PROCEEDINGS OF THE IEEE, VOL. 92, NO. 2, FEBRUARY 2004Gigabit Wireless: System-on-a-Package
Technology
RAO R. TUMMALA, FELLOW, IEEE AND JOY LASKAR, SENIOR MEMBER, IEEEInvited Paper    The system-on-a-package (SOP) concept is considered as the solution
of future communication modules, which would need more functionality, better
performance, low cost, and more integrity. In this paper we demonstrate how
SOP technology can address the integration platform for future communication
system, especially gigabit wireless communications. After the introduction
of the SOP concept, we introduce the critical design building blocks which
are required in a viable SOP technology: integrated passives, embedded RF
functions including high-performance filters and baluns, and integrated antenna
technologies. Second, we review how the three-dimensional deployment of core
elements such as baluns, lumped inductors, capacitors, and resistors, as well
as IF or low-pass filters enables RF-SOP module development. In the last section
we demonstrate how advanced radio architectures that include direct conversion,
antenna diversity, and collaborative signal processing are enabled using the
SOP technology format. Various ceramic and organic material based multilayer
packaging technology has been used for building those integrated modules as
well as circuit blocks. In the conclusion, the critical issues and challenges
for developing advanced communication platforms using SOP approach are discussed.
 Â
   Â
Keywords—Communication system, integrated module, integrated radio frequency (RF) passives, multilayer packaging, system-on-a-package (SOP).    Manuscript received May 12, 2002; revised November 12, 2003.
    The authors are with the Department of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta, GA 30332-0245 USA.
Digital Object Identifier:
10.1109/JPROC.2003.821902
0018-9219/04$20.00 © 2004 IEEE
I.  INTRODUCTION
II.  INTEGRATED PASSIVES
    A.  Multilayer Filter
    B.  Integrated Antenna
    C.  Multilayer Balun
III.  INTEGRATED RF-SOP
MODULES
    A.  Ku-Band Transmitter Module
    B.  PA Module With LTCC Embedded Filter
    C.  RF-Front-End Module Using MLO Package
    D.  VCO With High-Q Embedded Inductor
IV.  3-D RF MODULE DEVELOPMENT
    A.  5.8-GHz OFDM Module
    B.  2.4-GHz OFDM Module
V.  SUMMARY AND FUTURE DEVELOPMENTS
ACKNOWLEDGMENTREFERENCESI.  INTRODUCTION
    The dawn of the 21st century is witnessing a tremendous demand for wireless
(untethered) communications services such as paging, analog and digital cellular
telephony, and emerging personal communications services (PCS). Current commercial
cellular and PCS systems are concentrated at frequency bands around 900 MHz
and 1.8 GHz (S-band); future allocations for PCS systems are expected around
2.4 and 5.8 GHz (through C-band). Beyond the arena of mobile communications,
there are numerous wireless applications including RF identification (RFID),
satellite communications, local multipoint distribution systems (LMDS), and
wireless local area networks (WLANs) operating at frequencies extending into
the millimeter-wave regime (Ku to Q-band). The move to higher frequencies
has been motivated by the need for more and more bandwidth for multimedia
applications such as wireless real-time video transmission and iInternet access,
and by the increased overcrowding of the lower frequency bands.
    This rapid expansion of untethered communications services, along with
the need for low-cost, high-efficiency system implementations, has led to
an explosion in the development of IC approaches in the RF/microwave area.
Highly integrated components such as frequency converters, low-noise and power
amplifiers, and frequency synthesizers are now commonplace, replacing hybrid
circuits employing discrete semiconductor devices. These radio frequency integrated
circuits (RFICs) and monolithic microwave integrated circuits (MMICs) are
generally packaged together with very large scale integration (VLSI) digital
signal processing (DSP) and microprocessor (
) control chips on multilayer printed circuit boards (PCBs).
However, the system-on-a-chip (SOC) movement is ambushed by the cost of additional
mask layers needed to marry digital logic with memory and analog function
on one specific and optimum substrate.
   Â
| Fig. 1. Advanced
MCM solution for gigabit wireless module.
| |
Therefore, many believe that the system-on-a-package (SOP) approach
for the next-generation wireless solution is a more feasible option than SOC
considering the obstacles faced today by SOC developers. Recent development
of materials and processes in packaging area makes it possible to bring the
concept of SOP into the RF world to meet the stringent needs in wireless communication
area. RF-SOP is “to provide a complete packaging solution for RF module
by integrating embedded passives components and MMIC at the package level.�
SOP goes one step beyond multichip module (MCM) by enhancing overall performances
and adding more functionalities. Wireless devices implementing complex functionality
require a large amount of circuitry and consequently, require a large conventional
package or MCM real estate. Three-dimensional (3-D) integration techniques
using multilayer high-density architectures in different technologies with
vertical interconnect and embedded component integration are crucial for the
design and development of a single package to a MCM. SOP is the art of bringing
together at the system level ICs and embedded components following a codesign
philosophy
[1][2][3][4][5].
    A strong alternative to complete single-chip integration is some advanced
form of an MCM solution, such as the one proposed in
Fig. 1, which integrates antenna, filters, resonators, baluns
and other RF components. These can be integrated in either ceramic, such as
low-temperature cofired ceramic (LTCC), or organic, such as multilayer organic
(MLO).
    The basic philosophy here is to fabricate devices and circuits in the
optimum available technology for their particular function
(e.g., digital circuits in submicrometer CMOS, analog components in BiCMOS,
RF components in GaAs or InP) and then integrate them into a system or subsystem
using MCM technology.
| Fig. 2. Photograph
and configuration of the fabricated filter.
| |
High-Q passive structures could potentially be embedded directly
into the MCM structure. Multibeam antenna beamforming networks have been demonstrated
in MCM technology. For the active analog/RF and digital/DSP functions, individual
ICs can be interconnected using flip-chip bonding to eliminate bondwires (and
their associated parasitic) and allow more compact MCM floor planning. An
additional benefit of this approach is that complete system redesign is not
necessarily needed to change one part of the subsystem/module; individual
functional component dies could be upgraded as long as their interconnection
scheme and footprint were kept the same to remain compatible with the MCM
layout. This approach is referred to as
system on a package, in contrast to “system on a chip.� The incorporation
of microelectromechanical systems (MEMS), smart materials, smart antennas,
etc., with mixed-signal ICs in such an environment is now increasingly referred
to as “mixed-technology� or “heterogeneous� integration.
    In this paper we focus on the following topics which establish how SOP
technology establishes the preferred integration platform for future gigabit
wireless needs. First, we introduce
the critical design building blocks which are required in a viable SOP technology:
integrated passives, embedded RF functions including high-performance filters
and baluns, and integrated antenna technologies. Second, we review how the
3-D deployment of core elements such as baluns, lumped inductors, capacitors,
and resistors, as well as IF or low-pass filters, enables RF-SOP module development.
In the last section we demonstrate how advanced radio architectures which
include direct conversion, antenna diversity, and collaborative signal processing
are enabled using the SOP technology format.
II.  INTEGRATED PASSIVES
    A.  Multilayer Filter
    An on-package integrated multilayer filter offers a more attractive implementation
than on-chip and discrete filters. An RF image-reject filter was implemented
with six layers of LTCC in a stripline configuration whose 3-D view is illustrated
in Fig. 2 [6]. Layers 6 and 0 are the top and bottom metallization
which serve as the top and bottom ground planes. The two shunt inductors
and
were realized by the U-shaped strips fabricated on layers 4 and 3, which are
located two and three layers underneath the top ground plane, respectively.
The end of the strips are connected to both grounds through vias. There is
no metallization between layer 4 and 6 and, therefore, the top inductor strip
on layer 4 is 7.2 mils (2 layers) away from the top ground plane while the
bottom strip is 10.8 mils (3 layers) away from the bottom ground plane. The
required mutual inductive coupling is achieved by overlapping the
and
strips which are one layer (3.6 mils) apart. metal–insulator–metal
(MIM) capacitors with electrodes on layers 4 and 3 laid out beside the inductor
strips were utilized to implement
and
. The vertical interconnect capacitor
(VIC) topology was utilized to implement
,
which is realized by two series capacitors of capacitance
. Each of these capacitors is implemented in VIC topology
as a parallel combination of two capacitors with a value of
. Such implementation ensures the symmetry of the
structure desirable for high-frequency circuits. If
were implemented in MIM topology, the entire structure would
not have been symmetrical. The bottom plates of
and
are used as the top
plates of the VIC extended to layer 2 through via connections. The dumbbell-shaped
trace is inserted on layer 2 between layer 3 and 1 as the bottom plates of
the VIC. The extended top plates of the VIC on layer 1 are used as the top
MIM electrodes for the shunt capacitor
with the bottom ground on layer 0 as the bottom electrode. The filter prototype
is given in Fig. 2. Fig. 3 gives the measured insertion loss of 3 dB at
2.4 GHz with 40-dB rejection at 2 GHz.
   Â
B.  Integrated Antenna
    One of the major issues for developing RF-SOP is integrating an antenna
with a module efficiently.
| Fig. 3. Measured
and simulated magnitude of S21 for filter.
| |
| Fig. 4. Configuration
of CBPA.
| |
Fabricating an antenna directly on the package has the advantages
of reducing feeder loss and size of an entire module. However, for the frequency
range from 1 to 6 GHz, which is being used for the most of wireless mobile
and data communication, it is difficult to bring an antenna into the transceiver
module, since the size of the antenna becomes large. Also, the interference
between the antenna and other RF components in highly compact modules is another
main issue in this approach
[7].
| Fig. 5. Measured
and simulated return loss of the CBPA designed for 5.8 GHz.
| |
    For the compact module, the cavity-backed patch antenna (CBPA) has been
successfully adopted to LTCC packaging. The configuration of the CBPA is shown
in
Fig. 4. The antenna is designed
to be connected with the embedded RF blocks such as filter or duplexer switches.
The cavity structure is formed by surrounding multiple via connected to ground
plane. The radiator (patch) is made on the top surface and is connected with
the embedded filter through a via. The location of via feed is designed to
get impedance matching with an embedded filter. CBPA needs a smaller ground
plane, since the most of the energy is confined between the edge of the patch
and via-wall. Obviously, thicker substrate increases the bandwidth of the
antenna but the thickness should be determined moderately by considering the
entire module structure.
    Measured return loss of the CBPA designed for 5.8 GHz is shown in
Fig. 5. For comparison, a microstrip patch
antenna has been developed for the same frequency. The results shows that
bandwidth of CBPA is larger than the patch by 20%.
   Â
C.  Multilayer Balun   Â
| Fig. 6. Photograph
and configuration of the stripline multilayer balun.
| |
Baluns are required in a wide variety of microwave components such
as balanced mixers, push–pull amplifiers, multipliers, and phase shifters.
The balun has become one of the important RF components to improve performance
and reduce cost of the RF module by embedding it inside of the package
[8]. The multilayer balun
is presented in this section.
| Fig. 7. Measured
and simulated results of the stripline balun designed for 5.8 GHz.
| |
   Â
Fig. 6 shows the stripline type
multilayer Balun designed for LTCC. Two shorted lines are placed next to an
open line such that they couple energy from the open line as shown. The open
line is a half wavelength long at the operating frequency. Layers 0 and 4
represent the ground planes of the structure. The half wavelength open line
is on layer 2. A small portion of this line is on layer 3 to avoid overlapping
traces and in order to achieve the spiral space saving configuration described
previously. One of the quarter wavelength coupling sections is on layer 1
and the other on layer 3. Each of these lines is shorted to the ground planes
(through vias) above and below it, respectively.
| Fig. 8. Configuration
and picture of compact Ku-band transceiver module.
| |
    Return loss better than
![[$-{\hbox{15~dB}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166717.gif)
and insertion loss of 3.4Â dB is
measured as shown in
Fig. 7.
III.  INTEGRATED RF-SOP
MODULES
    A.  Ku-Band Transmitter Module
    The implementation of compact transmitter modules is a key issue for low
power and higher data rates for satellite communications in the Ku/Ka-band
range [9]. One obstacle
for a compact transceiver module is the large and heavy cavity filter that
cannot be easily integrated into the module.
 |
| (a) |
 |
| (b) |
| Fig. 9. Measured
gain of the module. (a) Overall system gain. (b) Gain variation at each level.
| |
    A highly integrated LTCC-based transmitter module using GaAs MESFET MMICs
for Ku-band satellite communication applications is presented in
Fig. 8. The figure shows the 3-D configuration and
picture of the module. The upconverter MMIC integrated with a VCO exhibits
a measured upconversion gain of 15 dB and an third-order input intercept point
(IIP3) of 15 dBm, while the power amplifier (PA) MMIC shows a measured gain
of 31 dB and a 1-dB compression output power of 26 dBm at 14 GHz. Both MMICs
were integrated on a compact LTCC module where an integrated front-end bandpass
filter (BPF) with a measured insertion loss of 1.8 dB at 14.5 GHz was integrated.
The BPF for the transmitter module was implemented in a coupled-line filter
topology on a multilayer LTCC substrate. The two stripline ground planes are
physically connected by vias and the actual input and output of the filter
are connected to the RF amplifier and driver amplifier of the transmitter.
Three segment-folded edge-coupled stripline filters, where the middle segment
was deployed perpendicular to the first and third segments for compactness,
have been designed to suppress the LO signal at 13 GHz as well as the harmonics
and spurious signals. This is a balanced stripline topology where the coupled-line
segments are sandwiched by two ground planes at an equal distance of 17.5
mils (four LTCC tape layers).
    The entire transmitter chain exhibits a total gain of 32 dB and output
power of 26 dBm incorporating the wirebond loss from 14 to 15 GHz and image
rejection of more than 30 dBc at 12 GHz as shown in
Fig. 9(a).
Fig. 9(b) shows
the overall system level diagram of the entire transmitter chain based on
the measurement of each transmitter blocks. The compact module was made possible
by embedding the filter and thereby saving more than 40% of real estate compared
to the module if it were implemented on a typical alumina substrate.
   Â
B.  PA Module With LTCC Embedded Filter    A 2.4-GHz SiGe heterojunction bipolar transistor (HBT) PA module with a
harmonic suppression filter implemented in LTCC substrate has been developed
[10]. The harmonic suppression
filter and output match network have been implemented completely in LTCC without
the use of external discrete components.
    The PA is designed in IBM's commercial SiGe HBT process. It consists of
three stages: two driver stages and one power stage; interstage, input, and
output match networks; and bias circuits for the three stages. It is packaged
in an eight-pin small-outline integrated circuit (SOIC) package.
Fig. 10 shows a picture of the developed PA. The harmonic
suppression filter and output matching networks are designed in Kyocera's
ten layers of LTCC process.
Fig. 11
shows the designed filter and output match network. The filter consists of
a transmission line in parallel with a capacitor.
| Fig. 11. Output
matching section and harmonic suppression filter built in LTCC substrate.
| |
    The output network is designed to give deep suppression of the second and
third harmonics. A
![[$\lambda /4$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166718.gif)
length RF-shorted
stub is used to achieve a short at the second harmonic. The output matching
network, which is also implemented in LTCC using transmission lines and tuned
in conjunction with the harmonic suppression filter, presents a short at the
second harmonic and nearly an open at the third harmonic.
| Fig. 12. Measured
performance of the PA module.
| |
| Fig. 13. Schematic
and picture of MLO-based integrated transmitter module.
| |
    The output power and efficiency of the PA module is shown in
Fig. 12. It achieves an output power of 27 dBm at 0–dBm
input with a power-added efficiency (PAE) of 45% at 2.4 GHz and
![[${\hbox{Vcc}}=3.3~{\hbox{V}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166719.gif)
. The linear gain of the
PA is 35 dB.
   Â
C.  RF-Front-End Module Using MLO Package    Organic material-based low-cost MCM-L technology is getting more attention
even in the RF area due to its low-cost feature, and despite such disadvantages
as high dielectric loss and too much process variation
[11][12][13]. A C-band transmitter module incorporating
an embedded BPF and an up-converter and a PA MMIC is shown in
Fig. 13.
    The transmitter MMIC provides an excellent LO and image rejection fabricated
in a commercial GaAs MESFET process. A miniaturized integrated square patch
resonator BPF with inset feed lines built in MLO technology has been incorporated
to realize a compact and highly integrated transmitter module suitable for
the low-cost network interface card (NIC), IEEE 802.11a WLAN applications
in the 5–6 GHz frequency band.
 |
| (a) |
 |
| (b) |
| Fig. 14. Measured
performance of up-converter chip, (a) Conversion loss, IIP. (b) Output spectrum.
| |
| Fig. 15. Measured
performance of embedded BPF (solid: S21; dashed: S11).
| |
    The up-converter MMIC consists of a double-balanced diode ring mixer (DBM),
a wide tuning range (20%) voltage controlled oscillator (VCO), a LO buffer
amplifier, a IF amplifier, and a RF amplifier. The up-converter MMIC integrated
with a VCO exhibits a measured up-conversion gain of 14 dB and an IIP3 of
15 dBm as shown in
Fig. 14. LO-to-RF
isolation is 45 dB and image rejection of 11 dBc without any bandpass filtering.
MLO-based BPF has been implemented and the measured data is shown in
Fig. 15. The BPF shows an insertion loss
of less than 3 dB and a return loss of 27 dB at 5.8 GHz as well as the image
rejection of 35 dBc at 3.8 GHz.
   Â
D.  VCO With High-Q Embedded Inductor    A C-band oscillator with external high-Q inductors was built in the MLO
package. All circuit blocks except the inductors are implemented in MESFET
MMIC
[14]. The inductors
embedded in the package have been designed to obtain a maximum quality factor
around the oscillation frequency.
 |
| (a) |
 |
| (b) |
| Fig. 16. Comparison
of the inductors, on-chip, embedded in MLO and wire-bonding inductors. (a)
Q factors. (b) Inductances.
| |
| Fig. 17. Phase
noise measurement result for the oscillator with MLO inductor.
| |
To validate the VCO with an embedded inductor, the same topology
of oscillators having the on-chip inductors, and wire-bonding inductors have
also been developed and compared. All the inductors have been designed to
get the inductance value of 1.5Â nH and optimized to get the highest quality
factor in C-band. The measured performances of the three inductors are shown
in
Fig. 16.
    The TriQuint 0.6
![[$\mu{\hbox{m}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1036431.gif)
GaAs MESFET
process is utilized to design and implement the oscillator circuit. To reduce
the
![[$1/{f}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166720.gif)
noise up-conversion, the cross-coupled
differential topology with the capacitive coupling feedback was used. Cross-coupled
transistors form a positive feedback to provide a negative resistance to cancel
the loss in the LC-resonators. The positive feedback is obtained through capacitors
that take the role of suppressing the
![[$1/{f}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166720.gif)
noise up-conversion as well as dc blocking in order to bias cross-coupled
transistors.
    The measurement result of the oscillator with MLO inductor is shown in
Fig. 17.
| Fig. 18. Proposed
3-D integrated RF front-end concept.
| |
On-wafer measurements of the oscillation frequency, output power,
and phase noise were performed using an Agilent 8563E spectrum analyzer. The
phase noises of the oscillator with the on-chip inductors and the embedded
inductor are
![[$-{\hbox{108~dBc/Hz}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166721.gif)
and
![[$-{\hbox{113~dBc/Hz}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166722.gif)
at 600-KHz offset frequency, respectively.
Using MLO inductors, the phase noise is 5 dBc/Hz better than the oscillator
with on-chip inductors and comparable to the oscillator with wire-bond inductors.
IV.  3-D RF MODULE DEVELOPMENT
    A.  5.8-GHz OFDM Module
    The 3-D integration approach is a very attractive option to increase density,
performance and meet the specifications of the next generation wireless communication
systems. Very compact 3-D integration technology based on various wafer stacking
methods and ultrathin packaging technology have been successfully proposed
for digital applications such as 3-D memory and smart cards. However, current
3-D RF-module integration is still based on low-density hybrid assembly technologies.
The development of wireless data communication systems in C-Band leads to
very stringent specifications for both IC and packaging performances. For
these portable and low-powered applications, a high level of integration is
also required to reduce essential features such as size, weight and interconnection
length.
    We also present a very compact 3-D-integration concept suitable for a RF
front-end module by means of substrate stacking method using the
ball process [15], [16]. Fig. 18 illustrates the
proposed module concept. Two stacked LTCC substrates are used and board-to-board
vertical transition is insured by
balls. Standard alignment equipment is used to stack the board and, thus,
provides a compact, high-performance and low-cost assembly process. Tests
structures have been designed for the characterization of
balls and an accurate
ball modeling shows insertion loss less than 0.1 dB and return
loss of
up to 10 GHz.
 |
| (a) |
 |
| (b) |
| Fig. 19. Embedded
filter. (a) Schematic. (b) Performance.
| |
    Multistepped cavities into the LTCC boards provide spacing for embedded
RF active devices and, thus, provide significant volume reduction by minimizing
the gap between the boards. Active devices can be flip chipped as well as
wire bonded. A cavity also provides integration opportunity for MEMS device
such as MEMS Switch. Passive components, off-chip matching networks, the embedded
filter, and the antenna are implemented directly into the LTCC boards by using
multilayer technology. Standard BGA balls insure interconnection of this high
density module with a motherboard such as an FR4 board. The top and the bottom
substrates are dedicated respectively to the receiver and transmitter building
blocks of the RF front-end module. The receiver board includes antenna, BPF,
active switch, low-noise amplifier (LNA), VCO, and down-conversion mixer.
The transmitter board includes up-conversion mixer, PA, and off-chip matching
networks. Ground planes and vertical via walls are used to address isolation
issues between the transmitter and the receiver functional blocks. Arrays
of vertical via are added into the transmitter board to achieve better thermal
management.
| Fig. 20. Cavity-backed
patch antenna layout and performance.
| |
    We have developed a high-performance second-order narrow-band BPF with
two cascaded coupled-line sections embedded in stripline configuration to
improve insertion loss. A schematic view and measured performance are presented
in
Fig. 19. Coupled lines have been
bent to fit into the module shape and via walls are used to connect ground
planes and reduce parallel plate modes. Filter performances have been measured
separately and exhibit a
![[$-{\hbox{2.9-dB}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166726.gif)
insertion loss,
![[$-{\hbox{20.8-dB}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166727.gif)
return loss,
and image rejection greater than
![[$-{\hbox{20~dB}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166728.gif)
as shown in
Fig. 19(b).
   Â
| Fig. 21. Rx
chipset used in the module. (a) LNA. (b) Mixer. (c) VCO.
| |
A via-fed stacked cavity-backed patch antenna has been designed for
IEEE 802.11a 5.8-GHz band as shown in
Fig. 20. The heights of the lower and upper patches (400
![[$\times$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1032418.gif)
400 mils) are 8 mils and 32 mils, respectively. The 10-dB
return-loss bandwidth of the antenna is about 4%, fully covering the required
band (5.725-5.825 GHz). Also, this antenna has a desirable gain (near 6 dBi)
and very low cross polarization (less than
![[$-{\hbox{35~dBi}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166729.gif)
).
   Â
| Fig. 22. Tx
chipset used in the module. (a) Mixer. (b) RF amplifier.
| |
We used the commercial GaAs MESFET process to implement the Rx and
Tx chipset. The Rx chipset consists of LNA, a down-converter mixer, and VCO
as shown in
Fig. 21. The Tx chipset
includes a double-balanced diode ring mixer and RF amplifier as shown in
Fig. 22    The LNA gain is about 13 dB; the LNA noise is about 2.2 dB.
Fig. 23(a) shows the output spectrum of the transmitter
mixer demonstrating image signal rejection of 11 dBc, and LO-to-RF rejection
of 40 dB, across the VCO tuning range between 4.5 and 5.5 GHz. It also shows
less than 11-dB conversion loss between IF frequency from dc to 1 GHz. The
RF amplifier demonstrates a measured gain of 7 dB as shown in
Fig. 23(b).
   Â
B.  2.4-GHz OFDM Module
 |
| (a) |
 |
| (b) |
| Fig. 23. Tx
module performances. (a) Mixer output spectrum. (b) RF amplifier
gain.
| |
| Fig. 24. Front-end
transceiver architecture.
| |
    In recent years, there has been a great deal of interest in realizing low-power,
low-cost, compact RF front-end modules for Bluetooth technology in the 2.4-GHz
industrial, scientific, and medical (ISM) band. LTCC technology is well suited
for compact wireless transceiver module realization and is becoming increasingly
popular for WLAN applications. However, it is quite challenging to integrate
the antenna with modules for WLAN applications at 2.4 GHz due to their large
size and isolation constraints at higher output powers. Commercial implementations
of Bluetooth using SOC technology provide output power levels of 0 and 4 dBm,
respectively. The SOP approach allows the development of high-quality passive
components on ceramic substrates, thereby providing much higher output power
levels (20 dBm). The choice of proper radio architecture and sharing of ground
plane among LTCC passive components enables compact implementation of the
RF front end (
Fig. 24)
[17]. Most of today's WLAN
solutions aim for very low intermediate frequency (VLIF) RF architectures
| Fig. 25. Three-dimensional
view of developed module: CMOS transceiver MMICs (LNA, PA, mixer) are on the
top layer connected by via structures to embedded passives in LTCC. Loop antenna
is separated by ground plane.
| |
for better integration of ICs in commonly available semiconductor
processes. For an IF frequency at half the channel bandwidth (500 kHz), the
image becomes an adjacent channel according to Bluetooth specifications, thus
leading to a looser image rejection specification. We present the development
of CMOS MMICs and LTCC multiplayer passive components. These building blocks
are then integrated in an LTCC substrate, and an antenna is implemented in
the same package.
    An exploded 3-D view of the developed module is shown in
Fig. 25. The same filter has been used in transmit
and receive paths to save space.
    A CMOS technology has been used in conjunction with commercially available
LTCC technology for the complete SOP solution. The 0.24-
![[$\mu{\hbox{m}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1036431.gif)
p–substrate twin-well CMOS technology is similar to
most commercially available deep-submicrometer CMOS technologies. The sheet
resistance of the silicon substrate is approximately 10
![[$\Omega $]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1039337.gif)
/
![[${\hbox{cm}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166730.gif)
. It is a five-metal-layer
process, and the gate is solicited to reduce gate resistance and transistor.
| Table 1 Summary
of Specifications of Building Blocks; RF Building Block Performances
| |
    Due to the constant envelope nature of the Gaussian-frequency shift keying
(GFSK) modulation scheme, the Bluetooth PA does not have strict linearity
requirements. The PA uses a compact three-stage cascade topology. The load-impedance
design is based on the load-pull measurement on the power transistor. This
PA utilizes on-chip input matching. The output matching is realized in LTCC
multilayer high-Q passive elements. Measurement results of the PA show an
output power of 20 dBm and a power-added efficiency of 20%. The PA is biased
at class-AB and exhibits an output 1-dB compression point of 16.8 dBm. It
utilizes a 2.4-mm (40 fingers with 60-
![[$\mu {\hbox{m}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166731.gif)
gate width) device size.
    The LNA has been designed using a cascode topology. Use of a cascode topology
helps in separate optimization of input and output networks, better stability,
and better high-frequency operation due to minimization of the Miller capacitor
effect. It uses n-channel MOS (NMOS) transistors of 250-
![[$\mu{\hbox{m}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1036431.gif)
widths to achieve low-noise performance. Output matching
was optimized to achieve high gain and high IIP3 simultaneously. The LNA has
a noise figure of 3 dB, gain of 11 dB, and IIP3 of
![[$-1 ~{\hbox{dBm}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166732.gif)
at 2.45 GHz.
    Double-balanced Gilbert cell topology has been chosen for the up/down conversion
mixer. Biasing the mixer core FETs near threshold allows reduced local-oscillator
(LO) power for driving the FETs like ideal switches. Two such balanced mixers,
with their LO in quadrature, generate a polyphase signal that can be used
for image rejection. It uses NMOS transistors of 250-
![[$\mu{\hbox{m}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1036431.gif)
widths for achieving the desired gain from the mixer. It
provides a gain of 2 dB and a IIP3 of 0 dBm at
![[$-{\hbox{5~dBm}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166733.gif)
of LO power. The LO to IF isolation is greater than 40 dB.
Fig. 3 shows a picture of the fabricated
chipset used in the module.
    A portion of the total development cost of a transceiver module is expended
on this front-end filter, which is typically realized by discrete components.
Integrating the filter-on-package is an attractive option, especially with
the availability of multilevel dielectric systems, such as that offered by
the LTCC process. There are two potential advantages: first, miniaturization
is possible by the vertical deployment of filter elements, since there are
a large number of dielectric layers available; second, integration essentially
eliminates the need for a discrete filter and, therefore, reduces the component
and assembly cost. The LTCC-integrated preselect two-resonator BPF described
in
Section II-A has been adapted
designed, developed, and characterized.
    The balun has been implemented using a multilayer coupling structure, occupying
a total of five dielectric layers (
Fig. 5). Two shorted transmission lines (A and B) are placed next to an open-circuited
line (C) such that they couple energy from the open circuited line that has
a length of half a wavelength at the desired frequency. This being the case,
a standing wave will form a short circuit at the center of the open-circuited
line. Here, the current will be at its maximum value. At equal distances away
from the center and on either side of it, the voltage will be equal in magnitude
but in opposite phases. The two short-circuited line sections are coupled
from both ends of the open-circuited line. Therefore, a signal incident at
port 1 induces signals at port 2 and port 3 with equal amplitudes but opposite
phases. The short-circuited lines have to be approximately a quarter-wavelength
long in order to achieve the desired characteristics.
    The antenna utilizes a 3-D loop structure with a vertically embedded loop
radiator in 20 LTCC layers for compactness and integration in the same package.
A microstrip-transmission line is used for the antenna feed. To minimize the
interference between the antenna and other components inside the package,
a vertical ground wall using a row of vias has been inserted between the radiator
and transceiver. The ground plane of the microstrip transmission-line feed
is implemented on layer 18 of the LTCC substrate. The antenna has a size of
![[$31 \cdot 14 \cdot 1.8~{\hbox{mm}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166734.gif)
but can be further
optimized by meandering the vertical loop radiator. The measured input-return
loss of the antenna is shown in
Fig. 7
and is 20 dB at 2.3 GHz. It shows a return loss of
![[$ -20~{\hbox{dB}}$]](http://mathfigs.ieeexplore.ieee.org/iel5/5/28340/1266919/1166735.gif)
at 2.3 GHz. This drift from the desired frequency can be
caused by a finite dielectric effect that was not considered in this method
of moment simulation.
   Â
Table 1 shows the specifications
of each of the building blocks in detail. The MMICs are connected to the embedded
passive structures by via interconnections.
V.  SUMMARY AND FUTURE DEVELOPMENTS
    We have presented the SOP concept applied for RF front-end module integration.
The development of an LTCC-based component for wireless RF-SOP applications
has been described. Various types of multilayer embedded passive components
including inductor, capacitor, filter, antenna, and balun suitable for the
highly integrated module have been presented. Development efforts on a 3-D
integrated module, which include a OFDM transceiver, a Ku-Band transmitter,
and 1.9-GHz DECT PA modules, have been demonstrated and then a novel 3-D integration
concept has been described for 2.4- and 5.8-GHz applications. Multilayer organic
packaging developed for SOP is finally reported. A high Q-factor inductor
and embedded filter have been presented as an example of the high performances
of multilayer organic package.
    The SOC movement is ambushed by the cost of additional mask layers needed
to marry digital logic with memory and analog functions on one specific and
optimum substrate. Rather than SOCs, the industry will move instead to 3-D
systems to meet the changing needs of the market toward the future third-generation
mobile phones as an example. The key will be integrate the optimum technologies
together in what one can call a SO3-D, a system-in-3-D package. The package
becomes up front part of the architecture, and it would require 3-D floor
planning and design.
    Furthermore, the spread of high-performances workstations and PCs in offices
and factories create emerging needs for very high data rate wireless transmission.
The inflexibility of hardwired networks makes wireless indoor communication
systems such as WLAN attractive because of their layout flexibility and portability
of terminals. For such short-range indoor broadband WLAN systems, the millimeter-wave
band at 60 GHz offers significant advantages supplying very wide bandwidth
for the multigigabit-per-second transmission of various multimedia content.
    The SOP evolution for gigabit wireless applications involves systems, technology,
and materials consideration. We are targeting 3-D high density module technologies
for SOP-based solutions for wireless communication applications.
    First, new materials have to be exploited to provide enhanced organic solution.
LCP films offer a unique combination of high-performance properties including
chemical resistance, barriers properties, temperature resistance, mechanical
strength, and stiffness and very good high-frequency properties. This places
LCP as an optimum candidate for advanced MCM-L technology without the need
of using a core substrate as a mechanical carrier.
    Special care is required in the interconnection technique for IC-to-board
and board-to-board transition to maintain ground, 50-
matching continuity, and address coupling and resonance issues.
It requires constant efforts in technology, characterization and modeling
in the field of flip-chip, micro-BGA, and also innovative approaches in embedded
IC technology for RF and opto-RF packaging.
    Multiband applications are also becoming extremely important within passive
development to realize multiple frequency bands for various wireless and optical
subcarrier multiplexing (OSCM) systems. Advanced passives design for millimeter
wave, smart antenna, and multibeam antenna require great attention as a next
step in the realization of completely integrated wireless communication front-end
systems.
    Quality factors and frequency limitation still limit RF front-end circuitry
to large number of discrete passive components and make RF front-end module
integration very critical. Furthermore, reconfigurable systems justify many
works using MEMS technology to fabricate high-Q passives as well as RF MEMS
switches on SOP modules. Design and fabrication have to be carried out to
demonstrate that these devices can be directly included in a complete MEMS-SOP
wireless integrated module.
ACKNOWLEDGMENT
    The authors would like to thank the National Science Foundation and
the Georgia Electronic Design Center for their support. The authors would
also like to thank the technical support of Drs. K. Lim and S. Pinel.
REFERENCES
Rao R. Tummala (Fellow, IEEE) received the Ph.D. degree
in materials science and engineering from the University of Illinois, Urbana-Champaign,
in 1968. Â Â Â Â He is a Distinguished and an Endowed Chair Professor in Electrical and
Computer Engineering and Materials Science and Engineering at the Georgia
Institute of Technology (Georgia Tech), Atlanta. He is also the Founding Director
of the National Science Foundation's (NSF) Engineering Research Center (ERC)
in SOP technology called the Packaging Research Center (PRC). The PRC is the
largest and most comprehensive microsystems packaging education and research
center. He has published 270 technical papers and holds 68 U.S. patents and
inventions. Â Prof. Tummala is a Fellow of IMAPS and the
American Ceramic Society, and a Member of the National Academy of Engineering
and President of the IEEE-CPMT Society. He has received many awards including Industry Week's award as one of the 50 stars in the United
States for improving U.S. competitiveness, the Third Millennium Award from
IEEE, the Daniel Hughes and John Wagnon Award from IMAPS, the Materials Engineering
Achievement Award from ASM International, the John Jeppson Award from the
American Ceramic Society, the Total Excellence Award in Electronics Manufacturing
(TEEM) Award from the Society of Manufacturing Engineers, and the European
Materials Award (DVM). Most recently, he received the Distinguished Faculty
Award from Georgia Tech. He is also an Eminent Scholar for the State of Georgia. |
Joy Laskar (Senior Member, IEEE) received the B.S.
degree in computer engineering from Clemson University, Clemson, SC, in 1985
and the M.S. and Ph.D. degrees in electrical engineering from the University
of Illinois at Urbana-Champaign in 1989 and 1991, respectively. Â Â Â Â He is the Joseph M. Pettit Professor of Electronics at the Georgia Institute
of Technology (Georgia Tech), Atlanta. He is also the chair for the Electronic
Design and Applications Technical Interest Group, the Director of Georgia's
Electronic Design Center, and the System Research Leader for the NSF Packaging
Research Center. He has authored or coauthored more than 200 papers, several
book chapters (including three textbooks), and has ten patents pending. His
research has focused on high-frequency IC design and their integration. At
Georgia Tech, Dr. Laskar heads a research group with a focus on integration
of high-frequency electronics with optoelectronics and integration of mixed
technologies for next generation wireless and optoelectronic systems. Â Â Â Â Dr. Laskar is the 1999 corecipient of the IEEE Rappaport Award, the recipient
of the faculty adviser for the 2000 IEEE MTT IMS Best Student Paper Award,
the 2001 Georgia Tech Faculty Graduate Student Mentor of the year award, the
2002 IBM Faculty Award, the 2003 Clemson's College of Engineering Outstanding
Young Alumni Award, and the 2003 Outstanding Young Engineer of the Microwave
Theory and Techniques Society Award. |