Non-Linearroblin/lsna/OSUresearch/roblin2.pdflarge-signal RF and microwave design process. With its strong presence in measurement, modeling and simulation, Agilent is uniquely positioned

'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

2003 DURIP Equipment Program

Non-Linear RF Research and Educational Laboratory

PI: Patrick Roblin (MISES, SSEP)

CoPI: Roberto Rojas (ESL)

CoPI: Mohammed Ismail (MISES)

CoPI: Len Brillson (SSEP)

CoPI: Wu Lu (SSEP)

CoPI: Lee Potter (IPS)

CoPI: Oscar Takeshita (IPS)

Department of Electrical Engineering

The Ohio State University

Agency:

Air Force O�ce of Scienti�c Research

Dr. Witt

MISES/SSEP Microwave Laboratory The Ohio State University

'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Presentation

� Overview of DURIP equipment proposal - Patrick Roblin

{ DURIP team

{ DURIP equipment requested

{ Need for equipment and examples of applications

� Impact of Equipment: from systems, to devices, to materials (a top down approach)

{ Signal processing research - Lee Potter

{ Wireless testbed - Oscar Takeshita:

{ RFIC design - Mohammed Ismail

{ Mixed signals & UWB radar-on-chip - Steve Bibyk

{ Phase array antenna - Roberto Rojas

{ Projects relevant to non-linear RF - John Volakis

{ Designing GaN devices - Wu Lu

{ Defects and traps in GaN material/devices - Len Brillson

{ Electronic traps in III-Nitride- Steve Ringel


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Non-Linear RF DURIP Team

Electromagnetism

(SSEP)Materials and Devices

(IPS)

Signal Processing

(MISES)

Microwave, RFIC

Antenna

(ESL)

Communication

RFLaboratory

Non−LinearMixed Signals

Group Full Name Number of Faculty Members

ESL ElectroScience Laboratory 11

IPS Information Processing System Lab 8

MISES Mixed Signal Electronic Laboratory 4

SSEP Solid State Electronics and Photonics 8

CEA Postdoc coming from Sept. 2003-2004.

EE Department committed release time for center development.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Equipment Requested

IsothermalDevice Models

Non−linear RFBehavioral

System Models

Device ModelsIsothermal

Non−linear RF

Thermal Modeling

LSNAPulsed−RF

Pulsed−IV

Probe Station

Infrared Camera

DUT


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

DURIP Equipment Budget

Equipment: Source Cost Revised Cost

Large-signal network analyzer: Agilent $ 367,246 $ 300,000

Pulsed RF system: Agilent $ 506,025 $ 358,000

Infrared&Cascade probe station: Cascade&Indigo $ 156,530 $ 50,000

Total Equipment $1,029,801 $ 708,000

OSU cost-share $ 514,878 $ 354,000

Requested DoD share $ 514,923 $ 354,000


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Location of LNSAs in the World

13 LSNAs are in use around the world now.

Country Institution Number of LSNA

US Agilent (demo, design, modeling) 3 LSNA

NIST 1 LSNA

Belgium Katholieke Universiteit Leuven, Dept. TELEMIC 1 LSNA

Vrije Universiteit Brussel, Dept. Electrical Eng. 1 LSNA

Agilent-Europe 3 LSNAs

France IRCOM - Universite de Limoges 2 LSNAs� + 1 pulsed-RF

Ireland National Micro-Electronics Research centre 1 LSNA

Singapore National University of Singapore 1 LSNA

� IRCOM has just acquired a second LSNA dedicated to system measurement. Their �rst LSNA is

dedicated to non-linear pulsed-RF measurement.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Wireless Technology Trends

Continuous push for:

� Higher frequencies of operation

� Wide and ultra-wide bandwith communication and radar systems

� Highly linear electronics to conserve spectrum resource

� High power handling (base-station, satellite)

� E�ciency (talk time, satellite, linearization)

� New modulation and code techniques (constant envelope, vector modulation)


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Non-Linearities in RF Systems

Non-linearities arise from the complex dependence of the device current sources (IVs) and

charges (QVs) upon their state-variables such as terminal voltages, traps state, internal device

temperature Tdev, body voltages Vb and so on.

−5 −4 −3 −2 −1 0 1 2 3 4 5−140

−120

−100

−80

−60

−40

−20

0

20

40IP5 Theoretical CDMA Spectrum × Bandwidth B

Pow

er L

evel

(dB

m)

Normalized Frequency (f−f0)/B

Pout=30 dBmPout=20 dBmPout=10 dBm

Non-linearities leads to intermodulation and spectral regrowth

Spectral regrowth depends on the 3rd and 5th derivatives of the current sources

and charges!


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Need for Higher Derivatives in Behavioral Modeling

−20 −15 −10 −5 0 5 10 15 20 25 30−120

−100

−80

−60

−40

−20

0

20

Pin

(a1)

Pou

t(a1)

IMD Internal Output Power

−20 −15 −10 −5 0 5 10 15 20 25 30−120

−100

−80

−60

−40

−20

0

20IMD Internal Output Power

Pin

(a1)

Pou

t(a1)

The LSNA provides the required phase and amplitude of the in-band

intermodulation products to account for the higher derivatives.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Need for Higher Derivatives in Device Modeling

−10 −5 0 5 10 15 20 25−40

−30

−20

−10

0

10

20

30

Input Power (dBm)

IMD

3 (d

Bm

)

Comparison of IMD3 between Different Models

OSUFETOSUMET

−10 −5 0 5 10 15 20−70

−60

−50

−40

−30

−20

−10

0

10

20

Input Power (dBm)

IMD

(dB

m)

Comparison of IMD between Different Models

MET OSUFET − k=4, it=30 OSUFET − k=5, it=1 OSUFET − k=5, it=300OSUMET

(a) (b)

(a): Absence of higher derivatives leads to a 10 dB error in IMD.

(b) Using 3rd order derivative continuity gives a closer prediction of the IMD.

The LSNA will provide the missing higher order derivatives.


Large Signal Network Analyzer

Characteristics:RF bandwidth: 0.6 to 20 GHzmax RF power: 10 WattModulation bandwidth: 8 MHzModulation: must be periodic

Building blocks:(1) test-set(2) down-converter(3) data acquisition(4) CW source

or modulation source(4)

(3)

(2)

(1)

Measurement Examples

CW in frequency domaina1, b1, a2, b2 (FET @ 2.4 GHz)

CW in time domainv1, i1, v2, i2 (FET @ 2.4 GHz)

-50-40-30-20-10

010

0 1 2 3 4 5 6 7 8

b2 /

dBm

Harmonic Index

-50-40-30-20-10

010

0 1 2 3 4 5 6 7 8

a2 /

dBm

-50-40-30-20-10

010

0 1 2 3 4 5 6 7 8

b1 /

dBm

Harmonic Index

-50-40-30-20-10

010

0 1 2 3 4 5 6 7 8

a1 /

dBm

0

8

16

24

32

0 0.5 1 1.5 2

Ids

/ mA

Time / period

0

0.5

1

1.5

2

0 0.5 1 1.5 2

Vds

/ V

-2

-1

0

1

2

0 0.5 1 1.5 2

Igs

/ mA

Time / period

0

0.5

1

1.5

2

0 0.5 1 1.5 2

Vgs

/ V

ModulationTime Domain Frequency Domain

Fundamental envelope3rd harmonic envelope

Spectral regrowth

2

What is large-signal analysis? S-parameters are commonly used to describe small signal device behavior. They are extremely valuable as they offer a mathematical representation of linear device behavior. The application of S-parameters allows circuit designers and modelers to achieve excellent correlation between simulation and measurement. However, S-parameters only apply to linear devices and systems, where the superposition principle is valid. Large-signal environments usually push devices into their nonlinear operating regions, and S-parameters and the superposition principle no longer apply. Researchers and design engineers face a significant problem when trying to translate high- level system specifications such as spectral regrowth, adjacent-channel-power-ratio (ACPR) and third-order-intercept (IP3) into concrete device and process parameters. There is an urgent need for a framework which deals with large-signal behavior from device to

system level in a coherent way, that can be applied to measurements, modeling and simulation. Imagine the increase of productivity in the design flow, if process engineers could work with circuit designers and system level engineers based on the same measurement tools and methods. A long-term cooperative effort between Agilent and several research institutions is under way with the goal of developing a new analysis framework for large-signal RF and microwave design process. With its strong presence in measurement, modeling and simulation, Agilent is uniquely positioned to lead this effort. Agilent’s large-signal network analyzer has allowed participating researchers to gain new

insight into their device’s behavior under large signal conditions. Now, for the first time, this measurement system is being made available to Agilent’s customers. Voltage and current relationships are the fundamental information used by computer-aided engineering (CAE) tools and device-level models. What has been missing is a way to accurately measure voltages and currents at the device under test’s (DUT’s) ports under realistic stimulus and load conditions. The large-signal network analyzer takes a first step towards bridging the gap between measurement, modeling and simulation, from device level to system level.

The large-signal network analyzer measures incident and reflected waves at fundamental, harmonic and modulation

frequencies.

ReferencePlanes

5 0 Ohm

Acquis it ionStimulus

Response

D UT

8

The concept of dynamic loadline is very useful for microwave power amplifier designers. The dynamic loadline represents current versus voltage at the output of a transistor for the fundamental and harmonics, under given bias conditions and input and output match. Once the dynamic loadline is

determined, the designer can easily tune bias parameters, input power and output impedance to optimize output power or power-added efficiency (PAE). Up to now, dynamic loadline information was only available from advanced simulators, and depended therefore on the accuracy of the large-signal models of

the device. The large-signal network analyzer makes it possible for the first time to accurately measure the dynamic loadline directly, without the dependence on accurate models. This allows faster and more efficient design cycles.

Time domain waveforms Frequency domain

gate drain

voltage

current

gate drain

Voltage - Current State Space

A GaAs HEMT is shown in voltage and current time domain view, as well as in voltage – current state space and frequency domain

For additional application information and technical papers about Large-Signal Network Analysis, please go to http:// www.agilent.com/find/lsna

'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Memory E�ects

Memory e�ects are low-frequency dispersion e�ects (DC to MHz) associated with

various physical e�ects:

� Deep level traps (GaN, SiC)

� Self-heating (thermal distributed network) (HBT, LDMOSFET, SOI-MOSFET)

� Parasitics bipolar transistor & impact ionization (SOI-MOSFET, HEMT)

Because they respond to the RF modulation envelope (DC-MHz), memory e�ects

are hampering the development of wide-bandwidth modulation (WCDMA, UWB

radar).

� They introduce time-varying non-linearities leading to further performance degradation.

� Present linearization schemes used in power ampli�ers are not e�ective for wide-bandwith

modulation due to memory e�ects.

� Present device models are not capable of accurately accounting for non-linearities and

memory e�ects, limiting their usefulness in circuit design.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Modeling of Memory E�ects in Devices

Internal Device

Q

T

thCinstP

TrapsSelf−heating

dev

Tsub

B

+

−

+_

+

Parasitic bipolar

Intrinsic FET

Drain

A

GI

Gate

DQG

IC

I ii

EmitterBase(Body)

Collector

Source

−

+

bd

Vds

Cbd

Cbs

Rbs

RTemperature

Rd

bs

Rth,dev

instP

−

+

V

Rth,sub

I wii−

� Both analytic and table (B-spline) representations can be used

� Models parameters extracted from pulsed-IV and pulsed-RF measurements

� Models implemented in ADS


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Parasitic Bipolar E�ect in SOI-MOSFET

0 0.5 1 1.5 2 2.5 3 3.5 40

100

200

300

400

500

600

700

Drain−Source Voltage (V)

Dra

in C

urre

nt (

uA)

Vgs = 1 to 5 V (0.5 V steps)

Circle: Iso-thermal DC-IV, Plain line: Pulsed-IV.

Both plots are isothermal! Traps and parastic bipolar-transistor remain frozen in pulsed-IV

measurements but not in isothermal DC-IV.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Self-Heating in a LDMOSFET

0 5 10 15 20 25 300

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

VDS

(V)

I D (

A)

Vgs

= 4 to 6.5 V (0.25 V steps)

60 60 60 61 61 61 62 62 63 63 64 64 656061 62 63 64 65 66 68 69 71 73 75

78

60

62 65 67 70 72 7679 83 88

93 99106

61

64 68 7378 83 89

96 104113 122

134 147

61

66 7280 88

96 106116 128

140 155170 186

61

67 7787 98

110 123137 151

167 184

61

6981 95

109 124140 157

174 193

61

69

86 103 120138 157 176

60

70

89 110 130 151 172

61

70

92 116 140 163 186

60

70

95 122 149 175

Electro-thermal DC IV characteristics Tsub = 60�C.

The device surface temperature Tdev is indicated.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Pulsed IV Measurements

0 5 10 15 20 25 300

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Dra

in C

urre

nt in

A

Drain Voltage Vds

in V

Pulsed I−V Characteristics for VGS

=4.75 V and VDS

=13.7 V and Tsub

=35 C

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1IV MEASUREMENT−DEV TEMP=45C to 105C

Vds

Ids

device temp=45 device temp=60 device temp=75 device temp=90 device temp=105

Pulsed-IVs showing the starting bias point Pulsed-IVs from 45 to 105 oC

Repetition rate 10 kHz, Pulsed duration 1�s, Duty rate: 1% Pulsed-IV have a double bias

dependence!


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Pulsed RF system

� OSU system (same principle as IRCOM's system, but 2-6 GHz)

� Agilent Pulsed RF system (2-20 GHz) (dedicated system, high-performance)

� Agilent LSNA can be used for doing pulsed RF (same dynamic range but slower). LSNA

acquires non-linear pulse response as well.

HP8592A

RF RFout in

A B

port1 port2

DUT

2GHz

test port1 test port2

HP85047A

pulse

S−parameter set

modulatorHP11720A

network analyzer

HP8753E

AV−1011−C−OP1Bpulse generator

spectrum analyzer


I. R. C. O. M. University of LIMOGES (FRANCE) PULSED LSNA 16/31IRCOM

LSNA WITH PULSED Fo SIGNAL (3/5)LSNA WITH PULSED Fo SIGNAL (3/5)

t

FiF

Pulses

The first sample point during the second pulse (B) should be at the phase of the first point outside (A), in order to obtain an array of samples suitable for FFT.Consequences :

- the LSNA calibration procedure from Agilent NMDG is suitable,- AdcF, Fo, sF, and To parameters are not independent,

- The AdcF clock must be phase-locked with RF synthesizer 10 MHz to avoid a large phase shift between A-1 and B point.

A : Next sample point without pulses B : Next sample point with pulses

'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Need for Infrared Thermal Imaging of Transistors

Modeling of distributed thermal e�ects in FETs

Au

P2 P3 P4 P5 P6 P7P1Chip

z

y

x

L

W

(Xs,Ys)

ZsZ1

k

k

1

2

deviceactive

columns

Package

20mm

Adiabatic Plane

R13

R12

C1R1 C2

R2

LDMOSFET

R23

C3R3

Thermal Model. . .

Electrical Model

Image method is applied to a 140mm 7 �nger FET


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Lateral Temperature Distribution Across the FET

−1000 −500 0 500 10000

10

20

30

40

50

60

70

Y axis (µm)

Tem

pera

ture

Ris

e (o

C)

Temperature Distribution of a Simulated 140mm LDMOSFET

Z=0Z=50µmZ=200µmZ=500µm

10 oC drop for a 70 oC rise.


Fig. 2 Image magnification sequence, from left to right, 1/5X, 1X, 5X and 25X, including visible overlay

Fig. 3 Quadrant display showing unpowered radiance image (upper left), powered radiance image (upper right), emissivity matrix(lower left) and “true temperature” thermogram (lower right)

. . .

Fig. 4 Typical InfraScope II screen display illustratingline scan and statistical display (upper left)

Fig. 5 Navigation image showing superposition of thermalinformation (color) over visible image (black and white)

'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Need for Higher-Order Distributed Thermal Network

Transient and steady-state responses of the device average temperature.

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

300

310

320

330D

evic

e te

mpe

ratu

re in

(K

)

(A)

0.094 0.095 0.096 0.097 0.098 0.099 0.1325

326

327

328

329

330

331

(B)

Time in (Sec)

Dev

ice

tem

pera

ture

in (

K)

A single RC thermal model is su�cient for DC and RF but NOT for modulated RF.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Example of Linearization Accounting for Thermal Memory

Thermal Memory Prediction

bb

IQ Modulator

dev

P

inst

sub

+

inst T sub

T

PP

bbbb dev dev

T

devT

I Q

Square−law detector

FilterLow pass

I (P ,T ) Q(P ,T )

PA

−

RF predistortion System

The real device temperature Tdev is estimated and used to update the temperature dependent

linearization while accounting for memory e�ect.


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Applications of DURIP equipment

The DURIP equipment requested will permit to

� qualify device/system linearity for wide-bandwidth system applications � (LSNA: Lu,

Takeshita)

� measure time constants associated with memory e�ects� (Pulsed RF: Brillson, Ringel, SSEP)

� develop DC to RF device models accounting for memory e�ects (Pulsed RF: Lu, Roblin)

� develop device models accurately accounting for non-linearities (LSNA+pulsed RF, Roblin)

� develop and test improved ampli�er linearization schemes accounting for memory e�ects

(LSNA: Serrani, MISES, IPS)

� develop and test improved non-linear system models (LSNA: IPS, MISES)

� develop and test non-linear RFIC (LSNA: Ismail, Rojas, Volakis)

� improved DAQ and RF signal processing (LSNA: Bibyk, Potter )

� test impact of non-linearities on wide-bandwidth modulation (LSNA: Takeshita, IPS)

� Application to GaN devices: Correlate memory e�ects observed in RF circuits with traps and

defects independently identi�ed in our SSEP labs (Brillson, Ringel) for improved material growth

(Ringel) and device fabrication (Lu).


'&

$%

OHIOSTATE

T . H . E

UNIVERSITY

Presentation

� Overview of DURIP equipment proposal - Patrick Roblin

{ DURIP team

{ DURIP equipment requested

{ Need for equipment and examples of applications

� Impact of Equipment: from systems, to devices, to materials (a top down approach)

{ Signal processing research - Lee Potter

{ Wireless testbed - Oscar Takeshita:

{ RFIC design - Mohammed Ismail

{ Mixed signals & UWB radar-on-chip - Steve Bibyk

{ Phase array antenna - Roberto Rojas

{ Projects relevant to non-linear RF - John Volakis

{ Designing GaN devices - Wu Lu

{ Defects and traps in GaN material/devices - Len Brillson

{ Electronic traps in III-Nitride- Steve Ringel


loadline.pdfattachmentOpen.pdfAgilentHardware OverviewWhat is large-signal analysis?

Voltage and current relationships are the fundamental information used by computer-aided engineering (CAE) tools and device-level models. What has been missing is a way to accurately measure voltages and currents at the device under test’s (DUT’s) portsThe large-signal network analyzer

CalibrationSoftware

Application examples

IRtransistor.pdfattachmentOpen.pdfAgilentHardware OverviewWhat is large-signal analysis?


CalibrationSoftware


IR2transistor.pdfattachmentOpen.pdfAgilentHardware OverviewWhat is large-signal analysis?


CalibrationSoftware


Documents

Non-Linearroblin/lsna/OSUresearch/roblin2.pdflarge-signal RF and microwave design process. With its strong presence in measurement, modeling and simulation, Agilent is uniquely positioned