A C-Band GaAs-pHEMT MMIC Low Phase Noise VCO for Space Applications Using aNew Cyclostationary Nonlinear Noise Model

Corrado Florian*, Pier Andrea Traverso*, Marziale Feudale\ and Fabio Filicori*

*DEIS - Department of Electronics, Computer Sciences and Systems, University of Bologna, 40136, Italy +TAS-I - Thales Alenia Space Italia, Rome, 00131, Italy

Abstract This paper describes the design and implementation of a C-band MMIC VCO developed in the framework of activities oriented to the improvement of products for space applications. The circuit exploits a single device with a microstrip integrated resonator coupled with varactors. The exploited technology is a space-qualified GaAs 0.25-um pHEMT process. The MMIC exhibits 350-MHz bandwidth at 7.3 GHz, with 14 dBm output power and -86 dBclHz single side-band phase noise at 100 kHz from the carrier. Measured performances are in good agreement with simulations.

The active device adopted for the design was characterized in terms of both low-frequency noise in quiescent bias-dependent operation and its up-conversion into phase noise under large­signal RF oscillating conditions, using in-house developed measurement setups. A new compact nonlinear noise model was identified, implemented and exploited for phase noise simulations. The model features cyclostationary equivalent noise generators. Comparisons between measurements and simulations show that the nonlinear cyclostationary modeling approach is more accurate than conventional noise models in oscillator phase noise analyses of pHEMT based circuits.

Index Terms - Voltage controlled oscillator, nonlinear noise model, oscillator phase noise.

I. INTRODUCTION

The main purpose of this work was to develop methodologies

and models to design low phase noise (LPN) VCOs for space

applications exploiting a pHEMT process. It is a matter of fact

that the best performance in terms of LPN capability can be

achieved (at least up to Ka band) by using HBT technologies,

which feature significantly lower levels of low-frequency (LF)

noise (i.e., flicker and generation-recombination (G-R) noise)

of the active device: in fact vertically conductive devices are

less sensitive to surface defects and trapping/de-trapping

phenomena, which are responsible for LF noise, rather than

microwave FET devices (e.g., HEMTs, pHEMTs, mHEMTs),

whose conduction is mainly along the device surfaces and

transition regions. Nonetheless, the interest in VCOs based on

pHEMT processes, especially for space applications, lies in

other important aspects. First, the possibility to integrate such

function in MMICs usually developed in PHEMT technology.

Indeed a VCO developed in PHEMT technology could be

conveniently integrated with many other components of a

receiver or a transmitter like mixers, LNA and power

amplifiers working up to very high frequencies. Another

978-1-4244-7732-6/101$26.00 ©201 0 IEEE 284

specific advantage lies in the possibility to use the same

process to develop more functions reducing also the cost of

qualification. Moreover it must be considered that pHEMT

has more heritage in space-qualification than HBT and, due to

its excellent characteristics in terms of RF low noise

performance, gain and power handling capabilities up to 60

GHz and beyond, it is expected that this process will easily

allow the integration of VCOs inside an integrated front end

with the advantage to operate at wider frequency ranges. The

VCO described in this paper is very attractive as a building

block for "on board" communications front ends and

frequency generation units, since with this kind of integration

it is possible to eliminate any external resonator and

consequently avoid time consuming tuning operations.

An important part of the activity is represented by the

characterization of the active device LF noise and the

development of a new nonlinear noise model aimed at the PN

simulation of the VCO.

II. TECHNOLOGY OVERVIEW

The selected technology is a 0.25-um gate length GaAs

pHEMT space-qualified process. The process is optimized for

HPA and LNA design up to 50 GHz. The main characteristics

are: Idss=300 mA/mm, gm=400 mS/mm, Vp=- l V, Vbd= 15

V, FT= 55 GHz. The process features also a complete selection

of passives: TaN and GaAs resistors, spiral inductors, via

holes, MIM capacitors. The substrate thickness is 100 urn. The

process features a minimum NF of 1 dB at 10 GHz. Three

metal layers are available for the micros trip lines, enabling the

development of high Q passive structures. Schottky diodes are

available for mixer design, which exploit the pHEMT device

gate Schottky junction, and they can also be conveniently

exploited as tuning varactors.

III. LF NOISE CHARACTERIZATION AND MODELING

For PN analyses of the circuit, a compact noise model is

needed, which adequately describes the active device LF noise

up-conversion to microwaves. In most of the traditional

approaches a set of equivalent noise (EN) generators (current

and/or voltage) is applied to the device deterministic model,

whose parameters are identified by means of LF measurement

of short-circuit noise current (or open-circuit noise Voltage)

IMS 2010

spectral densities at the device ports, at different quiescent

bias conditions. The obtained EN generators describe

(independently of the number, topology and analytical

definition chosen) the LF noise behavior of the device at the

intrinsic ports in terms of stationary stochastic processes,

whose statistical moments are controlled simply by the device

quiescent bias point. Such an approach involving stationary,

bias-controlled EN generators has demonstrated in many

occasions to be not adequate enough to the accurate prediction

of the nonlinear up-conversion of LF noise into PN, mainly

because it does not take into account the well-assessed

interaction of the oscillating device large-signal (LS) working

point with the microscopic LF noise sources distributed within

the device, which are equivalently described at the electrical

terminals through the EN generators. In order to take into

account the modulation phenomena suffered by the noise

sources in LS operation, a qualitatively different approach is

needed, by considering compact EN generators controlled by

the device LS working point, whose statistical properties are

thus non-stationary from a general standpoint, being

cyclostationary under periodic LS operating conditions. The

basic principles of the cyclostationary approach have been

previously applied to LF noise non-linear modeling of HBTs

by different authors [ 1]-[3], whereas it is here applied for the

first time to pHEMT devices. In particular, following a

generalized methodology which represents an extension to

FETs of the Charge Controlled Non-linear Noise (CCNN)

modelling approach described in [ 1] for bipolars, the intrinsic

pHEMT can be eventually represented (uniquely to the aim of

noise modelling) in terms of two noiseless nonlinear networks

(Fig. 1), describing its resistive and capacitive nonlinear

characteristics, plus four cyclostationary EN generators, which

are non-linearly controlled by the LS instantaneous working

point:

dV�)(t) = I wi�L[VG(t), vD(t)]-Xk,r(t) (k,r) (a=G,D) ( 1)

dig) (I) = I W�L[Va</), VD(t)]-Xk,r(t) (k,r)

In Eq. ( 1) Xk,r'S are elementary, independent stationary

"colored" stochastic processes, each featuring a normalized spectral density characterized by one of the possible "shapes" (index k) typical of LF noise (flicker, G-R with a given comer frequency) and assumed to act in a particular region of the device structure (index r) , while functions ware the nonlinear modulation laws that describe the dependence of the EN

sources from the instantaneous electrical regime [vG (I), vD (I) ] at the device intrinsic ports. Several simplifYing assumptions can be adopted and straightforward analytical/circuital transformations performed in order to achieve a simplified description for the EN generators, starting from the general formulation of Eq. (1).

978-1-4244-7732-6/101$26.00 ©2010 IEEE 285

S Fig. I. Topology of the proposed noise model EN generators.

In particular, in this work only three elementary processes

(XF,xGRl,XGR2) have been considered: one for flicker and two

having different comer frequencies (fGRl ,fGR2) for G-R

noise, respectively. The corresponding normalized spectral densities assumed are:

SXF =11 f

SxGRl = ( 1 + (f 1 laRl)2 r' ; SxGR2 = ( 1 + (f 1 laR2)2 r' (2)

As far as the modulation laws ware concerned, a simplified dependence on only the intrinsic conductive drain current

iDR (I) in the form B· iiR (I) has been considered. Finally,

for the given pHEMT technology employed in the VCO

design, the EN generator dV}j) has been found to be

negligible. As an example, the analytical expression for the EN current generator on the drain intrinsic port is provided:

dig) (I) = [ BD,F . i�/( (I) ] XF (I) + (3)

+ [ BD,GRl . i�RGRI (t) J XGRl (t) + [ BD,GR2 . i�RGR2 (I) ] xGR2 (I)

The set of cyclostationary EN generators can be preliminary

identified by means of conventional, bias-varying LF noise

measurement. However, on the basis of LF noise data only, a

family of noise models is actually identified, each reproducing

exactly the same LF noise behavior in quiescent operation, but

featuring very different PN predictions under LS oscillation.

More precisely, the unambiguous extraction of the noise

model parameters requires additional empirical information,

which can be obtained (e.g.) either from direct measurement

of PN of an oscillator [5], or residual PN of an amplifier [8],

made up with the OUT. Indeed only the presence of a LS RF

regime, capable of exciting the microwave response of the

device capacitive network, can allow to solve the ambiguity

on the different weights to be assumed for the EN generator

controlling functions w left by the LF noise measurements.

The LF noise characterization of the device in quiescent

operation was carried out by using an in-house developed

measurement setup summarized in [4], which is based on

transimpedance amplifiers to directly measure bias-dependent

short-circuit noise currents. The active bias tee featured by the

setup makes it possible to characterize the OUT even at the

IMS 2010

high current levels, which are dynamically stimulated at the

device ports under actual LS oscillating conditions. LF short­

circuit noise current spectra at V D=4 V are shown in Fig. 2 for

different drain bias, along with the corresponding simulations:

the fitting is very good up to high bias currents.

1E.13

ru

1E·14 � �

�� � � -----... -,.

� �� �"""'" . �

� � � �

'N 1E-15 ! � c Cii 1E-16

:!'!

1E_17 -�

1E2 1E3 1E4

Frequency [Hz)

bJ....

� ' .omA

46mA

� � �

�� �

8mA

4mA 2mA lmA

1E5

Fig. 2. Modeled and measured short-circuit drain noise current spectra of the pHEMT device used for the yeO design.

The noise data needed for the unambiguous, complete

characterization of the nonlinear noise model were obtained

instead by exploiting the laboratory setup described in [5],

which enables to force the DUT into different, highly

controllable LS RF oscillating conditions. By fitting the

measured PN spectra, the model optimization was possible

and the relative weights of the different EN generators could

be unambiguously estimated. From design considerations based on a trade-off between

the gain, power handling capability and noise characteristics, the device selected for the VCO design was a 400-um pHEMT (8x50 um). The nonlinear model (TOM3) available from the foundry design kit was made accessible at the device intrinsic ports: then the obtained cyclostationary EN generators were implemented in the CAD environment and applied to the actual deterministic nonlinear model.

IV. MMIC DESIGN DESCRIPTION

The topology for the VCO is a classic series feedback. As

shown in Fig. 3, the circuit can be divided into two distinct

parts at section S 1: the right part is the negative resistance

active bipole, whereas the left part includes the resonator with

the tuning varactors. The negative resistance is implemented

by placing a feedback capacitor in series with the device

source; a DC path for the source grounding is provided by a

square spiral inductor. The gate and drain integrated bias

networks feature spiral inductors and shunt MIM capacitors.

The device output matching network is composed of three

lumped elements (CLC) and is followed by a 4-dB resistive

attenuator to avoid oscillation frequency load pulling.

The active bipole was designed with the aim to guarantee

the self start-up capability of the circuit, by optimizing the

978-1-4244-7732-6/101$26.00 ©201 0 IEEE 286

negative resistance value and its bandwidth, and also selecting

an optimum LS working point of the device. Indeed, by means

of harmonic balance simulations, the LS voltage and currents

waveforms have been optimized to obtain the best

performance in terms of output power, gain compression,

amplitude and frequency stability and PN. Parametric LS

sensitivity analyses were exploited as well as a control of the

dynamic intrinsic device load line to avoid in particular the

device triode region and an excessive modulation of the noise

sources, which is harmful for PN as described in [ 1] and [6].

RESONATOR vo VD

OMN �1 - : -

I DCPAlH SI

--�1! cmc�Rl r�·

Fig. 3. Schematic of the monolithic yeo.

Fig. 4. Picture of the 2.4 mm x 1.5 mm monolithic yeo.

The microstrip resonator was privileged rather than a

lumped LC tank solution, since it had shown better Q factor

also thanks to the thick metal layer available. One end of the

resonator (Fig. 3) is shunted to ground by means of a via hole,

while at the other side a varactor is placed for the frequency

tuning. The choice of the varactor is justified by several

considerations and specifications: its capacitance affects the

total length of the resonator and its Q factor; also its losses are

important for the Q factor, whereas its maximum capacitance

ratio is related to the desired bandwidth. By considering these

factors a varactor composed by 8 fingers with 10 um width

was designed. The equivalent capacitance is between 107 fF

and 64 fF, with bias voltage ranging from 1 to 10 V. The

resulting capacitance ratio is only l.68, but it's enough for the

bandwidth requirements, while it optimizes the Q factor. The

varactor bias circuit is also integrated, and consists of a

resistor and a MIM capacitor shunted to ground. By adding

the varactor capacitance, the overall resonator is shortened

with respect to the initial A / 4 length. The resonator is

IMS 2010

coupled with the active circuit by means of a microstrip tap

followed by a series capacitor to block the DC (Fig. 3).

Moving the tap along the resonator it is possible to vary the

coupling factor and then the impedance presented by the

resonator to the active bipole.

V. VCO CHARACTERIZATION MEASUREMENTS

In Fig. 5 the frequency characteristic of the VCO vs. tuning

voltage is shown: the bandwidth is 4.8% and the central

frequency is 7.32 GHz.

7,60 7,55 7,50

¥ 7,45 � 7,40 >- 7,35 � 7,30 � 7,25 g- 7,20 u: 7,15

7,10 7,05 7,00

� � H ..... �

V ...I .... / ..) ....... Simulation --.V

// -+- Measurement --

(/ •

o 2 3 4 5 6 7 8 9 10 11

Tuning Voltage M

Fig. 5. VCO frequency characteristic vs. tuning voltage.

The small discrepancies between measurements and

simulations are due to the resonator and varactor modeling.

Nonetheless, the fitting is good and the circuit was within

design specifications. Measured output power is 14 dBm, with

collector current from 38 to 46 rnA.

10 0 - Measurements

-10 N -20 I "0 -30

D Cyclostationay nonlinear noise model

[lJ -40 � -so Q) (/) -60

'0 -70 z Q) -80 (/) -90 '"

.c: -100 11. [lJ -110 CfJ -120 CfJ -130

-140 -1S0

1E2 1E3 1E4 1ES 1E6 SE6

Offset Frequency [Hz]

Fig. 6. The proposed nonlinear cycIostationary noise model gives a better prediction of PN with respect to a stationary noise model.

The PN at the central frequency is -86 dBc/Hz @ 100 kHz

from the carrier, a value very close to simulations. The figure

of merit proposed by many authors was calculated to be

FOM = 191 dB, which places this design in a good position in

the comparative table presented in [7].

978-1-4244-7732-6/101$26.00 ©2010 IEEE 287

Comparisons between simulations and measurements were

carried out, which also confirm the validity of the proposed

noise model. Fig. 6 shows a comparison among measured PN

at 7.4 GHz, simulations with the proposed model and

simulations carried out exploiting a stationary noise model.

The accuracy of the proposed model is clearly better.

VI. CONCLUSION

The design and implementation of a MMIC LPN VCO for

space applications were described. The design was carried out

by exploiting a newly proposed compact cyclostationary

nonlinear noise model, whose accuracy is assessed by

simulation vs. measurement comparison. To the authors'

knowledge this is the first time a cyclostationary LF noise

modeling approach is applied to pHEMT devices. Also the

electrical performances are close to simulations, showing the

validity of the applied LS design methodology.

ACKNOWLEDGEMENT

The authors wish to acknowledge the support the Italian

Space Agency (ASI) for this activity.

REFERENCES [1] P. A. Traverso, C. Florian, M. 8orgarino, and F. Filicori, "An

empirical bipolar device nonlinear noise modeling approach for large-signal microwave circuit analysis," IEEE Trans. Microw. Theory Tech., vol. 54, no. 12, pp. 4341-4352, Dec. 2006.

[2] M. Rudolph et aI., "On the simulation of low-frequency noise upconversion in InGaP/GaAs HBTs," IEEE Trans. Microw. Theory Tech., vol. 54, no. 7, pp. 2954-2961, Jul. 2006.

[3] J. C. Nallatamby et aI., "An advanced low-frequency noise model of GalnP-GaAs HBT for accurate prediction of phase noise in oscillators," IEEE Trans. Microw. Theory Tech., vol. 53, no. 5, pp. 1601-1612, May 2005.

[4] C. Florian, P.A. Traverso, "Active bias network-based measurement set-up for the direct characterization of low­frequency noise currents in electron devices," 16th IMEKO TC4 Int. Symp., Florence, Italy, Sep. 2008, pp. 185-190.

[5] C. Florian, P.A. Traverso, "A Highly Flexible Measurement

Set-Up for the LF Noise Up-Conversion and Phase-Noise

Performance Characterization of Microwave Electron Devices"

2007 IEEE Proc. of Instrumentation and Measurement

Technology Conference. 1-3 May 2007 Page(s): 1 - 6.

[6) C. Florian, P.A. Traverso, "Investigation of Phase Noise

Generation in Microwave Electron Devices Operating in

Nonlinear Regime Exploiting a Flexible Load- and Source-Pull Oscillating Setup" IEEE Trans. Microw. Theory Tech., vol. 57,

issue. 12, part 2, pp. 3491 - 3504, Dec. 2009.

[7] Ferndahl, H. Zirath, "Broadband 7 GHz VCO in mHEMT technology," APMC 2005 Asia-Pacific Conference Proceedings, Volume 2, 4-7 Dec. 2005 Page(s):4 pp.

[8] O. Llopis, J.B. Juraver, et al. "Nonlinear noise modeling of a PHEMT device through residual phase noise and low frequency noise measurements," 2001 IEEE MTT-S Int. Microwave Symposium Digest, Vol.2, May 2001 Pag. 831 - 834.

IMS 2010