ivoi

Q ~10,000Disk

Supporting Stem

Smaller mass higher freq. range and lower series Rx

Smaller mass higher freq. range and lower series Rx(e.g., mr = 10-13 kg)(e.g., mr = 10-13 kg)

Young’s Modulus

Density

Mass

Stiffness

RE

mkf

r

ro

121

⋅∝=ρπ

Frequency:

R

VP

C(t)

dtdCVi Po =

Note: If VP = 0V device off

Note: If VP = 0V device off


DARPADARPA High-Q Resonator Needs

• High-Q best achieved via vibrating micromechanical resonators

Vibrating resonators switch themselves

Vibrating resonators switch themselves

No need for switches!

No need for switches!


DARPADARPA

Input Electrode

Output Electrode

Disk

Supporting Stem

R11 22

VP vLO

LxCx Rx

VPCo Co

11 22

0V

Open Circuit11 22

Reconfigurable Capacitive Transducer

33

vLO

vRF11

2233

ωIF

ForceInput

ElectricalSignal Input

ωRFωLO

ωRFωLOωIF

ω

ω

FilterResponse

( ) ( )tVVxCvvF LORFRFLORFLOd ωω −−

∂∂

−= cos~21 2


DARPADARPA

-100-98-96-94-92-90-88-86-84

1507.4 1507.6 1507.8 1508 1508.2

1.51-GHz, Q=11,555 NanocrystallineDiamond Disk μMechanical Resonator

• Impedance-mismatched stem for reduced anchor dissipation

• Operated in the 2nd radial-contour mode• Q ~11,555 (vacuum); Q ~10,100 (air)• Below: 20 μm diameter disk

PolysiliconElectrode R

Polysilicon Stem(Impedance Mismatched

to Diamond Disk)

GroundPlane

CVD DiamondμMechanical Disk

Resonator Frequency [MHz]

Mix

ed A

mpl

itude

[dB

]

Design/Performance:R=10μm, t=2.2μm, d=800Å, VP=7Vfo=1.51 GHz (2nd mode), Q=11,555

fo = 1.51 GHzQ = 11,555 (vac)Q = 10,100 (air)

[Wang, Butler, Nguyen MEMS’04]

Q = 10,100 (air)


DARPADARPA All Silicon “Hollow-Disk” Rings

Frequency

vi

Support Beam

vovo

Input Electrode

Output ElectrodeNotchedSupport VP

Frequency

vi

Support Beam

vovo

Input Electrode


Refill StemRefill Stem

Frequency

vi

Support Beam

vovo

Input Electrode


Frequency

vi

Support Beam

vovo

Input Electrode


Refill StemRefill Stem

[S.-S Li, et al., MEMS’04]

1204.3 1204.4 1204.5 1204.6 1204.7 1204.8 1204.9 1205 1205.1 1205.2 1205.3-101

-100

-99

-98

-97

-96

-95

-94

-93

-92

-91

ri=11.8μm ro=18.7μm VP=8Vfo=1.205GHz Q=14,603vRF=6.32VvLO=10VPP

Frequency [MHz]

Tran

smis

sion

[dB

m]

3rd Mode via mixing measurement

Notch in UHF

1204.3 1204.4 1204.5 1204.6 1204.7 1204.8 1204.9 1205 1205.1 1205.2 1205.3-101

-100

-99

-98

-97

-96

-95

-94

-93

-92

-91


Frequency [MHz]

Tran

smis

sion

[dB

m]



Frequency [MHz]

Tran

smis

sion

[dB

m]


Notch in UHF

Q ~ 14,600 at 1.2GHzQ ~ 14,600 at 1.2GHz

Reflect energy back into the ring structure Allow high Q

in an all-silicon structure

Reflect energy back into the ring structure Allow high Q

in an all-silicon structure

Direct Support

λ/4λ/4 λ/4λ/4

Longitudinal Mode Quarter Wavelength

Supports

Longitudinal Mode Quarter Wavelength

Supports

Notched Support

Mix

ed A

mpl

itude

[dB

m]


DARPADARPA

•Freq.-Q product rising exponentially over the past years

1990 1995 2000 2005 2010

1015

109

1010

1011

1012

1013

1014

Year

Freq

uenc

y-Q

Prod

uct

Intrinsic material Q limit nowhere

in sight?

Intrinsic material Q limit nowhere

in sight?

silicon

diamond

silicon

silicon

silicon

silicon

silicon

fo = 1.51 GHzQ = 11,555

fo = 1.51 GHzQ = 11,555

silicon

μMechanical Resonator fo-Q Product


DARPADARPA

1.5GHz, Q >10,000 in air [Wang, Nguyen, MEMS’04]


Breaking Down the Device Barriers

Thermal Stability

Thermal StabilityFrequency RangeFrequency Range

NoiseNoise

Absolute ToleranceAbsolute ToleranceQuality FactorQuality Factor

Aging StabilityAging

Stability

RF MEMS CktRF MEMS Ckt

Integration w/ Transistors

High ImpedancePower Handling

High ImpedanceHigh ImpedancePower HandlingPower

HandlingIntegration w/ Transistors


PackagingPackagingPackaging


DARPADARPA Voltage-Controllable Center Frequency

Gap


DARPADARPA Excellent Temperature Stability

Uncompensated μresonator

−1.7ppm/oC

100[Ref: Hafner]

Elect.-StiffnessCompensation−0.24ppm/oC

AT-cut Quartz

Crystal at Various

Cut Angles

Top Electrode-to-Resonator Gap

Top Electrode-to-Resonator Gap

Elect. Stiffness: ke ~ 1/d3

Elect. Stiffness: ke ~ 1/d3

Frequency:fo ~ (km - ke)0.5

Frequency:fo ~ (km - ke)0.5

Counteracts reduction in

frequency due to Young’s modulus temp. dependence

Counteracts reduction in

frequency due to Young’s modulus temp. dependence

Top Metal Electrode

Resonator

On par with quartz!

On par with quartz!


DARPADARPA

71MHz Array, Rx~480Ω[Demirci, Nguyen, Trans’05]71MHz Array, Rx~480Ω

[Demirci, Nguyen, Trans’05]

473MHz, Rx~80Ω[Piazza, Pisano, MEMS’05]

473MHz, Rx~80Ω[Piazza, Pisano, MEMS’05]

13.1MHz Osc. [Kaajakari, Seppa, EDL’04]

13.1MHz Osc. [Kaajakari, Seppa, EDL’04]

16kHz Osc. [Nguyen, Franke, Howe, ‘93-’02]

16kHz Osc. [Nguyen, Franke, Howe, ‘93-’02]



18ppm over 0-70oC [Hsu, Nguyen, MEMS’02]18ppm over 0-70oC

[Hsu, Nguyen, MEMS’02] <2ppm over 12 mos.[Kim, Kenny, Trans’05]

<2ppm over 12 mos.[Kim, Kenny, Trans’05]

Breaking Down the Device Barriers

Thermal Stability

Thermal Stability

Absolute ToleranceAbsolute Tolerance

Aging StabilityAging

Stability

Integration w/ TransistorsIntegration w/ Transistors


PackagingHigh ImpedancePower

HandlingHigh ImpedanceHigh ImpedancePower

HandlingPower

Handling

RF MEMS CktRF MEMS Ckt


DARPADARPA Solid Dielectric Capacitive Transducer

Antenna

r h

Gap = d

VP

Disk Electrode Electrode

Air Gap w/ ε = εo

Solid Dielectric w/ Solid Dielectric w/ εε = = εεrrεεoo in the Gapin the Gap

DiskΔGap κ timesPermittivity ε εr times

ΔGap κ timesPermittivity ε εr times

Rx εr2/κ times

( ) 222

4

221

hVd

Qk

rR

Po

rx εωπ

⋅=

377Ω

kr = 7,350,000 N/m

5kΩ

Mismatch!

ResonanceResonance

Opportunity: increase permittivity εsquare law reduction in Rx

Opportunity: increase permittivity εsquare law reduction in Rx


DARPADARPA

Micromechanical Resonator Oscillators


DARPADARPA

A, A’ connectedB, B’ connected

io

ffo x

yz

R1E

mk

21fo ⋅∝=

ρπE ⇒ Young’s Modulusρ ⇒ DensityR ⇒ Disk Radius

Vp

RL

A’

B B’ vo

ioWine GlassDisk Resonator

Output ElectrodeAnchor

Avi

SupportBeam

InputElectrode

Vp io

C(t)

Wine-Glass Disk Resonator

node

node

node

node


DARPADARPA

Output

Input

Input

Output

Support Beams

Wine Glass Disk Resonator

R = 32 μm

Anchor

Anchor

Resonator DataR = 32 μm, h = 3 μmd = 80 nm, Vp = 3 V

Resonator DataR = 32 μm, h = 3 μmd = 80 nm, Vp = 3 V

-100

-80

-60

-40

61.325 61.375 61.425

fo = 61.37 MHzQ = 145,780

Frequency [MHz]

Unm

atch

ed T

rans

mis

sion

[dB

]

Fabricated Wine-Glass Disk

• POCl3-doped polysilicon• Perimeter suspension at nodal

points alleviates anchor loss• Very high Q > 145,000 achieved!

[Y.-W. Lin, Nguyen, JSSC Dec. 04]


DARPADARPA Phase Noise in Oscillators

• Single Sideband Phase Noise Density to Carrier Power Ratio in Oscillator, L{fm}

Offset Frequency [Hz]

Phas

e N

oise

[dB

c/H

z]

100 1k 10k 100k10

-120

-100

-80

-60

-140

⇒ Carrier Frequency⇒ Offset Frequency⇒ Signal Power

of

mf

oP

•• High High QQreduces close-to-carrier phase noise

•• High High PPooreduces far-from-carrier phase noise

{ }2

m

o

om fQ

fPkTfL ⎟

⎟⎠

⎞⎜⎜⎝

⎛

⋅⋅∝

f

Po


DARPADARPA Wine Glass Disk Oscillator

Wine-Glass Disk Resonator (Q=48,000)

VDD = 1.65 V

VSS = -1.65V

M3M4

M1 M2

MRf

Vbias2

Vbias1

M11 M12 M13 M14

Vcm

M17M18

M16M15Output

Input

M5

Shunt-Shunt Feedback Tranresistance Amplifier

Common Mode Feedback Bias Circuit

VP

vi

io

Bond Wire ConnectionBond Wire Connection

MOS ResistorMOS

Resistor

VP=12V Rx=1.5kΩVP=12V Rx=1.5kΩ


DARPADARPA

OutputOutput Custom IC fabricated via TSMC 0.35μm process

Custom IC fabricated via TSMC 0.35μm process

InputInput

200 mV

20 ns

Frequency [MHz]

Pow

er [d

Bm

]

-100

-80

-60

-40

-20

0

61.05 61.1 61.15 61.2 61.25 61.3 61.35

Time & Freq. Domain Performance

Oscilloscope Waveform

Oscilloscope Waveform

Spectrum Analyzer Output

Spectrum Analyzer Output

[Y.-W. Lin, Nguyen, ISSCC’04]


DARPADARPA

-160-140-120-100

-80-60-40-20

0

1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06Offset Frequency [Hz]

Phas

e N

oise

[dB

c/H

z] 60 MHz, Wine-Glass Disk Oscillator

Down to 10 MHzFor Comparison

[Y.-W. Lin, Nguyen, JSSC’04]

1/f 3

GSM-Compliant Phase Noise!

-130 dBc/Hz @ 1 kHz-130 dBc/Hz @ 1 kHz

-147 dBc/Hz-147 dBc/Hz

Satisfies Global System for Mobile Communications (GSM)

phase noise specifications!

Satisfies Global System for Mobile Communications (GSM)

phase noise specifications!Requires only 300 μW

and 150 x 150 μm2!Requires only 300 μW

and 150 x 150 μm2!


DARPADARPA

Resonator Epi-Poly Cap

Substrate

ContactEpi-Poly Seal

Adv: lower pressureAdv: lower pressure

Adv: higher annealing temperatureAdv: higher annealing temperatureCleaner,

stress-free, more stable

devices

Cleaner, stress-free, more stable

devices

μMechanical Device

Transistor CircuitsVacuum Chamber

[Kim, Kenny Trans’05]

Freq. stays within

±2 ppm

Freq. stays within

±2 ppm

7 months7 months

Micromechanical Resonator Aging


DARPADARPA

Chip-Scale Atomic Clocks


DARPADARPA NIST F1 Fountain Atomic Clock

VolVol: ~3.7 m: ~3.7 m33


Stab: 3.3x10Stab: 3.3x10--1515/hr/hr


After 1 sec Error: 10-15 secAfter 1 sec

Error: 10-15 sec




DARPADARPA Benefits of Accurate Portable Timing

Secure Communications

Networked Sensors

Faster frequency hop ratesFaster frequency hop rates




interception


interception



Longer autonomy periodsLonger autonomy periods GPS

Faster GPS acquireFaster GPS acquire







Better TimingBetter Timing


DARPADARPA

State-of-Practice

State-of-Research


Power: 60 WAcc: 5×10–11


Power: 10 WAcc: 1×10–11

NISTNIST-- F1F1

CSACCSAC



Stab: Stab: 11××1010––1111/hr/hr

Miniaturizing Atomic Clocks

VolVol: ~3.7 m: ~3.7 m33


Stab: 3.3x10Stab: 3.3x10--1515/hr/hr

μs/dayμs/day

ps/dayps/day


DARPADARPA Atomic Clock Fundamentals

133Cs

m = 0f = 4

m = 0f = 3

m = 1

• Frequency determined by an atomic transition energy

Energy Band Diagram

Excite e- to the next orbital

Excite e- to the next orbital

Opposite e- spins

Opposite e- spins

ΔE = 0.000038 eV

ΔE = 1.46 eV

ν = ΔE/h= 352 THz852.11 nm

ν = ΔE/ħ= 9 192 631 770 Hz


DARPADARPA Miniature Atomic Clock Design

ν = ΔE/ħ= 9 192 631 770 Hz

HyperfineSplitting Freq.

HyperfineSplitting Freq.

Sidebands

ModulatedLaser

PhotoDetector

133Cs vapor at 10–7 torr

Mod f

μwave osc

VCXO4.6 GHz

9.2GHz

4.6GHz

Atoms become transparent to light at 852 nm

Atoms become transparent to light at 852 nm

Carrier(852 nm)

λ

Close feedback loop to lock

Close feedback loop to lock

vo


DARPADARPA Chip-Scale Atomic Clock

• Key Challenges:thermal isolation for low powercell design for maximum Qlow power μwave oscillator

Atomic Clock Concept Cs or Cs or RbRbGlassGlassDetectorDetector

VCSELVCSEL

SubstrateSubstrate

GHzGHzResonatorResonatorin Vacuumin Vacuum

MEMS andMEMS andPhotonic Photonic

TechnologiesTechnologies


Power: 30 Power: 30 mWmWStab: Stab: 11××1010––1111

Chip-ScaleAtomic Clock

Laser 133Cs vapor at 10–7 torr

Mod f

μwave oscVCXO

4.6 GHzvo PhotoDetector


DARPADARPA

Pros and Cons of Miniaturization


DARPADARPA

Rth= 18 K/WCth= 40 J/K

Rth= 18 K/WCth= 40 J/K Rth= 61,000 K/W

Cth= 4.8x10-5 J/KRth= 61,000 K/WCth= 4.8x10-5 J/K

P RthCth

T = P x Rth

Cth ~ volume

Rth ~ support lengthX-section area

Macro-Scale Micro-Scale

P (@ 80oC) = 1 mWP (@ 80oC) = 1 mW

Warm Up, τ = 3 sWarm Up, τ = 3 s

P (@ 80oC) = 3 WP (@ 80oC) = 3 W

Warm Up, τ = 12 min.Warm Up, τ = 12 min.

3,000x lower power3,000x lower power

240x faster warm up240x faster warm up

300x300x300 μm3

Atomic Cell @ 80oC

Long, Thin Nitride

Tethers w/ Metal Leads

T Sensor(underneath)

Heater

LaserInsulation

Macro-Oven(containing heater

and T sensor)2 cm3

Atomic Cell @ 80oC

Thermally Isolating Feet

Laser25oC

2 cm

Micro-Scale Oven-Control Advantages


DARPADARPA Challenge: Miniature Atomic CellLarge Vapor Cell Tiny Vapor Cell

1,000XVolumeScaling

Wall collision dephasesatoms lose coherent state

Wall collision dephasesatoms lose coherent state

Inte

nsity

Mod f9.2 GHz

SurfaceVolume

More wall collisions stability gets worse

More wall collisions stability gets worse

lower Qlower Q

lowest Qlowest QAtomic

Resonance


DARPADARPA Challenge: Miniature Atomic CellLarge Vapor Cell Tiny Vapor Cell

1,000XVolumeScaling

Inte

nsity

Mod f9.2 GHz

Atomic Resonance

Soln: Add a buffer gas

Soln: Add a buffer gas

Lower the mean free path of the atomic vapor

Lower the mean free path of the atomic vapor

Return to higher Q

Return to higher Q

Buffer Gas


DARPADARPA

1.5 mm

4.2 mm

1.5 mm

Laser

Optics

Cell

Photodiode

1 mm

Total Volume: 9.5 mm3 Stability: 2.4 x 10-10 @ 1sCell Interior Vol: 0.6 mm3 Power Cons: 75 mW

Total Volume: 9.5 mm3 Stability: 2.4 x 10-10 @ 1sCell Interior Vol: 0.6 mm3 Power Cons: 75 mW

1st Chip-Scale Atomic Physics Package

GlassND

SiQuartz

ND

Lens

Alumina

VCSEL


DARPADARPA Tiny Physics Package Performance

NIST’sChip-Scale

Atomic Physics Package

NIST’sChip-Scale

Atomic Physics Package

40 50 60 70 80 905.65

5.66

5.67

PD S

igna

l [V]

Frequency Detuning, Δ [kHz] from 9,192,631,770 Hz

7.1 kHz

Contrast: 0.91% 2.4e-10 Allandeviation @ 1 s2.4e-10 Allan

deviation @ 1 s

• Experimental Conditions:Cs D2 ExcitationExternal (large) Magnetic ShieldingExternal Electronics & LO Cell Temperature: ~80 ºCCell Heater Power: 69 mWLaser Current/Voltage: 2mA / 2VRF Laser Mod Power: 70μW

DimeDime

Open Loop Resonance:Drift to Be Removed in Phase 3

Drift to Be Removed in Phase 3

Sufficient to meet CSAC

program goals

Sufficient to meet CSAC

program goals

100 101 102 103 104 10510-12

10-11

10-10

10-9

Alla

n D

evia

tion,

σy

Integration Time, τ [s]

Stability Measurement:

Drift IssueDrift Issue

Rb (D1)Rb (D1) 1 day1 day1 hour1 hour

Cs (D2)Cs (D2)

Q =1.3x106Q =1.3x106

CSAC Goal


DARPADARPA Physics Package Power Diss. < 10 mW

Heater/Sensor SuspensionCesium cell

Frame Spacer

VCSEL Suspension

VCSEL / Photodiode 20 pin LCC

7 mm

0

2

4

6

8

10

12

0 20 40 60 80 100 120 140Temperature [oC]

Pow

er [m

W] Measured

Model

Only ~5 mWheating power

needed to achieve 80oC

cell temperature

Only ~5 mWheating power

needed to achieve 80oC

cell temperature

• Achieved via MEMS-based thermal isolation

Symmetricom / Draper Physics

Package Assembly

Symmetricom / Draper Physics

Package Assembly


DARPADARPA A 9.95 cc, 153 mW Atomic Clock w/ Chip-Scale Physics Package

3.94 cm

3.53

cm


Symmetricom’smeasured Allan deviation easily satisfies CSAC Phase 2 Goal

Symmetricom’smeasured Allan deviation easily satisfies CSAC Phase 2 Goal

0.47 cm

Packaged CSAC

Power Regulation 1 mW

Microprocessor 8 mW

Signal Processing 6 mW

RF 75 mW

VCSEL drive 2 mW

Heater Power (air) 51 mW

C-field 10 mW

Total: 153 mWPhysics PackagePhysics Package

1.43 cmPower Budget < 200 mW0.64 cm

9.95 cm3 Total Package Volume9.95 cm3 Total

Package Volume

MEMS-based thermal isolation allows low physics package power consumption

MEMS-based thermal isolation allows low physics package power consumption

5x10-10

@ 1 sec5x10-10

@ 1 sec

CSAC Phase 2 Goal


DARPADARPA Atomic Clock Technology Progression

State-of-Practice

State-of-Research

HP 5071AVol: 29,700 cm3

Power: 50 WAcc: 5×10–13


Power: 60 WAcc: 5×10–11


Power: 10 WAcc: 1×10–11

NISTPP Vol: 9.5 mm3

Power: 75 mW+elect.Stab: 10–11/hr

Symmetricom CSACVol: 9.95 cm3

Power: 153 mWStab: 5×10–11/100s

NISTNIST-- F1F1

VolVol: ~3.7 m: ~3.7 m33

Power: ~500 WPower: ~500 WAcc: Acc: 3.83.8××1010––1515

Stab: 3.3x10Stab: 3.3x10--1515/hr/hr CSACCSAC



Stab: Stab: 11××1010––1111/hr/hr

Stab = Allan deviation/integration time

Stab = Allan deviation/integration time



DARPADARPA Atomic Clock Power Hogs


VCXOVCO

÷ N MicrowaveSource

10 MHz4.6 GHz

LaserVapor Cell

Total Volume: 9.5 mm3

Cell Interior Vol: 0.6 mm3

Stability: 7x10-12 @ 100sPower Cons: 75 mW



Stability: 7x10-12 @ 100sPower Cons: 75 75 mWmW

Power Cons: 5 mWPower Cons: 5 5 mWmW

Problem: PLL consumes too much power (more than 30mW limit)

Problem: PLL consumes too much power (more than 30mW limit)

Soln: replace multi-oscillator w/ a single high-Q GHz oscillator

Soln: replace multi-oscillator w/ a single high-Q GHz oscillator

Soln: eliminate the need for a microwave oscillator

Soln: eliminate the need for a microwave oscillator

Power Budget:Cell < 10 mWControl < 10 mWOsc. < 10 mW



DARPADARPA Atomic Clock Power Hogs


4.6 GHz

LaserVapor Cell



Stability: 7x10-12 @ 100sPower Cons: 75 mW



Stability: 7x10-12 @ 100sPower Cons: 75 75 mWmW



Control Voltage

Voltage ControlledμMechanical Resonator

Oscillator

Q >10,000

No PLL much lower power consumption

No PLL much lower power consumption

Q >10,000 at GHz freqs. suppresses transistor thermal dependence while

allowing good phase noise performance

Q >10,000 at GHz freqs. suppresses transistor thermal dependence while

allowing good phase noise performance … but might still need

frequency division… but might still need

frequency division

Power Cons: 5 mWPower Cons: 5 5 mWmW


DARPADARPA

Integrated Micromechanical Circuits


DARPADARPA Micromechanical Filter Design Basics

RQ

vo

vi

RQ

VP

ω

xovi

ωo ω

xovi

ωo ω

vovi

ωo ω

vovi

ωo

Disk Resonator

Coupling Beam

Bridging Beam

Termination Resistor

Loss Pole

Loss Pole


DARPADARPA

-60

-50

-40

-30

-20

-10

0

8.7 8.9 9.1 9.3Frequency [MHz]

Tran

smis

sion

[dB

]

Pin=-20dBm

In Out

VP

Sharper roll-off

Sharper roll-off

Loss PoleLoss Pole

Performance:fo=9MHz, BW=20kHz, PBW=0.2%

I.L.=2.79dB, Stop. Rej.=51dB20dB S.F.=1.95, 40dB S.F.=6.45

Performance:fo=9MHz, BW=20kHz, PBW=0.2%

I.L.=2.79dB, Stop. Rej.=51dB20dB S.F.=1.95, 40dB S.F.=6.45

Design:Lr=40μm

Wr=6.5μm hr=2μm

Lc=3.5μmLb=1.6μm VP=10.47VP=-5dBm

RQi=RQo=12kΩ

[S.-S. Li, Nguyen, FCS’05]

3CC 3λ/4 Bridged μMechanical Filter

[Li, et al., UFFCS’04]


DARPADARPA

-60

-50

-40

-30

-20

-10

0

8.7 8.9 9.1 9.3

High-Order Micromechanical Filter

[Wang, MEMS’97]

High-Order Micromechanical Filter

[Wang, MEMS’97]HF Micromech. Filter[Bannon, IEDM’96]

HF Micromech. Filter[Bannon, IEDM’96]

fo = 340kHzIL < 0.6dBRQ = 364kΩ

fo = 340kHzIL < 0.6dBRQ = 364kΩ

Frequency [MHz]Tr

ansm

issi

on [d

B]

Sharper roll-offs

Sharper roll-offs

Bridged μMech. Filter[S.-S. Li, UFFCS’2004]Bridged μMech. Filter[S.-S. Li, UFFCS’2004]

fo = 9.3MHzIL < 2.8dBRQ = 12kΩ

fo = 9.3MHzIL < 2.8dBRQ = 12kΩ

fo = 7.8 MHzIL < 2dBRQ = 14.7kΩ

fo = 7.8 MHzIL < 2dBRQ = 14.7kΩ

Loss poleLoss pole

• MEMS Filters excellent insertion loss• Problem: High Impedance & poor power handling

Demo’ed Micromechanical FiltersMD2


DARPADARPA

RF Channel Selection



Antenna

Demodulation Electronics

The higher the Q of the Pre-

Select Filter the simpler the demodulation

electronics



electronics

Pre-SelectFilter in the GHz Range

Presently use resonators

with Q’s ~ 400


with Q’s ~ 400

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal



Antenna




electronics



electronics



with Q’s ~ 400


with Q’s ~ 400

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal


DARPADARPA Motivation: Need for Q’s > 10,000

Antenna




electronics



electronics



with Q’s ~ 400


with Q’s ~ 400

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal

If can have resonator

Q’s > 10,000


Q’s > 10,000



Antenna




electronics



electronics



with Q’s ~ 400


with Q’s ~ 400

Wireless Phone

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal


Q’s > 10,000


Q’s > 10,000



Antenna




electronics



electronics



with Q’s ~ 400


with Q’s ~ 400


Q’s > 10,000


Q’s > 10,000

Wireless Phone

Non-Coherent FSK Detector?(Simple, Low Frequency, Low Power)

Front-End RF Channel Selection

Front-End RF Channel Selection

Rec

eive

dPo

wer

FrequencyωRF

DesiredSignal

Substantial Savings in Cost and Battery PowerSubstantial Savings in

Cost and Battery Power


DARPADARPA RF Channel-Select Filter Bank

Bank of UHF μmechanical

filters

Bank of UHF μmechanical

filters

Switch filters on/off via

application and removal of dc-bias VP, controlled by

a decoder

Switch filters on/off via

application and removal of dc-bias VP, controlled by

a decoder

Tran

smis

sion

Freq.

Tran

smis

sion

Freq.Tr

ansm

issi

onFreq.

1 2 n3 4 5 6 7RF Channels


DARPADARPA Conclusions

•• Size reduction via MEMS have achievedSize reduction via MEMS have achievedQ’s >10,000 at GHz frequencies in sizes less than 20 μm in diameter and w/o the need for vacuum encapsulationTCf’s < -0.24 ppm/oC (on par with quartz)aging at least on par with quartzcircuit-amenable characteristics VLSI potential

•• MEMS technologies have now successfully harnessed atomic MEMS technologies have now successfully harnessed atomic properties pursuant to precise timingproperties pursuant to precise timing

<10cc atomic clocks consuming < 200mW have been <10cc atomic clocks consuming < 200mW have been demodemo’’eded with 5ewith 5e--11 @ 100s Allan deviation11 @ 100s Allan deviationwork on true chipwork on true chip--scale atomic clocks underwayscale atomic clocks underway

•• Time to turn our focus towards mechanical circuit design and Time to turn our focus towards mechanical circuit design and mechanical integrationmechanical integration

maximize, rather than minimize, use of high-Q componentse.g., RF channelizer paradigm-shift in wireless designdoes frequency domain computation make sense?

•• Need more involvement from the frequency control communityNeed more involvement from the frequency control community


DARPADARPA

• UC Berkeley:Roger Howe, Al PisanoSiGe & AlN resonators

• Hughes Research:Randy Kubenaquartz resonators

• Cal Tech: Michael Roukesnanomechanical resonators

• Georgia Tech:Farrokh AyaziSOI resonators

• Stanford:Tom Kennyepi/SOI resonators

fo = 224.6 MHzQ = 2,580

fo = 224.6 MHzQ = 2,580

fo = 1.03 GHzQ = 500

fo = 1.03 GHzQ = 500

Growing Vibrating RF MEMS Research

Documents

MEMS Technologies for Timing and Frequency Controlpeople.eecs.berkeley.edu/~ctnguyen/Research/PlenaryTalks/MEMSFr… · Faster GPS acquireFaster GPS acquire Higher jamming margin