Semiconductor devices for RF applications: evolution and current status

Transcription

1 Microelectronics Reliability ) 145±168 Introductory invited paper Semiconductor devices for RF applications: evolution and current status F. Schwierz a, J.J. Liou b, * a Institut fur Festkorperelektronik, Technische Universitat Ilmenau, PF , Ilmenau, Germany b School of Electrical Engineering and Computer Science, Universityof Central Florida, Orlando, FL , USA Received 18 May Abstract This paper reviews the history, evolution, current status, and applications of semiconductor devices for radio frequency RF) applications. The most important developments and major milestones leading to modern high-performance RF transistors are presented. Heterostructures, which are key elements for some advanced RF transistors, are described, and an overview of the di erent transistor types and their gures of merit is given. Applications of RF transistors in civil RF systems with special emphasis on wireless communication systems are addressed, and the issues of transistor reliability are also brie y discussed. Ó 2001Elsevier Science Ltd. All rights reserved. 1. Introduction * Corresponding author. Tel.: ; fax: address: liou@pegasus.cc.ucf.edu J.J. Liou). Since the invention of the bipolar transistor in 1947, semiconductor electronics had been advancing and evolving at an enormous pace. This can be attributed mainly to the dramatic reduction of the device dimensions and therefore the integration of more and more transistors onto a single Si chip, i.e. Si VLSI very large scale integration). Thanks to these advances, microprocessors now contain several tens of millions of transistors, Gbit DRAMs are commercially available, and the Si VLSI is a multibillion industry. The trend of ever increasing integration levels and decreasing device dimensions is expected to continue at least for the next ten years. Many reviews and studies have been published outlining the future development of integrated circuits ICs) and the limits of scaling of Si MOSFETs toward the nanometer range [1±5]. Besides Si VLSI, however, there are other emerging elds in microelectronics. Despite the fact that their current market share is much smaller than that of Si VLSI, some of these elds are in the state of dynamic growth. Among them, the radio frequency RF) electronics with RF transistors as its basic building blocks is likely the most prominent one. In this paper, the term RF is used to refer to electromagnetic waves with frequencies around and above 1GHz. The development of RF transistors went almost unnoticed until 1980 because, unlike Si VLSI, there were no mass consumer markets for RF systems. Most applications of RF transistors had been military or exotic scienti c projects. Examples for military RF applications are equipment for secure communications, electronic warfare systems, missile guidance, control electronics for smart ammunition, and radar systems. In the 80s, satellite television using low-noise transistors operating around 12 GHz in the receiver front-ends was the rst civil application of RF transistors with a market volume worth mentioning. Currently, we witness far-reaching upheavals in civil communication technology that have created mass consumer markets for RF systems. Mobile communications including cellular phones, mobile internet access, and new communication services will have an impact on human society at least as large as personal computers had in the past ten years. These new communication systems transmit, process, and receive great amounts of data in a very short period of time and in the GHz frequency range. RF transistors are the backbone of these modern communication systems. For example, the /01/$ - see front matter Ó 2001Elsevier Science Ltd. All rights reserved. PII: S )

2 146 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 widespread use of mobile phones during the 90s created the rst real mass market for RF transistors. In 1998, for the rst time, more mobile phones than PCs had been produced. The aforementioned shift from military to civil applications of RF systems becomes evident from the articles published in the two leading applicationoriented journals in the eld of RF devices and circuits, namely Microwave Journal Horizon House, Norwood, Mass) and Microwaves & RF Penton, Cleveland, Ohio). In the late 70s and early 80s, the majority of papers published in these journals dealt with military RF systems, and the term ``cost'' rarely appeared in the discussions. In the late 90s, however, a large number of papers were concerned with RF consumer products and cost was one of the most frequently addressed issues. One of the main di erences between Si VLSI and RF electronics is the choice of semiconductor materials and transistor types. Till date, CMOS is the standard device and Si is the only semiconductor used in VLSI. In RF electronics, on the other hand, a wide variety of di erent semiconductor materials i.e., Si, SiGe, GaAs, InP, and wide bandgap materials) and various transistor types, such as bipolar junction transistor BJTs), heterojunction bipolar transistor HBTs), metal-semiconductor eld-e ect transistor MES- FETs), high electron mobility transistor HEMTs) and metal-oxide-semiconductor eld-e ect transistor MOSFETs) have found their applications. The aim of this paper is to review the important developments leading to current high-performance RF transistors, to describe the state of the art of RF transistors, and to discuss aspects of the applications and design of these devices [6,7]. In Section 2, important gures of merit FOM) for RF transistors are introduced. Section 3 summarizes the history of RF transistors starting from the late 50s, with the major milestones in transistor development highlighted. In Section 4, the di erent types of current RF transistors and state of the art performance are discussed. Section 5 deals with the issues of applications, with special emphasis being placed on low-noise and power ampli ers for wireless communication systems as well as on problems of device choice for certain applications. Reliability problems of RF transistors are brie y discussed in Section 6. Summary of this work is given in Section Figures of merit for RF transistors To assess the capabilities and the performance of electronic devices, FOM are often used. FOMs are numbers or quantities that enable device and circuit engineers to estimate device performance and to compare the merits of di erent types of devices. In this section, the commonly used FOMs for RF transistors are introduced The problem of stability Active devices such as RF transistors are unconditionally stable at any operating frequency above a critical frequency f k. Unconditionally stable means that the transistor will not begin to oscillate independently from the value of the signal source and load impedances or from any additional passive termination networks at the transistor's input and output. At operating frequencies below f k, however, the transistor is conditionally stable and certain termination conditions can cause oscillations. The stability behavior of a transistor can be described by the stability factor k as introduced by Rollett [8]: k ˆ 2Re y 11 Re y 22 Re y 12 y 21 ; 1 jy 12 y 21 j where Re denotes the real part of the quantity in brackets and y 11 ;...; y 22 are the frequency-dependent Y parameters of the transistor. If k > 1, the transistor is unconditionally stable, and for k < 1, it works in the region of conditional stability where unintended oscillations may occur Power gain de nitions A key feature of a transistor is its ability to amplify currents and voltages, and thus to deliver larger amounts of power to the load than that received from the signal source. This property is described by the power gain. In general, the power gain is the ratio of the power P 2, delivered from the transistor output to the load, to the power P 1, delivered from the signal source to the transistor input. In practice, the problem is something more complex. The matching conditions between the signal source and transistor and between the transistor and load can in uence the power transfer. Furthermore, only a stable non-self-oscillating transistor can be used as an ampli er. There are several power gain de nitions commonly used to characterize RF transistors. If a transistor is to achieve maximum power gain, then power matching is required. For operating frequencies above f k, power matching is obtained when both the input and output of the transistor are conjugately impedance-matched to the signal source and the load, respectively. The power gain obtained under these matching conditions is the maximum available gain MAG) and can be calculated by MAG ˆ y21 k p k 2 1 : 2 y 12

3 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± dbš ˆ 10 log P 2 : 6 However, when the transistor is operated at frequencies below f k, auxiliary external admittances y 1 at the input) and y 2 at the output) have to be connected to the transistor to suppress its tendency to oscillate. Thus, an overall stability factor K of the network consisting of the transistor itself and of the admittances y 1 and y 2 of equal or greater unity can be obtained. The overall stability factor is given by [8] K ˆ 2Re y 11 Re y 1 Š Re y 22 Re y 2 Š Re y 12 y 21 : jy 12 y 21 j 3 P 2 P 1 P 1 It should be noted that the stability factors and all gains discussed above are calculated from small signal parameters. Therefore, strictly speaking these quantities are applicable only for the small signal case. Nevertheless, MSG, MAG, and U are also used as FOMs for power transistors under large signal operations Characteristic frequencies f T and f max If the admittances y 1 and y 2 are chosen such that at the operating frequency the overall stability factor is exactly equal to unity, and if the input and output of the whole network is conjugately impedance-matched to signal source and load, then the so-called maximum stable gain MSG) de ned as MSG ˆ y21 4 y 12 is achieved. Consequently, MAG is the maximum gain a transistor can achieve at frequencies above f K without any external network. MSG, on the other hand, is the maximum gain obtainable from an RF transistor in combination with external matching impedances under the condition of K ˆ 1at the operating frequency. For MSG, there is no restriction concerning the operating frequency. This means that the operating frequency may be either lower or higher than f k. A comparison between Eqs. 2) and 4) shows that MAG is always lower than MSG except for the case that the stability factor of the transistor is equal to one. The unilateral power gain U [9] is another frequently used measure for the maximum gain attainable from an RF transistor. In general, U is the gain of a two-port network having no output-to-input feedback, but with input and output conjugately impedance-matched to signal source and load, respectively. A zero outputto-input feedback would mean that the output is completely isolated from the input. Because any RF transistor has a non-zero feedback from output to input, a lossless network must be added to cancel the feedback. The resulting network i.e., the transistor in combination with the lossless network canceling the feedback) will not oscillate unintentionally since only an output-toinput feedback not equal to zero can cause oscillations. Thus, the unilateral power gain is de ned over the whole frequency range irrespective of the value of the stability factor k of the transistor. U can be calculated using jy 21 y 12 j 2 U ˆ 4Rey 11 Re y 22 Re y 12 Re y 21 Š : 5 Power gains such as MAG, MSG, and U are commonly given in decibels db): The cuto frequency f T and the maximum frequency of oscillation f max are the most important FOMs for the characterization of the high-frequency performance of RF transistors. The cuto frequency, often also designated as the gain-bandwidth product, is related to the short-circuit current gain h 21. This current gain is de ned as the ratio of the small-signal output current to input current of the transistor with the output short-circuited. It is frequency dependent, and its magnitude rolls o at high frequencies at a slope of ±20 db/dec for any transistor. The cuto frequency is de ned as the frequency at which the magnitude of h 21 decreased to unity. The maximum frequency of oscillation f max is the frequency at which the unilateral power gain U rolls o to unity or 0 db). Therefore, f max is the maximum frequency at which the transistor still provides a power gain. The somewhat misleading designation, maximum frequency of oscillation, stems from the fact that it is also the highest frequency at which an ideal oscillator would still be expected to operate [10]. Like the shortcircuit current gain h 21, U also rolls o at ±20 db/dec. Fig. 1shows the measured current gain and unilateral Fig. 1. Measured current gain h 21 and unilateral power gain U, and extrapolated f T and f max of an RF GaAs MESFET with a channel length of 0.2 lm after Ref. [11]).

4 148 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 power gain of an RF GaAs MESFET with a gate length of 0.2 lm [11]. Also shown in the gure are the characteristic frequencies f T and f max. Sometimes in the technical literature, f max is referred to as the frequency at which the maximum available gain MAG, rather than the unilateral gain U, decreases to unity. This is not entirely correct, but the frequencies at which U and MAG become unity in most cases are not signi cantly di erent from each other. Moreover, it can be shown that the expressions of U and MAG derived from H parameters are very similar [12]. As will be shown in Section 4, the record f max is always higher than the record f T when considering all RF transistors. The value of f max for a speci c RF transistor, however, may be either higher or lower than the value of f T. Transistors with f max > f T can have useful power gains also at frequencies above f T and up to f max.a simpli ed explanation for this is that the current gain lesser than one is compensated by a voltage gain greater than one in the frequency range between f T and f max. Thus, a power gain greater than one is possible. Transistors with f max < f T, however, can achieve power gain only at frequencies up to f max and cannot be used as power ampli ers at frequencies between f max and f T.In the case of bipolar RF transistors, there existed a tradeo between f T and f max. A bipolar transistor designed for maximum f T commonly shows a relatively low f max and vice versa. A frequently asked question is which of the two characteristic frequencies, f T and f max, is more important for RF transistors. There is no unequivocal answer. The commonly cited statement that f T is the more important FOM for digital circuits while for analog applications f max is most signi cant is far too simple. The importance of f T and f max depends on the speci c application of the transistor. Manufacturers of RF transistors often strive for f T f max so that the devices are useful for a large number of di erent applications. Another important issue is that of the maximum operating frequency f op of an RF transistor with certain f T and f max. Again, there is no de nite answer. A rather conservative rule of thumb is that both f T and f max of the RF transistor should be at least ten times higher than the operating frequency of the system in which the transistor is to be used. A less stringent requirement is that the operating frequency of an RF system should not exceed 50% of the cuto frequency of the transistors used [13]. For power transistors, f T must be at least equal to f op and f max should be at least three times that of f op [14]. Clearly, the requirements di er from application to application Minimum noise gure and the associated gain An RF transistor used as an amplifying device receives the signals and noises at its input terminals. Because the transistor cannot distinguish between signal and noise, both will be ampli ed. Besides the external noises, there are noises generated in the transistor. For front-end ampli ers where the signal-to-noise ratio is small, the noise produced by the transistors must be kept as small as possible. An FOM describing the amount of noise produced in RF transistors is the noise gure NF. The NF commonly given in db) is de ned by NF ˆ 10 log P Si=P Ni P So =P No : 7 Here, P Si and P So are the signal powers at the input and the output, and P Ni and P No are the noise powers at the input and output, respectively. The magnitude of NF is dependent of the matching conditions at the input of the transistor, bias condition, and frequency. Unfortunately, both the matching and the bias conditions for minimum noise gure NF min are di erent from those for maximum power gain. Therefore, if the transistor is to be operated under conditions for minimum noise, it will possess a power gain lower than MAG, MSG, and U. The power gain obtained by the transistor biased and matched for minimum noise is called the associated gain G a. For RF eld e ect transistors FETs), one can make a general statement that a transistor with high f T and f max will have a low NF min at high frequencies. In the case of bipolar transistors, the picture is less clear. For example, InP HBTs have higher f T and f max than its SiGe counterparts, yet InP HBTs are noisier than SiGe HBTs Output power and power-added e ciency The output power P out and the power-added e ciency PAE) are two FOMs relevant to RF transistor used in power ampli ers. For such ampli ers, the amount of RF power that can be delivered to the load is of primary importance while the NF is of no concern. P out is dependent of the frequency and of the type of the ampli er circuit, e.g. class A ampli er or class B ampli er. Frequently, the output power is reported in terms of power density. Commonly used FOMs are the output power per mm gate width for FETs and the output power per lm 2 emitter area for bipolar transistors. Despite the fact that these FOMs are not measures of the total output power of a certain transistor, they do give a general idea of the power handling capability of the transistor and allow for the comparison of power performance for di erent transistors or even di erent types of transistor. In power ampli ers where heat dissipation or battery power is of concern, the PAE is an important FOM. It is de ned by

5 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± PAE ˆ Pout rf P in rf ; 8 P in dc where P out rf ) and P in rf ) are the transistor's RF input and output powers and P in dc) is the dc power delivered by the power supply that transformed into heat in the transistor Mean time to failure Reliability is another important issue for semiconductor devices [15]. Regardless of how good and sophisticated the device is, it is useless unless the device possesses a relatively long lifetime. The FOM for the device reliability is called the mean time to failure MTTF). To achieve a high switching speed and operating frequency, most RF transistors are fabricated with III±V compound semiconductors, such as GaAs and InP. For the purpose of discussion, let us consider a widely used RF device, the AlGaAs/GaAs HBT. GaAs is a relatively weak material and is susceptible to electrical and thermal stresses. In addition, GaAs has a poorer thermal conductivity than silicon. As a result, during the AlGaAs/GaAs HBT operation, heat generated by the current ow cannot be dissipated quickly to the ambience and therefore resulted in a high lattice temperature in the HBT. Consequently, stress-induced defects are often generated. Depending on the location of the defects generated, the base current in di erent bias regions is increased. The degree of base current increase is also a function of stress condition, such as the stress time, stress temperature, and stress current level. As the base current is increased along with the stress time, the current gain of the HBT is decreased which is a phenomenon called the long-term current gain drift. An HBT is considered failed when its current gain is reduced by 15% of its initial value. The time it takes to reach this condition is the MTTF. For an HBT under typical and continuous operating conditions, the MTTF ranges from a few hundred to a few thousand hours. For FETs, the MTTF is normally de ned as the time when the transconductance is degraded by 15% of its initial value [16]. This decrease stems from the high electric eld near the drain junction associated with the applied drain voltage, which gives rise to a high temperature in the channel region and stress-induced defects near the oxide±silicon interface generated by the hot carriers. Another important measure of the FET reliability is the gate-oxide breakdown. Depending on the quality of the oxide and the stress condition applied to the gate oxide, soft breakdown SBD) or hard breakdown HBD) can take place in the oxide layer, and a sudden increase in the gate leakage current can occur. In general, the SBD is reversible, whereas the HBD is destructive. 3. Historical review of RF transistors 3.1. The earlyyears In the rst half of the 20th century, vacuum tubes were used exclusively as active devices in RF applications. These tubes were bulky, unreliable, and consumed a large amount of power. Since the invention of the solid-state device, engineers have spent a lot of e orts to increase the operating frequencies of transistors and to replace vacuum tubes in high-frequency applications. Ge BJTs developed in 1958±1959 were the rst transistors operating above 1GHz. As in the rest of semiconductor electronics, however, the dominance of Ge transistors in RF applications declined very quickly. By 1963, Si BJTs began to become competitive, and in 1970, almost all RF transistors were Si BJTs [17]. Fig. 2 demonstrates the output power and NF min obtained from a state-of-theart Si BJTs reported in It became clear in the early 60s that Si is not the optimum semiconductor for RF transistors. GaAs having an about sixfold electron mobility and much higher maximum electron drift velocity compared to Si is a far better material for high-speed transistors. However, there were several fruitless attempts to develop GaAs RF BJTs. The work on GaAs FETs, on the other hand, had been successful and led to new types of RF transistors revolutionizing the whole RF electronics industry. In 1966, Mead presented the rst GaAs MESFET [18]. This transistor operated like a junction FET and consisted of an n-doped active layer and a Schottky contact as the control electrode. Despite the fact that this device was not designed for RF applications originally, it marked a major milestone towards current RF FETs. Since then, the GaAs MESFET evolved rapidly. The rst GaAs MESFET reported in 1967 with GHz capability showed a f max of 3 GHz [19]. In 1970, the record f max of GaAs MESFETs increased to 30 GHz, which clearly exceeds the RF performance of all the Si BJTs at that time [20]. Fig. 2. Record NF min and output power of RF transistors reported in 1970 after Ref. [17]).

6 150 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 Fig. 3. Schematic of a typical RF Si BJT. Fig. 5. NF min vs. frequency obtained from GaAs MESFETs and Si BJTs in Table 1 Output power and maximum frequency of oscillation of GaAs MESFETs and Si BJTs reported in 1980 Transistor type P out f max Si BJT 60 W at 2 GHz 35 GHz 6 W at 5 GHz Fig. 4. Cross-section of a GaAs MESFET. Si BJTs and GaAs MESFETs were the only two RF transistor types in use in the 70s and early 80s. During that period, rapid improvements of GaAs MESFET performance had been obtained while progress in the matured Si technology took place only gradually. Figs. 3 and 4 show the basic structures of an ion-implanted Si RF BJT and an epitaxial GaAs MESFET. The critical dimensions to obtain good RF performance are the base width w B i.e. the thickness of the base layer) in BJTs and the gate length L for GaAs MESFETs, which should be as small as possible. At frequencies below 4 GHz, Si BJTs were commonly used, whereas in the frequency range between 4 and 18 GHz, the GaAs MESFET was the device of choice. Fig. 5 shows the noise performance of the RF transistors in GaAs MESFETs typical gate length of 0.25 to 0.5 lm) with NF below 2 db at frequencies up to 18 GHz and Si BJTs with NF around 2 db between 2 and 4 GHz could be achieved. The tting of the NF min of FETs shown in Fig. 5 can be obtained by [21] NF min ˆ 10 log 1 c n f ; 9 where f is the frequency and c n is a constant containing the in uences of material properties, device design, and technology maturity on NF min. Progress in GaAs MESFET development during the 70s becomes evident GaAs MESFET 10 W at 10 GHz up to 100 GHz when comparing c n for the best experimental devices reported between 1975 c n ˆ 0:072 GHz ±1 ) and 1980 c n ˆ 0:03 GHz ±1 ) see tting curve in Fig. 5). The output power and the maximum frequency of oscillation of Si BJTs and GaAs MESFETs reported in 1980 are listed in Table Development of heterostructures for RF transistors Despite the fact that aggressive device scaling shrinking of FET gate length and BJT base width) always plays an important role in improving the transistor high frequency performance, only the use of heterostructures after 1980 o ered the opportunity of tremendous progress in RF transistor performance. Since then, heterostructures have been the key components in modern high-performance RF transistors such as HEMTs and HBTs. A heterostructure is a combination of at least two layers of di erent semiconductors with distinct bandgaps grown epitaxially one on top of the other. The use of heterostructures in high-speed devices was materialized by the progress in epitaxial growth based on molecular beam epitaxy MBE) [22]. Using MBE, it is possible to grow extremely thin layers with a thickness of only a few nm and with sharp interfaces between the adjacent layers. During the late 70s, intensive work was

7 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± Fig. 6. The energy band diagram of the heterostructure in an AlGaAs/GaAs HEMT. Fig. 7. The energy band diagram of the heterostructure in an AlGaAs/GaAs HBT. done at Bell Labs to grow and characterize sequences of thin GaAs and AlGaAs layers, called the superlattices. Measurements of the Hall mobility in such layer sequences consisting of n-doped AlGaAs and undoped GaAs showed that mobilities clearly exceed those of bulk GaAs or AlGaAs at room and lower temperatures [23]. It was logical then to utilize these enhanced mobilities for fast FETs, and research groups at di erent labs started to create transistors taking advantage of the AlGaAs/GaAs heterostructure. The rst device of this kind came from Fujitsu [24] and was called HEMT. Shortly afterward, researchers from other labs such as Bell Labs, the University of Illinois, Cornell University and Thomson CSF also reported transistors based on the same concept but with di erent names: selectively doped heterostructure eld-e ect transistor SDHT), modulation doped eld-e ect transistor MODFET) and two-dimensional electron gas eld-e ect transistor TEGFET). The heterostructure physics exploited in AlGaAs/ GaAs HEMTs is shown in Fig. 6. The di erent bandgaps DE G of AlGaAs and GaAs results in band o sets DE C and DE V in the conduction and valence bands, respectively, at the heterointerface. In HEMTs, a large DE C is desired which stimulates the transfer of electrons from the n-doped AlGaAs larger bandgap) to the low doped or undoped GaAs smaller bandgap). The transferred electrons are con ned in a two-dimensional electron gas 2DEG) only a few nm thick in the GaAs layer near the heterointerface. Because the 2DEG electrons are spatially separated from the donor impurity, scattering is suppressed and the electron mobility in the channel is increased. The term 2DEG stems from the fact that the electrons can move freely only in the two spatial directions parallel to the interface but not across it. The second type of RF transistors using heterostructures is the HBT. The idea of the HBT is almost as old as the bipolar transistor itself. In 1948, Shockley outlined the advantage of incorporating a heterostructure into a bipolar transistor [25]. In 1957, Kroemer formulated the basic HBT theory [26]. Because, during that time, high-quality heterostructures could not be grown, it was not until the early 80s when practical AlGaAs/GaAs HBTs could be successfully fabricated [27]. The heterostructure physics exploited in HBTs is shown in Fig. 7. The key part of an HBT is the emitter± base heterojunction with the bandgap of the emitter material being larger than that of the base. Because of the bandgap di erence DE G, electrons moving from the emitter to the base encounter a smaller energy barrier to be surmounted than holes traveling from base to emitter. Thus, hole injection from the base into the emitter is strongly suppressed and higher current gains compared to homojunction BJTs can be obtained. The dc current gain of an HBT can be approximated by b N Ew E D B N B w B D E exp DE G kt ; 10 where N E and N B are the emitter and base doping concentrations, w E and w B, the emitter and base widths, D E and D B, the minority-carrier di usion constants in emitter and base, k, the Boltzmann constant, and T, the absolute temperature in the device. Due to the exponential term in Eq. 10), it is possible to dope the base much higher than the emitter without signi cantly decreasing the current gain. Using a high base doping density has two positive e ects on the transistor performance: 1) a low base resistance resulting in high f max and low NF min, and 2) a very thin base without running the risk of base punchthrough) leading to a short base transit time and thus a high cuto frequency. Furthermore, by varying the composition of the base e.g. by varying the Al content in the base of an Al- GaAs/GaAs HBT from higher values at the emitter± base junction to zero towards the collector), a graded bandgap in the base can be engineered. The resulting accelerating base eld decreases the time the electrons

8 152 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 needed to transport across the base and thus increases f T. When considering the lattice constants of the semiconductors used in a heterostructure, three di erent heterostructure types can be found: lattice-matched, pseudomorphic, and metamorphic heterostructures. The rst HEMTs and HBTs made use of the material system AlGaAs/GaAs on GaAs substrates). The lattice constants of AlAs and GaAs are nearly the same. Therefore, Al x Ga 1±x As layers of any Al content x can be grown lattice-matched lm) on GaAs substrates. Because of the existence of the so-called DX centers deep-level states in doped AlGaAs), however, the Al content in AlGaAs/ GaAs heterostructures in typical RF transistors is restricted to about 0.3. Fig. 8 shows the bandgap of commonly used semiconductors as a function of lattice constant. As can be seen, another lattice-matched system is Al 0:48 In 0:52 As/In 0:53 Ga 0:47 As/InP on InP substrates). This system o ers larger DE G values than the AlGaAs/ GaAs counterpart, which makes heterostructures on InP substrates more promising for high-performance RF transistors. The drawbacks of fragility, availability of only small diameter wafers and high price, however, hamper the use of InP substrates in high volume commercial applications. In the beginning of heterostructure research, it was suspected that only materials with nearly the same lattice constant could result in heterostructures meeting the requirements of electron devices. It turned out, however, that it is possible to grow such heterostructures also from materials with di erent lattice constants provided the thickness of the grown layer does not exceed a certain critical value t c. If the grown layer is thinner than t c, its crystalline structure accommodates to that of the substrate material. This causes a lattice deformation of the grown layer and a pseudomorphic pm or p) layer, often also called strained layer, is created. The amount Fig. 8. Bandgap vs. lattice constant plots for commonly used semiconductors. Fig. 9. Pseudomorphic heterostructure on the right) consisting of the layer sequence AlGaAs±InGaAs±GaAs. of strain depends on the lattice mismatch between the substrate and the layer, and on the layer thickness. When the thickness exceeds t c, the grown layer will relax causing a large number of dislocations at the interface. The critical thickness is about 32, 14.5, and 8.5 nm for x values of 0.1, 0.2, and 0.3 in the In x Ga 1±x As/GaAs system, and 780, 150, and 55 nm for x ˆ 0:1, 0.2, and 0.3 in the Si 1±x Ge x /Si system, respectively. Fig. 9 shows the schematic of a pseudomorphic AlGaAs/InGaAs/GaAs heterostructure. Since 1986, pseudomorphic AlGaAs/InGaAs/GaAs heterostructures with In contents in the range of 15% to 25% were successfully grown and pseudomorphic HEMTs on GaAs substrates phemt) have been reported [28]. At about the same time, pseudomorphic heterostructures on InP substrates also became popular for use in phemts [29]. Another pseudomorphic heterostructure frequently used in RF transistors is strained SiGe on Si for SiGe HBTs [30]. The third and newest kind of heterostructures currently used in RF transistors is the so-called metamorphic mm) type. The basic concept is to use a substrate material e.g. GaAs) and to overgrow a graded bu er layer e.g. InGaAs) with a thickness much greater than t c [31]. The bu er layer serves as a relaxed pseudosubstrate for the actual device layer. Because the bu er is extremely thick, dislocations arising at the interface of substrate/bu er barely in uence the electrical properties of the device layer on top of the bu er. Metamorphic InGaAs/InGaAs-bu er/gaas structures with In contents above 0.5 have been demonstrated in metamorphic HEMTs mmhemts) [32]. The main advantage of the metamorphic approach is that inexpensive GaAs substrates can be used to obtain high DE C values, and thus

9 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± InP-HEMT-like performance can be attained with GaAs Recent developments Two new orientations in RF transistor research that occurred in the 90s are worth mentioning. The rst is the application of the standard device of Si VLSI, i.e. the Si MOSFET, as an RF device. Despite the fact that Si MOSFETs had not been considered seriously for RF applications due to their relatively low speed, the continuous FET scaling and increasing maturity of Si MOS technology in recent years has led MOSFETs to become a strong candidate. In fact, the topic of RF CMOS was frequently discussed in all major device conferences around the world in the second half of the 90s. Meanwhile, Si power MOSFET transformed into well accepted RF transistors at frequencies up to about 2.5 GHz. Much e ort has also been spent on the development of low-noise Si MOSFETs. The second direction is the use of wide bandgap semiconductors such as SiC and III-nitrides for RF power transistors with large output powers in the GHz range. The wide bandgap of these materials also allows operating temperatures far exceeding those for Si and III±V transistors. This could open up many new applications of RF transistors, e.g. in automobiles and aircrafts. Fig. 10 summarizes the major milestones of the development of RF transistors during the last 45 years. The frequency limits of RF transistors extended considerably during that period. It is worth mentioning that both bipolar and FETs made from III±V compound semiconductors with cuto frequencies exceeding 200 GHz and maximum frequencies of oscillation above 500 GHz have been reported in late 90s. 4. Current state of the art of RF transistors The evolution of RF transistors from 1980 to 2000 was discussed in Section 3. The introduction of new device structures and semiconductor materials added numerous novel and potential devices for RF application. The frequency limits and thus the operation frequencies of RF transistors have increased steadily. Fig. 11 shows the evolution of the maximum values of f T and f max obtained from the most advanced RF transistors fabricated for research and development purposes. Both f T and f max increased continuously, and further improvement is expected in the years to come. In the following, the recent developments of the various RF transistors is presented III±V eld e ect transistors Both the HEMT and GaAs MESFET are widely used RF devices due to their device simplicity and superior high-frequency performance. Fig. 12 shows the evolution of reported record f max performance of different types of III±V FETs. Also included are the record values for Si MOSFETs, which will be discussed later. For GaAs MESFETs, f max improved only slightly since 1985, while more signi cant improvements have been observed for HEMTs. In other words, the evolution of the ``older'' GaAs MESFET has already saturated, Fig. 10. Major milestones in the development of RF transistors. Fig. 11. Record f T and f max of RF transistors from 1960 to 2000.

10 154 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 Fig. 12. Evolution of f max for di erent basic types of microwave FETs from 1980 to whereas InP HEMTs and mmgaas HEMTs are still in an early phase of development. This trend applied generally to all semiconductor devices, including RF transistors. The evolution of a device started when the principle of a test vehicle in the research laboratory is successfully demonstrated. This is followed by a period of technology maturization and device structure optimization, which leads to rapid advances in device performance. At a certain stage, improvement of device FOMs becomes more and more di cult and saturation occurs. The record values of f T are evidently lower than those of f max. Till date, 0.12 lm gate length GaAs MESFETs with f T and f max of 121 GHz and 160 GHz, respectively, as well as NF min of 0.73 db at 12 GHz and 0.9 db at 18 GHz have been realized [33]. A record f T of 168 GHz has been obtained from a 0.06 lm gate length GaAs MES- FET [34]. At 26 GHz, a 0.13 lm gate length MESFET showed an NF min of 1.3 db [35]. A power module containing four GaAs MESFETs delivered an output power of 125 W at 2.2 GHz [36]. In terms of the GaAs MES- FET output power density, no signi cant progress has been made since The highest output power density reported for GaAs MESFETs is 1.4 W/mm at 8 GHz [37]. The most important developments in the eld of III± V FETs resulted from HEMTs. Fig. 13 shows the crosssection of a typical HEMT. All early HEMTs consisted of GaAs substrate, bu er and channel, and an n-doped AlGaAs barrier layer. A characteristic feature of many RF FETs both MESFETs and HEMTs) is the crosssection of the gate. When the gate length is reduced down to the deep submicron range, the cross-section of the gate decreases. Thus, the resistance of the small gate strip becomes large, which has a negative in uence on the gain and noise behavior at high frequencies. Therefore, the so-called mushroom-shape gate, as shown in Fig. 13, is frequently used to achieve a short gate length and a small gate resistance. Fig. 13. Cross-section of an advanced HEMT after Ref. [38]). The high free-carrier mobilities in the 2DEG channel of AlGaAs/GaAs heterostructures raised much hope that AlGaAs/GaAs HEMTs could surpass both the noise behavior and the frequency limits of GaAs MES- FETs. Experimental work during the 80s revealed, however, that these expectations had been too optimistic. A high 2DEG mobility in HEMT channels does not automatically lead to superior device performance. Moreover, the 2DEG must have a high electron sheet density n s and good electron con nement. These two properties require a conduction band o set larger than that found in the AlGaAs/GaAs system. Therefore, since late 80s, the focus of HEMT research has been shifted to systems that o er large conduction band o sets, such as pseudomorphic heterostructures on GaAs and both lattice-matched and pseudomorphic structures on InP. Another added advantage is that the mobility in the In x Ga 1±x As channel layer is larger than that in the GaAs counterpart. Tables 2 and 3 compare the properties of di erent III±V heterostructures frequently used in HEMTs. The much higher electron sheet densities in heterostructures on InP substrates and in pseudomorphic heterostructures on GaAs substrates result from the fact that two heterojunctions with large Table 2 Bandgap discontinuities and conduction band o sets for different III±V heterojunctions, where lm and pm denote latticematched and pseudomorphic respectively Heterojunction type DE G ev) DE C ev) Substrate Al 0:3 Ga 0:7 As/GaAs GaAs lm) In 0:52 Al 0:48 As/ In 0:53 Ga 0:47 As lm) InP In 0:53 Ga 0:47 As/ InP InP lm) In 0:52 Al 0:48 As/ In 0:65 Ga 0:35 As pm) InP In 0:52 Al 0:48 As/ In 0:8 Ga 0:2 As pm) InP

11 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± Table 3 2DEG mobilities and electron sheet concentrations in HEMTs Heterojunction l cm 2 /V s) n s cm 2 ) Substrate type Al 0:3 Ga 0:7 As/ GaAs GaAs lm) Al 0:25 Ga 0:75 As/ GaAs In 0:15 Ga 0:85 As pm) Al 0:25 Ga 0:75 As/ GaAs In 0:22 Ga 0:78 As pm) In 0:52 Al 0:58 As/ InP In 0:53 Ga 0:47 As lm) In 0:52 Al 0:48 As/ InP In 0:7 Ga 0:3 As pm) In 0:35 Al 0:65 As/ In 0:65 Ga 0:35 As pm) InP DE C contribute to the 2DEG in these structures. In the case of an In 0:52 Al 0:48 As/In 0:53 Ga 0:47 As/InP heterostructure, for example, the upper In 0:52 Al 0:48 As/In 0:53 Ga 0:47 As heterojunction DE C 0:53 ev) and the lower In 0:53 Ga 0:47 As/InP heterojunction DE C 0:21eV) form the boundaries of the channel. In the case of the conventional AlGaAs/GaAs HEMT, only one heterojunction with an DE C of around 0.24 ev existed. As can be seen from Table 3, in pseudomorphic heterostructures on InP, sheet concentrations and mobilities about three times higher than those in conventional AlGaAs/GaAs structures have been obtained, thus resulting in much better RF performance of InP HEMTs compared to AlGaAs/ GaAs HEMTs. During the 90s, a lot of experimental work has been done in the eld of mmhemts on GaAs. The aim was to achieve InP-like HEMT performance but to avoid expensive InP substrates. In 1999, for example, a 3 in. diameter InP substrate costs about $700 compared to $170 for a 4 in. diameter GaAs substrate [32]. Table 4 summarizes the record f T and f max values reported for the di erent III±V HEMTs. Fig. 14 is a Fig. 14. Cuto frequencies vs. gate length of experimental InP HEMTs. compilation of the cuto frequencies of a large number of experimental InP HEMTs collected from the technical literature [6,7]. Connecting the best f T values for di erent gate lengths results in an empirical f T vs. gate length plot, which represents the upper f T limit of InP HEMTs in the year For the gate length ranging from 0.3 to 1 lm, the f T increase is almost linear in the log±log plot. This behavior corresponds to the L ±2 dependence of f T mentioned in various recent FET review papers. For FETs with extremely short gates 0.1 lm and below), f T increases only slightly with decreasing gate length. The reason is that parasitic e ects, such as gate fringing capacitances and gate bonding pad capacitances, become comparable to the actual gate capacitance of the FET. We also constructed the upper limit plots for GaAs MESFETs, AlGaAs/GaAs HEMTs, and phemts on GaAs, which are shown in Fig. 15. The aforementioned saturation of f T occurs for all types of FETs and the general trends of the upper limit plots are similar. As expected from the superior properties of InP heterostructures Tables 2 and 3), InP HEMTs are the best III±V FETs for RF applications in terms of both f T and f max. Surprising trends are found when comparing the cuto frequencies obtained from GaAs MESFETs, AlGaAs/GaAs HEMTs, and GaAs Table 4 Record cuto frequencies and maximum frequencies of oscillation obtained with experimental HEMTs FET type f T GHz) L lm) Reference f max GHz) L lm) Reference AlGaAs/GaAs [39] [40] HEMT phemt on GaAs [41] [42] mmhemt on [43] [44] GaAs lmhemt on InP [45] [46] phemt on InP [47] [48]

12 156 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145±168 Fig. 15. Upper f T limits for di erent types of RF III±V FETs. phemts. Above 0.3 lm gate length, GaAs MESFETs show lower f T than the GaAs-based HEMTs. For 0.1 lm gates, however, the cuto frequency of MESFETs is higher than that of AlGaAs/GaAs HEMTs and GaAs phemts. This behavior might be a result of the decreased in uence of mobility on device speed in shortgate FETs. The current status of III±V FET noise performance is shown in Fig. 16. The NF min of Si MOSFETs are also included. Experimental results published in the literature and Eq. 9) have been used to calculate F min shown in Fig. 16. InP HEMTs possess by far the best noise behavior of all III±V FETs. This can be attributed to the excellent transport properties extremely high mobility and peak velocity) of the 2DEG channels with high In contents. NF min values are below 1dB up to more than 60 GHz, and even at 94 GHz, NF min is only slightly above 1dB. Also, mmhemts on GaAs showed very low NF up to 24.5 GHz. Although the NF of mmhemts on GaAs are slightly higher than InP HEMTs, the inexpensive GaAs substrate makes this device attractive. Despite the fact that phemts on GaAs have higher NF than InP and mmhemts, these transistors are widely used in low noise ampli ers at Fig. 17. Output power vs. frequency for experimental III±V power FET. frequencies up to about 50 GHz. NF as low as 1.2 db have been obtained at this frequency. The reasons for the current popularity of GaAs phemts are that their fabrication is easier and less expensive compared to InP HEMTs and mmhemts on GaAs and that GaAs phemt technology already reached a high level of maturity. The best reported NF of AlGaAs/GaAs HEMTs are lower than those of GaAs MESFETs and only little higher than those of GaAs phemts. The good noise performance of GaAs HEMTs stems from the very high mobilities in the channel. Useful power ampli cation up to 200 GHz using monolithically integrated InP HEMT ampli ers has been reported [49]. The output powers reported for GaAs MESFET, AlGaAs/GaAs HEMT, InP HEMT and phemt on GaAs are shown in Fig. 17. The following relation holds between RF output power and operating frequency: P out f 2 op ˆ c p: 11 In principle, the output power at low frequencies can be increased simply by increasing FET periphery, i.e. by using a larger gate width W. At high frequencies, however, there is an upper limit for W beyond which output power decreases with increasing operating frequency. Around 2 GHz, the highest output powers reported are 200 W for AlGaAs/GaAs HEMTs and 125 W for GaAs MESFETs. At 94 GHz, GaAs phemts and InP HEMTs are the only two transistor types delivering useful power gain. The output powers, however, are orders of magnitude lower than those at 2 GHz Bipolar junction transistors and heterojunction bipolar transistors Fig. 16. NF min for di erent types of RF FETs. Si BJTs are still used in RF systems, and some developmental e orts have been made to improve the high-

13 F. Schwierz, J.J. Liou / Microelectronics Reliability ) 145± Fig. 18. Cross-section of a GaAs HBT after Ref. [50]). frequency performance of these devices. Currently, 100 GHz f max and 84 GHz f T are available from experimental Si BJTs [50]. However, there is a trend to shift research and development to HBTs for RF applications. The rst successfully realized HBTs for microwave applications were based on GaAs. Much work on InP and SiGe HBTs has also been done recently. Fig. 18 shows the cross-section of a GaAs HBT consisting of an AlGaAs emitter and GaAs base and collector regions. The dashed lines in Fig. 18 separate the intrinsic part within the dashed lines) and the extrinsic part outside the dashed lines) of the HBT. The extrinsic HBT does not have any useful e ect on transistor operation, but rather it deteriorates the power gain, f max, as well as the NF min due to the external components of the collector± base capacitance C CB and the base resistance R B. The in uence of R B and C CB on f max can be clearly seen from the expression s f T f max ˆ : 12 8pR B C CB GaAs HBTs are commercially available today and commonly used in RF power ampli ers, while InP HBTs are still less mature. Table 5 summarizes the state of the art of GaAs and InP HBTs in terms of f T and f max. These transistors have a structure similar to that shown in Fig. 18. Both GaAs and InP HBTs show f T and f max values of about 200 GHz, but with InP HBTs having a slight edge over the GaAs HBTs. Moreover, double heterojunction InP HBTs with InP collectors have a higher breakdown voltage compared to GaAs HBTs. Recently, an interesting and novel InP HBT produced by a substrate transfer process has been reported [57]. In these transistors, the size of the extrinsic device part is dramatically reduced, thus leading to a very small external collector±base capacitance and external base resistance and resulting in an extraordinarily high f max. A transferred substrate HBT with a record f max of 820 GHz has been successfully fabricated in 1999 [58]. This is the highest f max ever reported for a three-terminal device till date. Despite the fact that transferred substrate HBTs are not for cost-e ective mass production, they o er the possibility of realizing useful power ampli cation up to extremely high frequencies. The major disadvantage of InP HBTs is, as already discussed before, the brittle and expensive InP substrate. Also the technology of InP HBTs is relatively immature compared to GaAs HBTs. A main advantage of HBTs compared to III±V FETs is that high f T and f max can be obtained without the limitation of photolithography. The lateral minimum device dimensions in HBTs are usually around or above 1 lm, while in FETs with comparable frequency limits, the gate length as the smallest lateral dimension is 0.25 lm or less Table 4). In bipolar transistors conventional homojunction bipolar transistors and HBTs), the electrons carrying the collector current are perpendicular from emitter to collector. Therefore, the device dimensions critical for transistor speed are not lateral but vertical and thus are independent of the photolithography process. The high-frequency noise behavior of III±V HBTs is worse compared to III±V FETs, especially in the higher GHz range. However, the use of HBTs in low-noise ampli ers in wireless communication systems operating at frequencies up to 2.5 GHz becomes an interesting option. State of the art NF of both GaAs and InP HBTs are listed in Table 6. The fact that between 1989 and 2000, no signi cant improvement of the noise behavior of InP HBTs could be obtained is worth mentioning. The main application of III±V HBTs is RF power ampli ers. HBTs o er much higher power densities than Table 5 Reported f T and f max of GaAs and InP HBTs Table 6 NF min of III±V HBTs HBT type NF min db) GaAs 0.9 at 2 GHz InP 0.46 at 2 GHz InP 0.6 at 2 GHz Transistor type f T GHz) f max GHz) Year of publication Reference NF min db) Year of publication Reference Si [51] GaAs [52] GaAs [53] GaAs 171 ± 1990 [54] InP [55] InP [56] 1.73 at 18 GHz 3.33 at 18 GHz 3.6 at 18 GHz 1998 [59] 1989 [60] 1999 [61]