1382
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 36, NO. 9, SEPTEMBER 2001
`
`Technology Developments Driving an Evolution of
`Cellular Phone Power Amplifiers to Integrated
`RF Front-End Modules
`
`Rik Jos, Member, IEEE
`
`Abstract—Millions of cellular phone power amplifiers (PAs) are
`produced every day worldwide using a great diversity of technolo-
`gies. This is true both for the active devices, where various silicon
`as well as GaAs transistors are used, and for the PA architecture, in
`which MMIC, module, and discrete solutions compete. This paper
`gives an overview of the various technological and architectural
`choices and discusses their influence on the PA performance, no-
`tably the power efficiency and linearity. It sketches the future of
`PA development toward more functional integration, which may
`be obtained by two paths: integration on chip and added function-
`ality in modules.
`
`Index Terms—Efficiency, front-end modules, linearity, PA archi-
`tecture.
`
`I. INTRODUCTION
`
`C ELLULAR phone power amplifiers (PAs) usually come
`
`in three variants: discrete transistor line-up, monolithic
`microwave integrated circuit (MMIC), and power amplifier
`modules (PAMs). The discrete line-up is the oldest and cheapest
`solution and is still widely in use in, for example, AMPS PAs
`[Fig. 1(a), a discrete GSM PA is indicated by white dots]. The
`technology of choice is silicon bipolar for cost reasons. A major
`disadvantage is that the telephone set maker has to design the
`PA himself. Due to the large area that is required for a discrete
`solution, many parasitics are present and the RF design is often
`cumbersome, especially as peak powers increase while supply
`voltages drop and frequencies get higher. The setmaker requires
`extensive in-house RF competence and the time to market
`(TTM) takes a long time. These facts make discrete line-ups
`unattractive for fast development of PAs for systems above
`1 GHz. That is where the MMIC gets in. Once mostly made
`of GaAs MESFETs, the present-day state-of-the-art MMICs
`are using GaAs HBT [Fig. 1(b), indicated by white dots]. The
`main reason for HBT’s popularity is its single supply voltage.
`Although epigrowth for HBTs is more expensive than for
`p-HEMT, the processing is cheaper because of a more relaxed
`lithography. Output and sometimes also interstage matching
`circuits are not integrated on chip. The setmaker needs only
`limited RF knowledge. The parasitics in a GaAs IC are small
`and well known, facilitating a fast development and often a
`design that is right the first time. Price, power added efficiency
`
`Manuscript received December 12, 2000; revised March 28, 2001.
`The author is with Philips Discrete Semiconductors, 6534 AE Nijmegen, The
`Netherlands (e-mail: Rik.Jos@philips.com).
`Publisher Item Identifier S 0018-9200(01)06106-6.
`
`(PAE), and linearity are well balanced. However, MMICs still
`do not provide the entire PA function and price and PAE can
`in practice be compromised by external components needed
`for matching. PAMs, on the other hand, deliver a complete
`RF function, minimizing the required RF competence of the
`setmaker and in principal lowering the cost of ownership of
`the PA function [Fig. 1(c)]. Currently PAMs are on the market
`using silicon bipolar, silicon LDMOS, or GaAs HBT. The
`main advantage of a PAM is the possibility to combine various
`technologies which makes it easy to add extra functionality
`, antenna switch, antenna, load
`like output matching to 50
`switching, power control loop or circulators. Candidate tech-
`nologies include Si, SiGe, GaAs active devices, and a broad
`range of options for passive devices made on silicon, on glass
`or in the substrate in a multilayer technology.
`Future developments in PAMs will be twofold: adding more
`functionality to the module and increasing the performance to
`meet the requirements of broad-band systems like wide-band
`CDMAs. The intrinsic performance of Si and GaAs devices
`is probably good enough to do the job at frequencies up to
`2.4 GHz. SiGe HBT will offer no significant intrinsic benefit
`over Si BJT in a PA function. However, it offers the possibility
`of a more relaxed lithography combined with a low base sheet
`resistance, thereby avoiding current crowding. Application of
`both Si and SiGe technologies is seriously hampered by para-
`sitic elements which makes it difficult to compete with GaAs.
`Development efforts will therefore focus on increased perfor-
`mance by reduction of parasitics, which can for example be done
`in a Si technology by a very radical isolation technique called
`Silicon On Anything (SOA), which is based on a technology
`previously published [1].
`
`II. ACTIVE DEVICES
`
`It is difficult to compare the capabilities of various device
`technologies for PA applications by considering only the
`usual figures of merit like
`,
`,
`and power density.
`Performance of PAs is typically measured by efficiency (PAE),
`ruggedness (the ability to survive power mismatch conditions
`at
`the output using specific dc and RF drive conditions),
`linearity (i.e., nonlinear signal distortion like intermodulation
`and ACPR) and of course, price and size. In the process of
`selecting a suitable active device, the problem is to relate the
`PA performance parameters to the active device parameters,
`which is not a trivial task. We will treat the most important
`items and focus on bipolar technologies for active devices.
`
`0018–9200/01$10.00 © 2001 IEEE
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1382
`
`

`

`JOS: EVOLUTION OF CELLULAR PHONE PAs TO INTEGRATED RF FRONT-END MODULES
`
`1383
`
`Fig. 1. Boards with PAs in different shapes: (a) discrete line-up; (b) dual band GaAs MMIC’s; and (c) PA module.
`
`A. Power Added Efficiency
`The efficiency of a PA is mainly determined by efficiency
`of the power transistor and losses in the output matching net-
`work. The latter strongly depends on the choice of technology
`and PAMs with their integrated matching generally score worse
`than MMICs with a matching circuit on the board. This solution
`has fewer losses because components like strip lines and capac-
`itors can be made much larger, of course at the cost of size. In a
`PAM, losses can vary between 0.9 dB, which is equivalent to a
`loss of 900 mW from a 5 W transistor output for 900-MHz GSM,
`to 0.3 dB, for which improved technologies like plated bonded
`copper and lossless MOS capacitors are needed. Power transis-
`tors can run with power added efficiencies anywhere between
`50%–80%. There are several factors influencing this. Power
`gain limits the PAE if the gain is low, say below 10 dB. In prac-
`tice, however, this is usually not very important when compared
`to two other factors. One is the output capacitance in combina-
`tion with losses in the Si substrate [2]. The output capacitance
`is partly collector–substrate or drain–substrate capacitance and
`partly interconnect capacitance. The associated substrate losses
`limit the efficiency. GaAs scores significantly better than Si
`where substrate removal techniques or very highly or very lowly
`doped substrates are needed to achieve competitive parasitics.
`The other factor determining PAE is the harmonic impedance
`at transistor input and output and especially the second-har-
`monic termination. If both input and output second-harmonic
`terminations are shorted, the device operates in an ideal class
`AB mode, achieving efficiencies of 78% in theory if parasitic
`losses are negligible. With nonshorted second harmonics, the
`device can have high efficiency when operating in a kind of in-
`verse class AB. In this case, the current waveform is more or
`less sinusoidal and the voltage wave has a half-sine form. Fig. 2
`shows the inverse class AB waveforms and Fig. 3 the PAE as
`a function of the phases of the second-harmonic reflection co-
`efficients at input and output. Full simulation details are de-
`scribed in [3]. Experiments have shown that it is possible to run
`Si bipolar devices, when optimally designed to reduce parasitic
`losses, with power-added efficiencies approaching the theoret-
`ical limit of 78% in inverse class AB [3]. Although high effi-
`ciencies can be obtained, the inverse class AB has a higher peak
`
`V ) and current (I ) versus time.
`Inverse class AB voltages (V
`Fig. 2.
`Note that the shape of I resembles a sine wave, whereas that of V resembles
`the positive half of a sine wave.
`
`;
`
`Fig. 3. PAE versus phases of second-harmonic source and load reflection
`coefficients. Class AB is only reached at one point, in which both input and
`output second-harmonic impedances are shorted. At random impedances,
`chances are good that one operates in inverse class AB.
`
`.
`voltage at the output than normal class AB by a factor of
`This means that the device needs to have a higher breakdown
`voltage to meet the normal ruggedness standards and this is at
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1383
`
`

`

`1384
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 36, NO. 9, SEPTEMBER 2001
`
`TABLE I
`DATA FOR DEVICES
`
`and less gain. So there is a tradeoff be-
`the cost of a lower
`tween efficiency, ruggedness, and gain. The better the product
`of breakdown voltage and transition frequency (
`) is,
`the better this tradeoff can be made and the easier a PA de-
`sign will be. Fig. 4 shows
`values for emitter–collector
`(
`) and base–collector (
`) breakdown as a function
`of
`and
`, respectively, for GaAs, InP, SiGe, and
`Si bipolar devices. Data were taken from literature or design
`
`manuals [see Table I]. There is no fundamental difference be-
`tween Si and SiGe transistors. However, in very shallow doping
`profiles, when
`V and
`V, the influ-
`ence of nonlocal avalanche becomes notable and the
`product reaches values above 200 for
`and above 500
`for
`. Such values are reached by low-voltage SiGe
`HBTs that have epitaxially grown, shallow doping profiles. The
`product of these HBTs is comparable to the values
`reached by GaAs HBTs at higher voltages. In today’s power am-
`plifiers, using battery voltages of 3 to 4.5 V, peak output voltages
`reach values as high as 18 V during mismatch.
`is not im-
`portant in this respect since there is always a low ohmic path
`between base and emitter. The real limiting parameter is
`which consequently has to be higher than 18 V. This means that
`the transistor needs a relatively thick collector and no benefit can
`be obtained from nonlocal avalanche. Si and SiGe bipolars per-
`form similarly whereas GaAs HBTs have higher
`prod-
`ucts. Although silicon LDMOS and GaAs FET and p-HEMT
`score very poorly on
`, very good PAs can still be made
`with these devices, indicating that this performance parameter
`is not the last word in PA performance. GaAs p-HEMTs have
`high efficiencies, up to 65% at 2 W and 1.8 GHz [4]. The dis-
`advantage of p-HEMT is that it needs a switch to turn it off.
`This is an expensive, space-consuming MOS device that re-
`duces the overall PA efficiency by about 5%, thereby annihi-
`lating the p-HEMT intrinsic efficiency advantage. LDMOS de-
`vices suffer from a high intrinsic output capacitance between
`drain and source. At low supply voltage, the output impedance
`gets very low and matching problems arise. This can be solved
`by splitting the output transistor into two or more sections and
`matching these separately at a higher impedance level. Some
`sort of power combining is necessary to obtain the full power of
`all sections.
`
`B. Linearity
`
`In general, linearity or nonlinear signal distortion in an elec-
`tric circuit is an interplay between the nonlinear elements in de-
`vices, the linear device elements including parasitics, and the
`circuit elements. Improving circuit performance is not merely
`a matter of exchanging an active device for a more linear one.
`The role that the intrinsic nonlinearities of active devices play is
`often obscured by the linear elements, i.e., the parasitics as well
`as the matching and feedback components, since these tend to
`linearize the total circuit behavior. This is the reason why bipolar
`devices in practice usually score rather well in comparison to
`FETs which are intrinsically more linear. The usual parameter
`to describe small-signal distortion is IP3/Pdc, i.e., third-order
`intercept point divided by the dc power consumption of the de-
`vice. IP3/Pdc is not adequate to describe class AB behavior,
`which is the operating mode of the power amplifier output tran-
`sistor and its driver. In AB mode, clipping behavior of the device
`plays an important role. This is determined by knee voltage (e.g.,
`) and quasi-saturation. Clipping results in gain compres-
`sion and amplitude–amplitude (AM–AM) modulation. Simula-
`tion results of two-tone intermodulation as a function of input
`power show that as soon as gain compression sets in, the dis-
`tortion rises considerably above the level predicted by the in-
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1384
`
`

`

`JOS: EVOLUTION OF CELLULAR PHONE PAs TO INTEGRATED RF FRONT-END MODULES
`
`1385
`
`Fig. 4. Literature data on BV f versus BV and BV
`(open circles) bipolar transistors.
`
` f versus BV
`
`for SiGe (crosses), Si (solid diamonds), GaAs (solid triangles), and InP
`
`Ge content in the base and the same collector thickness as the
`BJT. The ACPR (IS95) simulations yield
`46.3 dB for the Si
`BJT, which is exactly the same as measured on an actual de-
`vice, and
`46.8 dB for the SiGe HBT. There are currently no
`measured data on the SiGe HBT available. Both simulation and
`measurement were done at the 1-dB compression point, which
`is 12.7 dBm at 3.3-V supply voltage. The device emitter size is
`78 m by 0.8 m. A description in detail of these simulations
`can be found in [7] and [8]. At this moment, it is still unclear
`whether Si or SiGe can meet CDMA or W-CDMA linearity re-
`quirements at rated output power.
`
`III. ARCHITECTURE
`
`Cellular phone systems are and will be dominated by GSM in
`the forthcoming five years by volume. However, linearity and ef-
`ficiency requirements will be more stringent on wide-band sys-
`tems (2.5 G and 3 G), and these will drive the development in
`PA performance. Next to this, other developments will focus on
`adding functionality to the power amplifier. This stimulates a
`trend toward modules.
`Fig. 6 shows typical dual-band GSM cell phone RF func-
`tions. The antenna is followed by a diplexer, which is essen-
`tially a low-pass and a high-pass filter to separate the 900- and
`1800-MHz bands. Transmit and receive signals can be sepa-
`rated by a switch, because GSM is a time-multiplexed system.
`In the receiver path, a SAW filter eliminates all out-of-band
`noise to prevent overloading of the LNA. The LNA is either
`stand-alone or integrated into the transceiver IC. The transmit
`path may consist of a power VCO, also either integrated in the
`transceiver IC or stand-alone, which directly feeds into the PA.
`There are also other options for the path between the IC and PA
`but we will not discuss those in this paper. The output power
`of the PA is monitored and a feedback loop takes care of the
`powercontrol. Higher harmonics, caused by class AB operation
`of the power stage of the PA, are filtered out. Several functions
`will be integrated into an RF front-end module: power detector
`and control, antenna switch (Tx/Rx), diplex filter, lumped filters
`and, possibly, antenna. Fig. 7 shows an example of a triple-band
`GSM PA on an LTCC substrate with integrated antenna switch,
`diplexer, and harmonic filters. For the matching circuits and the
`diplexer, silicon chips are used with passive integration tech-
`nology. This is a cheap process with five mask layers that al-
`lows for capacitors and inductors. Q-factors well above 40 have
`been demonstrated for inductors at 2 GHz in this technology.
`The accuracy of the matching circuits is much better than what
`
`Fig. 5. Doping profiles for Si and SiGe used in ACPR (IS95) simulations. The
`base of the SiGe HBT is uniformly doped with 20% Ge. Collector thicknesses
`are identical for Si and SiGe.
`
`tercept point method [5]. AM–AM effects are especially impor-
`tant in modulation schemes with high peak to average power
`ratios like W-CDMA. Also, amplitude–phase (AM–PM) mod-
`ulation can contribute to intermodulation even well below the
`onset of gain compression if the amplifier is in deep class AB.
`There are two types of transistor nonlinearities, one linked to
`AM–AM and one to AM–PM modulation. Nonlinear transfer
`functions, like, e.g., transconductance, are related to AM–AM
`modulation. For instance, adjusting the quiescent current influ-
`ences the transconductance. This can be seen in class AB most
`notably at low output power, where gain compression is associ-
`ated with distortion. Also, clipping behavior can be considered
`as a kind of nonlinear transfer function at high output power. The
`fact that FETs and bipolar devices behave not very differently
`in class AB shows that clipping is often dominant. Nonlinear
`charge storage, which for instance plays a role in base-push out
`and quasi-saturation, is related to AM–PM modulation. In gen-
`eral, one can say that if the variation in the stored charge in the
`device as function of current and voltage ( Q/ V and Q/ I) is
`smaller, then the
`is higher and the distortion is better because
`the AM–PM modulation is reduced. The
`(
`product)
`of GaAs is superior to Si or SiGe due to a higher mobility and
`velocity overshoot. This is an intrinsic benefit for GaAs devices
`and they are therefore the first choice when it comes to linearity.
`It is not expected that SiGe will perform better on distortion than
`Si under large-signal conditions. This is corroborated by ACPR
`simulations performed with the mixed level simulator MAIDS
`[6] on transistors under CDMA conditions. Fig. 5 shows doping
`profiles of a Si BJT and a SiGe HBT, having 20% uniform
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1385
`
`

`

`1386
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 36, NO. 9, SEPTEMBER 2001
`
`Fig. 6. GSM front-end building block structure. The 900-MHz block diagram is shown in detail, and the 1800-MHz section is similar. Transceiver and base-band
`ICs are used for both 900 and 1800 MHz.
`
`in a front-end module. The filters are of ceramic resonator
`or SAW types which are large and expensive. Furthermore,
`SAW filters above 2.5 GHz are expected to be very expensive
`because either diamond substrates or submicron lithography is
`needed. A possible alternative is a Bulk Acoustic Wave (BAW)
`filter. Such a filter is made using BAW resonators as shown in
`Fig. 9. It consists of two electrodes with a piezo-electric film
`in between, in which acoustic waves are electrically excited.
`The resonator is made by depositing several layers onto a
`substrate and the actual resonator is isolated from the substrate
`by an acoustic reflector to avoid losses. The piezo-electric
`film, e.g., aluminum nitride, can be deposited as an oriented
`polycrystalline layer. The substrate can be silicon, so the BAW
`technology could in principle be integrated with the passive
`integration technology. A filter is made by using two resonators,
`one series and one parallel, the resonance frequencies of which
`are shifted with respect to each other (see Fig. 10). Such a shift
`can be established by extra mechanical loading of one of the
`two resonators by applying a thicker top electrode metal. For
`further reading on BAW filters, see [9]–[11].
`
`IV. FUTURE DEVELOPMENTS
`
`The key factors that drive PA development are performance,
`cost, and size. The set maker is not so much interested in how the
`PA scores itself on these factors but more in the score of the total
`PA or RF function in the telephone. In the near future, active de-
`vice technology and predominantly GaAs HBT technology will
`not be a roadblock for meeting the requirements for new sys-
`tems. Active device development will therefore be more evolu-
`tionary than revolutionary in the forthcoming years, especially
`for existing systems. Performance, cost, and size will benefit
`most from further integration in an RF front end module as said
`before. Integration cannot be done by merely taking components
`off the board and putting them inside a module, because there is
`no overall benefit to this approach. The costs would shift from
`set maker to module supplier but would not be reduced. What
`limits integration in a cost-effective way is the large variety of
`technologies used in the separate RF functions in a telephone.
`To do something about this, alternative technologies which are
`
`Fig. 7. Example of an integrated front-end module including triple-band GSM
`PA, antenna switch, diplex filter, and harmonic filters. The module is made
`on LTCC using Si passive integration chips for output matching and diplexer
`function. Total size for this experimental module is 140 mm .
`
`can be achieved by placing SMD capacitors on a substrate since
`it is only determined by spreading in the capacitor dielectric
`layer thickness. Besides higher accuracy, passive integration
`also offers a size and price reduction compared to conventional
`matching circuits.
`In Fig. 8, a typical CDMA RF front-end architecture is
`shown. Since CDMA is not time-multiplexed, it is essential
`to keep the transmit signal out of the receive path. Therefore,
`transmit and receive signals are separated by a duplex filter that
`meets very stringent requirements, since Tx and Rx channels
`are only 20 MHz apart. For the same reason, it is unlikely that
`the receiver and transmitter will ever be integrated together in
`a CDMA front-end module. Also, linearity requirements are
`severe for CDMA. To guarantee those under all circumstances,
`the PA is followed by an isolator, which protects the PA
`from mismatch conditions and prevents an increase in signal
`distortion during mismatch at the antenna. For the same reason,
`there is no power control in the PA. Power control is done in a
`separate small-signal amplifier directly after the upconverter.
`The SAW filter is needed to filter out the noise accumulated in
`the Rx band to protect the receiver LNA. From this picture, it
`is clear that the PA is surrounded by components that are im-
`possible to integrate on-chip but also very difficult to integrate
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1386
`
`

`

`JOS: EVOLUTION OF CELLULAR PHONE PAs TO INTEGRATED RF FRONT-END MODULES
`
`1387
`
`Fig. 8. CDMA front-end building block structure. An isolator is used to guarantee PA linearity during antenna mismatch. SAW and duplex filters prevent unwanted
`signals from the transmit path to enter the receive path.
`
`Fig. 9. Bulk-acoustic wave resonator.
`
`A. Parasitic Elements and Number of Components
`
`Reduction of parasitics can be achieved by several methods.
`Flip-chip assembly is an obvious choice, not only for Si but
`also for GaAs, because it eliminates bonding wire inductances
`and metallized via holes in GaAs devices, and because total PA
`module area can be shrunk. Thermal behavior is, however, often
`worse than in conventional assembly methods. How serious this
`is, depends on the phone system (e.g., GSM or W-CDMA) and
`on a careful thermal design.
`Another line of attack on parasitics in the module, especially
`in the SMD components, is passive integration, of which two
`versions exist: one by incorporation of passive elements in the
`multilayer substrate and one by integration on a separate car-
`rier substrate (e.g., a high-ohmic silicon wafer), resulting in a
`component that is flip-chipped onto the module substrate. The
`advantages of the first type are a simpler assembly with less
`components and possibility for large area capacitors and induc-
`tors for, e.g., dc decoupling. The advantages of the second op-
`tion are higher component accuracy, because components are
`defined by lithography, possible co-integration with devices like
`PIN diodes and BAW filters, and lower cost. Both versions re-
`duce the number of SMD components and the total module area.
`The reduction of active device parasitic elements is most
`acute in Si or SiGe since they score poorly on this aspect in
`comparison to semi-insulating GaAs. Too high parasitics will
`
`Fig. 10. Two BAW resonators are used to construct a BAW filter. The
`resonance frequency of the serial resonator is tuned to coincide with the
`antiresonance frequency of the parallel resonator. Tuning can be done for
`example by applying different mass-loading to each resonators.
`
`more suited to integration need to be found for the various func-
`tions. Some of these we have already mentioned: ceramic an-
`tennas that can be integrated, BAW instead of SAW filters, and
`multilayer substrates with well-defined second and higher order
`harmonic matching circuits. Another obstacle to pursue integra-
`tion effectively is the abundance of parasitic elements in the ac-
`tive devices, as well as in the passive (SMD) components, and
`the mere size and number of the SMD components.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1387
`
`

`

`1388
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 36, NO. 9, SEPTEMBER 2001
`
`Fig. 11. SOA process eliminates all parasitics associated with the substrate in silicon-based processes at the same time providing a sort of chip-scaled package.
`
`compromise performance in linearity and efficiency. Even
`though high efficiencies have been demonstrated in Si bipolars,
`parasitics still hamper PA design tradeoffs.
`A very rigorous processing called Silicon On Anything
`(SOA) aims an all-out attack on parasitics. Fig. 11 gives a cross
`section of the SOA process and devices. The devices are made
`in an SOI starting wafer, in which a blanket buried layer is used.
`Device isolation is done by deep trench isolation (DTI) which
`also eliminates parasitics due to silicon underneath inductors.
`After device processing, the Si wafer is glued on a glass wafer,
`the Si substrate is complete removed up to the buried oxide
`and contact openings are made in this oxide to the back of the
`device for collector contact and to the front metallization for
`all other contacts. Thick Cu is plated, providing connections
`to the contact openings and options for interconnect and high
`Q inductors. Note that the large Cu collector contacts provide
`also extremely good thermal contact to the devices. The entire
`SOA IC can be directly PbSn soldered onto the module (or even
`on the board in case of a nonmodule solution). This scheme
`virtually eliminates all parasitics usually associated with Si
`IC’s. Although of course the intrinsic device performance, e.g.,
`, is better in GaAs than in Si, other properties like
`thermal contacting or Q factors for inductors and capacitors
`may actually be better in SOA. Therefore SOA is expected
`
`to give Si the possibility to compete with GaAs, with the
`exception of device linearity.
`
`V. SUMMARY
`
`The performance of several active devices in cellular phone
`power amplifiers has been treated concerning efficiency and lin-
`earity. The conclusion is that GaAs is the preferred technology
`when it comes to linearity, as is required by future wide-band
`(2.5 G and 3 G) systems. Efficiency is limited by parasitic output
`capacitance and second-order harmonic input and output termi-
`nation. Here also GaAs has a distinct advantage over Si because
`it has less parasitics. However, new technologies aimed at re-
`ducing parasitics will bring Si back into the game.
`We have presented a view of the future of PAs which will
`see a development in technologies, not so much in the intrinsic
`active devices as in the peripheral RF devices like antennas,
`filters, and SMD components. Further integration of possibly
`all RF functions in the telephone to inside a module will drive
`technology development. What is needed for that are, for in-
`stance, integrated (multiple) antennas, BAW filters, multilayer
`substrates, passive integration on substrate or on silicon and Sil-
`icon On Anything. These and other technologies will emerge in
`the near future to reshape the RF content of cellular phones.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1388
`
`

`

`JOS: EVOLUTION OF CELLULAR PHONE PAs TO INTEGRATED RF FRONT-END MODULES
`
`1389
`
`ACKNOWLEDGMENT
`
`The author would like to thank F. van Rijs, R. Dekker,
`R. Mahmoudi, N. Pulsford, F. van Straten, P. Lok, M. Klee, and
`P. Löbl for their valuable contributions and fruitful discussions.
`
`REFERENCES
`[1] R. Dekker, P. Baltus, M. van Deurzen, W. V. D. Einden, H. Maas, and
`A. Wagemans, “An ultra low-power RF bipolar technology on glass,” in
`Proc. IEDM’97, Washington, DC, Dec. 1997, pp. 921–923.
`[2] F. van Rijs, H. A. Visser, and P. H. C. Magnee, “Record power added
`efficiency of bipolar power transistors for low voltage wireless applica-
`tions,” in Proc. IEDM’98, San Francisco, CA, Dec. 1998, pp. 957–961.
`[3] F. van Rijs, R. Dekker, H. A. Visser, H. G. A. Huizing, D. Hartskeerl,
`P. H. C. Magnee, and R. Dondero, “Influence of output impedance
`on power added efficiency of Si-bipolar power
`transistors,” in
`IMS2000. Boston, MA: MTT, June 2000.
`[4] D.-W. Wu, R. Parkhurst, S.-L. Fu, J. Wei, C.-Y. Su, S.-S. Chang, D. Moy,
`W. Fields, P. Chye, and R. Levitsky, “A 2W, 65% PAE single-supply
`enhancement-mode power HEMT for 3V PCS applications,” in Proc.
`IEEE MTT-S’97, Denver, CO, June 1997, pp. 1319–1322.
`[5] S. C. Cripps, RF Power Amplifiers
`for Wireless Communica-
`tions. Norwood, MA: Artech House, 1999, p. 196.
`[6] L. C. N. de Vreede, W. V. Noort, H. C. de Graaff, J. L. Tauritz, and J.
`W. Slotboom, “MAIDS, a microwave active integral device simulator,”
`in Proc. ESSDERC’97, Stuttgart, Germany, Sept. 1997, pp. 180–183.
`[7] R. Mahmoudi, J. L. Tauritz, and J. N. Burghartz, “Spread spectrum com-
`munication system performance otimization based on collector epilayer
`engineering,” in IEEE Topical Meeting on Silicon Monolithic Integrated
`Circuits in RF Systems, Garmisch, Germany, Apr. 2000, pp. 167–173.
`
`[8] R. Mahmoudi and J. L. Tauritz, “A systematic simulation of large-signal
`on-chip amplifier modules excited by WCDMA,” in IEEE 55th Auto-
`matic RF Tech.Group Conf. (ARFTG), Boston, MA, June 2000, pp. 1–9.
`[9] H. P. Löbl, M. Klee, O. Wunnicke, R. Kiewitt, R. Dekker, and E. V. Pelt,
`“Piezoelectric AlN and PZT films for micro-elecronic applications,” in
`1999 IEEE Ultrason. Symp., Lake Tahoe, CA, 1999, p. 1031.
`[10] H. P. Löbl, M. Klee, R. Milsom, R. Dekker, C. Metzmacher, W. Brand,
`and P. Lok, “Materials for bulk acoustic wave (BAW) resonators and
`filters,” in Microwave Materials and Applications MMA 2000 Int. Conf..
`Bled, Slovenia, Aug. 2000.
`[11] K. M. Lakin, “Thin film resonators and filters,” in Proc. 1999 IEEE
`Ultrason. Symp., Lake Tahoe, CA, 1999, p. 895.
`
`Rik Jos (M’00) was born in Laren, The Netherlands,
`in 1954. He received the M.Sc. and Ph.D. degrees in
`physics from the University of Utrecht, The Nether-
`lands, in 1982 and 1986, respectively.
`In 1986, he joined Philips Discrete Semi-
`conductors, Nijmegen, The Netherlands, where
`he developed silicon processes and devices for
`RF applications. He is particularly interested in
`semiconductor device physics and modeling with
`emphasis on nonlinear distortion in applications
`like high-efficiency power amplifiers, highly linear
`LDMOS devices, and CATV modules. Since 1995, he has headed the RF
`Device Technology Group of Philips Discrete Semiconductors
`Dr. Jos was a Power Device Subcommittee member of the BCTM in 2000.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on July 07,2020 at 14:31:59 UTC from IEEE Xplore. Restrictions apply.
`
`Petitioners Lenovo Holding Co., Inc., Lenovo (United States) Inc. and
`Motorola Mobility LLC - Ex. 1015, p. 1389
`
`