Previous Page

MICROWAVE INTEGRATED CIRCUITS 2773

MICROWAVE INTEGRATED CIRCUITS

NIROD K. DAS DONALD M. BOLLE
Polytechnic University

In this article a general overview and basic principles of operation of a class of highly integrated analog devices and circuits, used for applications in the microwave and millimeter-wave frequency range, are presented. Although there are subgroups of sister technologies that evolved over the years, having different acronyms such as MIC (microwave integrated circuits in hybrid form), MMIC (monolithic microwave integrated circuits), and MCMs (multichip modules), their basic principles of operation are similar, with their objectives and scopes rapidly overlapping. The discussions in this article may be directed to a broad class of such integrated circuits, referred to in general as microwave and millimeter-wave integrated circuits, with MMIC as a generic acronym. Essential building blocks of MMICs, such as the substrate material and parameters, transmission line geometries, passive and active devices, integrated antennas, integration architectures, and packaging concepts, are explained at a fundamental level for readers new to the subject. Fundamental design considerations and modern analytical and computer-aided design tools for the design of MMICs are introduced. Current trends and future directions of the technology are also discussed. More knowledgeable readers are referred to a selection of significant technical articles for further reading. As indicated, microwave and millimeter-wave integrated circuits refer to a special group of highly integrated analog circuits, operating in the microwave and millimeter-wave frequency range. In this frequency range the various circuit functions that were usually implemented in the past using bulky metal waveguides and coaxial lines, can now be implemented using printed microstrip lines or other forms of planar transmission lines. These planar circuits can also be fabricated together with semiconductor active devices on a single chip, employing a technology similar to that used in microelectronic circuits. As a result, quite complex microwave and millimeter-wave circuits and systems have been realized in a compact, reliable, and cost-effective manner. In many ways, this class of modern integrated circuits has opened the promise and potential for microwave and millimeter-wave communication, much like what silicon digital integrated circuit technology has done for computers. From a historical perspective, after the experimental demonstration of electromagnetic waves by Heinrich Hertz in 1888, and then the successful achievement of transatlantic communication by Guglielmo Marconi in 1901, signal distribution and circuit components in the microwave frequency range were implemented using rectangular metal waveguides. These waveguides were essentially hollow rectangular metal pipes capable of guiding microwave signals, and are sometimes referred to as uniconductor waveguides. They are so named because the rectangular hollow waveguides use only one conductor,

which is fundamentally different from conventional signal transmission in the very-low-frequency range using two conductor transmission lines. Although the uniconductor waveguides had the advantages of low-loss propagation compared to two-conductor lines, due to their limited bandwidth of operation, dispersion and, above all, their bulky physical size, interest later shifted to two-conductor lines, such as coaxial lines, for microwave circuits. However, due to the inconvenience of fabricating circuit components in coaxial form, attened versions of coaxial lines were then introduced. Soon after, attempts were made to implement two-conductor lines by laminating metal strips on a hard dielectric surface in order to greatly simplify the fabrication process. The stripline and the microstrip line were the candidates of choice. The geometries of various waveguiding media used for microwave applications leading up to the printed microstrip line are shown in Fig. 1. A good discussion of the historical developments leading to MMICs is presented in Ref. 1. Sometime in the 1960s, the concept of microwave integrated circuits was introduced. Instead of building individual microwave components separately and then connecting them on a piece-by-piece basis, it was thought cost effective to laminate or print an entire circuit on a single dielectric substrate with individual components (lter, coupler, etc.), connected to each other in a continuous

Hollow rectangular waveguide (a)

Coaxial line (b)

Dielectric substrates Flattened coaxial line (c) Stripline (d)

Air

Dielectric substrates Metalization Microstrip line (e)

Figure 1. Evolution of waveguide geometries (cross sections) used for microwave circuits, from (a) hollow (uniconductor) metal waveguide, to (b) coaxial line, to (c) attened coaxial line, to (d) stripline, and currently to (e) microstrip line. Microstrip line is now the most commonly used transmission line for MMICs. Interest in other forms of transmission lines, such as slotline and coplanar waveguides, came later in order to meet specialized needs.

2774

MICROWAVE INTEGRATED CIRCUITS

integrated fashion. Miniaturization of the circuit was possible by meandering the connecting microstrip lines. Also by using high dielectric-constant substrates, the same electrical size could be achieved while maintaining smaller physical dimensions. The transmission-line components were printed on a hard dielectric substrate by photolithographic processes, and they constituted a major portion of the circuit. Other passive components, such as chip capacitors or chip resistors, and any active components, such as diodes or transistors, were discretely mounted on the circuit board. In this sense the MICs are really hybrid integrated circuits. The substrate materials commonly used include alumina, sapphire, low-loss plastics (ber-reinforced), and ceramics. Although such MICs are much more cost-effective and compact, compared with bulky waveguide circuits, the density of circuits that can be implemented is strictly limited by the precision required in the manual placement of discrete components. Small to moderately complex circuits are implemented in this manner. Complexities could be increased, however, by using double-sided or multilayered circuit boards. It was natural, then, to try direct integration of microwave semiconductor devices together with the printed transmission line components on the same substrate (i.e., in a monolithic fashion), in order to implement active as well as passive circuit functions. This generation of integrated circuits was called MMIC (monolithic microwave integrated circuits) [2]. The result was a dramatic reduction in size, allowed increased circuit complexity, and reduced cost. The substrate material needed for MMICs must be a semiconducting material, such as GaAs or Si, on which both active and passive components can be printed. As a result, the fabrication cost is increased, while allowing batch processing of significantly more complex and compact circuits, compared to the hybrid MIC. A comprehensive discussion of monolithic microwave integrated circuits, specifically those based on GaAs material, is available in Refs. 3 and 4. As the acronyms suggest, the MIC and MMIC are meant for applications in the microwave range. The basic concepts of the technology are similar for applications in the millimeter-wave range, except for the need to maintain tighter dimensional tolerance in the fabrication process due to smaller wavelengths at these frequencies. A more specialized category of monolithic integrated circuits was then developed specifically for applications in a broader frequency band covering the millimeter-wave range. Though MMICs may generically refer to the microwave as well as the millimeter-wave range, a different acronym, MIMIC (microwave and millimeter wave monolithic integrated circuits) is sometimes used to cover millimeterwave applications. MICs, MMICs, and MIMICs are now rapidly taking on new meanings and employing new materials, 3D integration, integration of antennas, optical and optoelectronic components, high-speed digital circuits, and other specialized functions. Circuits and systems of greater complexity may be implemented on a single chip or module, consisting of multiple MMIC or MIMIC chips interconnected and packaged together in a hybrid MIC form [called a multichip module (MCM)] [1]. The entire circuit or system may

also be integrated through a batch process using multilevel processing technology. With the objective merging of these sister technologies, all related technologies for largescale integration of circuits and systems operating in the microwave to millimeter-wave range are generally referred to as microwave and millimeter-wave integrated circuitsgenerically called by the common acronym MMIC. The applications may even cover the lower radio frequency (RF) range. In the following sections the essential building blocks of MMIC, such as the substrate material and parameters, transmission-line geometries, passive and active components, integrated antennas, integration architectures, and packaging concepts are discussed.

1. THE SUBSTRATE The choice of a proper substrate for MMICs is conditioned by several factors, including dielectric constant, resistivity, thermal characteristics, mechanical strength, and fabricational tolerance. For hybrid MICs the required characteristics are low-loss, low-cost, and mechanically rigid insulating materials, such as alumina or soft-plastic substrates. Alumina is a ceramic-type material with relative dielectric constant, er 9.7. Teflon and similar types of soft-plastic materials can provide er values ranging from 2 to 11. Usually the higher dielectric-constant substrates are preferred because they reduce the wavelength of propagation, which results in smaller-size circuits. However, for higher-frequency applications (20 GHz or higher), where the wavelength is already small and fabrication of very small-sized circuit components is a problem, a highdielectric-constant substrate may not be desirable. Substrates of lower dielectric constant may be more useful in this high-frequency range. This results in increased wavelength allowing the design of larger-size circuit components, so that inaccuracies in dimensions during the fabrication process can be better tolerated. For monolithic microwave circuits, where active devices have to be fabricated together with passive components, the substrate will have to be a semiconductor. Si and GaAs are the most common types of substrate materials in use. Two factors become important in the selection of a semiconductor substrate for MMICs: (1) higher substrate resistivity, in order to achieve low propagation loss and (2) higher carrier mobility, in order for the active devices to operate at higher frequencies. Si and GaAs, in their semiconducting states, cannot maintain high resistivity. Therefore, the base substrate must be in an insulating (or semiinsulating) state with higher resistivity levels, on which the passive microwave circuits can be printed. Then active devices are grown on the same base substrate in isolated regions, using ion implantation or epitaxial techniques. Table 1 lists the material parameters of various substrates pertinent to MMICs. It is seen that the electron mobility of GaAs is more than 5 times that of Si, and that the semiinsulating GaAs has a much higher (100 times or more) resistivity compared with that of Si, thus making GaAs a better choice for MMICs [3,4]. However, compared with GaAs, Si fabrication technology is much more ma-

MICROWAVE INTEGRATED CIRCUITS Table 1. Properties of Semiconductors and Insulators Used in Microwave and Millimeter-Wave Integrated Circuits Type of Substrate Semiconductor Si SemiinsulatingSi GaAs SemiinsulatingGaAs InP Insulator Alumina Sapphire Soft-plastic PTFE/glass
a

2775

Relative Dielectric Constant (er ) 11.7 11.7 12.9 12.9 12.6 9.7 11.6 210

Resistivity (r, O cm) 103105 107109 10111014 41014 41013

Electron Mobility mn, cm2/V 800a 4300a

Thermal Conductivity k, W/cm 1K 1.45 1.45 0.45 0.45 0.37 0.46 0.0020.004

At 1017/cm3 doping.

ture, owing to its extensive use in digital electronics. Also, as discussed earlier, in modern MMICs it is desirable to fabricate digital circuits for peripheral processing and control functions together with microwave circuits. The above situation makes it more compelling to try to use Si for MMICs. In this pursuit, the higher propagation loss due to lower values of resistivity of semiinsulating Si is a major hurdle. This problem is overcome by using siliconon-sapphire (SoS) technology, where the base substrate is made out of low-loss sapphire, instead of the lossy semiinsulating Si. Even then, the SoS technology, due to lower carrier mobility of Si, usually nds application at lower frequencies (several gigahertz), leaving GaAs as the principal choice for the millimeter-wave range. Assuming a printed microstrip line as the transmission line-of-choice for signal distribution in a MMIC, Fig. 2 shows the signal attenuation constant a for different
0.8

0.6 Si 0.4 GaAs 0.2 SOS

Alumina

10 Frequency (GHz)

20

substrates as a function of frequency. A set of parameters of practical interest to MMICs were chosen for these data, assuming copper to be the conducting medium. As seen in Fig. 2, the semiinsulating Si is the most lossy, SOS and GaAs are comparable in their loss performance, whereas alumina substrate provides the lowest loss. It turns out that for SOS, GaAs, and alumina substrates, the loss is dominated by the metal loss, not by loss in the substrate material. For semiinsulating Si, however, the loss in the substrate material contributes significantly to the total loss. Besides material loss in the substrate, the substrate parameters inuence the power lost to radiation in the form of surface waves generated at various transmission-line junctions and circuits. The substrate thickness and the dielectric constant are the governing parameters. Figure 3 shows the effective dielectric constant, ee l0 =l2 , of the fundamental and the rst higher-order surface-wave modes, where l0 is the wavelength in free space, and l is the wavelength of the surface-wave mode [5]. The fundamental mode propagates for all frequencies, whereas the higher-order mode has a cutoff frequency. In order to avoid excessive surface-wave loss, the cutoff thickness (d dc) of the substrate at which the second higherorder mode is excited is often used as a reference value for design of the substrate thickness. As a general rule, up to one-third of the critical value dc can be safely used and will yield reasonable levels of surface-wave loss. This amounts to maximum practical thicknesses of about 725 mm at 10 GHz, and 244 mm at 30 GHz, for GaAs, and somewhat higher thicknesses for Si and alumina. These thicknesses do not usually pose a manufacturing problem for monolithic circuits, but can be an important consideration for mechanical strength in hybrid MICs.

Figure 2. Attenuation (a) in a microstrip line with different substrate materials [semiinsulating Si and GaAs, indicated simply as Si and GaAs, SOS (silicon-on-sapphire), and alumina] as a function of frequency. All lines have approximately 50 O characteristic impedance. Alumina substrate thickness H 500 m (for hybrid circuits), and for all other substrates H 100 m. Resistivity of SOS 1014 O cm; for semiinsulating GaAs 107 O cm; for semiinsulating Si 103 O cm; and for alumina 1011 O cm. Line width W: SOS and semiinsulating Si, 80 m; semiinsulating GaAs, 70 m; alumina, 500 m. Conducting medium is assumed to be copper.

(dB/cm)

2. CHIP SIZE AND CIRCUIT COMPLEXITY An estimate for the circuit complexity that one can achieve in a MMIC of a certain size, at two selected frequencies3 GHz for lower-frequency and 30 GHz for higher-frequency applicationsis presented here. A Si substrate is assumed for calculations, but the estimates should be close for GaAs and alumina, since these materials have similar dielectric constants. At 3 GHz, a

2776

MICROWAVE INTEGRATED CIRCUITS

GaAs 3.0 Microstrip line 2.5 )2 Coplanar waveguide

0/

=(

2.0 TM (fundamental) 1.5 TE (first higher order mode) 0 0.1 d/

0

Slotline Metalization Substrate

Coplanar stripline

1.0

0.2 0.3 TE-cut-off (0.26)

0.4

Figure 4. Different congurations of printed transmission lines (cross sections) currently used in MMICs. The slotline and coplanar waveguides may sometimes use a conductor backing under the dielectric substrate for added mechanical support, but require careful design [8,9,30]. Other variations of these transmission lines with multilayer substrates are also useful.

Figure 3. Dispersion characteristics for the fundamental (TM) and the rst higher-order (TE) mode of a GaAs substrate with a metallized ground plane on one side, used in MMICs. The cutoff p value of er d=l0 for the rst higher-order TE mode is 0.26. The dispersion characteristics for Si and alumina substrates are very similar to that of GaAs, with the corresponding cut-off numbers for the rst TE modes equal to 0.261 and 0.264, respectively.

3. TRANSMISSION MEDIA Figure 4 shows various congurations of printed transmission lines that are used in MMICs [6,7]. Other variations of these transmission lines with different arrangements of the dielectric substrates or metal planes are also useful. For example, the slotline or coplanar waveguide may be used with a conductor backplane for added mechanical support and increased signal isolation [8,9]. Among the geometries in Fig. 4, the microstrip line is the most commonly used transmission line for MMICs. However, under certain situations other geometries may be more suitable. Table 2 compares the practical features of various transmission lines. Specific technical details of the transmission lines can be obtained for microstrip [10,11], coplanar waveguide [12,13], and slotline [14]. In the following, basic design data are provided, and important design considerations for MMICs are discussed, based on requirements for the transmission line used. Microstrip line is assumed as the transmission line of choice. Figures 5ae present data for various useful transmissionline characteristics for GaAs, Si, and alumina substrates. Figure 5a shows variation of the characteristic impedance Zc of a microstrip line as a function of W/H, while keeping the substrate thickness H xed, for different substrate materials. Data are plotted for values of W/H around 1, where the characteristic impedance is nominally 50 O. The characteristic impedance reduces for larger values of W/H, owing to an increase in the effective capacitance between the line and the ground plane. Figure 5b shows the attenuation constant owing to material loss for

microstrip transmission line on Si has a guide wavelength l 4 cm. The distributed subcircuits are assumed to be l=4 l=4 1 cm 1 cm in size (these are typical dimensions for distributed circuits) and therefore occupy most of the substrate area. If a mixture of 20% distributed functions and 80% lumped-circuit functions (of size 1 1 mm) are used, a 2 2-in. substrate can accommodate about 120 circuits or, equivalently, about 60 circuit functions if a factor of 2 is used to account for additional space required between components, to avoid intercomponent coupling. At 30 GHz, on the other hand, the distributed and lumped circuits are comparable in size: l=4 l=4 1 mm 1 mm. This will yield about 1300 circuit functions on a 2 2-in. substrate, taking into account a factor of 2 for additional intercomponent spacing. These gures are indicative of a low level of integration density in MMICs, compared with that achievable in digital integrated circuits. This fact strongly motivates the use of multilevel integration by stacking circuit layers with proper electrical isolation between layers. Although this stacked integration leads to other constraints, it can potentially increase the effective circuit density several times, while also allowing convenient integration as well of other functions (digital, optical, antennas, etc.) on independent layers.

Table 2. Characteristic Features of Common Printed Transmission Lines Type of Line Microstrip line Coplanar waveguide (CPW) Coplanar stripline (CPS) Slotline Zc (O) 10100 20150 40250 60250 Loss Low Medium Medium High Dispersion Low Medium Medium High Connect Series Element Easy Easy Easy Difcult Connect Shunt Element Difcult Easy Easy Easy Discontinuity Radiation Low Low High High

MICROWAVE INTEGRATED CIRCUITS

2777

100 Alumina 80

1.0 Si 0.8

Zc ()

50 40

GaAs

(dB/cm)

60

Si,SOS

0.6

0.4

GaAs

20

0.2 SOS Alumina

0 0 1.0 W/H (a) 10 GaAs 9 2.0

0 0 1.0 W/H (b) 2.0

1.6

1.2 = ( 0/ )2 8 Si,SOS 7 Alumina 6 0.4 GaAs

(dB/cm )

0.8 Si

1.0 W/H (c) 0.4

2.0

100 Substrate thickness, ) H( (d)

200

(10 GHz)

0.2 GaAs,Si,SOS 0

= 0.2 Alumina 0.4 0 10 Frequency (GHz) (e) 20

Figure 5. Variation of (a) characteristic impedance Zc; (b) attenuation constant a; and (c) effective dielectric constant ee l0 =l2 , as a function of W/H for the substrates of Fig. 2, but with frequency 10 GHz. Variation of a for semiinsulating Si and GaAs substrates (characteristics for SOS and alumina are close to that of GaAs) as a function of substrate thickness H (with W selected for different H to have Zc 50 O) at 10 GHz is plotted in (d). Frequency variation for the ee of (c) is plotted in (e), in order to show dispersion behavior.

e e

2778

MICROWAVE INTEGRATED CIRCUITS

the same parameters of Fig. 5a. As can be seen from Fig. 5b, the loss increases sharply for smaller values of W/ H, making small values of W/H undesirable. In order to maintain compactness of integration the upper limit of W/ H is restricteda reasonable upper limit for W/H is 4. The aforementioned constraints in W/H limit the range of Zc values that can be practically attained to about 10100 O. Figure 5d shows the variation of attenuation constant for different values of substrate thickness H, while maintaining a 50-O line (with properly adjusted linewidth W). These data show that material loss sharply increases as the substrate thickness is reduced. Therefore, the choice of substrate thickness H below a certain limit is not desirable. It may be recalled that the upper limit for H is restricted, in order to minimize excitation of substrate modes. The microstrip line, like other printed transmission lines used in MMICs, does not support the TEM (transverse electromagnetic) mode. This is so because the material medium around the transmission line is not uniformit is partly air and partly the substrate material. As a result, the transmission line is dispersive. The effective dielectric constants ee of microstrip lines on different MMIC substrates are shown in Fig. 5c. As should be expected, the effective dielectric constant lies between 1.0 and er of the substrate material. The actual value depends on the electrical lling factor of the substrate. This lling factor increases with increased W/H. This is because as W/H increases, the fraction of the total electric eld conned in the dielectric material increases, allowing only a small fraction to fringe out into the air medium. Figure 5e shows the dispersion behavior of ee for the same parameters of Fig. 5c, but keeping the linewidth W xed. Notice that the dispersion is stronger for the line on an alumina substrate. This is because the alumina substrate chosen for the data in Fig. 5e is much thicker than is the case for Si or GaAs. Dispersion worsens for thicker substrates. This is also a consideration that limits the substrate thickness to smaller values. Besides the restriction on the linewidth and substrate thickness, it is important to consider the constraint on the spacing between two adjacent lines. The lower limit of the line-to-line separation determines the minimum level of isolation that can be maintained between nearby circuits. This consequently restricts the compactness of integration. Figure 6a plots the electric eld of a 50-O transmission line on a GaAs substrate as a function of distance D (normalized with respect to the linewidth W) from the center of the transmission line in the transverse direction. The eld rapidly drops beyond the region below the line. At a distance 4 times the linewidth W the eld strength drops to about 3% of its peak value. In order to achieve this high level of isolation, one must, therefore, maintain a center-to-center separation S between two lines of about 8 times the linewidth W (7W for edge-to-edge separation). For a 10% eld isolation the corresponding value for center-to-center separation is about 4W. Figure 6b illustrates how this eld coupling translates to coupling of signal power, as a function of edge-to-edge separation S between lines. For S/W 3 one can achieve better than about 25 dB isolation, which can be increased to about 30 dB

1.0

0.75 Ez/Ez ( = 0)

0.5

98% Field confinement region

0.25

0 6.0

3.0

0 /W (a)

3.0

6.0

0 Maximum coupling (dB)

10

20

30 0

2 S/W (b)

Figure 6. (a) Variation of the eld strength of normal (to the substrate) component of electric eld of a 50-O microstrip line on a 100-m-thick GaAs substrate, sampled at the bottom ground plane, plotted as a function of the distance D from the center of the line. Width W of the line is 70 m, and frequency 10 GHz. (b) Level of power that can be coupled from a signal line to a nearby line as a function of edge-to-edge separation S between the lines. It is assumed here that the signal line is matched to the source and terminated by the characteristic impedance of the line. The coupled line is also match terminated at both ends. Here the maximum level of power that can be coupled between the lines, which occurs when the line lengths are odd multiples of l/4, is plotted. Such characteristics also generally apply to other substrates and frequency, and are useful to determine the minimum separation one must maintain between lines in an MMIC in order to maintain a minimum level of isolation.

for S/W 4. In most practical applications an S/W ratio greater than 3 provides reasonable isolation. 4. PASSIVE COMPONENTS Some of the passive components commonly used in MMICs include (1) resistors, (2) capacitors, and (3) inductors. These components can be implemented in lumped form if their physical size can be maintained sufciently small (ol/10) compared to the operating wavelength l. Otherwise, distributed behavior becomes more pronounced, and therefore the components no longer operate as normally expected. The lumped condition is more easily

MICROWAVE INTEGRATED CIRCUITS

2779

satised at lower microwave frequencies (where l is larger), making the use of such lumped components more practical in this range. Except for resistances, the inductive and capacitive elements can also be realized in distributed form, using a transmission-line stub of a certain length. As a basic principle, however, the lengths of such distributed components are proportional to the operating l. At a low operating frequency, where the operating wavelength may be too large, the required large length of the distributed components can make it difcult to implement in an MMIC. Therefore, the use of the distributed elements in MMICs is limited only to the higher-frequency range (where the wavelength is sufciently small). In the intermediate frequency range (around 1020 GHz) a mixture of lumped and distributed components may prove useful. Besides implementing simply inductances and capacitances in distributed form, a variety of other circuit functions, such as delay lines, couplers, resonators, and lters can be implemented only in distributed form. Transmission-line segments are interconnected in a variety of ar-

rangements to achieve circuit functions that are useful in many microwave applications. However, all such distributed-circuit elements are essentially transmission-line metallizations, which can be fabricated in a MMIC similar to other metal interconnections. The various lengths and widths of transmission-line segments required can be designed using distributed-circuit theory [15,16]. In the following only lumped-circuit elements as used in MMICs will be discussed. Figure 7 shows various congurations of MMIC lumped components, some of which may be relevant to hybrid-type integration, and others to monolithic integration. 4.1. Capacitors Figure 7a shows useful congurations of capacitance components used in MMICs: (1) edge-coupled, (2) end-coupled, (3) interdigitated, (4) end-coupled overlay, (5) overlay, and (6) chip capacitance. Of these, (4) and (5) are relevant only to monolithic integration, (6) only to hybrid MIC conguration, and the others can be implemented in either

(i)

(ii)

Dielectric film (iii) (iv) Dielectic film Metalization

Metalization (i) Substrate Air-bridge (ii)

Dielectric material Metalization

Substrate (v) (a)

Metal plates (vi) (iii) (b) Resistive film

Metalization

Resistive material

Substrate (i) Metal terminals (ii) (c)

Figure 7. Various congurations of lumped passive components used in MMICs. (a) Capacitors in different forms: (i) edge-coupled, (ii) end-coupled, (iii) interdigitated, (iv) end overlay, (v) overlay, and (vi) chip capacitor. (i)(iii) are planar forms showing the top view of the metallizations, whereas (iv)(vi) are nonplanar components showing their side views. (b) Planar inductors in different forms: (i) using a straight section of a high-impedance transmission line, (ii) meander-type line, and (iii) spiral inductor with an overbridge connection. (c) Resistances in two forms: (i) lm resistance and (ii) chip resistance.

2780

MICROWAVE INTEGRATED CIRCUITS

monolithic or hybrid form. All capacitances shown in Fig. 7a are essentially series-type circuit elements, but a shunt-type capacitance can be realized by connecting a via hole to one of the terminals. Up to about 1 pF capacitance can be achieved using an edge-coupled or end-coupled design, with some higher values possible from an interdigitated conguration. Capacitance values in the range of 1030 pF can be realized by overlay-type designs, which are useful for RF-bypass or dc-blocking applications. Much higher values can be implemented only in discrete chip form, which can be used in hybrid MICs, but not in monolithic form. The edge-coupled, end-coupled, and interdigitated capacitors are implemented in MMICs as two metallized lines with a suitable gap(s) maintained between them. No additional dielectric lm is required. In all other designs in Fig. 7a an additional dielectric lm is needed. In a monolithic fabrication process such lm capacitances are realized by controlled deposition of dielectric lms of required thickness. Some important considerations for the dielectric lms to be used include (1) dielectric constant of the material (and hence the capacitance values that can be achieved per unit area), (2) compatibility with monolithic fabrication process, (3) microwave losses, and (4) breakdown eld. The capacitance would exhibit some resistive behavior in the microwave frequencies, due to (1) losses in metal and dielectric lm, and (2) radiation into the freespace and/or substrate medium. Q factors of the order of 50100 can be achievable in the X band (10 GHz). Distributed effects are always present, to some extent, resulting in deviations from lumped behavior of the device. These effects may be taken into account through the use of computer-aided modeling and design. 4.2. Inductors Figure 7b shows different congurations of MMIC inductors: (1) using a high-impedance line, (2) a meanderline type, and (3) a spiral type, among which the spiral type allows a higher range of inductance that can be achieved. All such inductors are implemented in a planar metallized form, and, thus, can be used in hybrid or monolithic integration. In the spiral-type inductor one would need an airbridge to connect to the center of the spiral. This would require an additional fabrication step in monolithic form, or could be manually established in a hybrid MIC. The operation of all inductance elements is based on the production of strong magnetic stored energy in the vicinity of the device (equivalent to the operation of a coil in audio or RF circuits). The goal here is to achieve high inductance within a small physical space. Unavoidable capacitive effects are also present in the planar inductor design, due to edge coupling between lines and the presence of the ground plane under the metal lines. This is in addition to resistive effects owing to material loss (metal and dielectric), as well as radiation. Therefore, the planar inductor does not behave like an ideal inductance, but needs to be treated as an RLC resonant circuit, with a dominant inductive effect in the operating frequency range. All the nonideal factors must be carefully accounted for, requiring the use of accurate

computer-aided modeling tools. Inductance values on the order of 10 nH can be achieved using planar spiral inductors, with Q factors on the order of 50 in the X band. These values are useful for RF isolation/bypass purposes. The use of inductors requiring higher inductance values should be avoided in MMICs. 4.3. Resistors Figure 7c shows two general classes of MMIC resistors: (1) the resistive lm element and (2) the chip resistor. Chip resistors nd application only in hybrid circuits, while the lm resistors are convenient to implement in monolithic circuits. The resistors are useful in resistive loading and match termination elements. Resistances requiring high power dissipation (e.g., in DC biasing) should be avoided in MMICs. The resistive lm used in a lm-type resistor can be realized by the deposition of a lossy metal lm or a semiconductor lm. A lossy metal lm of an appropriate material can be deposited in the MMIC fabrication process, in a manner similar to the fabrication of a lm capacitance. Important considerations for the choice of the lossy metal lm are: (1) sheet resistivity (which determines the resistance per unit length), (2) thermal variation of resistivity, and (3) compatibility with MMIC fabrication. The resistive lm may also be realized in an MMIC using a semiconductor process, similar to that used for active devices. The doping level in the semiconductor controls the resistance value, which can be realized using an epitaxial or implantation technique. The use of metal lms for resistors is usually preferred over semiconductor lms, owing to nonlinearity behavior of the latter at high current values and poor thermal stability. Nonideal effects common to both types of lm resistors include additional capacitive effects between the lm and the ground plane, inductive effects of the metal connection, and some radiative effects. This results in frequency dependence of the performance, which must be properly accounted for in the design process. 4.4. Other Passive Elements Besides the aforementioned passive circuit components, other passive circuit congurations such as (1) junctions between transmission lines, (2) transmission line bends, (3) vias, and (4) open ends, which are frequently used in MMICs, need to be considered. These are generally called discontinuity elements, whose presence is not desirable but inevitable in MMICs. The discontinuity elements can be modeled as RLC equivalent circuits, using a computer simulation or some approximate theoretical approach. Whenever possible, their unwanted effects should be minimized by proper design of the discontinuity itself, or through compensation within the design of the rest of the circuit. It is desirable to integrate antenna elements together with other microwave circuits in an MMIC. This approach is particularly attractive in large integrated phased-array radars, in order to avoid the complexity of fabricating the circuits and antenna elements separately and then manually connecting them. Although current applications of

MICROWAVE INTEGRATED CIRCUITS

2781

MMICs in the wireless communication industry are growing, large phased-array radars are still the major driving force behind MMIC technology. The microstrip antenna [17] is the most suitable candidate for such integration. This is mainly because of the planar nature of microstrip antennas, which can be fabricated with an MMIC process in a manner similar to other metallizations. The basic geometry of the metallization structure of a microstrip antenna, which can be connected to the rest of the MMIC by a microstrip line, is shown in Fig. 8a. The radiation from this microstrip antenna is along the broadside direction (outward, perpendicular to the substrate). Sometimes it may be preferable to have the antenna element radiate

along the endre direction (along the substrate plane). Figure 8b shows one such printed antenna conguration, called a tapered-slot antenna [18]. The tapered-slot antenna shown in Fig. 8b is fed by a slotline, and radiates to the right along the plane of the substrate. One of the drawbacks of integrating antennas with MMICs is that it often occupies significant space on the valuable semiconductor substrate. Another problem is that since the MMIC substrate is normally optimized for circuit functions in order to minimize radiation from the circuit components, the same substrate cannot at the same time be optimal for antenna radiation. Certain techniques may sometimes be used to provide a compromise between

Substrate (bottom side metalized)

Microstrip antenna

Substrate (bottom side not metalized)

Top metalization Tapered slot antenna

Slotline feed

Microstrip feed (a) (b)

Slot Microstrip feed Microstrip patch antenna

Top view Dielectric substrate Semiconductor substrate Microstrip feed (c)

Figure 8. Geometries of printed antennas that can be integrated together with MMICs: (a) a microstrip-line-fed printed antenna (shows top surface) for radiation normal to the substrate; (b) a coplanar stripline-fed printed tapered-slot antenna (shows top surface), which radiates along endre direction, (toward the right side along the substrate plane). In (a) and (b) the antenna and microwave circuitry are printed on the same substrate, whereas (c) shows an aperture-coupled microstrip antenna, where the antenna is printed on a different substrate layer, coupled to the microwave circuit layer through an aperture on a common ground plane.

Slot

Common ground plane Side view

2782

MICROWAVE INTEGRATED CIRCUITS

both functions. In such a situation, a multilevel integration, as shown in Fig. 8c is desirable, where the antenna is fabricated on a cheaper dielectric substrate, independently optimized for antenna radiation, retaining the valuable semiconductor substrate only for circuit integration. The common metal plane between the antenna and circuit layers serves as the ground plane for both. It may be noticed in Fig 8c that the antenna and circuit layers are electrically isolated from each other by the common ground plane, while they are electromagnetically connected between each other only through a small slot etched on the ground plane [19]. This idea can be naturally extended to more layers, with possibly more than one antenna layer, or an additional dielectric layer for distributed feeding circuitry, thus reserving the bottom semiconductor layer mostly for active functions.

5. ACTIVE DEVICES AND PROCESSING For hybrid-type integration there is exibility gained in using heterogeneous active components such as bipolar or eld-effect devices, and two- or three-terminal devices. However, uniformity must be maintained in the type of active device one might use in monolithic integration. In the following the active device and fabrication process used in monolithic MMICs will be discussed. Some early forms of MMICs used two-terminal active devices, such as Schottky and Gunn diodes. Modern MMICs use MESFETs (metal semiconductor eld-effect transistor) as a versatile active component (20). MESFETs are convenient to fabricate in an MMIC process, and are known to provide good performance in implementing a large class of active circuits, including specialized ampliers with low-noise, high-gain, and broadband features, as well as mixers, switches, oscillators, and phase shifters. The power levels that can be achieved from a single MESFET amplier or oscillator are low to moderate. Power may be increased by use of on-chip power combiners. However, operation of MMICs should be limited to moderate power levels, to avoid problems associated with heat dissipation. Power outputs on the order of 10 W can be realized from a single chip using power-combining techniques. For a detailed theory of MESFETs one may refer to texts on semiconductor devices [2022]. Theoretical work and practical techniques for the design of microwave active circuits can be found elsewhere [15,2325]. The MMIC fabrication process starts with a good-quality substrate wafer, followed by more than 3040 individual processing steps. For GaAs MMIC, the substrate material is semiinsulating GaAs, whereas in SoS (silicon-on-sapphire) MMIC the base material is sapphire. The specific processing steps for the two cases are different, though they more or less share a major set of common processing techniques. Only the major processing techniques will be briefly discussed. First, an active semiconducting layer, which is needed for active device fabrication, is formed on the substrate. There are two techniques commonly used in forming the active layer: (1) epitaxy and (2) ion implantation, which may be used independently or in combination. In the epitaxial technique,

a doped single-crystal semiconducting layer can be deposited on top of a crystalline base substrate. An intermediate high-resistivity buffer layer is used in the epitaxy process, in order to screen out any diffusion of impurity atoms from the substrate into the active layer. There are different types of epitaxial growth processes: VPE (vapor-phase epitaxy), MBE (molecular-beam epitaxy), and LPE (liquid-phase epitaxy), each having different basic advantages and drawbacks. In the ion implantation technique, on the other hand, the dopant atoms can be implanted directly onto a semiinsulating semiconductor substrate, using high-energy impurity ions. This process is quite versatile, and even selective doping proles at different locations on the substrate may be possible. However, this process requires a base semiconductor substrate with a high state of purity. Therefore, if direct implantation is not practical, an epitaxially grown buffer layer on top of the primary substrate may be used as the implantation medium. Once the active layers are formed using one or a combination of the above techniques, the active device areas are isolated from the surrounding regions, leaving mesas. This is important for microwave circuits, in order to reduce parasitic coupling to the active components. Mesa isolation may be achieved by deep etching of the substrate around the active area. The mesa-etching process may be substituted by implantation of oxygen onto the epitaxial layer surrounding the active device region. The oxygen implantation creates the required high-resistivity barriers. This process is called isolation implant, which essentially uses the epitaxial deposition in combination with selective ion implantation. Next, ohmic contacts to the active areas are made to provide source and drain contacts for the MESFET. The standard approach involves alloying of Au (gold) or AuGe (goldgermanium) onto the substrate. This results in a heavily doped region under the contact, which facilitates establishing the ohmic junction. The gate regions of the MESFET are then processed. The gates are Schottky-type contacts, which are formed by depositing Ti-Pt-Au (titaniumplatinumgold compound). Because the gate regions are usually small (r1 mm), this calls for high alignment accuracy in the lithographic patterning process. Some form of lithography (optical or electron-beam) is needed here, as well as in other fabrication steps, for accurate definition of devices and the interconnection layout. At this stage the active device processing is essentially completed. The active devices are now subjected to on-wafer DC and microwave tests. Wafers that do not meet process control specications are rejected, in order to avoid any subsequent processing costs. Following this, the rst layer of metallization is deposited for external contacts, transmission line interconnections, spiral inductors, and distributed circuits. Then, thin lms of appropriate resistive or dielectric materials are deposited to form thin-lm resistors and capacitors. Resistors may also be realized using semiconductor material in the initial epitaxial or implantation process. Then a second layer of metallization is deposited to connect thin-lm capacitors, and to form crossovers or other miscellaneous connections. The twolevel metallization process allows topological exibility in

MICROWAVE INTEGRATED CIRCUITS

2783

(a)

(b)

Figure 9. Photographs of commercially used hybrid and monolithic microwave /millimeter wave integrated circuits: (a) A balanced amplier operating in the 28 GHz frequency range, consisting of printed Lange coupler, FETs and associated power supply and biasing circuits (lm/chip resistance, chip capacitance, and printed inductance) integrated in a hybrid MIC form. Actual size is 3 8 in. 1 in. (Picture courtesy of Mini-Circuits, Brooklyn, NY.) (b) A GaAs traveling-wave MMIC 8 amplier (7 dB gain, 18 dBm output power level) operating over a broad bandwidth of 2 GHz to 18 GHz, consisting of six stages of GaAs FETs, printed transmission lines, biasing lm resistance, capacitance and printed inductance, integrated in a monolithic MMIC form. Actual size 0.11 in. 0.086 in. (Picture courtesy of MITEQ Inc., Hauppage, NY.)

the circuit layout. Some of the functionalities of the rstand second-level metallizations may be interchanged as needed. The nal steps in the MMIC fabrication involve backplating of the substrate, in order to provide the ground plane for the circuit. The thickness of the substrate is critical for maintaining correct values of characteristic impedance of the transmission-line components. Therefore, the back side of the wafer must be thinned in a controlled manner. The entire backside is then metallized. Ground connections from the bottom metal plane to the circuits on the top surface are provided through via holes. This is possible by the etching of through-holes at required locations and then metallizing the inner surface of the hole. Wet-chemical etching, reactive-ion etching (RIE), or combinations thereof are used. It may be mentioned that much of the technologies used for MMIC fabrication have been adapted from wellestablished techniques used in the silicon digital IC industry. Each processing step needs to be optimized for MMIC application, howeverparticularly for GaAs MMIC [4]. Figure 9 shows photographs of two integrated circuits fabricated in hybrid and monolithic processes.

6. COMPUTER-AIDED DESIGN It is virtually impossible to design complex MMICs through an experimental trial-and-error procedure. Except for a few simple MMIC components, analytical

formulas are not available for accurate design. This is owing to the complexity of electromagnetic interactions in and between different MMIC components. Because of the nature of MMIC fabrication, any tuning after fabrication would also be quite difcult or impossible. From the above considerations, computer-aided simulation and design [26] play a critical role in the successful design of MMICs. The computer-aided simulation tools for MMICs that are currently available may be classied into four broad categories: (1) purely circuit-based tools, (2) circuit 2D tools, (3) circuit 2.5D tools, and (4) full EM-based tools or 3D EM tools. Purely circuit-based tools perform simulations employing simple circuit theory, which requires the user to provide an equivalent-circuit model for individual components, valid for the microwave frequency range of operation. As a result of the relative simplicity of the circuit theory used in simulation, such tools are computationally fast (particularly for linear circuits), but electromagnetic interactions cannot be properly modeled. This approach has only limited use, because accurate equivalent-circuit models for MMIC components are often not available. On the other hand, purely EM-based tools can rigorously model all electromagnetic interactions in an MMIC, and can handle arbitrary geometries of components, package structures, and microstrip antennas. This is accomplished by treating the entire chip as a 3D (threedimensional) electromagnetic system and, therefore, such tools may be called 3D EM tools. However, compared with purely circuit-based tools, the 3D EM tools are often computationally quite time-consuming. A suitable

2784

MICROWAVE INTEGRATED CIRCUITS

compromise between the purely circuit-based and purely EM-based (or 3D EM) tools is provided by the circuit 2D and circuit 2.5D tools, which are based on hybrid circuit EM models. Compromise between speed and rigor is also achieved, having subgroups among the 3D EM tools, depending on the level of rigor, type, and size of geometries the software tools can handle. In the circuit 2D tools the parameters of the transmission lines are calculated by treating them as innitelength lines, which simplies the problem to a 2D structure. An approximate quasistatic or a more accurate full wave approach may be used to this effect [6,15]. Coupling between nearby transmission lines can also be incorporated via coupled-line analysis. However, electromagnetic effects of transmission line discontinuities, such as junctions, bends, and open/short circuits, coupling among lumped/distributed components, and the effect of

Radome

Antenna

Antenna module

Slots

Electro-magnetic coupling Ground plane

Primary feed network Coupling via slotline Multilayer circuits module

the surrounding package structure, are ignored. The circuit 2.5D tools, on the other hand, model the electromagnetic effects of the discontinuities, in addition to the transmission line parameters, but still fail to model the electromagnetic interactions among components and package. The term 2.5D in circuit 2.5D tools appropriately suggests that the rigor of EM modeling used here lies somewhere between the circuit 2D and 3D EM tools. The various commercial CAD tools that are currently available mainly provide simulation capability, where the operator needs to manually iterate the simulation procedure to arrive at a nal design. In addition to the simulation, limited design capability may also be available in some cases, where a nal design with user-dened parameters can be reached, starting with an approximate design specied by the user. However, owing to the excessive computation needed in design algorithms, such designoriented tools are mostly circuit-based, with very limited EM-modeling. The EM-modeling required for different types of MMIC CAD tools can be performed using a variety of numerical techniques. The major techniques include (1) moment method (MM), (2) nite-element method (FEM), (3) transmission line matrix method (TLM), and (4) nite-difference method (FDM). The individual techniques have their own advantages and drawbacks, and have specific strength in being able to handle specific classes of problems. For example, the moment method may be suitable for planar geometries, whereas FEM can be useful for nonplanar components, such as lm capacitors, via holes, and so on. A comprehensive review of commercial EM simulation tools currently available, their modeling techniques, and scope of application, is presented in Ref. 27. Major suppliers of product lines for MMIC CAD include Hewlett-Packard Co. (HP-EESOF), Westlake Village, CA; Ansoft Corporation, Pittsburg, PA; and Sonnet Software Inc., Liverpool, NY.

Ground plane

Dielectric Active circuits and secondary feed network

Metal film Conventional microstrip

Dielectric Ground plane Coupling slot Optoelectronics, digital control circuitry Control module

Figure 10. Conceptional sketch of a multifunctional MMIC conguration with microwave/millimeter wave active and passive circuits integrated together with printed antennas, optoelectronics, digital circuits, and possibly other specialized functions, in multiple levels. The next generations of MMICs may take such forms in order to realize high-density, reliable, and multifunctional integration on a single package.

Metal film

Packaged microstrip

Figure 11. Geometry of a packaged microstrip line, in contrast to a conventional microstrip line, which can be useful for avoiding parasitic coupling between nearby components. Similar packaged congurations for other types of printed lines are also possible.

MICROWAVE INTEGRATED CIRCUITS

2785

7. CURRENT TRENDS AND FUTURE DIRECTIONS In current MMIC technology there is greater emphasis on advanced interconnects and packaging of MMICs in the form of multichip modules (MCM). In addition to compact, highly integrated chips, future system applications will require greater packaging density and increased functionality at the MCM level. The effect of the packaging structure is recognized to have a critical role in the overall performance of MMICs. Packaging of multichip modules in a tile architecture, interconnected in multiple levels using layer-to-layer transitions, and between chip to substrate or chip to chip using solderless connections, is an attractive approach to achieve the high-density, low-cost, and multifunctionality demands of the future. A sketch of one such architecture is shown in Fig. 10. The multichip modules may be fabricated similarly to the chip itself on a larger substrate, using a common semiconductor processing technique. The functional and fabricational concepts of chips and MCM will therefore merge, expanding the scope of MMICs to also include digital, optoelectronic, and other functionalities. A review of current trends and future directions of MMIC research and technology can be found in Ref. 1. In a multilevel architecture, particularly for applications in higher frequencies, there may be fundamental problems owing to a nonconventional power leakage phenomenon. Under certain conditions power can leak or couple from the printed transmission lines to the surrounding substrate structure [28]. This can cause attenuation of the signal along the transmission line, and also result in undesired coupling to the surrounding structure. Such problems should be carefully considered in advanced MMIC designs. The undesired effects can be minimized or eliminated by using shorting pins, properly designed multilayer substrate arrangement, or new types of transmission media [8]. Besides the nonconventional leakage problem, conventional parasitic coupling between various planar trans-

Top layer Active devices and/or passive devices Heat dissipation PPDW layer Passive devices, power distribution, interconnects Low-loss Bottom layer Active devices and/or passive devices Heat dissipation

Figure 13. Schematic of a hybrid-integration architecture in multiple levels, with parallel-plate dielectric waveguide (PPDW) in the middle level(s) for signal distribution, and printed metal lines in the top and bottom levels for connection to active components. Such architectures will be attractive for millimeter and submillimeter ranges for low-loss signal distribution. Conventional circuits with metal lines (e.g., microstrip line) will prove to be too lossy at high frequencies.

Central dielectric guide

Ground planes

mission line discontinuities may also be prohibitive at higher millimeter and submillimeter frequencies, requiring new techniques to achieve greater isolation between components. Approaches to minimizing the parasitic coupling, by placing components physically farther apart, will not be desirable due to space limitations. Lines fabricated in a boxed manner, with metal walls surrounding the central line (see Fig. 11) may be useful. This can be achieved in semiconductor processing using micromachining technology [29]. Ironically, this concept is equivalent to a attened coaxial line, which in the past was rejected for use in MMICs because of perceived inconvenience in fabrication. Another concern is the metallic loss experienced by printed lines at higher frequencies. This loss can be significantly reduced by using dielectric-type guides (see Fig. 12), where a significant fraction of the eld is conned to a dielectric medium (which will have lower loss than metals). However, the dielectric waveguides will not be suitable for connecting to active devices that require metal connections for their operation. A hybrid architecture with combinations of metal lines and dielectric guides, as schematically shown in Fig. 13, can solve this problem. Designs combining the dielectric waveguide concept, together with other ideas discussed above, may allow us to meet diverse and fundamental challenges.

BIBLIOGRAPHY
b
0 s 0 c 0 s

a External substrate
Figure 12. The geometry of a dielectric guide that exhibits lowloss characteristics, compared with metal lines for millimeter and submillimeter applications. The guide is not truly a dielectric guide, because it has metal planes on top and bottom (sometimes called NRD guide [31], H-guide [32], or PPDW [33], operating in different preferred modes). The metal planes are useful for isolation from top and bottom, permitting integration of circuits in multiple levels on top of one another. Most of the signal is contained in the dielectric medium between the metal planes, resulting in lower attenuation, compared with commonly used metal lines (e.g., microstrip line or coplanar waveguide).

1. N. K. Das and H. L. Bertoni, eds., Directions for the Next Generation of MMIC Devices and Systems, Plenum, New York, 1997. 2. C. Mahle, MMICs in communications, IEEE Commun. Mag. 23(9):815 (1985). 3. R. A. Pucel, Design considerations for monolithic microwave circuits, IEEE Trans. Microwave Theory Tech. MTT29(6):513534 (1981). 4. R. A. Pucel, ed., Monolithic Microwave Integrated Circuits, IEEE Press, New York, 1985. 5. R. F. Harrington, Time Harmonic Electromagnetic Fields, McGraw-Hill, New York, 1984. 6. K. C. Gupta, R. Garg, and I. J. Bahl, Microstrip Lines and Slotlines, Artech House, Norwood, MA, 1979. 7. T. Itoh, ed., Planar Transmission Line Structures, IEEE Press, New York, 1987.

2786

MICROWAVE ISOLATORS Trans. Microwave Theory Tech. MTT-29(11):11881192 (1981). 32. F. J. Risher, H guide with laminated dielectric slab, IEEE Trans. Microwave Theory Tech. MTT-18(1):59 (1970). 33. N. K. Das et al., Multilayer integration of microwave and millimeter-wave circuits: New interconnect methods and design considerations, in N. K. Das and H. L. Bertoni, eds., Directions for the Next Generation of MMIC Devices and Systems, Plenum, New York, 1996, pp. 8396.

8. N. K. Das, Methods of suppression or avoidance of parallelplate leakage from conductor-backed transmission lines, IEEE Trans. Microwave Theory Tech. MTT-44(2):169181 (1996). 9. Y. Liu and T. Itoh, Leakage phenomena in multilayered conductor-backed coplanar waveguides, IEEE Microwave Guided Wave Lett. MTT-39(11):426427 (1993). 10. H. A. Wheeler, Transmission-line properties of a strip on a dielectric sheet on a metal plane, IEEE Trans. Microwave Theory Tech. MTT-25(8):631647 (1977). 11. E. J. Denlinger, A frequency dependent solution for microstrip transmission lines, IEEE Trans. Microwave Theory Tech. MTT-19(1):3039 (1971). 12. C. P. Wen, Coplanar waveguide: A surface strip transmission line suitable for nonreciprocal gyromagnetic device applications, IEEE Trans. Microwave Theory Tech. MTT17(12):10871090 (1969). 13. R. W. Jackson, Considerations in the use of coplanar-waveguide for millimeter-wave integrated circuits, IEEE Trans. Microwave Theory Tech. MTT-34(12):10211027 (1986). 14. S. B. Cohen, Slotline on a dielectric substrate, IEEE Trans. Microwave Theory Tech. MTT-17(10):768778 (1969). 15. D. M. Pozar, Microwave Engineering, Addison-Wesley, Reading, MA, 1990. 16. R. E. Collin, Foundations for Microwave Engineering, McGraw-Hill, New York, 1992. 17. D. M. Pozar, Microstrip antennas, IEEE Proc. 80:7991 (1992). 18. K. S. Yngvesson et al., Tapered slot antennaa new integrated element for millimeter wave applications, IEEE Trans. Microwave Theory Tech. MTT-37(2):365374 (1989). 19. D. M. Pozar, A reciprocity method of analysis of printed slots and slot-coupled microstrip antennas, IEEE Trans. Anten. Propag. AP-34(12):14391446 (1986). 20. S. M. Sze, Physics of Semiconductor Devices, Wiley, New York, 1981. 21. S. Y. Yngvesson, Microwave Semiconductor Devices, Kluwer Academic Publishers, Boston, 1991. 22. S. Y. Liao, Microwave Devices and Circuits, Prentice-Hall, Englewood Cliffs, NJ, 1990. 23. G. Gonzalez, Microwave Transistor Ampliers, Analysis and Design, Prentice-Hall, Englewood Cliffs, NJ, 1984. 24. T. T. Ha, Solid-State Microwave Amplier Design, Wiley, New York, 1981. 25. G. D. Vandelin, Design of Ampliers and Oscillators by the SParameter Method, Wiley, New York, 1982. 26. K. C. Gupta, R. Garg, and R. Chadha, Computer-Aided Design of Microwave Circuits, Artech House, Norwood, MA, 1980. 27. M. S. Mirotznik and D. Prather, How to choose EM software, IEEE Spectrum Mag. (Dec. 5358, 1997). 28. N. K. Das, Power leakage, characteristic impedance and mode-coupling behavior of nite-length leaky printed transmission lines, IEEE Trans. Microwave Theory Tech. MTT44(4):526536 (1996). 29. R. F. Drayton and L. P. B. Katehi, Development of self-packaged high frequency circuits using micromachining techniques, IEEE Trans. Microwave Theory Tech. MTT43(9):20732080 (1995). 30. N. K. Das, Characteristics of modied slotline congurations, IEEE Microwave Theory Tech. Symp. Digest, 1991, pp. 777780. 31. T. Yoneyama and S. Nishida, Nonradiative dielectric waveguide for millimeter-wave integrated circuits, IEEE

MICROWAVE ISOLATORS
VERNON E. DUNN
Space Systems/Loral

Isolators are used extensively in microwave equipment to prevent interaction between other components of the system. This article describes the principles of operation of the most common types of microwave isolators and typical operating characteristics of these isolators. In addition, references are provided for further information on the theory of operation and design details. An isolator is a two-port device that has low insertion loss from port 1 to port 2 and high insertion loss from port 2 to port 1, as shown schematically in Fig. 1. An ideal isolator is represented by the following scattering matrix, which indicates that ideally the device is also perfectly matched at the input and output ports: " S 0 S21 0 0 # where =S21 = 1 1

Isolators nd wide application in microwave systems eliminating interactions between components, for instance, as in Fig. 2a, between a transmitters power amplier and an antenna. The output from the amplier is transmitted with low loss to the antenna, but energy reected from the antenna is absorbed by the isolator. Other typical applications are to reduce interaction between stages of amplication (Fig. 2b) or between a local oscillator and a mixer (Fig. 2c). An ideal isolator would have no insertion loss in the forward direction, innite loss in the reverse direction, and a perfect match at the input and output. It would maintain these characteristics over a wide frequency range and would be able to handle high-power signals in both the forward and reverse directions without limiting or distorting the signal by generating nonlinear intermodulation products. Practical, commercially available isolators typically are expected to have less than 0.5 dB

Figure 1. Schematic representation of an isolator.

MICROWAVE ISOLATORS

2787

properties to form practical devices, most are included in the following general categories:
Transmitter Isolator (a) Antenna

1. 2. 3. 4.
Amplifier

Terminated circulators Faraday rotation isolators Resonance isolators Field displacement isolators

Amplifier

Isolator (b)

Oscillator

Isolator (c)

Mixer

Figure 2. Typical isolator applications: (a) isolating transmitter from antenna; (b) isolating two stages of an amplier; (c) isolating local oscillator from mixer.

All these isolators differ from the ideal. In evaluating these different types of isolators, important characteristics that must be considered are forward and reverse insertion loss and the frequency bandwidth over which those values are obtained as well as the power handling capability. Some approaches are more applicable to some frequency ranges than others, and they differ in their applicability to different transmission media such as waveguide, coaxial line, or microstrip. The following discussions describe the operation of these different types of isolators and compare their characteristics.

insertion loss in the forward direction, greater than 20 dB loss in the reverse direction, and return loss of the input and output ports greater than 20 dB over frequency bandwidths up to an octave. If the isolator must operate over a wider frequency range, some sacrice in insertion loss and isolation performance must be expected. Conversely, better performance may be required for some applications and is obtainable over narrow bands. In a typical application illustrated in Fig. 3, the effect of the less-than-ideal characteristics of the actual isolator would be analyzed by considering the reection at port b resulting from the mismatch between the load and the output impedance of the isolator. This reected signal is then attenuated by the isolation of the isolator, and the resulting signal at port a is added vectorily to the reection resulting from the mismatch between the source impedance and the input impedance of the perfectly terminated isolator. For example, if the isolator has an insertion loss of 0.5 dB, isolation of 20 dB, and an input and output return loss of 20 dB and the signal at port b is totally reected back to the isolator, the reected signal returning to the input port will be 20.5 dB below the incident signal and will be added vectorily to the reection resulting from the input port mismatch, which is 20 dB down from the incident signal. The resulting total return loss will be between 14.2 and 45 dB depending on the relative phases. Microwave isolators make use of the nonreciprocal microwave properties of a ferrite material biased by an applied direct-current (DC) magnetic eld. Of the many different approaches that have been devised to use these

1. TERMINATED CIRCULATORS Microwave circulators are described in detail in another article of this encyclopedia and are devices with three or more ports with low loss, for example, from ports 1 to 2, 2 to 3, and 3 to 1 as illustrated in Fig. 4a. Ideally they are lossless and are described by 2 0 0 0 S32 S13 3

6 S 6 S21 4 0

7 0 7 where =S13 = =S21 = =S32 = 1 2 5 0

Terminating port 3 of a three-port circulator (Fig. 4b) in a matched load results in an isolator between ports 1 and 2.
1 2

3 (a) 1 2

Load (b)
Figure 4. Connecting a matched load to port 3 of a three-port circulator forms a two-port isolator.

Figure 3. The total reected signal at a is the vectorial sum of the reection resulting from the mismatch of port a plus the signal reected at b reduced by the isolation of the isolator.

2788

MICROWAVE ISOLATORS

The isolation depends on the match between the termination and port 3 of the circulator. The circulator may be constructed in any of the many forms described elsewhere, such as waveguide, stripline, microstrip, or lumped constant, and at frequencies from UHF to millimeter wavelengths. In addition to being compatible with the wide variety of transmission media, the terminated circulator approach has a significant advantage over other types of isolators in that the nonreciprocal function performed by the circulator is entirely separated from the problem of dissipating the energy of the signal propagating in the reverse direction. Thus, in applications where a large amount of reverse power must be dissipated, a high-power load, with external cooling if necessary, can be provided; or in low-power applications, a miniature termination such as a chip resistor can be integrated with the circulator to form a compact device. The successful development of junction circulators in many forms and over a wide frequency range has resulted in the terminated circulator being the most common type of ferrite isolator.

2. FARADAY ROTATION ISOLATOR The Faraday rotation isolator (1) was one of the rst types of microwave ferrite devices. Its operation can be described with reference to Fig. 5. The Faraday rotator section consists of a ferrite rod at the center of a circular waveguide with its axis parallel to that of the waveguide. A DC magnetic bias eld is applied along the axis of the ferrite rod. It is a property of the Faraday rotator that, if the input to the rotator section is a signal in the TE11 mode of circular waveguide, the orientation of the E eld will rotate as the signal propagates through the rotator as shown. This Faraday rotation can be demonstrated theoretically by considering the linearly polarized eld in the circular waveguide section to be composed of two counterrotating circularly polarized modes. The magnetized ferrite can be shown to present a different microwave permeability to the two counterrotating modes, which therefore propagate with different velocities, resulting in the rotation of the total eld pattern [2].
E field at output of rotator E field at input of rotator Faraday rotator Rotator Circular-torectangular waveguide transition Output b

Resistive card Rectangularto-circular waveguide transition

Output a

Figure 5. A Faraday rotation isolator.

To make an isolator from this Faraday rotator, rectangular-to-circular waveguide transitions are placed at the input and output with matching provided to the ferriteloaded section. Resistive cards are placed across the circular sections of the guide at both the input and output. An incoming signal from port a will go through the rectangular-to-circular waveguide transition transforming to the TE11 mode in the circular guide with the E eld perpendicular to the resistive absorber. Thus the signal is transformed with low loss to the ferrite-loaded section. The ferrite-loaded portion of the guide is terminated at the point where the E eld has been rotated by 451, where it passes another resistive absorber oriented perpendicular to the E eld and enters the transition back to the rectangular guide. Thus, the signal passes with low loss to the output port b. On the other hand, if a signal is applied to port b, it again passes with low loss to the ferrite-loaded section, where it is rotated so that at port a it is oriented with the E eld parallel to the resistive absorber. The signal is then attenuated by the absorber. The Faraday rotation isolator was one of the rst microwave ferrite devices to be introduced, but it suffers from several performance limitations. The rotation in the basic device of Fig. 5 is frequency-dependent, so the isolator is narrowband. In addition, the rotator is inherently a lowpower device because of its geometry. The ferrite and the absorber are thermally isolated so that the power that can be handled, especially in the reverse direction, is limited. Techniques have been developed to increase the bandwidth over that of the basic device [2,3]. For example, dielectric loading, ridged waveguide, and cascading of stagger-tuned sections have been used to produce acceptable performance over bandwidths of several gigahertz at X band, but because of its limitations the Faraday rotator has largely been superseded by the other isolator approaches discussed here. One version of the Faraday rotation isolator that still nds application was described by Barnes [4]. In this approach, instead of using a circular waveguide loaded by a ferrite rod, the conducting waveguide walls are eliminated, and the ferrite rod is increased in diameter so that the ferrite becomes a dielectric waveguide. The high dielectric constant (typically between 12 and 15) of the relatively large ferrite rod connes most of the energy to the ferrite. The details of this type of isolator are described in detail by Barnes [4]. The rotator consists of a short length of ferrite rod suspended in a nonconducting housing by plastic supports and is coupled to the input and output rectangular waveguides by dielectric tapers that protrude into the waveguides. The attenuator lms that provide the reverse loss are deposited metal lms sandwiched into the tapered dielectric waveguides. Barnes shows that, because the dielectric rod guide with approximately 95% of the energy conned to the ferrite behaves much like an innite ferrite medium, the Faraday rotation is less dependent on frequency than in a ferrite-loaded metal waveguide. Figure 6 shows the performance reported by Barnes for a 5060-GHz isolator, along with the performance of a conventional Faraday rotation isolator. Because of its broadband performance

MICROWAVE ISOLATORS

2789

45 Reverse loss (dB) 40 35 30 25 20 35 Return loss (dB) 30 25 20 15 Forward loss dB 2.0 1.0 0 50 52 54 56 Frequency (GHz) 58 60 1.03 1.06 1.12 1.22 1.45

innite ferrite medium f0 gHdc 3

Figure 6. Performance of the broadband isolator of Barnes (solid curves) compared with a conventional Faraday rotation isolator (dashed curves) [4]. (Copyright r 1961 IEEE.)

and low magnetic bias eld requirement, this approach has proved to be useful for broadband millimeter wavelength isolators.

3. RESONANCE ISOLATORS The resonance isolator makes direct use of the phenomenon of ferromagnetic resonance, characterized by the precession of the magnetization vector in a ferrite about the direction of an applied DC bias eld [2]. In an unperturbed state, the direction of the magnetization vector of the ferrite is aligned with an applied bias eld. Any disturbance, such as a momentary magnetic eld applied perpendicular to the bias eld, will cause the magnetization to precess about the direction of the bias eld, as illustrated in Fig. 7. After the perturbation is removed, the precession will decay at a rate that depends on the magnetic losses of the material, until the magnetization is again aligned with the DC eld. The frequency of the precession, the ferromagnetic resonance frequency, is proportional to the DC eld. In an

where, in the commonly used units, g is 2.8 MHz/oersted. In the case of a nite ferrite element such as a plate or rod, the demagnitizing factor of the sample must be taken into account to relate the applied bias eld to the internal eld that determines the resonant frequency. For readily achievable magnetic elds, this resonant frequency is in the microwave region. Microwave signals near the resonant frequency with magnetic elds perpendicular to the bias eld will interact strongly with the ferrite magnetization. In particular, circularly polarized elds in the plane perpendicular to the bias eld will interact strongly if the sense of polarization corresponds to the direction of the resonant precession but only weakly if the sense of circular polarization is opposite to the direction of the precession. Now consider a rectangular waveguide operating in the fundamental TE10 mode. Figure 8 shows the magnetic eld of this mode, looking down on the broad face of the guide. At point A in the guide, the magnetic eld will be circularly polarized in the counterclockwise direction for a signal traveling from left to right and in a clockwise direction for a signal traveling from right to left. If a ferrite is placed in the waveguide at this point and biased by a DC eld, as shown in Fig. 9, the ferrite will interact strongly with a signal in the waveguide with a frequency near the ferromagnetic resonant frequency when the direction of propagation is such that the sense of circular polarization corresponds to direction of the ferromagnetic precession. The interaction will be weak for a signal traveling in the opposite direction. By proper design, the parameters such as the dimension and location of the ferrite, ferrite material properties, and bias eld can be chosen so that, at frequencies near the ferromagnetic resonance frequency, excellent isolator performance can be achieved. In principal resonance isolators can be constructed in any transmission line where circularly polarized radiofrequency (RF) magnetic elds exist. For example, they have been built with various degrees of success in dielectricloaded coaxial line, nline, dielectric waveguide, and image line; however, by far the largest application of the resonance isolator approach is in rectangular and ridged waveguide where extensive analysis and optimization has resulted in devices with very attractive properties.

Hb

Waveguide walls A

Hrf

Hrf

Figure 7. When perturbed from the steady state, the magnetization M will precess about the bias eld Hb at the ferromagnetic resonant frequency.

Figure 8. At point A the RF magnetic eld for a signal traveling from left to right in the fundamental mode of the rectangular waveguide will be circularly polarized in the counterclockwise direction.

2790

MICROWAVE ISOLATORS

Hb

Permanent magnets

Dielectric slab Ferrite

Figure 9. In this basic resonance isolator, the ferrite will interact strongly with a signal in one direction but not with one in the opposite direction.

Ferrite a

Ferrite b

Figure 11. A resonance isolator using hexagonal ferrites.

Of the possible geometries illustrated in Fig. 10, Fig. 10b has demonstrated advantages over Fig. 10a. Placing the ferrite in a thin strip along the broad waveguide wall allows efcient dissipation of the heat produced in highpower operation. In addition, the magnetic elds are more nearly circularly polarized in the conguration of Fig. 10b than in Fig. 10a, and experiments of Weiss [5] have shown Fig. 10b to have a superior ratio of isolation to forward insertion loss. This ratio of isolation to forward insertion loss is a gure of merit used to compare different isolator congurations. The gure of merit for Fig. 10b was determined by Weiss to be 75 compared with 14 for Fig. 10a. Further improvement can be achieved with dielectric loading as illustrated in Fig. 10c. The dielectric concentrates the energy in the vicinity of the ferrite and increases the gure of merit to 150. Resonance isolators in rectangular waveguide have been developed to give good performance over a full waveguide bandwidth (e.g., 30 dB isolation, 1 dB forward loss, and 1.15 VSWR over 8.212.4 GHz, or 12.418 GHz). Such resonance isolators are large and heavy because of the permanent magnets required to bias the ferrite to ferromagnetic resonance at microwave frequencies (approximately 1.5 kg in the case of the previously mentioned 8.212.4-GHz and 12.418-GHz isolators). This becomes a serious problem at higher frequencies. Because the DC eld required to bias the ferrite to resonance is proportional to frequency, at millimeter wavelengths this eld becomes quite large, on the order of 1.53.5 T or more. In order to minimize the size and weight of millimeter

(a) Ferrite Dielectric (b)

wavelength resonance isolators, devices have been developed making use of magnetoplumbites, often called hexagonal ferrites [6]. Such materials have a strong anisotropy eld. All ferrite materials exhibit a certain amount of magnetic anisotropy because, in the individual crystallites that make up the material, the crystalline structure produces a preferred direction for the orientation of the magnetic moment vector. In a typical polycrystalline ferrite, however, these microscopic crystallites are randomly oriented with respect to each other, so that in the bulk material the effect of the anisotropy of the individual crystallites averages out, resulting in an isotropic material. The materials most useful for resonance isolators are the so-called uniaxial compounds that possess a large anisotropy eld along the C axis of the hexagonal crystals. By special processing used to produce the ferrite material, for instance by pressing the slurry of the appropriate material in the presence of a large magnetic eld before ring, it is possible to produce a material in which the individual crystallites are oriented in a particular direction. This results in a large anisotropy eld that in some ways is indistinguishable from an externally applied bias eld. In this way, it is possible to make self-biased materials, or materials that require very little applied eld to bias them to resonance at millimeter wavelengths. Such an isolator is illustrated in Fig. 11. Thin slices of the hexagonal ferrite material are bonded to the dielectric slab, which serves to concentrate circularly polarized magnetic elds in the ferrite. In order to achieve wide bandwidth, the isolator may consist of a cascade of several sections of ferrite materials with different anisotropy elds and dielectric characteristics. Additional small permanent magnets can be used to achieve the correct resonant frequencies for the different sections. Figure 12 shows the performance of an isolator of this type weighing only 160 g. Materials have been developed enabling such isolators to be built at frequencies to 110 GHz.

4. FIELD DISPLACEMENT ISOLATORS Field displacement isolators can be of many types but are all based on the fact that, in a transmission line loaded by a magnetically biased ferrite, the eld pattern may exhibit nonreciprocal behavior by being distinctly different for the two directions of propagation. An isolator can then be built by judiciously locating dissipative material in an area where the elds are intense for one direction of propagation but weak for the other.

(c)
Figure 10. Three different congurations for a resonance isolator in rectangular waveguide.

MICROWAVE ISOLATORS

2791

40 Isolation (a + 1) Attenuation (dB) 30 20 1.5 1.0 0.5 0 35 37 Insertion loss (a 1) Strong RF fields 43 45 39 41 Frequency (GHz) 47 49 50 Dc magnetization Weak fields Q -Band isolator VSWR (max) = 1.15 B B E E

Ferrite

Figure 12. Performance of a 3350 GHz isolator using hexagonal ferrites [6]. (Copyright r 1963 IEEE.)

Figure 14. Fields in a ferrite-lled stripline as analyzed by Hines [9]. (Copyright r 1971 IEEE.)

Figure 13 illustrates an early type of isolator based on this principle [7]. As in the resonance isolator, a ferrite slab is introduced into the rectangular waveguide, and a DC magnetic eld is applied across the narrow dimension of the waveguide; however, in this case the bias eld is less than that required for ferromagnetic resonance. With an appropriate choice of dimensions, ferrite properties, and magnetic bias, the resulting eld pattern for the two directions of propagation can be as shown with a concentration of electric elds at one surface of the ferrite in one direction and minimal electric eld at that surface for the other direction of propagation. If a sheet of resistive material is placed at this surface of the ferrite, it will dissipate energy for one direction and have very little effect on the other. An isolator of this type can produce, for example, isolation of greater than 30 dB over the 5.96.4 GHz band with a forward loss of less than 0.25 dB [8]. The isolator of Fig. 13 was one of the earliest types of microwave ferrite devices, but it has largely been supplanted by the other types of isolators. One more recent type of eld displacement isolator that has significant unique advantages, particularly in regard to broadband operation, is the peripheral mode, or edgeguided, isolator. This device uses the edge-guided mode analyzed by Hines [9]. Circulators and isolators using this mode of propagation were patented by Anderson [10]. The analysis by Hines revealed that in a stripline, or microstrip transmission line, with a wide center conductor and ferrite as the dielectric medium, and with a magnetic bias eld perpendicular to the ground plane(s), a mode of

propagation exists in which the energy concentrates toward one edge of the center conductor. The elds of this mode in stripline are illustrated in Fig. 14. The elds are similar to transverse electromagnetic (TEM) modes except for their displacement to one side or the other of the center conductor, depending on the direction of the bias eld with respect to the direction of propagation. Hines has shown that for a very wide center conductor in comparison to the ground-plane spacing, with a weak bias eld, sufcient only to saturate the ferrite, this mode of propagation is free of dispersion and has a constant characteristic impedance over all frequencies. Because the energy is concentrated at one edge of the center conductor, an isolator can be constructed by placing lossy material along one edge of this conductor. One direction of propagation will be unperturbed by this material, while a signal in the opposite direction can be strongly absorbed. Such an isolator is illustrated in Fig. 15. In the idealized case, this isolator would have innite bandwidth. In practice, the bandwidth is limited by low eld losses at low frequencies, by higher-order modes at high frequencies, and by the difculties in matching at the input and output to or from a conventional stripline or coaxial line. Nevertheless, very broadband isolators can be achieved using this approach. Hines reported the results shown in Fig. 16 for a basic microstrip conguration and in Fig. 17 for the case of added capacitive loading along the low-loss edge. Thus the approach yields performance acceptable for some applications over a multioctave band. The isolators described in this article are those that have found significant application in microwave systems.
Hb

Ferrite

Resistive film

Forward

Resistive film

Reverse

Ferrite Microstrip transmission line

Figure 15. A broadband isolator using the edge-guided mode of a microstrip transmission line on a ferrite substrate.

Hb

Electric field strength

Figure 13. A eld displacement isolator in a rectangular waveguide.

2792

MICROWAVE LIMITERS

35 30 Forward loss 25 20 Forward loss left scale 15 10 6 7 8 9 10 Frequency (GHz) 11 Backward loss right scale

60 50 40 30 20 10 12 Reverse loss

6. D. R. Taft, G. R. Harrison, and L. R. Hodges, Millimeter resonance isolators utilizing hexagonal ferrites, IEEE Trans. Microwave Theory Tech. 11:346 (1963). 7. S. Weisbaum and H. Seidel, The eld displacement isolator, Bell Syst. Tech. J. 35:877 (1956). 8. R. L. Comstock and C. E. Fay, Operation of the eld displacement isolator in rectangular waveguide, IRE Trans. Microwave Theory Tech. 8:605 (1960). 9. M. E. Hines, Reciprocal and nonreciprocal modes of propagation in ferrite stripline and microstrip devices, IEEE Trans. Microwave Theory Tech. MTT-19:442451 (1971). 10. R. Anderson, Gyromagnetic Device Having a Plurality of Outwardly Tapering Members, U.S. Patent 3,555,459 (Jan. 12, 1971). 11. L. E. Davis and D. B. Sillars, Millimetric nonreciprocal coupled-slot nline components, IEEE Trans. Microwave Theory Tech. MTT-34(7) (1986). 12. A. Beyer and K. Solbach, A new n-line ferrite isolator for integrated millimeter-wave circuits, IEEE Trans. Microwave Theory Tech. MTT-29(12) (1981). 13. J. M. Owens et al., W-band ferrite-dielectric image-line eld displacement isolators, 1989 IEEE MTT-S Int. Microwave Symp. Digest, 1989. 14. G. M. Smith et al., Design, analysis and application of high performance permanently magnetised, quasi-optical, Faraday rotators, 1994 IEEE MTT-S Int. Microwave Symp. Digest, 1994.

Figure 16. Measured performance of a microstrip isolator similar to that of Fig. 15 [9]. (Copyright r 1971 IEEE.)

35 30 25 Loss (dB) Isolation 20 15 10 5 3.0 5.0 Field 1900G Field 2450G Forward insertion loss Field 2450G

MICROWAVE LIMITERS
7.0 9.0 11.0 Frequency (GHz) 13.0 15.0

ROGER KAUL
U.S. Army Research Laboratory

Figure 17. Measured performance of a microstrip isolator similar to that of Fig. 15, but with added capacitive compensation along the low-loss edge [9]. (Copyright r 1971 IEEE.)

In recent years development of microwave isolators has continued with most effort being devoted toward development of isolators compatible with newer types of transmission lines, particularly for application at millimeter wavelengths. For example isolators have been developed for use with nlines [11,12], image lines [13], and quasioptical transmission lines [14]. These isolators will become important as these transmission media nd application. BIBLIOGRAPHY
1. C. L. Hogan, The microwave gyrator, Bell Syst. Tech. J. 31:1 (1952). 2. B. Lax and K. J. Button, Microwave Ferrites and Ferrimagnetics, McGraw-Hill, New York, 1962. 3. P. H. Vartanian, J. L. Melchor, and W. P. Ayres, Broadbanding ferrite microwave isolators, IRE National Convention Record, 1956, Part 5, p. 79. 4. C. E. Barnes, Broad-band isolators and variable attenuators for millimeter wavelengths, IRE Trans. Microwave Theory Tech. 9:519 (1961). 5. M. T. Weiss, Improved rectangular waveguide resonance isolators, IRE Trans. Microwave Theory Tech. 4:240 (1956).

A microwave limiter is designed to allow low-power signals to pass through it, while attenuating high-power signals. Stated another way, microwave limiters are power-dependent attenuators that prevent intense microwave energy from interfering with susceptible microwave components in the latter stages of a cascade. Most limiters operate by reducing the impedance of a transmission line when the incident power is above the threshold power level, reecting or absorbing the incident power. Although, in most cases, a limiter reects the majority of the intense incident power back toward the power source on the transmission line, where a circulator, an isolator, or a hybrid coupler may divert or absorb it, a microwave limiter may also be designed to absorb the incident power in the same manner as a microwave switch. In most applications, microwave limiters (sometimes referred to as receiver protectors and terminal protection devices) are located near the antenna port. However, limiters have also been used in intermediate-frequency channels, where signals from a number of channels are combined to create an intense signal (e.g., the intermediate frequency circuit of a phased-array radar). In radar applications, the term duplexer is sometimes used to refer to the front-end receiver protection circuitry.

MICROWAVE LIMITERS Table 1. Common Microwave Limiter Technologies Technology Solid-state Gaseous Ferrite Common Implementations pin diode MESFET Waveguide Stripline Filter Stripline

2793

Many microwave limiter technologies have been investigated since the midtwentieth century. The common technologies are identied in Table 1. The solid-state limiter, which was rst designed using the varactor diode, today employs the Si pin (or PIN) diode, because of the pins lower capacitance per unit area resulting in better thermal characteristics, in addition to the fact that it requires no external power supply. Also shown in the table is the fact that a MESFET-based, MMIC-compatible microwave limiter is being developed for integration with the lownoise ampliers common in MMIC designs. Furthermore, although gaseous limiters are widely used in high-power radars as receiver protectors, ferrite limiters have not found the same wide use.

regions), the pin diode has significantly lower capacitance per unit area at zero bias than a signal diode. This feature allows the pin diode to be a high reactance connected in shunt across a transmission line even at microwave frequencies, yielding low insertion loss. The feature also provides more volume than a signal diode for dissipating heat from intense incident pulses. The i region changes the terminal currentvoltage relationship, as compared to other minority-carrier (e.g., signal) or majority-carrier (Schottky) diodes. Varactor diodes have carrier distributions similar to thin i-region pin diodes. For this reason, varactor diodes were used as limiter diodes until the special doping prole of the pin was developed. Leenov [2] studied the pin diode conguration and determined that, at frequencies lower than the inverse transit time of the i region, the diode recties with a low series resistance. At very high frequencies, the charge distributions at the edges of the i region oscillate with the applied terminal voltage; however, since carriers do not have time to transit the i region, the current is primarily a displacement current and the diode impedance remains high. Leenov found that, when excited with a sine wave, the diode resistance is R   Z0 h O qD=2pf 1=2 Pi 104 kT 1

1. SOLID-STATE LIMITERS At radiofrequencies, it is common to place back-to-back signal diodes (e.g., 1N914) shunted to ground across the input transmission line of a radio receiver, as shown in Fig. 1a. When the peak voltage on the line exceeds the forward conduction voltage (typically, 0.7 V) the transmission line voltage is clipped. Below the threshold voltage, the diodes appear as shunt capacitors across the line. The high capacitance of the IN914 prevents the extension of this design to microwave frequencies, where the capacitive reactance becomes very low, yielding an undesirable impedance discontinuity, shunting the transmission line that must be tuned out to retain a small voltage standing-wave ratio (VSWR) at low line voltages. The additional reactance narrows the bandwidth of the transmission line, which may be unacceptable. 1.1. pin Diode Limiter The pin diode has a lower capacitance for a given crosssectional area than other diode designs, because the distance between the p and n regions is separated with the i (intrinsic) region, whereas in other signal diodes, only the depletion region separates the two highly doped regions. The i-region thickness of a pin diode is typically in the 1200 mm range, whereas other diode depletion regions are typically less than 2 mm. A good measure of the i-region thickness (h) is the reverse breakdown voltage (Vb) at a few microamperes. Several relations are used, including Vb 36 h0.81 [1] and the simpler rule of thumb Vb 20 h for hZ5, where Vb is in volts and h is in micrometers. Since the capacitance of a parallel-plate capacitor is inversely proportional to the separation distance of the charged plates (represented by p and n charge

where k is Boltzmanns constant (1.38 10 23 W s/K), T is the absolute temperature (K), q is the electron charge (1.6 1019 C), D is the diffusion coefcient (15.6 cm2/s for Si), Z0 is the impedance of the transmission line in ohms, and Pi is the incident power in watts. Garver [3] uses Eq. (1) to show that the attenuation a provided by a single diode across a transmission line is a 10 log q=kT2 DZ0 Pi 8 108 pfh2 dB

The attenuation is proportional to the log(1/fh2) for a given impedance and pin device material. At higher frequencies, the h must be reduced, resulting in a higher shunt capacitance per unit area for the diode and increased lowpower level attenuation or bandwidth limiting. To compensate for this effect, the cross-sectional area of the diode is reduced. Brown [4] showed that the device thickness should be less than about 25 mm at 0.1 GHz to 2.5 mm at 10 GHz in order to avoid high spike leakage, high power absorption during the transition from the high-impedance diode state to the low-impedance diode state, and low insertion loss at low signal levels. Spike leakage is the momentary power that passes through the limiter before the diodes impedance reduces, thus reecting the incident power back toward the source. Fast-risetime, incident pulses will appear to have a spike of leakage power at the output of the limiter. Since the performance of the pin diode depends on the ability of carriers to transit the i region, transient effects occur that are dependent on frequency and other parameters (e.g., i-region doping density). For example, Fig. 2 shows the voltagetime waveform of a limiter consisting of

2794

MICROWAVE LIMITERS

20-mm Si pin diodes closely spaced in shunt with a 50-O transmission line as shown in Fig. 1a with l 0 at four incident voltage amplitudes. Note that this circuit is shown only to present the concept; actual limiters might use thinner i-region diodes arranged in another conguration (see Figs. 1b1f). All these circuits allow rectied current to ow through both diodes. Leenov [2] showed that a DC current is much more efcient in lowering the pin diodes impedance than an RF current. Unless this

rectied current is allowed to ow in a low-impedance circuit (typically less than 5 O), the pin diode resistance may not be reduced to a few ohms (typical). The voltage time waveforms on the left in Fig. 2 are the incident voltage at 1.1 GHz, and the waveforms on the right are the voltage following the dual-diode limiter. The pin diodes do not clip the incident wave as would a high-speed signal diode; rather, their impedance is reduced by injection of carriers into the i region. At the lowest incident voltage Vi,

dc bias l

Z0

Vb1

Vb2

Z0

Z0

XL

Z0

Zw

Zw

(a)

(b)

(c)

Z0 delay line

L Z0 Z0 Z0 Schottky diode (d) (e) Z0

3 dB hybrid coupler To antenna Ls Z0 Rs Isolator Z0 To receiver Cp Cj Rj

To transmitter (f)
Figure 1. Diode limiter equivalent circuits.

(g)

MICROWAVE LIMITERS

2795

Z 0 = 50

Spike leakage period 30 15 0 15 30

Recovery time Flat leakage

30 15 0 15 30

Amplitude (V)

15 0 15 15 0 15 15 0 15 Vi

15 0 15 15 0 15 15 0 15 0 100 Vo

20

40 60 Time (ns) (a)

80

20

40 60 Time (ns) (b)

80

100

Figure 2. Incident and output voltagetime waveforms for a pin diode limiter.

the attenuation (insertion loss) of the limiter is 20 log10 Vo =Vi % 1 dB, where Vo is the output voltage. As the Vi is doubled, the Vo tends to show an initial transmission transient (termed spike leakage), followed by a relatively constant output voltage (termed at leakage). It requires tens of nanoseconds for the two diodes to lower their impedance below the Zo of the transmission line. However, the spike leakage period would be much more rapid if the i region were thinner or the 20-mm diodes were excited at a lower frequency. The amount of energy in this period (i.e., the integral of the powertime prole) is the spike energy that may destroy susceptible microwave devices in cascade with the limiter. As a rule of thumb, lownoise, microwave ampliers will be destroyed if the spike energy exceeds 1 mJ, and destruction of microwave mixers will occur at the 10 mJ level. If the spike energy must be reduced, the second limiter diode with a thinner i region is located with a quarter-wavelength behind the rst limiter diode as shown in Fig. 1a with l 1 wavelength. 4 A commercial limiter using three pin diodes shunting a soft-substrate, 50-O transmission line is shown in Fig. 3. The thinnest pin diode is located near the output receiver port, of the limiter and activates rst, setting the threshold for limiting. The 10-turn coil allows the rectied

current to ow through the diode(s). The standing wave reected from the thinnest diode excites the middle diode, and the thickest diode is activated at higher incident power levels. At signal levels below threshold, the bandwidth (dened by VSWR 1:6 : 1) of this limiter design is 28 GHz, with an insertion loss less than 1.3 dB. The limiter is specied to sustain a 3 W continuous incident power, with 0.1 W output level. The inputoutput curve measured at 2 GHz for the commercial limiter (shown in Fig. 4) conrms the at leakage level at 0.1 W, and the insertion loss at less than 1 dB at the low end of the operating band. The input 1 dB compression point was measured to be 11 dBm, and the input third-order intercept was 18 dBm at 3 GHz and was 15 dBm at 7 GHz. During pulsed operation, the unit will sustain a 1000 W pulsetrain, with a 1 ms length and a 1% duty cycle. The recovery time (dened as the time to return to low insertion loss after the high incident power is removed) is specied as less than 1 ms. 1.2. Spike Leakage The data in Fig. 2 show that the peak spike leakage power increased and the duration of the spike decreased with increasing incident voltage. The measured spike energy

2796

MICROWAVE LIMITERS

Ground pad 10-turn coil Input 50 Microstrip line

dc blocking capacitor

Thickest p-i-n diode Soft substrate Baseplate

Thinnest p-i-n diode

Output

Figure 3. Sketch of a multistage pin diode limiter.

for 1.5 and 5-mm i-region diodes is shown in Fig. 5. Calculations supporting these results (1) show that a slightly p-doped (1015 cm 3) intrinsic region would exhibit less spike leakage than does the usual n-doped (1014 cm 3 or less) intrinsic region. This result arises because a higher density of the lower mobility holes yields a lower impedance i region than in the usual case with the n-doped i region 1.3. Diode Limiter Circuits Figure 1 shows many congurations of diode limiter circuits. Figure 1a has been discussed previously. Figure 1b is a single-diode circuit in which the rectied current ows through the inductor. Note that since the diodes impedance is lowered at high signal levels throughout the cycle, the single diode does not rectify the positive or negative peak voltages on the transmission line; if that were the case, the maximum isolation would be only 3 dB. The inductors reactance XL should be high at the operating frequency, but have a low inductance at frequencies associated with buildup of the rectied current to allow the rectied current to build up rapidly, thereby reducing spike leakage. In waveguides, the pin diode is mounted parallel to the electric eld lines in the lowest-order mode on an inductive post. The equivalent circuit in Fig. 1c shows the reactances of the post in shunt with the waveguide impedance Zw. For self-activated limiter operation, the lowpass lter for inserting the DC bias is not needed. The

limiters shown in Figs. 1a1c reect most of the incident power back toward the source. If this reected power is undesirable (e.g., an application in which a limiter is used for stealth purposes), a nonreective limiter (i.e., a powerdependent attenuator, as shown in Fig. 1d) may be required. A nonreective design developed by Glenn et al. [5] is self-activating but does not provide the same amount of isolation that reective limiters provide. The nonreective design is parallel-resonated by the L and C at high signal levels, for added isolation. Nonresonant, nonreective circuits using resistors are possible; however, resistors provide only moderate levels (1015 dB) of isolation. Where spike leakage is a significant problem at high power levels, the delay-line limiter (Fig. 1e) recties a sample of the large signal and applies a DC bias to the pin diode, lowering its impedance before the intense signal arrives. The delay line may be implemented in a low-loss coaxial cable yielding an approximate 3 ns delay per meter. Figure 1f shows the duplexer application for a pulsed, monostatic radar receiver protector. At normal receive signal levels, the balanced duplexer design employs two 3 dB, 901, hybrid couplers to split the input signal between two diode limiters. Since the limiters are in their highimpedance state, the signals are recombined at the output into the receiver input. If an intense receive signal is present, the pin diodes conduct, and the signal is reected back into the transmitters circulator stage, where

1.0
50 Output power (dBm) 40 30 20 10 0 0 A Limiter input-output Hysteresis characteristic Insertion loss 10 20 30 Input power (dBm) 40 50 Zero insertion loss Isolation

5 m diode at 8.7 GHz 1.5 m diode at 8.7 GHz 5 m diode at 1.5 GHz

Spike energy ( J)

0.1

0.01

10 10

0.001 0.1

10 Input power (W)

100

1000

Figure 4. Inputoutput power characteristic of a pin diode limiter.

Figure 5. Spike energy for 1.5 and 5-mm-thick i-region pin diodes.

MICROWAVE LIMITERS

2797

it is absorbed in the matched load. During transmit, the pin diodes are low impedance and reect the power out of the antenna port. If the antenna VSWR is high, the transmitter power is rereected by the diodes and is absorbed in the load of the three-port circulator. This duplexer circuit uses all passive components. If desired, the pin diodes can be externally biased during the transmitters pulse. If a DC path for the rectied pin diode current is not available through the 3-dB hybrids, one must be provided via shunt inductors or other bias circuitry. In Fig. 1, all diodes assume no package parasitics. These package parasitics (see Fig. 1g) can have a significant effect at microwave frequencies. The pin diode is represented by the variables Cj and Rj, where Cj is small (typically less than 1 pF) and Rj is large at low signal levels. At high signal levels, Rj is small (typically 12 O), thereby shunting Cj. Rs represents the series contact resistance of the diode and the resistance of the inductive bondwire Ls to the diode chip. Cp is the package capacitance. Additional details of limiter design in microstrip and waveguide congurations are found in White [6] and Garver [3]. 1.4. Unexpected Effects Some limiters exhibit a hysteresis effect when operated CW, or with long pulses, as sketched in Fig. 4. The sudden increase in the isolation above an input power threshold is retained as the input power is reduced, until the input power equals the value marked A in Fig. 4. If the power is increased again, the original inputoutput curve is retraced and the threshold can be observed. However, if the input power is not reduced to point A, the lower input output curve is followed, and a threshold cannot be observed. A plausible explanation, based on space-charge effects, is given in Ward et al. [1]; however, the hysteresis effect needs further experimental investigation. When connected to high-Q circuits (e.g., lters), limiters may exhibit the nonlinear dynamic effects (chaos) of period doubling and noisy behavior. This behavior was rst observed with a limiting lter that utilized a pin diode as a capacitive reactance at the output of a microwave lter structure developed by Tan [7]. Unexpected signal generation by a limiter appears to be avoidable above 1 GHz by using i-region thickness exceeding 3 mm and a circuit Q less than 100. Further experiments and analyses are needed to fully understand and alleviate this devicecircuit interaction.

1.5. Limiter Burnout Levels The CW burnout level of a commercial pin diode limiter is usually only a few watts. It is recommended that the CW power specication not be exceeded since the diode may be operating above its maximum junction temperature or the mounting solder could melt. The same power limitation applies to pulse lengths long compared to the thermal response time of the diode. Most limiters are also specied for pulsed operation with 1 ms pulselength at 0.1% duty cycle (1000 pps). In general, these ratings are conservative and can usually be exceeded by 3 dB. Table 2 shows the burnout level results of a limited number of experiments in 50-O coaxial circuits for pin diode limiters with varied-thickness i regions. Damage was observed in three stages: (1) reverse current increases causing reduced reverse breakdown voltage, (2) as the diode impedance became lower the insertion loss increased, and (3) eventually fusing occurred and the diode became an open circuit, ceasing its limiting action. GaAs pin diode limiters have been fabricated and their performance has been measured. There appears to be little advantage to using GaAs, since its lower thermal conductivity cannot transfer the heat generated in the diode to the heatsink as effectively as Si can. However, for GaAs monolithic microwave integrated circuit (MMIC) devices, where high volumes make the specialized assembly of Si technology expensive, GaAs pin devices may be a viable alternative. Fabrication of the GaAs i region is usually an additional process in manufacturing GaAs MMIC devices, incurring higher cost and possibly leading to lower MMIC yields. The reliability of GaAs pin diode limiters have been questioned. GaAs limiters designed for high-power, high-duty-cycle, pulsed operation have degraded and become lossy after a few thousand hours of operation. However, majority-carrier GaAs devices (e.g., MESFETs) do not appear to exhibit this form of degradation. 1.6. MESFET Limiters MMICs are nding increasing application in todays designs when uniform performance and high quantities can justify the relatively high nonrecurring design engineering costs. Use in applications such as phased-array radars and high-volume consumer products is typical. In these applications, low-noise devices with relatively small physical dimensions are used, resulting in a susceptibility to burnout due to incident short single- or multiple-pulse

Table 2. Single-Pulse Damage Levels for pin Diode Limiters i-Region Thickness 0.5 1 1 2 5 10 Pulselength 10 ns 1 ms 1 ms 10 ns 10 ns 1 ms Frequency (GHz) 2.7 9.4 3.3 2.7 2.7 9.4 Damage Level (kW) 30 0.2 1.1 135 4330 10 Damage Observed Vb reduced 1 V Vb reduced to 12 V Vb reduced to 25 V Vb reduced to 10.5 V No damage up to 300 kW Vb reduced to 9 V

2798

MICROWAVE LIMITERS

50 Transmission line C R G R Vg C D S

MESFET widths allow more current with increased insertion loss. Simple limiter circuits have been rened for specific applications by Vasile [9] and Podell and Stoneham [10].

2. GASEOUS LIMITERS Gaseous limiters are able to operate over a wide range of incident power levels from a few watts to megawatts. For this reason, gaseous limiters are the technology of choice for the highest-power applications throughout the microwave spectrum. Most gaseous limiter designs have a few percent bandwidth and are suitable for radar receiver protector applications. However, a new TEM design has low insertion loss over several octaves of bandwidth. Gaseous limiters use the breakdown of a gas in a high electric eld to change the impedance across a transmission line. The typical high-power waveguide design shown in Fig. 7 places a quartz tube lled with a noble gas (typically Ar) across the waveguide gap at the point of the highest electric eld. The capacitance of the posts is canceled by the inductance of the iris in the waveguide. These two reactances limit the bandwidth to a few percent. Other designs use waveguide windows to contain the gas in the post region. The window design has a lower lifetime because of the increased absorption cross section of the gas (or gas cleanup) with the metal (typically Ni) walls, resulting in reduced gas pressure and performance. Because the quartz absorption cross sections are several orders of magnitude smaller than those made of metals, gaseous limiters using quartz gas reservoirs have useful lifetimes usually exceeding 20 years. At electric eld intensities below the arc threshold, the posts and iris appear to be a resonant circuit across the waveguide. The gaseous pressure, mixture, and electric eld intensity in the presence of seed electrons set the threshold for the arc. When the arc occurs, the increased conductance across the posts presents a severe mismatch to the waveguide impedance, resulting in significant

Figure 6. Schematic of a simple MESFET limiter.

energies from 0.1 mJ to 10 mJ. (Most devices can sustain a CW incident power up to 0.1 W without degradation.) A MESFET limiter that operates as a switch has been built using standard MMIC technology, to allow its fabrication along with the circuit that it must protect. This cost-effective approach avoids the requirement for employing mixed technologies (e.g., using an Si limiter with a GaAs MMIC, or adding an i-region fabrication step to the GaAs fabrication process). The GaAs MESFET limiter circuit in Fig. 6 shunts the transmission line to ground when the gate voltage allows current from drain to source. Note that the location of the drain and source change each half-cycle, based on the instantaneous polarity at the limiters terminal connected to the transmission line. The MESFET is operated in a bidirectional mode, since no dc bias is required on the transmission line. The RC voltage divider network that biases the gate is high impendence (typically, 40Zo O), thereby shunting the line. This bias port (Vg) may be used as a switch in applications for which the presence of the highpower incident pulse is known a priori (e.g., the transmitters pulse in a phased-array radar). When Vg is grounded, a depletion-mode MESFET exhibits low impedance across the transmission line, protecting the front end when not in use. An enhancement-mode version of the MESFET limiter has been developed [8]. It operates in a similar fashion to the pin diode limiter and requires no external bias in the low-loss state. When the gate is backward-biased (backbiased) at small signal levels, the MESFET represents a small capacitance, consisting of the drain-to-gate and gate-tosource capacitances in series, in parallel with the drainto-source capacitance. These capacitances are proportional to the gate width for a given MESFET technology. When the gate is not backbiased, the saturated drain-to-source current (Idss) and the drain-to-source resistance are also proportional to the gate width. As a result, the designer performs a tradeoff of the amount of shunt MESFET capacitance allowed across the transmission line, to the peak current that the MESFET can pass (which, in turn, sets the power limit rating) in order to determine the gate width of the MESFET. A typical design (e.g., using a 1 mm gate width) will have an insertion loss less than 1 dB at 10 GHz and sink 0.2 A peak. The insertion loss decreases at lower frequencies because the MESFETs capacitive reactance shunting the transmission line increases. Larger

Gas reservoir

Quartz envelope

Iris

Iris

Waveguide wall

High E-field region Waveguide post

Figure 7. Cross section of a gaseous limiter in waveguide.

MICROWAVE LIMITERS

2799

reected power. The arc absorbs about 7% (10%, worst case) of the incident power, resulting in heat that must be conducted through the quartz tube to the walls of the waveguide posts. Because quartz has a high melting point, pulsed operation with hundreds of kilowatts incident is possible. The gas pressure and mixture is adjusted according to the Paschen curve (see discussion of gas-discharge tubes) for the desired arc threshold. The seed electrons are provided by a radioactive source or a small microwave oscillator. The tritium source (typically 100 mCu) emits electrons with a half-life of 12.6 years. Since only a few electrons are needed to initiate the arc, the tritium is useful for three half-lives. The tritium is positioned to irradiate the gas between the posts. Goldie and Patel [11] used a small microwave oscillator that continually excite enough gas molecules to provide the seed electrons. When Ar is used for the gas the threshold is several watts, with a recovery time of several milliseconds. If the Ar recovery time is too long, a chlorineoxygen mixture may be used in which the electrons and ions recombine faster (typically within 100 ns), and the arc loss is lower, but the arc threshold is higher (10 W to 20 W). If the at leakage of the gaseous limiter is too great, pin limiters may be cascaded to remove the spike leakage and lower the at leakage to an acceptable level. The wideband gaseous limiter operates on the same principle as the narrowband device. Patel et al. [12] congured a suspended, 50 O, stripline with the gas mixture surrounding the transmission line. Units with 68 W threshold and 50 W of average power have been built. The device operates over a 30% bandwidth primarily limited by coax-to-stripline transitions at the ports.

Low-loss epoxy Direction of external magnetic field

50 Stripline Stripline outer metalization Single-crystal YIG

Gadolinium gallium substrate

Figure 8. Cross section of a frequency-selective limiter.

3. FERRITE LIMITERS Ferrites [e.g., yttrium iron garnet (YIG)] are used for tunable lters and other applications. These lters are narrowband devices in which the magnetic spin vectors (magnetic dipoles) of the ferrite are oriented by an externally applied magnetic eld (typically, 100 Oe) that can be varied to change the resonant frequency of the lter. The spins on the lattice sites are also coupled to the magnetic eld of a microwave signal. If the strength of the signals magnetic eld exceeds a threshold, energy from the signals magnetic eld is coupled to the spin vectors of the xed lattice ions, thereby creating spin waves. These spin waves are able to transfer energy to heat the lattice, removing energy from the incident microwave signal. The process is nonlinear, with respect to incident microwave magnetic eld strength, and yields substantially different performance as a microwave limiter. Lax and Button [13] stated that limiting thresholds vary from 25 dBm to 50 dBm. Carter and McGowan [14] developed a limiter consisting of a ferrite slab mounted against the narrow wall of a waveguide with a 1500 Oe externally applied magnetic eld. This ferrite limiter was able to dissipate 10 kW incident pulses. The insertion loss over the 8.99.5 GHz range was less than 1 dB. The threshold power and at leakage power levels were 28 W. At 10 kW incident power, the spike energy was 3 mJ with a spike power level of

2.9 kW. The recovery time was less than 20 ns, which is much shorter than that of a gaseous limiter. The threshold and at-leakage power levels were reduced to o30 mW with the addition of a varactor (thin i-region pin diode) limiter behind the ferrite limiter. Because of their tunability by varying the externally applied magnetic eld by at least an octave, ferrite lters can be used as frequency-selective limiters. They have the ability to attenuate a signal at a frequency selected by the magnitude of the external eld, while providing minimal attenuation at nearby frequencies. Adam and Stitzer [15] determined that the bandwidth of the limiting frequencies (typically 50 MHz) is related to the linewidth of the spinwave coupling. A frequency-selective limiter will attenuate an intentional jamming signal, while an electronic warfare receiver is able to listen to signals on nearby frequencies. A stripline conguration of the frequency-selective limiter was developed by Adam and Stitzer [15] using singlecrystal YIG in place of the usual microwave dielectric material, as shown in Fig. 8. The limiting threshold was about 20 dBm. The ultimate limiting capability of this technique appears to be in the 1520 dB range. By cascading limiter, amplier, limiter, and so on, they demonstrated that an incident power range of 60 dB could be compressed to a dynamic range of less than 5 dB. Ferrite limiters require a significant amount of peripheral support, since they require an external magnetizing eld; their insertion losses are usually high (except in the waveguide conguration), necessitating an external amplier; they are sensitive to temperature variations; and they are large and heavy, when compared to other technologies. For these reasons, they have been used only in applications that require tunability.

4. OTHER LIMITER TECHNOLOGIES 4.1. Multipactor Experimentation has been taken place using numerous other technologies, with mixed success. A summary of some of the technologies appears in Table 3. The multipactor takes advantage of secondary electron resonance (multipacting) to provide low impedance across a waveguide. The multipactor is usually congured as the rst stage of a two-stage limiter, since it limits the power to only several watts. The second stage is a pin diode limiter, with a at leakage of less than 100 mW.

2800

MICROWAVE LIMITERS

Table 3. Other Limiter Technologies Technology Multipactor Bulk Window switch array Electrooptic Superconductor Field emitter array limiter Varistor paint Temperature-dependent resistor Performance Excellent, requires external biases Excellent millimeter wave switch Fair, requires laser and optics Fair, 50 ns turnon Unknown Unknown Fair Status Implemented in waveguide congurations Limiter performance untested, difculty making thin i region Relatively high insertion loss Demonstrated at 10 GHz Needs fabrication on low-loss substrate Initial attempts failed Only analyzed to date

Multipacting is an electron avalanche phenomenon operating in a vacuum. The multipacting region allows electrons in an alternating electric eld to ow across a gap in less than a one-half cycle of the eld. Both surfaces of the gap are coated, so that the secondary electron emission coefcient d is greater than 1. The incident electric eld accelerates the electrons toward one of the gap surfaces. As these initial electrons Ni strike the surface, the RF eld changes sign, and the secondary electrons dNi are accelerated toward the opposite surface. The speed at which the multipactor ignites is N Nid2tf, where t is the time and f is the frequency (in hertz) of the electric eld. Electron multiplication continues until a space charge in the gap inhibits additional electrons from being emitted from the gaps surfaces. At 10 GHz, the saturated electron density of 3 1010 electrons/cm3 is achieved in 0.6 ns. The impinging RF eld is reected by the lower impedance in the gap region and is absorbed by conversion to heat in the electron cloud. The multipactor gap consists of low-Q resonators in a combline lter conguration. An electron source emits enough electrons (Ni) into the gaps to cause the multiplication to begin on the rst few cycles of RF. A supply of oxygen is leaked into the gap region so that the metal oxides continue to have a coefcient d, since prolonged electron bombardment of the surface reduces the oxides to the metal with low d. To counter this leak of oxygen, a small ion pump is required. The pump maintains a vacuum pressure suitable to enable the electrons to be accelerated across the gap without colliding with the oxygen molecules. Clearly, the multipactor is a more complex limiter than the others discussed here. Therefore, it has found only limited application where biases for the ion pump are already available. The multipactor operates within the bandpass of the combline lter, while the lter protects the receivers front end from intense signals out of band. Measured performance of a 9.6-GHz multipactor had a 12% bandwidth (VSWR o1.6:1), would attenuate a 50 kW pulse to a 50 mW at leakage with a 2 mJ spike energy, and had a recovery time of less than 15 ns. Below threshold, the insertion loss was 1.5 dB, due to the combline lter [16]. 4.2. Bulk Window Waveguide Switch Array This switch array (or the monolithic diode array) was developed by M/A-COM Semiconductor Products, Burlington, MA, as a low-loss millimeter-wave switch for use in waveguide applications. The switch consists of a matrix of Si pin diodes grown monolithically onto an insulting Si

substrate. The substrate acts as a carrier, which is attached to a waveguide ange with suitable contacts to apply bias to the switch. The few attempts to operate the switch array as a self-activated limiter have not been successful, apparently because the i region could not be designed thin enough at millimeter-wave frequencies to allow the microwave energy to lower the impedance of the pin diodes. However, with external bias, the switch array has sustained operation at 1 kW pulse and 20 W levels at 94 GHz, with a low signal-level insertion loss of only 1 dB [17]. 4.3. Electrooptic The electrooptic limiter uses a coplanar waveguide transmission-line conguration on an electrooptically active, semiinsulating semiconductor as the switched medium. When illuminated with photons that have sufcient energy to excite electronhole pairs, the coplanar transmission line becomes lossy and absorbs most of the incident power. This limiter design is complicated because it requires an intense light source (e. g., a laser) to create enough electronhole pairs to provide the level of conductivity modulation of the substrate. The light source is activated above a threshold established by external circuitry coupled to the coplanar line, similar to the delay-line limiter shown in Fig. 1e. An implementation of this limiter design in semiinsulating Si yielded more than 30 dB isolation at 1.7 GHz, with over 100 mW of optical power. The isolation bandwidth was 25%. With no optical illumination, the same limiter had a high insertion loss of 6.5 dB, which was probably due to the high series resistance of the center conductor of the coplanar transmission line. 4.4. Superconductor High-Tc superconductor (HTSC) lms with a superconducting-to-normal transition at 86 K have been fabricated that show a surface resistivity change of 105. Gaidukov et al. [18] fabricated a two-element lter in a 8.012.4-GHz rectangular waveguide. The lter used resonant irises spaced a quarter-wavelength apart, fabricated with HTSC lm strips across the opening of each iris. In the superconducting state, the lm strip appears as a resonant inductor across the iris, with an insertion loss of E1 dB. The 3 dB bandwidth is E1 GHz. In the normal state, the lm strip is resistive, reducing both the Q and the resonant frequency of the iris, and yielding over 20 dB of isolation. When operated at 65 K with 0.8 ms pulses, the limiting began at 50 mW, and the pulseshape was unaffected by the limiter up to 0.5 W. With increased power

Next Page
MICROWAVE LIMITERS 2801

levels, the pulseshape showed more attenuation later in the pulse, due to the heating of the HTSC lm with a time constant ranging from 0.1 to 0.5 ms. The insertion loss increased about 5 dB during a pulse. If the HTSC lm strips are heated with an external current, the dual iris assembly can be used as a switch. 4.5. Field Emitter Array Field emission from a cathode incorporating a matrix of TaSi2 rods with nal tip radii of curvature in the range of 110 nm has been measured at DC. Kirkpatrick et al. [19] showed that the parameters of the FowlerNordheim relation depend on the magnitude of the electric eld at the tip. The emission is fast enough to be useful at microwave frequencies, when congured as a suspended 50-O microstrip (cathode) above a ground plane (anode). The TaSi2 rods are mounted in an insulating Si substrate. For the limiter application, Glenn et al. [20] used a conguration similar to that shown in Fig. 8, with a vacuum replacing the YIG and a microstrip-to-ground plane spacing of 40 mm. When the electric eld at the tips exceeded a threshold, electrons were emitted that shunted the transmission line, thereby operating as a limiter. As expected, the device worked with a DC threshold voltage of E200 V, providing limiting above 400 W on a 50-O transmission line. The device showed a high insertion loss at 1 GHz, which was attributed to losses in the Si substrate. Two other limiter circuit technologies have been proposed, but have not been demonstrated to date: varistor paint and temperature-dependent resistor. The rst technology uses the coplanar waveguide conguration with a varistor paint, shunting the center conductor to the coplanar ground planes. Initial attempts to build this device failed because the fringing electric eld from the coplanar line was too low to activate the varistor paint. The second technology is a highly temperature-dependent resistor (e.g., tungsten) with a small thermal mass, in series with a transmission line. The limiter could be used in series with, or in place of, the bondwire of a susceptible microwave component. However, experimental results indicated that the shunt capacitance of the temperature-dependent resistor was too high, reducing the current heating in the resistor. 5. FUTURE ACTIVITIES While pin diode limiter technology is rather well developed, several investigations still remain to be conducted. These investigations include: determining why inputoutput power hysteresis occurs for continuous and long-pulse operation; experiments to determine whether a slightly pdoped i-region diode will have lower spike leakage than do present designs; and determining why GaAs pin devices appear to degrade with intense pulses. MESFET limiters must be designed to carry higher pulsed currents to enable operation above 100 W, without the requirement for a large gate width. Also, limiters must be developed using the low-noise PHEMT technology, to make them compatible with the ampliers they must protect.

Although self-activated, high-power, millimeter-wave limiters are not presently available, the monolithic diode array concept appears to be the most likely to have low insertion loss, if the thickness of the i region can be reduced. Furthermore, the monolithic diode array does not increase the loss at low signal levels, due to the distribution of the carriers at the p i and the n i transitions. BIBLIOGRAPHY
1. A. L. Ward, R. J. Tan, and R. Kaul, Spike leakage of thin Si PIN limiters, IEEE Trans. Microwave Theory Tech. 42:1879 1885 (1994). 2. D. Leenov, The silicon PIN diode as a microwave radar protector at megawatt levels, IEEE Trans. Electron. Devices ED11:5361 (1964). 3. R. V. Garver, Microwave Diode Control Devices, Artech House, Norwood, MA, 1978, Chap. 9. 4. N. J. Brown, Design concepts for high-power PIN diode limiting, IEEE Trans. Microwave Theory Tech. MTT-1:732742 (1967). 5. C. M. Glenn et al., Nonreective Limiter, U.S. Patent 5,345,199 (1994). 6. J. F. White, Semiconductor Control, Artech House, Norwood, MA, 1977, Chap. 7. 7. R. J. Tan, Limiting Filter, U.S. Patent 5,280,256 (1994). 8. C. Trantanella, M. Pollman, and M. Shifrin, An investigation of GaAs MMIC high power limiters for circuit protection, IEEE MTT-S Int. Microwave Symp. Digest, Denver, CO, June 813, 1997, pp. 535538. 9. C. F. Vasile, FET Adative Limiter with High Current FET Detector, U.S. Patent 5,157,289 (1992). 10. A. F. Podell and E. B. Stoneham, Input Protection Circuit, U.S. Patent 5,301,081 (1994). 11. H. Goldie and S. Patel, An rf-primed all-halogen gas plasma microwave high-power receiver protector, IEEE Trans. Microwave Theory Tech. MTT-30:21772183 (1982). 12. S. D. Patel et al., Microstrip plasma limiter, IEEE MTT-S Int. Microwave Symp. Digest, Long Beach, CA, June 1315, 1989, pp. 879-882. 13. B. Lax and K. J. Button, Microwave Ferrites and Ferrimagnetics, McGraw-Hill, New York, 1962. 14. J. L. Carter and J. W. McGowan, X-band ferrite-varactor limiter, IEEE Trans. Microwave Theory Tech. MTT-17:231232 (1969). 15. J. D. Adam and S. N. Stitzer, Frequency selective limiters for high dynamic range microwave receivers, IEEE Trans. Microwave Theory Tech. 41:22272231 (1993). 16. T. P. Carlisle, X-band high-power multipactor receiver protector, IEEE Trans. Microwave Theory Tech. 26:345347 (1978). 17. A. L. Armstrong and Y. Anand, A limiter for high-power millimeter-wave systems, IEEE Trans. Microwave Theory Tech. 31:238241 (1983). 18. M. M. Gaidukov et al., Microwave power limiter based on high-Tc superconductive lm, Electron. Lett. 26:12291231 (1990). 19. D. A. Kirkpatrick, A Mankofsky, and K. T. Tsang, Analysis of eld emission from three-dimensional structures, Appl. Phys. Lett. 60:20652067 (1992). 20. C. M. Glenn et al., Microwave eld emitter array limiter, U.S. Patent applied for March 6, 1996.