7/30/2019 2004mar01 Icd Rfd Ta

1/3

By Freek van Straten

Senior PrincipalRF Product Concept DevelopmentPhilips Semiconductors

E-mail: freek.van.straten@

philips.com

Integration of RF wireless sys-tems is fundamentally differentat lower and higher frequencies.At HF, the most important sys-tems are those for wireless con-nectivity. CMOS will eventuallybe the preferred process tech-nology because it can offer theneeded bandwidth better thanbipolar technology. RF CMOS

will in general not be integratedwith digital CMOS on one chip.At LF, where the most impor-tant systems are for cellularcommunication, RF integra-tion for these systems will focusat the passives. This gives anoverview on the needs and op-tions for integrating passivecomponents and technologiestogether with the RF activeparts in a multichip package or

module approach. Only con-sumer handset applications aretreated because integration incellular base stations has a lowpriority.

RF functions are always partof larger communication sys-tems that transfer informationfrom one point to another. TheRF function is usually physicallyseparated from other functionsin such systems. RF transmit-ting and receiving is generally

performed by different ICs. Re-duction of system size and costdrives a trend to integrate RFwith other functions of the sys-tem, notably DSPs. Besides thistrend to integrate RF with non-RF, there are also other integra-

tion trends within the RF partitself. These arise from the factthat, depending on the system,different technologies areneeded and used for the required

RF function.For example, some systemsrequire sharp filtering of the re-ceived signal before it is passedto a LNA. A receive filter, usu-ally either ceramic or SAW, isneeded for this, which cannot beintegrated in a transceiver IC.

An important distinction be-tween systems at LF and those atHF is that the latter only oper-ate when transmitter and re-

ceiver are in the line of sight ofeach other, whereas at LF, directsighting is not necessary thusproviding larger coverage area.While there is no sharp bound-ary, the transition takes placesomewhere in the 2GHz to5GHz range and depends on sys-tem characteristics such astransmitter output power and

receiver sensitivity. We will use2.4GHz as the transition fre-quency for this article. At HF,the line-of-sight systems can besubdivided into long and short

haul systems. Long haul systemslike radar, satellite links,basestation links, fixed wirelessbroadband access have a needfor higher transmit powers thanshort haul wireless connectiv-ity systems like Bluetooth,802.11a and b, and WLAN.

Integration at HFTargeting the consumer mar-ket, short haul wireless connec-tivity systems have cost and sizepressure compounded the need

for ever-increasing data ratesfrom voice, via data to stream-ing video. In general, these sys-tems are portable and batterydriven, which specific require-ments lead to long stand-by andtalk time.

Working at HF (>2GHz) havecertain advantages: availabilityof larger frequency bandwidthsneeded to obtain higher datarates and moderate receiver se-lectivity due to few transmitters

within range. For the same rea-son the receiver SNR can behigher and/or the transmitteroutput power lower. As an ex-ample: the 802.11b WLAN stan-dard offers 11MBps at 2.4GHzand the 802.11a standard even54MBps at 5GHz. Using widerbands or more complex modu-lation schemes puts strongerdemands on signal linearity,which is es pe cial ly re le vant

RF integration strategies for HF, LF systemsfor the transmitter.

HF has consequences for thechoice of technologies to beused in the systems.

Assuming that fmaxis directly

related to obtainable operatingfrequency, CMOS obviously of-fers good possibilities for theseapplications. Also, CMOS canfulfill the relaxed specificationson selectivity, SNR and outputpower but at continuous lowersupply voltages, reducing dy-namic range.

Although this is generallytrue, many systems operate infrequency bands that are opento all kinds of applications. Thiscan increase the number of

transmitters even within theline of sight. Microwave ovensthat interfere with Bluetoothapplications are notoriousexamples.

Despite the advantages ofCMOS at HF, there may be goodtechnical reasons to useBiCMOS. This includes betterdeveloped RF models for bipo-lar technology, transistor pa-rameter matching and otherpractical reasons such as more

experience in BiCMOS designs.Size does not matter: a Blue-tooth transceiver function inCMOS 0.18m or BiCMOShave similar sizes.

If CMOS is the technology ofchoice, the trend is standarddigital CMOS and not to addmultiple options on top of thisalready multi-masklayer pro-cess. The digital function willoccupy the largest chip areahence will have the highest costcontribution.

When us in g ma in st re amCMOS, does the integration ofdigital and RF functions in asingle chip make sense? Theanswer is two-fold. From a tech-nical perspective, it may be pos-sible to use standard CMOSwith technology modificationsfor RF purposes such as highresistive substrates to reducecrosstalk through the substrateand thick dielectrics to achievehigher quality factors for pas-

sive components. From an inte-gration perspective, there isnot much benefit in trying touse standard CMOS for RF andto integrate digital and RFfunctions in one chip because ofthe fundamental differences in

240

180

120

60

fmax/GHz

1996 1998 2000 2002

CMOS

BiCMOSiCMOS

0.25m

0.18m

0.12m

0.10m

Development of the maximum oscillation frequency fmaxof CMOS and BiCMOS technologies.

A Bluetooth module with embedded antenna.

7/30/2019 2004mar01 Icd Rfd Ta

2/3

models and libraries for digitaland RF. Digital circuits are of-ten designed in VHDL orVerilog languages. Redesign ina shrunken technology version

is straightforward. Digital li-brar ies for CMOS technologyare usually available before thetechnology itself while follow-ing the road map from one gen-eration to the next so the de-signer can make digital designsbefore the next process genera-tion is released.

From an RF design perspec-tive, models and libraries onlybecome available after the pro-cess has been released and RFcomponents have been charac-terized. Since RF functions ingeneral do not contain 1:1 reus-able blocks, they have to be de-veloped almost from scratch.The availability of RF librariesgenerally lags one to two yearsbehind the ir digi tal counter-part. Meanwhile, a new genera-tion of CMOS technology willemerge. Using mainstreamCMOS for RF implies a lag ofone generation. Integration ofdigital and RF in one chip will

therefore lead to an older gen-eration CMOS for the digitalfunctions usually at a highercost. Also passive componentslike inductors and RF/analogfunctions do not really scalewith the technology so the sur-face area of the RF part willgrow relative to the digital parton the chip in each generation.

Other obstacles of single-chip integration of digital andRF are:1.Crosstalk via the substrate

between the digit al and RFparts has to be controlled;

2.Very high mask costs of ad-va nc ed CM OS pr oc es se s,leading to prohibitive costsin development of digital RF

integrated chips due to theinherently higher numberof design iterations for RFdesigns;

3.RF IC yields usually are de-

sign determined resulting ina significant lower yield com-pared to the parametric deter-mined yield for digital ICs;

4.Packages used for digitalCMOS hamper RF efficiencythrough high lead inductance.

Technically the best solutionfor short-haul systems at HF aremultichip packages and mod-ules in which digital and RFfunctions are made in separateICs and in separate BiCMOSprocesses. While these pre-ferred solutions are available tover tic all y integ rat ed compa-nies, multichip packaging andmodules are not readily avail-able to fabless companies thatwork with foundries. Thereforeit is likely that fabless compa-nies will advance toward digi-tal-RF integration in one chip.

Wireless connectivity sys-tems also need antennas andswitches for band selection,

transmit-receive switching andantenna diversity. To embedthese odd components, moduleintegration is preferred over

multichip packaging.

Integration at LFAt frequencies below 2.4GHz,cellular systems are by far themost widespread and importantapplication. Due to lower costand smaller size requirementsof cellular handsets, further in-tegration is in demand. Thecomponents used are diversebe cause of th e stringe nt re-quirements of cellular systems.

On the receiver side highsensitivity and selectivity is re-quired, commonly achieved byincorporating a receive filter,for example, a SAW filter.Large SNR is achieved by theLNA in which inductors areused for emitter degradation toachieve the best trade-off be-tween noise and gain match-ing. Often this LNA function is

integrated in a single chip

transceiver IC. Baseband func-tionality is always realized inmainstream CMOS ICs. Thetransceiver function is tradi-tionally in BiCMOS but CMOSis increasingly drawing atten-tion. At the same time multi-band or system integ ration isevolving. This low risk routeleads towards a single base-ba nd /p ro to co l en gi ne wi thseparate RF functions.

Another challenge is thetransmit path. High output

power of 24dBm to 33dBm is re-quired for these omni-direc-tional, non-line-of-sight sys-tems. The technology of choice

for the power amplifier functionis Si bipolar or GaAs HBT (SiLDMOS) because of ease of ap-plication, added power effi-ciency and good performance.SiGe HBT can also be used, butdoes not offer much improve-ment over silicon bipolar junc-tion transistors. After the finalamplifier stage a low-loss outputmatching circuitry is needed,which is technically difficult torealize on-chip. Often this func-tion is partly integrated in thesubstrate in combination withdiscrete SMDs or realized bymeans of special low cost passiveintegration (PI) chips.

Technologies used includeGaAs HBT for the power ampli-fier, Si BiCMOS for the poweramplifier-driver, biasing stagesand power control loop Si PIchips for output matching.

Todays cellphones are

multiband and multimode re-quiring extensive switching andfiltering capabilities betweenthe power amplifier, receivingpath and the antenna. Theswitching part usually is doneby GaAs pHEMT or pin diodesand, later, RF-MEMS. Duplexfilters (Rx-Tx separation), di-plex filters for band selectionand harmonic filters completethe passive front-end part to-wards the antenna. The logicalsecond step in forward integra-

tion after the multiband PAM isa Tx-FE module.

The next step after this is thefull radio module, adding thetransceiver function to thepackage. Cost-effective inte-gration of all these technolo-gies in cellular systems is ex-tremely challenging. System onsilicon (SOS) integration can bedone for the transceiver func-tion, including LNA. However,a receive filter still needs to beplaced off-chip. The power am-plifier and, in general, the RF-front-end cannot be put on onechip. The challenge is in thepassives and multi-technologypackaging. A modular integra-tion on an LTCC or organic

Variable gain SAW

Power amplifier

Power

control

Duple

x

filter

IF

IF

SAW

Isolator

Block diagram of a CDMA RF front-end.

BGY284, quad band GSM PA module. BGY281, tr ip le band GSM TX-FEM.

1st metal

2nd metal

5m Al metal

Capacitor Via to bottom metal

High-ohmic Si

Cross section of PASSItechnology.

7/30/2019 2004mar01 Icd Rfd Ta

3/3

laminate substrate is preferred.All FE and power amplifier sup-pliers are taking this path.

A crucial step towards thereduction of passive compo-nents advancing passive inte-gration is PASSI technology.This process features capaci-tors with 145pF/mm2 and a 4percent (3) accuracy, plus in-ductors with Q-factors over 50.It also serves as a platform tointegrate lateral pin diodes,

high-density capacitors and, inthe future, MEMS variable ca-pacitors and switches.

Another related develop-ment is the bulk acoustic wave(BAW) technology, which is ca-pable to replace ceramic andSAW technologies in filters.

Advantages of BAW overSAW technology are perfor-mance, losses, thermal charac-teristics, size and cost especiallyat frequencies above 1GHzwhere SAW technology requiresthe use of submicron litho. Thelosses of SAW filters quickly in-crease above 2GHz due to thesubmicron structures, whileBAW technology can be used atleast up to 10GHz. Due to thecosts associated by adding extra

mask steps and yield limitations,it may not be attractive to inte-

grate the BAW technology ontoa BiCMOS process.

Integration on modules andPI on-chip can offer the de-signer the most flexibility insystem partitioning and inachieving the best trade-off be-tween performance and cost fora particular system. This con-sists of an active transceiver ICflipped on a passive IC, con-taining RF and decoupling ca-pacitors, inductors and resis-

tors needed for the transceiverfunction, the total sandwichthen being flipped in a HVQFNpackage.

This combination can beused as RF-subsystem on a

module substrate that alsohouses the power amplifiercontrol loops, matching, RFswitching and filter functions,providing a complete RF sys-tem solution.

A non-technical, trend in thecellular market is the outsourc-ing of RF functions and fullsystem solutions are becom-

ing an accepted businessmodel. Trends in forward inte-gration as described above willcontinue into the baseband andpower management domain.Full system solutions will takeoff if the function (for example,a GSM phone) is so mature thatOEMs regard it as a commodity,not something that providesthem a competitive edge. Untilnow we have assumed a systempartitioning as it is valid today,with a clear boundary between

RF and baseband functionality.However if the trends in con-nectivity standards are success-ful and prove feasible for cellu-lar applications as well, RF in-tegration might stop at Tx-FEMtype products, not deliveringfull radio functions.

At HF, mainstream CMOStechnology will offer the re-quired bandwidths. However,this RF CMOS may at least dif-

Cross section of ChipScalePackaged BAW filter.

Ultra integration system in package.

fer from standard digital CMOSby lagging one generation be-hind as far as RF is concerned.One-chip integration of digitaland RF may be technically op-timal but is the best design phi-losophy for fabless companies.

In the cellular domain, inte-gration issues focus mainly onthe numerous passive compo-nents that are needed in thismulti-band, multi-system envi-ronment. Dedicated processesfor active technology includingGaAs HBT are used and for pas-sive integration include PASSI.Single-chip digital and RF inte-gration might become techni-cally feasible. Today, RF inte-gration is from the PA to theantenna on the module levelsince this offers the neededmulti-technology integ rationand optimal performance at

minimum cost with maximumflexibility. The transceiver,power management and base-band function is built-up withSMDs around.

Overall, module integrationsystems in package may domi-nate as they enable multi-tech-nology integration and addi-tional features like embeddedantennas and plug and playfunctions at low cost.