AN INTEGRATED 94GHz MONOPULSE TRACKING RECEIVER. by Curis;Vhi h-shan fLing V:: A dissertatib'n ubm:itteidz in'ftail; lfillment of the requ1re me'ts, dgree ot D QCto 6f bvPh16b.h v' (Electri~ ~pngl ig) in The University of Michigan 1993 Doctoral Committee: Associate Professor Gabriel M. Rebeiz, Chairperson Research Scientist Jack East Associate Professor Linda P. IKatehi Professor Fawwaz Ulaby Assistant Professor Kim Winick

Ir\ Is

Curtis C. Ling 1993 All Rights Reserved

To Chung-Li and May

ACKNOWLEDGEMENTS My advisor, Prof. Gabriel Rebeiz provided the opportunity, inspiration and abundant funding for my graduate work. He always had uplifting advice during the failures and setbacks that are an integral part of a graduate student's existence. Aside from his plethora of ideas and incisive understanding, he had the ability to tolerate me throughout these years, and in spite of me, was able to start a successful research group. I express my thanks to Prof. Linda Katehi, who was willing to listen to my many questions and provided me with important insight. I also want to thank each member of my committee for their comments and suggestions for my thesis. Prof. David Rutledge was my undergraduate research advisor at Caltech, and I am greatly indebted to him and his research group for their guidance and support. Experimental work can often be tedious and unforgiving. From this standpoint, I want to start by thanking each member of my research group, TICS, especially Steve Gearhart, Dr. Walid Ali-Ahmad, and Brian Kormanyos, for being essential to getting our group off the ground, for their assistance, ideas, technical advice, as well as for their friendship. I also want to thank Terry Hull for her patience in helping our group get started with using the cleanroom facilities. Similarly, I would like to thank Prof. Tony England and Joe Landry, for providing the expertise in, and sharing the use of their Fourier Transform Spectrometer, resulting in several successful terahertz measurements presented in this thesis. Though this research relies primarily on experimental techniques, the theoretical

analysis provided by several individuals was important in validating the approaches to a number of problems investigated here. In particular, I appreciate the excellent analysis of the integrated horn structures by Dr. George Eleftheriades. I also would like to thank Dr. Nihad Dib for sharing his expertise in planar transmission lines, and John Natzke for helping me better understand the electromagnetic behavior of resistive sheets. Special thanks goes to the Rad Lab and EECS staff, particularly Ron Hartikka, Brenda Hadley and Jim Morgan, for their fine work. I wish to acknowledge the generous financial support I received from the NASA/Center for Space Terahertz Technology, Jim Mink with the Army Research Office, as well as the Rackham School of Graduate Studies. The strong friendship and encouragement of many, especially Joe Landry and Scott Ohman were vital to my well-being during graduate school. Joe, thank you for helping me finish this and all those other marathons. Dan and Barbara Balbach welcomed me as part of their family in Ann Arbor: their friendship and love will always be treasured. Finally, my deepest gratitude is reserved for my father, mother, and my brothers, Horace, Bryant, and Maurice, who have been with me from the beginning. I ~ ~ ~ ~ ~ ~ ~ ~ ~D~l

PREFACE This thesis presents the work of two separate projects. A 94GHz integrated monopulse tracking receiver is discussed in Chapters I-III. Chapter IV discusses the development of a wideband integrated millimeter-wave and terahertz power sensor.

TABLE OF CONTENTS DEDICATION................................. ii ACKNOWLEDGEMENTS......... o................ iii PREFACE............................... v LIST OF FIGURES......................... viii LIST OF TABLES.............................. xv LIST OF APPENDICES......................... xvi CHAPTER I. Introduction.......................... 1 1.1 Tracking Methods........................ 3 1.2 Monopulse Implementation................... 6 1.3 Millimeter-Wave Technology....-........... 8 1.4 Power Measurement at Millimeter-wave and Terahertz Frequencies............................. 11 II. Monopulse Receiver Design.................. 12 2.1 Subharmonic Receivers................ 12 2.2 Methods for Mixer Analysis...........15 2.3 Antenna and Mixer Design................... 17 2.4 Antenna Modeling.......................... 23 2.5 Antenna and Mixer Optimization.......29 2.6 Receiver Fabrication.................... 37 III. Tracking Receiver: Implementation and Measurements... 43 3.1 Oscillator Design...................44 3.2 LO Amplifier Circuit Design and Measurement o...... 53 3.3 Diode Characterization................... 64 3.4 Antenna and Mixer Measurements.............75 vi

3.4.1 LO Power Investigation................ 78 3.4.2 Discussion of Losses in the Antenna..... 80 3.5 IF Monopulse Processor.............. 81 3.6 Conclusion and Future Work................ 86 IV. An Integrated Millimeter-Wave and THz Power Sensor... 90 4.1 A Brief Technology Review................... 90 4.2 Introduction to the Power Sensor............. 91 4.3 Fabrication...92 4.4 Modeling and Calibration.................... 94 4.4.1 Determining an Electromagnetic Model...... 94 4.4.2 Finding the Bolometer Responsivity Using a LF Network.................... 95 4.5 Millimeter-wave Measurements................. 101 4.6 Absolute Output Power of a Millimeter-wave Tripler..... 107 4.7 THz Laser and Spectroscopy Measurements........ 108 4.7.1 FTS Measurements.............. 110 4.7.2 FIR Laser Measurements............... 111 4.8 Conclusion............................ 113 APPENDICES.................................. 116 BIBLIOGRAPHY................................ 160

LIST OF FIGURES Figure 1.1 An imaging array........................ 4 1.2 Monopulse antenna patterns: sum (Z) and difference (A)..... 4 1.3 Two types of monopulse detection. Amplitude (left), and phase monopulse (right). Note that the amplitude monopulse can also be used as a phase monopulse...................... 5 1.4 A waveguide-fed monopulse four-horn array........... 6 1.5 A generic planar monopulse antenna................... 1.6 A quasioptical diplexer using a polarization grid for injecting the local oscillator power........................ 9 1.7 An extended hemispherical substrate lens (a). A thin dielectric membrane used to support an antenna (b).......... 11 2.1 Antiparallel diodes: LO voltage and large-signal conductance waveform g(t) (above), and large-signal equivalent model (below). Cp is the parasitic capacitance of the diode pair, and is included with the linear circuit....................... 14 2.2 Large-signal harmonic balance analysis. N is the number of LO harmonics to be included in the numerical computation...... 16 2.3 Integrated horn antennas..................... 18 2.4 Etched 90pi-thick substrate with via-trenches used to suppress substrate modes...................... 20 2.5 Antenna and mixer circuit dimensions. The dashed lines show the location of a polyimide insulating patch used in overlay capacitors. The cross section of the horn corresponds to 0.36A. All dimensions are in micrometers..................... 22

2.6 Illustration of the monopulse array layout and LO distribution scheme. Dimensions are in micrometers. Detailed dimensions of the transitions, and an equivalent circuit diagram are shown in Figure 2.7 24 2.7 The CPW-CPS transition and power splitter circuit. Dashed lines show the outline of insulating polyimide patches. All dimensions are in micrometers.......................... 25 2.8 Calculated resonant modes within a dielectric of e, -12 in a square waveguide of cross-section of 0.53Ao [34].............. 27 2.9 The transmission line model for analyzing the dielectric layer... 27 2.10 Coaxial quarter-wave balun...................... 29 2.11 Conversion loss as a function of series resistance, for a constant set of LO embedding impedances, and 6mW LO power........ 31 2.12 Contour plot of conversion loss including RF mismatch, as a function of the LO (23GHz) embedding impedances for various power levels. The X-axis and Y-axis denote the real and imaginary parts of the LO aembedding impedance, respectively. Power level plays a crucial role in optimal mixer performance. The other embedding impedances are held constant, at the values listed in Table 2.2......... 33 2.13 Contour plot of conversion loss including RF mismatch, as a function of the LO (23GHz) embedding impedances for 6dBm LO power. Rs = 3Q (left) and R8 = 62 (right). Diode series resistance severely degrades second-subharmonic mixing performance......... 34 2.14 Antenna/mixer embedding impedances obtained from a 2.7GHz microwave model. The 10% bandwidths at each LO harmonic are plotted................................... 35 2.15 Microwave scale model measurement of the impedance ZLO, the LO feed to the CPW-CPS transition marked on Figure 2.7....... 36 2.16 The two-step etching process. Dashed lines indicate the positions of the antenna and membrane. Arrows mark the direction of undercut. 39 2.17 Top view of the etched cavities for wafer #4. The thick line circumscribing the features indicates the edge of the wafer, which is diced or scribed............................... 41

2.18 Sideview of receiver wafers..................... 42 3.1 The reflection amplifier...................... 45 3.2 Smith chart illustration of the oscillation conditions and the first stability criterion. Arrows indicate the direction of trajectory as frequency increases............................ 47 3.3 The second stability criterion minimizing the phase noise produced by changes in oscillator signal amplitude. Circles mark the points at which the oscillation condition is satisfied. Arrow indicates the direction of the trajectory of 1/S11 as signal amplitude increases. 0 should be 90~ to minimize AM-to-PM noise............ 47 3.4 Circuit diagram and layout of the VCO design. The CPW\ transmission line sections all have an impedance of 40Q............ 49 3.5 1/S11 and output match S-parameters of the VCO design. Arrows indicate direction of change as frequency is increased. C, is the capacitance of the two gate varactors in parallel......... 51 3.6 Change in the trajectory of 1/Sl1 as the length of the transmission line connected to the source of the HEMT is changed....... 51 3.7 23GHz VCO output spectrum................... 53 3.8 23GHz VCO tuning data................. 54 3.9 Dimensions of the lines used the TRL calibration set and capacitor test fixture................................ 56 3.10 Measured losses for CPW line and DC-blocking capacitors..... 57 3.11 Layout of a single-transistor amplifier shown with T-junction corrections, and the equivalent circuit................. 58 3.12 S-parameters for a single-stage amplifier based on the FLR016XV transistor biased at VDS = 7V, Vcs = OV, IDS = 60mA....... 60 3 23GHz LO 3-stage amplifier circuit layout.............. 62 3.14 The modeled small-signal performance of the 3-stage LO amplifier. 63

3.15 The active wafer, sandwiched by horn sidewall wafers stacked above and below......................... 65 3.16 The fabricated active wafer of the 94GHz planar subharmonic monopulse receiver................................. 66 3.17 Experimental setup for RF diode measurements. Video detection (a), and subharmonic mixer measurements (b)............ 68 3.18 Dimensions of the planar self-complimentary log-periodic antenna and substrate lens.......................... 69 3.19 Diode model used in video detection calculations. P,, is defined as the incident RF power density (S) multiplied by the physical area of the aperture, Aphys........................ 70 3.20 Calculated and measured video responsivities as a function of bias current for the following parasitic values: Rf - 6Q, Cjo = 75pF, Cp, 35fF, bi = 0.75, n=l.l, and I, = 1.2 x 10-14............ 72 3.21 Subharmonic coniversion loss at 80.36GHz using a 20GHz LO source, for a MA/COM 40422 antiparallel Schottky diode pair mounted in a self- complementary log-periodic antenna, as a function of available LO power. The diode parameters used in the calculations are the same ones listed in Figure 3.20. The conversion loss is defined on page 74................................ 74 3.22 E and H-plane patterns for the individual horns in the monopulse array. The antennas are numbered as indicated in Figure 3.16... 76 3.23 Conversion loss for two monopulse channels, as a function of relative LO power level (dB below maximum LO power)........... 80 3.24 Calculated IF impedance and conversion loss for the monopulse tracking receiver, as a function of LO power, using harmonic balance analysis. (R, = 6Q, Cjo = 75fF, Cp = 35fF, n = 1.1, I 1.5 x 10-14, qbi- 0.75V)...................... 81 3.25 Block diagram of the 200MHz IF processor.............. 82 3.26 Measured monopulse amplitude patterns at 94GHz......... 83 3.27 Measured monopulse phase patterns at 94GHz........... 84

3.28 Measured antenna patterns of the monopulse receiver when used ill conjunction with a 10cm, F/D-0.9 lens........8........ 7 4.1 A monolithic wideband quasi-optical power sensor........ 93 4.2 The equivalent transmission line circuit for the power sensor for a normally incident plane wave................... 96 4.3 The low-frequency (0.5 to 5MHz) calibration network...... 97 4.4 A signal in the LF spectrum turned on at t=0 and applied to the bolometer (a). The bolometer's thermal response (b)........ 98 4.5 The high-pass and low-pass filters used in the LF calibration netw ork................................. 100 4.6 The millimeter-wave quasi-optical measurement set-up....... 100 4.7 The single-pole thermal bolometer response versus modulation frequency.................................. 101 4.8 Bolometer response at 140GHz as a function of polarization angle for normal incidence................. 103 4.9 The equivalent transmission line circuit for E and H-plane incidence. 104 4.10 H-plane pattern at 140GHz for a 7x7mm bolometer with a sheet resistance of 98Q.................... ii4 4.11 Relative absorbed power versus mirror position for a 4x4mm bolometer at 140GHz (top) and 240GHz (bottom).............. 105 4.12 Large area bolometer, shown with the transmission line circuit equivalent at 1THz............................. 110 4.13 FTS transmission spectra for the dielectric membrane with and without the thin film bismuth. Simple transmission line model theory (dashed lines) is shown for comparison. Squares and triangles mark 802GHz and 2.54THz respectively................. 114 4.14 FIR laser beam profile at 802GHz, at several distances from the laser output coupler. PO is the peak incident power density....... 114 4.15 FIR laser beam profile at 2.54THz, at several distances from the laser output coupler. Po is the peak incident power density... 115

A.1 A monolithic H-plane monopulse antenna............. 11S A.2 Side view showing sum and difference antenna planes. The detector of the difference antenna is integrated very close to the center the crossover network. Note that the two antennas are not drawn to scale................................ 120 A.3 (a) Measured 1.2GHz input impedances. (b) Isolation between Z and A channels.... 123 A.4 Measured 3.3GHz E and A H-plane patterns, shown with theoretical A pattern.......................... 124 A.5 Measured 92GHz E and A H-plane patterns, shown with the theoretical A patterns..................... 125 A.6 Modified difference antenna. The enlarged center pad allows a hybridmounted Schottky diode to be attached to the CPS line. The sum antenna did not require modification..127 A.7 Measured A pattern with horn extension at 92GHz, compared with theory. The difference pattern without the machined section is also shown for comparison.. 128 A.8 Layout of 94GHz sub-subharmonic integrated monopulse receiver, difference channel......................... 129 B.1 Substrate surface profiles for initial thicknesses of 5000A/3000A/7500A. After 20 seconds BHF etching (resulting in thickness of 4670A/3000A/7500 ) (a), and 40 seconds BHF etch (resulting in thicknesses of 4330A/3000A/7500A ) (b). 134 C.1 The equivalent circuit used in the reflection algorithm....... 139 C.2 Illustration of the symmetry used in Kerr's analysis..141 C.3 Diagram of the nonlinear circuit analysis for antiparallel Schottky diodes............................. 143 C.4 Typical waveforms for the total current in an antiparallel Schottky diode pair being driven by a single-tone large-signal excitation... 144 D.1 Low-noise 200MHz IF amplifier circuit having a noise figure of 1.7dB and a gain of 45dB..................... 146 xiii

D.2 Amplitude and phase trim circuits for a single channel. The TFAS-1 is a matched variable attenuator.................... 147 D.3 The IF monopulse processor, which processes the signals from the four channels after amplification (Fig. D.1 ) and amplitude and phase-trimming (Fig. D.2 )................... 148 D.4 Circuit diagrams for a differential amplifier (above) and an IF logarithmic amplifier (below)....................... 150 D.5 Circuit diagram for the phase discriminator............. 151 D.6 Square-law detector with a sensitivity of 5600mV/mW at 200MHz. 152 D.7 Approximating the normalized monopulse slope from the normalized difference voltage pattern.................. 153 E.1 Bond wire over a ground plane. s = 2h is the center-to- center separation between the bond wire and its image........ 157 E.2 Diagram illustrating the dimensions used in the VCO bond wire calculations. Manufacturer's test fixture with HEMT residing in a 150pm-deep trench (above). Surface-mounted HEMT used in the receiver (below)..................... 159 xiv

LIST OF TABLES Table 2.1 Diode parameters used in mixer design. The values lie in the center of the manufacturer-published ranges for each parameter... 31 2.2 Mixer embedding impedances at each LO harmonic. -.. 32 3.1 DC diode parameters extracted using I-V curve fitting, listed in pairs for each diode................... 67 3.2 Measured total conversion loss for the four monopulse receiver channels, defined as the IF power divided by the total power incident on the aperture........................... 8 3.3 Measured IF impedances for the four monopulse receiver channels. 79 4.1 Absolute measured power density of an incident plane wave using four different bolometers at 90G11z, 140GHz and 20GHz. The results at 90GHz agree well with the Anritsu power meter measurements. 106 4.2 Measured output power of a Millitech MUT-03 tripler. The bolometers with the largest measurement discrepancy over the 220-280GHz range are presented here. Notes- 1 Limited by a maximum available source power of 6mW; 2 Measured at 277GHz........... 108 B.1 Prefurnace Clean... 13-5 B.2 LPCVD SiO2 and Si3N4 Deposition Parameters.......... 136 D.1 Active components list................ 154 D.2 Passive components list........................ 155 xv

LIST OF APPENDICES Appendix A. A Single-Horn 94GHz Integrated Azimuthal Monopulse Antenna. 117. A.1 Antenna Description.. 117 A.2 Fabrication... 119 A.3 Theory................. 121 A.4 Microwave Measurements....... 122 A.5 Millimeter-wave Measurements... 124 A.6 A Receiver Implementation.. 12.5 A.7 LO and Mixer Design.......... 197 A.8 Discussion..!3U B. Fabrication of Thin Dielectric Membranes on a Silicon Substrate... 131 B.1 Determining Suitable Layer Thicknesses. 131 B.2 Thermal SiO2... 133 B.3 LPCVD S-3N4.............. 135 B.4 Comments........... 1335 C. Harmonic Balance Analysis Supplemental....138 D. IF Processing and Monopulse Accuracy.... 145 D.1 Amplitude Processing..146 D.2 Phase Processor..... 149 D.3 Monopulse Error Due to Noise........... 152 D.4 Component List.............. 153 B. B ond Viie Chiaracterization.......... 156 Xvit~~illC~ L~v ~I,,o....

CHAPTER I Introduction Tracking systems are essential components in numerous radar applications. Obvious military applications include projectile guidance and tracking. Vehicle collision avoidance, robotics navigation and object location, and telemetry are some important civilian applications. Disturbances caused by a target are detected by various types of sensors or receivers in order to yield information about the target's position and velocity. These disturbances include electromagnetic radiation produced by or reflected from the target. The work completed in this thesis describes the implementation of a novel tracking receiver (the sensor or front-end), designed to track the target's radiation at millimeter-wave frequencies. The receiver is designed to operate at 94GHz, but is scalable to frequencies from 50GHz to above 100GHz. Radiation in the frequency spectrum ranging from low-frequency radio waves to visible light and beyond has been used in tracking systems. Passive tracking utilizes radiation which is emitted from the target. Active tracking systems illuminate the target with a source in order to produce a stronger, coherent signal which can be used for post-sensor processing such as Doppler processing and beam-steering. Lower frequencies (f < 10GHz) have the advantage of efficiency, simplicity of power production, and range. Such systems include most X-band radars, over-the-horizon

LF radars, and telemetry systems such as those used in spacecraft. Operation at optical wavelengths (A < 10m) allow compact imaging and highl-gain systems to t)e constructed, since the very short wavelengths allow very small yet efficient sensors to be constructed. Examples of such systems include laser radars used in imaging and missile tracking [1] and infrared and radar missile guidance systems. Each of these systems has certain drawbacks. Low-frequency systems produce radiation at long wavelengths and therefore require large, cumbersome antennas in order to obtain high-gain radiation patterns. They also may be more susceptible to multi-path effects. On the other hand, laser and infrared optical systems experience a significant degradation in performance when operating in weather conditions such as clouds, fog, and dust, which strongly attenuate or scatter radiation at these wavelengths. Operating at frequencies lying between radio and optical frequencies, millimeter-wave systems (10mm < A < 1mm) attempt to utilize the benefits of microwave and optical systems while avoiding their drawbacks. Radiation in this portion of the spectrum can penetrate many optically opaque conditions (such as poor weather, clouds or dust), and encounters much less RF interference and noise than other types of systems. The large bandwidth offered by millimeter-wave systems also allows spread spectrum techniques, and greater Doppler frequency shifts compared to microwave systems. The wavelengths at these frequencies are also small enough to allow the easy fabrication of compact high-gain antennas together with other integrated circuit components. The work in this thesis therefore concentrates on systems operating at millimeter-wave frequencies.

1.1 Tracking Methods The geometry and associated capabilities of the receiver system determine how the information it provides will be used to locate a target. An imaging system utilizes a large number of closely spaced sensor elements arranged in an array in order to obtain the requisite imaging resolution. This array is used in conjunction with a lens in order to capture an image of the far-field (Fig. 1.1). An example of an imaging system is a CCD-based optical tracker. Information from each sensor element (or pixel) must be processed using image-processing algorithms. These include centroid detection and edge enhancement in order to locate a target or its path [2]. Such systems are restricted to frequencies lying within or above the IR and visible spectrums. At microwave frequencies, the longer wavelengths correspond to larger sensor elements, making imaging systems excessively large. The shorter wavelengths (A < 4mm) at frequencies greater than 80GHz allow compact imaging arrays to be fabricated, and much research is currently being done in this area [3, 4, 5]. Tracking systems can also use fewer elements, such as a single antenna or laser system coupled to a lens, to locate an object using direction finding techniques [6]. One method for using a single antenna to track an object is to modulate the motion of the antenna by a circular or conical scan. By comparing the phase and amplitude of the modulation in the return signal with the transmission modulation, the target's position with respect to boresight of the scan can be determined. Another method involves sequentially switching between several lobes with boresights at fixed relative angles to achieve the same effect. These methods have the advantage of simple implementation and lower cost: any antenna can be used with little or no modification in the scanning system. However, resolution and performance in adverse conditions suffers

4 Figure 1.1: An imaging array. Figure 1.2: Monopulse antenna patterns: sum (E) and difference (A). rom several errors. For example, any variation in signal strength which is not due to the scan modulation will introduce some pointing error. This variation can be caused by target scintillations, movement by the target, or weather conditions. The monopulse technique was developed in order to circumvent these problems. In order to locate a target in one coordinate (the azimuth, for example), it uses two antenna patterns: one pattern with a sharp null on boresight (the difference pattern), and a second pattern with a broad lobe on boresight (the sum pattern) collocated with the sharp null (Fig. 1.2). The monopulse processor uses the relative amplitude and phase of the signals received from the two patterns (referred to as the sum and difference signals) to obtain tracking information. The amplitude of the difference signal is divided by the amplitude of the sum signal in order to produce a "normalized" difference pattern. This normalized difference pattern is invariant with

Amplitude or Phase Monopulse Phase Yonopulse Figure1.3: Two types of monopulse detection. Amplitude (left), and phase monopulse (right). Note that the amplitude monopulse can also be used as a phase monopulse. respect to the absolute signal strength: it produces an output which only depends on the angle-of-arrival of the signal, not its power level. When angular location of the target is close to the null, the magnitude of the normalized difference pattern is proportional to the angle between the target's location and antenna boresight. Two types of monopulse processing techniques have been developed. In a phase monopulse system, the phase difference between the sum and difference signals contains odd symmetry with respect to boresight, and is thus used in order to eliminate any ambiguity concerning the target's location relative to the null. The boresights of the elements in the monopulse antenna may be parallel to each other, a;, li. d piualli array used without a lens, and the system essentially operates as an interferometer. The amplitude monopulse technique in contrast, utilizes separate antennas with boresights which are nonparallel, as in a horn array feeding a lens (Fig.1.3). The processor compares the amplitudes of the signals received by the separate antennas in order to determine which side of the null the target resides on. Note that the phase difference between the sum and difference patterns in the amplitude monopulse system is also an odd function about the boresight, and therefore the amplitude monopulse system can also be used as a phase monopulse, but not vice versa.

Figure 1.4: A waveguide-fed monopulse four-horn array. 1.2 Monopulse Implementation Monopulse systems are more robust and accurate than scanned systems because they are less susceptible to errors caused by fluctuations in absolute signal amplitude. They also eliminate the mechanical complexity required in scanning systems, along with the wear that rapid scanning produces. Target tracking is more rapid because a single pulse can provide complete information about the target's location. An analysis of tracking error in monopulse and conical scanning tracking systems due to the presence of noise can be found in [7]. In order to obtain the target location in both the azimuth and elevation directions, the antenna system need to produce two difference patterns (one for azimuth and one for elevation) in conjunction with the sum pattern. In a single-horn monopulse, orthogonal modes can be excited within a single horn cavity to produce the three antenna patterns [8]. This method is elegant but hard to build in a planar realization. The monopulse patterns can also be produced by combining the radiation collected from several radiating elements arranged in an array. Though technically speaking, only three elements are needed to track in both the elevation and azimuth directions, four elements are usually used to produce symmetric patterns and simplify processing (Fig.l.4). The array can be used alone or to feed a reflector.

C, F, A Figure 1.5: A generic planar monopulse antenna. The monopulse array can consist of individual horn antennas, quadrants of a planar antenna array (Fig. 1.5), or a phased array. The RF signals from the individual elements in the monopulse array are then processed using hybrid combining components to produce the monopulse signals [9]. In a phased array, the monopulse patterns can be formed and pointed electronically in order to track several different targets simultaneously in real-time. Note that a planar antenna array with identicai elements must operate as a phase monopulse since each element has the same boresight. When combined with a lens which maps the pattern of each element's boresight to a different direction, the array can be used as either a phase or an amplitude monopulse. A single, multi-modal horn on the other hand, must be used as a phase monopulse since the sum and difference processing takes place within the horn, and thus individual element signals do not exist. The only disadvantage of the monopulse technique is the increased complexity of the antenna and receiver processing elements. Increased complexity can also mean increased system size, weight and cost, and reduced reliability. These are drawbacks which can be overcome with the advent of integrated millimeter-wave technology.

1.3 Millimeter-Wave Technology Millimeter-wave radar systems rely on waveguide components which are difficult and expensive to produce especially at higher frequencies. Waveguide systems are also heavy and hard to realize in a compact form. Part of this problem is due to the fact that each sub-system (local oscillators, mixers, and antennas, for example) are each individually designed and assembled, and then combined together using standard waveguide components such as couplers, attenuators and isolators. This is done because it is prohibitively expensive to custom manufacture each system using waveguide technology. One way to circumvent these difficulties is to integrate the system on a semiconductor wafer. At millimeter-wave frequencies, antennas and sub-systems can be designed to be small enough so that standard integrated circuit processing techniques can be used to mass-produce a rugged, reliable and inexpensive system. Planar circuits operating at millimeter-wave frequencies are not yet well characterized. Also, traditional planar circuits, such as those based on microstrip and coplanar waveguides [10, 11], are lossy at millimeter-wave frequencies due to increased radiation, dielectric and metal losses. Thus RF monopulse processing networks, which using planar transmission lines to realize combiners and couplers, are very hard to build. One way to overcome this difficulty is to integrate mixers directly at the antenna elements. This would allow the system to mix the RF and local oscillator signals directly at the antennas in order to process the signals at the intermediate frequency. However, this system is faced with the difficulty of distributing the LO power among the mixer elements. Because LO is difficult to produce at millimeter-wave frequencies, this design must overcome the equally challenging task

planar polarization antenna grid [ -LO+RF \ RF Quasioptical LO injection LO (9d polarization) Figure 1.6: A quasioptical diplexer using a polarization grid for injecting the local oscillator power. of achieving low circuit losses in LO distribution. A possible solution is to inject the LO into each mixer through the antennas using a quasi-optical technique and balanced mixers [121. This is a bulky nonplanar approach which requires precision manufacturing and alignment. Another solution, based on the novel use of planar subharmonic mixers, is presented in this thesis. In addition to the problem of low-loss RF signal processing and LO distributions efficient millimeter-wave integrated antenna designs are difficult to achieve. At these frequencies, wavelengths become comparable to the thickness of the substrate. The antenna radiates preferentially into the dielectric substrat6, especially on semiconductor substrates with high dielectric constants 113, 14]. Consequently, the antenna excites "substrate modes" which trap power within the substrate. Because of this, such antennas typically suffer from narrow band impedances and poor radiation patterns. They also produce undesired coupling to other portions of the system. Conventional methods for suppressing substrate modes include placing the planar antenna on a substrate lens (Fig. 1.7). This allows the radiation from the antenna to propagate nearly normal to the air-dielectric interface. Substrate lenses have been

extensively used with dipoles, slots, as well as log-periodic and spiral antennas with good results up to 800GHz [15]-[18]. One drawback of this techniques is that the lens must be machined out of materials with the same dielectric constant as the substrate, making it hard to manufacture and expensive when materials such as silicon or GaAs are used. Multi-layered substrates can also be used to effectively match the antenna to free-space [19, 20]. However, they suffer from air-gap problems, are difficult to fabricate with standard semiconductor processing techniques, and result in narrowband designs. Another approach to circumventing problems with substrate modes is to integrate the antenna on a thin dielectric membrane. Dielectric membranes 1-1.5-m thick grown or deposited on a substrate are used to suspend the antenna so that it radiates as if in free-space. Several antennas have been developed using this technique [21, 22, 23], but it is difficult to monolithically integrate semiconductor devices such as Schottky diodes or HEMT devices on the thin membrane. A compact, efficient yet inexpensive monolithic planar system is thus difficult to realize at millimeter-wave frequencies. The tracking receiver presented in this thesis is based on antennas which not only circumvent substrate modes, but also are compatible with standard IC fabrication techniques. The work in the first part of this thesis describes methods for designing and fabricating a planar millimeter-wave tracking receiver with the RF antennas, LO and mixers defined on a single semiconductor chip. The receiver is based entirely on inexpensive, easily produced semiconductor components, and its fabrication is compatible with standard semiconductor fabrication techniques, making it well suited for low-cost mass production. Chapter 2 describes the main ideas in the receiver design, which is based on subharmonic mixers. Chapter 3 describes fabrication and

R Jf d ~ut-wsay view (b) Figure 1.7: An extended hemispherical substrate lens (a). A thin dielectric mem-'brane used to support an antenna (b). measurement of the system at 94GHz. 1.4 Power Measurement at Millimeter-wave and Terahertz Frequencies Chapter IV, the second part of this thesis, describes the investigation of a wideband quasi-optical power sensor which accurately measures absolute power densities at frequencies from 90GHz to nearly 3THz. It is based on monolithic integrated circuit construction, and is easily calibrated using low-frequency techniques. The power meter is compared to commercial instruments at millimeter-wave frequencies. It is also used to determine the output power of a commercial millimeter-wave tripler. The sensor structure is analyzed using transmission spectroscopy, and an array of sensors is used to measure the characteristics of a far-infrared laser at 802GHz and 2.54THz.

CHAPTER II Monopulse Receiver Design The purpose of this chapter is to describe the analysis and design of the monopulse antennas and mixers, which are the basis for the receiver architecture. These two -ormnopepn. ari monllocated. and must therefore be designed and optimized together The receiver is realized using hybrid-mounted devices, in order to simplify fabrication. It is based on subharmonic mixers which are an integral part of the antennas. Sections 2.1 and 2.2 motivate the mixer design. Methods for modeling and simulating the antennas and mixers are then discussed in Sections 2.3 and 2.4. In Section 2.5, the antenna/mixer circuit design and optimization process is presented. The final section covers the fabrication of the receiver. 2.1 Subharmonic Receivers The subharmonic mixer (SHM) is a fairly well understood technology [24] and has been used especially at millimeter and submillimeter-wave frequencies to obtain noise temperatures and conversion loss which are only a few dB above the performance of fundamental mixers [25, 15]. Local oscillator (LO) power generation at millimeter} wave frequencies is a difficult task, and device output power decreases as the square of the frequency [26]. A main design challenge is thus to produce sufficient LO power

to drive the mixer at millimeter-wave frequencies. The SHMN was therefore developed in order to circumvent this difficulty. The power-producing capabilities at millimeterwave frequencies has advanced due to improved device technology, resulting in more available power at higher frequencies. Thus, the frequency at which millimeterwave SHM's are commonly used has also increased, to the point where designers at frequencies below 100GHz seldom resort to subharmonic designs. At frequencies above 100GHz, the SHM is more commonly found, and can be potentially used up to terahertz frequencies. The subharmonic receiver utilizes a mixing element with a nonlinear, antisymmetric current-voltage characteristic which allows it to mix efficiently at even harmonics of the LO frequency. This allows the LO to be produced at lower frequencies, where much more power is available. A number of devices exhibit such an I-V characteristic, such as the planar-doped barrier diode [27]. The most commonly used device consists of two Schottky diodes connected together in an antiparallel configuration (Fig. 2.1). At millimeter-wave frequencies, planar Schottky diodes are extensively used since they can be fabricated with cutoff frequencies above 1THz [28]. The LO power drives each diode's nonlinear conductance into the forward-bias region at different times in the LO waveform. During the positive half-cycle, one diode is forward-biased, causing the conductance to rise, while the other diode is in reversebias. During the negative half- cycle, the other diode is turned on while the first diode is reverse-biased. Since the two diodes are connected in an antiparallel configuration, the total conductance of the diode pair is changing at twice the frequency of the LO making it an efficient mixer at twice or even four times the LO frequency. With high-quality Schottky diodes, conversion efficiencies of 8dB are typical for a 94GHz mixer being driven at 46GHz [15].

14 VLO Vi g(t) - {p~j~-_ t Nonlinear circuit -----------------------------— 7 r —-' —------- r —------------— I Linear cuidiode 2 diode 1 Circuit c VLO C Vcj L______________________________J L ~___________J L______ _~ __J Figure 2.1: Antiparallel diodes: LO voltage and large-signal conductance waveform g(i) (above), and large-signal equivalent model (below). Cp is the parasitic capacitance of the diode pair, and is included with the linear circuit. The 94GHz tracking receiver described in this thesis utilizes four sensor elements (antennas) arranged in a two-by-two array. These sensors are based on SHM's driven by a LO with a frequency one fourth the RF, roughly 23GHz. Signals received by each antenna element are mixed down and processed at the IF to produce the three monopulse patterns. The use of SHM's is central to the receiver design: it allows the LO to be produced and distributed using low-loss planar circuits. This, in turn, makes attractive the integration of mixers directly with antennas, in order to perform monopulse processing at the intermediate frequency using low loss, conventional signal processing circuits. Thus each part of the receiver can be realized in an efficient yet compact, planar form.

2.2 Methods for Mixer Analysis A solid foundation of nonlinear techniques has been developed for analyzing mixers in order to obtain designs which are optimized for noise or conversion loss [29]. These techniques can be used to analyze a large number of nonlinear devices numerically, and the algorithms rapidly converge to solutions when implemented on a personal computer. A technique known has "harmonic balance analysis" (HBA) is particularly useful for looking at circuits which have either strong or weak nonlinearities, and which are excited by a single tone. Such circuits include power amplifiers, multipliers and mixers. When applied to mixer analysis, the single tone corresponds to the LO pump frequency. It is important to start with the mixer analysis, to show that the subharmonic mixer has the potential to achieve very good performance at 94GHz. HBA is aptly named, because it essentially finds the voltage components at each harmonic of the single tone signal necessary to balance, or equal, the voltage waveform produced by the nonlinear part of the circuit. The HBA technique involves first calculating the behavior of the nonlinear circuit under large-signal excitation. When applied to mixer analysis, the results of this calculation provide complete information of how the nonlinear device is behaving as a function of time [30]. These time-varying conductances are then used to perform small-signal analysis, which yields conversion matrices describing the small-signal conversion loss and noise characteristics of the mixer. A "small-signal" is one whose magnitude is small enough to have a negligible effect on the time-varying impedances of the diode junction. This is a very suitable description of typical RF signals being processed by mixers, which are indeed much smaller than the injected LO signal. The way HBA works is by separating

~~~~li ln] iil i ics to be included in the numerical computation. the circuit into linear (embedding) and nonlinear components (Figs. 2.1 and 2.2). An initial guess for the voltage v is applied across the nonlinear device to obtain a time-domain steady-state solution to its current zi(t). This current only contains a periodic boundary condition on the circuit. The embedding circuit also contains sigurenals which conLarge-signast of harmonics of thbalance anaLO. The current vector i (containing eachrmonof the irLO hcuit into linear (embedding) and non in the embedding circuit) are subtractedand 2.2). from i,nl (the LO harmonic current components in the nonlinear circuit) to form an error vector Ze (Fig. 2.2). If the solutions are correct, the current in the nonlinear device will be equal to that existing in the embedding circuit, and the error vector ie is zero at each frequency component. If ie is nonzero, then the initial guess v, is systematically modified, and the process is iterated until the error vector is zero. A number of algorithms can be used to modify the solutions in order to minimize the error vector with few iterations and stable convergence. One such algorithm is the reflection algorithm. It is robust, requires few initial parameters to be set, and utilizes low amounts of computer memory by ax iriding matrix manipulation. The HBA and reflection algorithm were implemented in computer code published in earlier versions of [30]. The reflection algorithm is described in Appendix C, and was used by Held

and Kerr [31] to analyze loss and noise characteristics of single-diode mixers. \Nork in [24] extended this work to balanced and subharmonic mixers by taking advantage of symmetry of the diodes and their embedding impedances. This simplified the analysis and reduced the amount of computation by allowing the calculations for a single diode to be used to find the mixing performance of the diode pair. Work done in this thesis treats the back-to-back diode pair as a single nonlinear device, and incorporates the resulting model in the HBA code. This different approach to analyzing the diodes was used to verify Kerr's work, as well as to permit the analysis of asymmetric diode pairs. This implementation, and comparative results with Kerr's method are presented in Appendix C. The results of the comparison between the two methods show close agreement in diode impedances and conversion loss, and gives confidence in both methods since they each utilize different models of the nonlinear device. 2.3 Antenna and Mixer Design The performance of the receiver is very dependent on the embedding impedances of the mixer circuit. The tracking receiver design employs mixers which are an integral part of the antenna. This eliminates the losses-associated with propagating the RF power from the antenna to a separate mixer circuit using lossy 94GHz transmission lines. Thus in order to understand the mixer, it is first necessary to discuss the planar antenna design. Therefore, this section digresses from the discussion of mixer analysis in order to talk about the modeling and fabrication of the antenna structures. The antenna design is based on integrated horn cavities. These are fabricated using anisotropic etching of silicon which results in a fixed flare angle of 700 (Fig.

Dielectric Membrane Antenna IF Out Space for processing electronics Figure 2.3: Integrated horn antennas. 2.3). Several wafers are etched with different opening sizes in order to form the cavity, which is constructed by stacking the wafers above and below a wafer containing the circuit. In order for the circuit to couple to modes within the horn cavity, elementary antenna structures such as dipoles or slots are deposited (using standard IC metallization techniques) on a thin dielectric membrane, which suspends the antennas within the horn cavity. The thin membrane is grown directly on the silicon wafer using fabrication techniques discussed in Appendix B. Because the membrane is much thinner than a wavelength at millimeter-wave frequencies, this allows the antennas suspended on the membrane to radiate as if suspended in free space within the horn cavity. This greatly simplifies the analysis and microwave modeling of the antenna structure. The membrane is rugged enough to easily withstand planar processing techniques. Thus metallized antenna structures can be integrated on the dielectric membrane and then aligned to the horn cavity side-wall wafers. TI integrated horn antenna design was first developed by Rebeiz and Rutledge [23], and similar structures were subsequently used in antennas such as the integrated corner cube [32] and the single-horn integrated monopulse antenna [33] (Appendix A). Appendix A presents a design which consists of radiating structures suspended

at different planes within the horn cavity to couple to different Imodes within the horn. The membrane is an elegant approach to designing efficient millimeter-wave antennas, by completely avoiding the substrate mode problem discussed in Chapter I. However, it has certain drawbacks. First, semiconductor elements are very difficult to integrate on the membrane with the antenna. Thus, devices must be hybridmounted with antennas on the membrane. Furthermore, the membrane process is not yet compatible with standard GaAs processing, and although rugged, it is much more fragile than the substrate itself. In order to circumvent these difficulties, a second method for coupling to the horn cavity was developed at the University of Michigan [34]. This method, referred to as the "modified horn antenna," replaces the membrane with a thin substrate with a thickness of about O.1Ad. When using silicon or GaAs as a substrate material at 94GHz, this corresponds to a substrate which is approximately 90tm thick. The antenna and any semiconductor devices can be integrated on the thin substrate and, via trenches, or grooves, are etched around this area, which is aligned with the stacked cavity sidewalls to complete the horn antenna. The substrate thickness is chosen so that the frequency of operation lies between resonant modes of the thin substrate. Work in Chapter 7 of [34] discusses the preliminary analysis of such a structure. The analysis, and the microwave modeling of this horn configuration is discussed in Section 2.4. A 90pm-thick substrate is too thin for practical fabrication and handling. In order to overcome this practical difficulty, only the substrate directly underneath the antenna is etched away until only 90im remains (Fig. 2.4). Via grooves surround the 90Om region, and thin silicon beams attach the thin substrate with the rest of the wafer. The structure shown in Figure 2.4 was micromachined in silicon using a

20 groove. 9,um' 9...1650/m —- Sia e-iew Via groove Figure 2.4: Etched 90y-thick substrate with via-trenches used to suppress substrate modes. two-step etching process which is described in Section 2.6. Mlicrowave measurements on a scale model of the modified horn excited by a simple dipole Nyields tuneable input impedances for the dipole, with moderate bandwidth (~10%o). These measurements are discussed in detail in Section 2.4. The first approach to designing a monopulse antenna using the modified horns was to try to excite even and odd modes within the same horn cavity by integrating planar antennas on dielectric layers located at different planes within the cavity. A similar design was first used successfully at 94GHz with membrane-based horn cavities, and is fully described in Appendix A. When applied to the modified horn antenna during microwave modeling however, this approach yielded antennas with narrow band impedances which were difficult to tune, especially in the case of the sum antenna, which is located closer to the apex of the horn.

Because of this difficulty, an alternative approach using a four-horn monopulse receiver architecture was used instead, where each of the horn cavities is excited by a simple dipole. By placing mixing devices directly at the apex of the dipole, the mixer becomes an integral part of the antenna. The width and length of the dipole, and its position relative to the apex of the horn, can be adjusted to tune the input impedance of the antenna. A coplanar strip feed is attached to the apex of the antenna in order to extract IF or inject LO signals to the mixer (Fig. 2.5). The feed of the antenna includes a quarter-wave RF capacitive choke which isolates the RF signal received by the antenna from the rest of the circuit. The length of the choke, and the capacitance used, can be used as a tuning circuit for adjusting the embedding impedances seen by the mixer, though' a simple quarter-wave choke was used in this design.. The RF choke capacitance also has the effect of lowering the LO embedding impedance seen by the diode, and thus cannot be chosen to be too large. The value for the RF choke capacitance is chosen to yield a susceptance of j50Q at the LO frequency, and j12Q at the RF frequency (Fig. 2.5), and its distance from the dipole is 0.25Ag.'"he impedance of the CPS on the 90ym-thin silicon substrate is estimated to be about 90Q, with an estimated guided wavelength of 2040pm. Both the impedance and propagation constant of the CPS transmission line were calculated using approximate analytical formulas developed by Gupta et al. [35]. The choke has an estimated return loss at RF of about 0.2dB, neglecting conductor and dielectric losses. The antenna/mixer design is shown in Figure 2.5, and is the result of an experimental optimization process described in Section 2.5. The four antennas are driven by the same LO source through a power distribution circuit shown in Figure 2.6. The 23GHz LO power is fed using a CPW transmission line and is produced by oscillator and amplifier circuits presented in Chapter 3.

22 020 << X ~ ~ iio1 Metal 2 < XZ~~~~~ Y~~8888 L Metal 1 101 Polyamide 90Om Silicon MA/COM 40422 / > A: >(Back-to-Backr C C C LS Silicon / Schottky Diodes' ~~~//~~ ~~~ m~RF Isolation / Overlay Cap. (O 25pF) CPS Za 0.25pF To Zcps90' Zcps 4G0 LO/IF Va( 90'(94GHz) Circuits Figure 2.5: Antenna and mixer circuit dimensions. The dashed lines show the location of a polyimide insulating patch used in overlay capacitors. The cross section of the horn corresponds to 0.36A. All dimensions are in micrometers.

The power distribution circuit contains a CPW-to-slotline transition (balun), which also function as a power divider. The design of the transition is straightforward, and is based on a quarter-wave choke. A similar balun with a design frequency of 15GHz was measured by Cahana [36], and exhibited a 1dB insertion loss across a 35% bandwidth. The balun design is adapted so that it drives two equal loads (the mixers) via the slotlines. The layout and equivalent circuit are shown in Figs. 2.6 and 2.7. The circuit dimensions are determined from the microwave modeling and optimization discussed in Sections 2.4 and 2.5. The slotlines easily transition to CPS, which feeds the antennas/mixers. The IF is isolated from the LO by a 20,um-wide break in the slotline. At LO frequencies, signals travel unimpeded across this break via a capacitive overlay whose value is selected to present a very low impedance at the LO frequency of 23GHz, and a very high impedance at the IF frequency of 200MHz. Once the design of the antennas and mixer circuits is finalized, the input impedance of the four antenna/mixers driven in parallel must be matched to the output impedance of the LO drive circuits introduced in Chapter III. 2.4 Antenna Modeling Preliminary modeling of a dipole radiating on a thin dielectric substrate (O.lAd) surrounded by via-trenches indicates that operation is well above the TE0o resonance, and that TE12 mode occurs when the thickness of the substrate approaches 0.25Ad (Fig. 2.8). The structure can be approximated as a simple waveguide model, a reasonable assumption since the fields produced by the resonant mode are mostly confined to the dielectric and die off rapidly as a function of distance from the substrate, similar to the situation encountered with a dielectric waveguide (Fig. 2.9). Within the air-filled sections of waveguide, higher-order modes are evanes

24 CPW X O 5 Slotline XW////////S////.B/// g 0 f W/////~~~~/////// HF sf IF PS~P 4100 LO0 _ON —~~~ ""LO/IF isolation _ — ~~ /xo~verlay capacitors 74\\\\\\\\\\\\\SS 0 l 1650 ~H / ~/ _'and~~~~~~~~~~~~~~~~~ol IF ]I ~IF Figure 2.6: Illustration of the monopulse array layout and LO distribution scheme. Dimensions are in micrometers. Detailed dimensions of the transitions, and an equivalent circuit diagram are shown in Figure 2.7.

+ VIF-V 1230 0 \I\ F 86 LO-IF 510 solation'Overlay WCap.\ >(6pF)> LO 7 LO \2 1 ( \\\> 210 0 in70 945 Figure2.7: The - transition andpwerslittrciruit.aso120 Polyamide 1120 cr o600 e 6pF 85' 6pF Antna/,Slotline Slotline Antenna/ Zs=900 Zs=900 Antenna/ Mixer 62' 62' Mixer Zc= 6001 LO in 65' the outline of insulating polyimide patches. All dimensions are in micrometers.

cent. Within the dielectric, higher-order propagating modes can exist, and couple to the evanescent modes in the air-filled waveguide. The impedance of the evanescent waveguide modes on either side of the dielectric layer, Za, is of course, purely imaginary. The impedance of the propagating mode in the dielectric is Zd. Za is transformed to the impedance Z+ through the dielectric section through simple transmission line analysis, illustrated in Figure 2.9. Thus the resonance condition resulting in the curves shown in Figure 2.8 can be written as Za+Z+ = O (2.1) Full-wave analysis of the dielectric within the horn cavity is needed to obtain the antenna input impedance. At the time this work was performed, such analysis had not been yef completed, so antenna impedances were found by constructing a 2.7GHz microwave scale model (corresponding to a second subharmonic LO frequency of 675MHz) of the cavity structure. Stycast with e, = 12 was used to model the semiconductor substrate. A 125 mil-thick layer of stycast was used to model the O.lAd substrate. Strips of copper tape attached to the stycast are used to construct the dipole antenna and the feeding CPS transmission line. Since the dipole radiates preferentially into the substrate, this suggests that the dipole should be placed facing the apex to allow radiation to propagate toward the horn aperture. This is verified by measurements which show that the antenna input impedance has a wider bandwidth in this configuration than when placing the dipole on the opposite face of the dielectric substrate. RF antenna impedances could also be adjusted over a fairly wide range of values, from l5fQ to 75Q by adjusting the width of the dipole as well as the distance of the dipole from the apex of the horn cavity, while the thickness of the dielectric

27 0.70.................,. 0.60 -oI ----..TEl. h- 0.50 L, -. —----- TE12 0.40 T30 0.30 f,, 0.20 0.10 0.00.... 0.90 0.95 1.00 1.05 1.10 Frequency f/fo Figure2.8: Calculated resonant modes within a dielectric of r = 12 in a square waveguide of cross-section of 0.53Ao [34]. eo Eo Xd-Yd d Za Zd,/ Za Z+ Figure 2.9: The transmission line model for analyzing the dielectric layer.

was held constant. The antenna impedance measurements were performed using an 0.086" o.d. semirigid coaxial cable fed through the apex of the horn and incident normal to the face of the substrate, where the coaxial conductors were attached to the copper strips of the antenna. It was found that this measurement configuration does not disturb antenna impedance measurements. The microwave scale model was constructed to include the LO distribution circuits (the CPW-CPS transition shown in Figure 2.7) and the two dipole antennas connected to the two outputs of the transition. The coaxial cable was attached to the apex of one of the dipoles, and the other antenna was terminated using a 25Q chip resistor to approximate the load impedance of the diode. A balun is used to allow the unbalanced coaxial line to drive the balanced dipole antenna and CPS transmission line. This balun is based on a quarter-wave choke (Fig. 2.10), and must be altered for measurements at the different LO harmonics. The LO distribution circuit had little effect on the antenna impedance at 2.7GHz due to the RF choke capacitor (Fig. 2.5), which was modeled as a 5pF capacitor. As the frequency of the LO harmonic decreases, the influence of the LO distribution circuit on diode embedding impedances progressively increases. One factor which simplifies the measurement is that the antenna is very reactive at frequencies other than the design frequency of the antenna (2.7GHz for the microwave scale model, corresponding to 94GHz in the actual receiver). This means that the antenna does not radiate into the horn cavity very much at the lower frequencies, and thus the horn need not be included in the impedance measurements at 675MHz, 1.4GHz, and 2GHz (f = fLo,2 x fw,and 3 x fLo). The capacitance of the LO-IF isolation capacitor results in a susceptance of less than jlQ at 2.7GHz, and can be safely neglected in the microwave modeling. The antenna and mixer circuits are then

29 Stycast 2.8GHz (125mil) / Microwave Model Measurements LO Circuits Figure 2.10: Coaxial quarter-wave balun optimized using the procedures described in the next section. 2.5 Antenna and Mixer Optimization The mixer design procedure consists of obtaining preliminary antenna embedding impedances at the LO harmonics including the RF, and employing these impedances in the HBA code with a commercially available device. The code utilizes LO drive level, embedding impedances at several harmonics of the LO, and the parameters of the mixer diode in order to calculate the impedance of the diode pair under LO excitation. The diode parameters include series resistance, junction capacitance, ideality factor, parasitic capacitance and inductance, and anode area. These values are usually specified by the diode manufacturers, and are critical to the performance of a design. There exists a surprising dearth of commercially available antiparallel diodes for use at millimeter-wave frequencies. In fact, the diode pair with the highest operating

frequency, the MA/COM 40422, was originally designed for operation as a K-band mixer. The reason for the scarcity of planar millimeter-wave antiparallel diodes is the lack of commercial demand rather than limited diode technology. This is demonstrated, for example, by the planar antiparallel Schottky diodes fabricated by research institutes such as the University of Virginia, which have been used efficiently at frequencies up to 205GHz and higher [15]. The parameters for the 40422 diode were used in conjunction with mixer and antenna modeling in order to determine what range of conversion losses could be achieved. The diode series resistance is a very important diode parameters in determining the performance limits of a second-subharmonic mixer: a change by one ohm can mean a change by several dB in the conversion loss of the mixer. The junction and parasitic capacitances are crucial in determining the optimum LO embedding impedances, but do not intrinsically limit the mixer performance as long as good LO and RF impedance matching can be obtained. For example, a change in capacitance from 50fF to 75fF has a large effect on the optimum LO embedding impedance, but have much less impact on the optimum performance than a change in series resistance from 2Q to 3Q, which has an exponential effect on the conversion loss (Fig. 2.11). The increase in series resistance also makes the conversion loss less sensitive to variations in the embedding impedances. Another parameter which has a large impact on subharmonic mixing is the parasitic inductance, L,. This inductance has the effect of reducing the effects of the junction and parasitic capacitances, thereby improving the RF match to the diode. Oftentimes, diode manufacturers will specify a value for L8 based on low-frequency measurements, with an uncertainty of 50%. Since the mixer design depends greatly on knowing the value of L8, such uncertainty increases the difficulties a designer faces. Thus, even before commencing a system

31 Rs C 4j n Lp (Pbbi I X 3Q 60fF 1.1 0.1nH 40fF 0.75V 1 x 10A-' Table 2.1: Diode parameters used in mixer design. The values lie in the center of the manufacturer-published ranges for each parameter. m14 > 10 - L, O 2 4 6 8 10 Series Resistance (Ohms) Figure2.11: Conversion loss as a function of series resistance, for a constant set of LO embedding impedances, and 6mW LO power. atic and careful design, the diode must be independently measured in a test fixture or circuit. Unfortunately, MA/COM40422 diodes were not readily available at the time the mixer was designed, so parameters specified by the manufacturer were used (Table 2.1). In order to optimize the design, initial values for the embedding impedances based on the first results of microwave modeling are used with the code in order to determine a ball-park figure for the RF and LO impedances of the diode. The embedding impedances are then adjusted in order to improve the RF and LO matching, and the harmonic balance analysis is repeated to see how the diode impedances and con

LO 2xLO 3xLO RF 15 + j21Q 72 - j215Q 20 - j50Q 16Q Table 2.2: Mixer embedding impedances at each LO harmonic version loss are affected. This process is iterated in order to optimize for conversion loss. In general, the embedding impedances at the fundamental, second and fourth harmonics are the most critical in determining mixer performance, though the code includes the effects of embedding impedances up to the 8th LO harmonic. For a mixer with an LO at the second subharmonic, it is desirable that the impedances at 2x LO and 3x LO be terminated reactively since any noise or signals present at these frequencies should be rejected by the mixer. The second subharmonic mixer performance is affected most by the LO impedance (which determines how well matched the diode is to the LO source) and the impedance at the 2nd LO harmonic (which behaves like an idler frequency in a multiplier). The RF impedance primarily influences the RF mismatch loss, and has a straightforward effect on the total conversion loss. Optimizing mixer performance as a function of three impedances plus the LO drive level seems to be a very difficult task, one which is made much simpler Ylv limits on the range of obtainable impedances by the planar antenna-mixer structure. Furthermore, the HBA code can be automated to simulate the mixer performance for a wide range of embedding impedances and power levels. Figure 2.12 shows the conversion loss (including RF mismatch) of the design process for different LO embedding impedances and power levels. The antenna/mixer must be designed such that;t present an LO impedance which yields good intrinsic conversion loss and minimiz itF mismatch. Since the 40422 diode junction capac tances are large, it is not surprising that a low RF impedance is needed to provide a

33 2dBm 4dBm 10 20 30 40 50 10 20 30 40 50 50 50 50 50 40 40 40 -1 40 30 30 30 30 20 20 0 20 102 10 10 10 0 0 0 10 20 30 40 50 10 20 30 40 50 6dBm 8dBm 10 20 30 40 50 10 20 30 40 50 50 50 50 40 40 40 40 30 30 30 30 20 20 20 20 10 10 10 10 20 0 0 -10 -10 0- -10 10 20 30 40 50 10 20 30 40 50 Figure 2.12: Contour plot of conversion loss including RF mismatch, as a function of the LO (23GHz) embedding impedances for various power levels. The X-axis and Y-axis denote the real and imaginary parts of the LO aembedding impedance, respectively. Power level plays a crucial role in optimal mixer performance. The other embedding impedances are held constant, at the values listed in Table 2.2

34 6dBm, Rs=3 Ohms 6dBm, Rs=6 Ohms 10 20 30 40 50 10 20 30 40 50 40 40 40 40 30 - 30 30 i 30 20 - 20 20 20 0 - 0 0 -10 -10 -10 -10 10 20 30 40 50 10 20 30 40 50 Figure 2.13: Contour plot of conversion loss including RF mismatch, as a function of the LO (23GHz) embedding impedances for 6dBm LO power. R, = 3Q (left) and R, = 6Q (right). Diode series resistance severely degrades second-subharmonic mixing performance. good match to the diode, and thus the dipole dimensions are fairly wide and short. Even so, because the diode parasitics result in an RF impedance of about 6Q, a RF mismatch of about 2dB is difficult to avoid. The embedding impedances at the dipole apex obtained from the measurement of the microwave model are plotted on a Smith chart (Fig. 2.14). The resonant antenna impedance is 16Q at RF (4 x LO) and the antenna/mixer shows a comfortable 10% bandwidth. Once the antenna design was optimized, the input impedance ZLO at point A marked on Figure 2.7, was measured by connecting- a SMA female bulkhead connector directly to the CPW of the microwave scale model (Fig. 2.15). The diodes have a calculated LO impedance of about 30 - jlOOmega, and were approximated by 25Q resistors at the apex of each dipole. ZLO is combined in parallel with an identical impedance (corresponding to the other antenna/mixer pair), and the resulting

35 4x/'' \// Figure2.14: Antenna/mixer embedding impedances obtained from a 2.7GHz microwave model. The 10% bandwidths at each LO harmonic are plotted. impedance is matched to the output of the LO-producing circuits which drive the mixers (described in Chapter III). It is important to note that the coaxial-to-CPW transition transforms the 50Q coaxial transmission line to a 70Q CPW transmission line t35]. Thus any S-parameters measured at the coaxial-to-CPW transition must be converted from a 70Q2 characteristic impedance Smith chart to a 50Q Smith chart. The estimated LO power needed to drive each mixer is between 5 and 8mW: Figure 2.12 shows that as the LO power rises, low conversion loss can be obtained with a wider range of LO embedding impedances. Thus, between 20 and 30mW of LO power is needed to drive the four subharmonic mixers. In Chapter 3, this figure is used with measurements of circuit element losses to estimate the required performance of the LO-producing circuits. From the contour plot, the predicted conversion loss of this design and diode characteristics, including RF mismatch, is about 13dB. The design of the LO circuits, and the fabrication and measured results

36 ZLO T h, LO f n ca al be me —d 4at 67tMHz T Starto 40MHzt',-'/'/ feed to the CPW-CPS transition marked on Figure 2.7 of the entire receiver are discussed in the Chapter 3. This could improve the efficiency of the mixer as well as result in a more compact design since the transmission line lengths would be reduced. There are two reasons the design in this thesis instead utilizes a LO frequency of 23GHz. First, the wafer probing capabilities of the available facilities have an upper frequency limit of 40GHz. Secondly, hybrid mounting of transistors becomes extremely difficult at frequencies above 30GHz, due to the increased parasitics associated with bond wires and packages. A simple method for estimating the bond wire inductance is included in Appendix E. Nevertheless, it is worthwhile to consider the potential performance of a receiver using an LO at 46GHz, since MMIC fabrication of such a receiver can realized using current fabrication technology. Harmonic balance analysis of an unoptimized mixer design with an LO of 47GHz using commercially available diodes

shows that a SSB conversion loss of 7dB is easily obtainable. This differs by only 3dB compared to fundamental mixer performance [30], and does not degrade system performance significantly since a 3dB (50%) degradation in conversion loss only translates to a decrease of 0.8dB (20%) in radar range. 2.6 Receiver Fabrication The receiver is based on hybrid-mounted, inexpensive commercial components, allowing the chip to be fabricated on silicon. First, the active wafer is etched using a two-step process in order to produced rectangular etched areas of 90pm-thin substrate, on which the mixer circuits and antennas are to be located, while simultaneously producing the via-trenches and supporting silicon beams depicted in Figure 2.4 on page 20 of Chapter II. This etching process is illustrate in Figure 2.16. The membrane layers described in Appendix B are conveniently suited for allowing wafers to be etched to differing degrees on different areas of the wafer, using a single masking step, thus avoiding the difficulty of processing on a non-planar wafer. First, rectangular holes are define on the back side of the wafer, where the antennas are to be located, and the top layer of silicon oxide and silicon nitride is removed in these areas (Fig. 2.16). Next, a second mask is used to etch through the remaining oxide layer in the areas where the via-trenches are to be located, and the wafer is etched anisotropically using a solution of ethylene diamine pyrocatechol (EDP) or potassium hydroxide (KOH). (See Appendix B for detailed information on the etching solutions.) The anisotropic etching naturally stops where the 700 etching planes intersect. The remaining silicon oxide is removed using buffered hydroflouric acid (BHF), and the wafer is etched until the via trenches have completely etched through the wafer. The thickness of the thinned silicon is monitored with a microscope until

the desired thickness is obtained. The masks are designed to account for undercuting which occurs in the protective peninsulas of nitride and oxide which are essential to the formation of the supporting silicon beams. An undercut rate of 200mm per hour for the silicon material located at the corners is typical. Thus, in two or three hours, the protective membrane peninsula is completely undercut, leaving only a protruding ridge of silicon which has an observed etch rate 1.7 times more rapidly than a flat silicon surface. Once the etching process is complete, metallization and insulating layers are placed on the active center wafer, to which additional etched wafer pieces (sidewall wafers) are later attached to form the horn cavity. The fabricated circuit is first DCtested to ensure that the overlay (metal-insulator-metal, or MIS) capacitors are not shorted out. The insulating used in the MIS capacitors is photosensitive polyimide (HTR3-50 [38]). It is processed using manufacturer-specified procedures, including resist spinning at 6000RPM and curing in an oven for 30 minutes and 4 hours at 1500 and 200~ respectively in laboratory atmosphere. This results in a thickness of 1.24m, with an estimated dielectric constant of 3.3. After processing, it was found that the insulating polyimide had a large number of pinholes which resulted in short-circuits in a number of the bias lines. The short circuits were easily removed by injecting several hundred milliamps of current through the bias lines to effectively burn out the short circuits. The polyimide also caused two major fabrication complications. First, wire bonding of the hybrid components to the circuit was found to be very difficult. Second, metal-on-metal contacts in the circuit were unreliable. Both complications were related to residue deposited by the polyimide process after development. In order to ensure the removal of this polyimide residue, it was necessary to first place the circuit in a plasma etcher to "descum" the surface. This was done using an

39 V - - - - - /' —- - - -------- r — / iS I 3N — Si3 N4 IR~~~~~~ / / / /////// i Figure 2.16: The two-step etching process. Dashed lines indicate the positions of the antenna and membrane. Arrows mark the direction of undercut.

oxygen plasma at 250 milliTorr with 100W of power for 2 minutes, and resulted inll a reduction in the thickness of the polyimide of approximately 200 A. Further cleaning in a mild wet-etching solution consisting of H2SO4 and H202 with a ratio of 1:5 was required for 30 seconds. These two cleaning procedures resolved the fabrication problems, but also made the polyimide more susceptible to pinholes. A side view of the assembled integrated horn antenna with dimensions for each sidewall wafer is shown in Figure 2.18. The sidewall wafer #4 also takes advantage of the two-step etching process, which permits a design which allow wafer #4 to be attached to the active wafer while avoiding severely affecting circuit performance. The etching process creates cavities approximately 300p/m deep in the silicon areas located above the CPW lines and hybrid-mounted elements (Fig. 2.17). The thin walls of silicon separating adjacent-cavities are removed using a scribe. The sidewall wafers were metallized with 4500A of evaporated gold and then attached to the circuit wafer to complete the horn cavities. The sidewall wafers were attached using thin doublesided tape which allowed the walls to be removed in order to observe their effects (,,r the mixing performance of the circuit. The wafer was then attached with epoxy to a printed circuit board fixture which contained low-noise amplifiers and biasing circuitry. In a GaAs MMIC design, the receiver circuits can be fabricated before etching the substrate beneath the antenna and mixer. This can be done because GaAs is much more easily etched that silicon, and can be masked with photoresist. The mask step determining the location of the thinned substrates for the antenna can then be aligned to the circuits using infrared back-side alignment. This procedure has a precision of better than 10tm, more than sufficient for integrated antenna designs at 94GHz. A number of wet etching solutions (such as H202: H2S04 1:1)

300m-deep etched cavity//////// 300,m-deep etched cavity Figure 2.17: Top view of the etched cavities for wafer #4. The thick line circumscribing the features indicates the edge of the wafer, which is diced or scribed.

42.- 4100 - --- 4100; —-- 31n300 g//X///g 2100 1360 /ACtive waferhigh(uncoated15Si; 2000.cm) 3 90 Figure 2.18: Sideview of receiver wafers. etch GaAs rapidly at low temperatures. This allows the via-holes to be separately defined using simple, front-side etching, thereby eliminating problems with undercut associated with the two-step silicon etching process.

CHAPTER III Tracking Receiver: Implementation and Measurements This chapter presents the implementation of the receiver. The LO generation circuit operates at the second silhbarmonic of 94GHz, which is approximately 23GHz. This allows the LU power to be produced and distributed using planar circuits with low losses. Furthermore, it permits the use of hybrid-mounted devices usinE conventional wire bonding, thereby simplifying the fabrication of the receiver. First, the design of the LO system is presented and includes the voltage-controlled oscillator (VCO) and the LO power amplifier circuits. The entire LO system is fabricated using uniplanar transmission lines, primarily coplanar waveguide (CPW) but also slotline and coplanar striplines (CPS). CPW is used instead of microstrip for several reasons. First, the ground plane for CPW lines is on the front side of the wafer, eliminating the need for precision via-holes for connecting the ground terminals of active devices. Also, the substrate need not be thinned in order for the transmission line to operate properly at high frequencies. CPW also allows simple on-wafer RF measurements of the circuits to be made. Finally, CPW circuits can easily transition to slotline and CPS transmission lines, which are used to feed the antennas and mixers in the receiver.

Receiver fabrication details are discussed, and losses in the LO circuit elements are measured in order to estimate the LO amplification requirements. Measurements of individual components in the LO system are presented, followed by those of the complete receiver. An IF monopulse processor design and its implementation are discussed, and the patterns produced by this processor are presented. Finally, improvements to the receiver design are suggested, and future work is discussed. 3.1 Oscillator Design The 23GHz LO power is produced using a VCO utilizing a commercially available HEMT, the NE32100 manufactured by NEC [39]. The transistor is hybrid-mounted in a uniplanar, CPW transmission line circuit and connected using 0.7mil gold bond wire. This results in a design which produces lower output power than a two-terminal device such as a Gunn or an IMPATT diode, but allows a simpler, easily fabricated planar circuit to be used. Two-terminal devices are transit-time devices which are inherently non-planar, and exhibit very low negative resistances [40]. This makes them difficult to incorporate in a planar design since planar circuit losses can be significant compared to the device negative resistance. The uniplanar HEMT oscillator is designed using the reflection amplifier technique presented by Boyles [41]. Figure 3.1 helps to illustrate this technique. The transistor is placed in a one-port network, which is connected to a load. The conditions for oscillation are as follows: 1l/Sll < Irl (3.1) arg(1/S11) = arg(r) (3.2) For a passive load, this means that S11 must be greater than 1. Note that when an inequality exists in equation 3.1, a situation exists where the magnitude of the

Reflection. Output Amplifier Matching Circuit Figure 3.1: The reflection amplifier signal between the reflection amplifier and the output matching circuit will increase indefinitely with each reflection. Clearly such a situation is not physically possible, and in an actual oscillator, the device gain begins to decrease (a phenomenon referred to as gain compression) when the output power exceeds a certain level. Eventually the gain decreases to the point where equality holds, and the magnitude of the oscillations reaches a stable point, given certain stability criteria discussed below. The reflection amplifier design technique not only gives an unambiguous condition for whether oscillation should occur, but also gives information about the potential stability of the design. The first stability criterion is illustrated by plotting 1/S11 and F on the Smith chart and using arrows to indicate the trajectories of each variable as frequency increases (Fig. 3.2). To ensure that oscillations can take place at only one frequency, the conditions specified in equations 3.1-3.2 must be satisfied only at one frequency. This means that, as frequency changes, the direction of change of 1/Sl1 should be opposite that of r, at the point where oscillation conditions are satisfied. The second stability crite ririon gives conditions which minimize the phase noise produced by amplitude modulation in the signal generated by the oscillator (referred to as "AM to PM noise"). When the voltage between the reflection amplifier and the

output match builds up to a certain point, the device gain saturates (resulting in gain compression) and equality in equation 3.1 holds. At this point, a small amount of AM jitter exists due to noise in the device, and this AM jitter modulates the phase of the oscillation condition, thereby producing phase noise. The condition which minimizes AM to PM noise is written below: a(1/S11) ar (33) av af This simply means that the trajectory of S11 as a function of signal amplitude must intersect normal to the direction of the path that F takes as frequency changes. Direct measurement of the transistor, or nonlinear models simulating the gain compression characteristics of the device can be used to ensure that the conditions specified by equation 3.3 are met (Fig. 3.3). The oscillator design in the receiver uses CPW transmission line tuning stubs attached to the source and gate of the transistor. Its circuit diagram and layout are shown on Fig. 3.4. The CPW transmission line dimensions used in the layout are calculated using a commercially available software (EEsof/Linecalc [42]), which calculates the impedances and propagation constants for CPW lines on a dielectric substrate with finite thickness. The relative dielectric constant and substrate thickness used is 11.7 and 360tm, respectively. The effective relative dielectric constant is determined using EEsof/Linecalc to be 6.25. The reflection coefficient looking into the drain of the transistor (Sil, marked on Fig. 3.4) is calculated using ideal transmission line models and the linear S-parameters of the transistor which are published on the device data sheets for several values of drain current, drain-to-source voltage, and gate- to-source voltage (IDS, VDS, and VGS). The two-port S- parameters of the transistor c.an be converted to an indefinite scattering matrix [43]. Transistor

47 Figure 3.2: Smith chart illustration of the oscillation conditions and the first stability criterion. Arrows indicate the direction of trajectory as frequency increases. Y lf / r\,/ \ \.. Tr, ectory \f --,'/ as a function of / signal amiitUe. \ Figure 3.3: The second stability criterion minimizing the phase noise produced by changesigure 3.2: Smith chart oscillustration of the oscillation condit ions and the points at whicstathe oscillationty conditerion is satisfied. Arrow indicate the direction of trajetory as frequency mincrimize AM-to-PM noises. minimize AM[-to-PM. noise.

S-parameters are usually given as a two-port matrix, which is measured with one of the device terminals (such as the source or emitter of a transistor) to ground. An indefinite scattering matrix is one for which the ground terminal is undefined. This effectively adds a port to the network by treating the normally ground terminal into a network port. Thus the transistor treated as an indefinite circuit yields a three-port matrix, which allows circuit elements (such as transmission lines, in the case of the VCO design) to be connected to the source. The lengths of the CPW shorted stubs connected to the gate and source are adjusted so that a large resonant peak in S11 at a single frequency is obtained. The frequency and magnitude of this resonance, and the direction of its trajectory on the Smith chart are adjusted across a wide range of frequencies by varying the impedance and lengths of the transmission line stubs. This can also be done electrically by attaching two varactors to the gate transmission line (Fig. 3.4), and is discussed on page 50. The S-parameters of the NE32100 measured by the manufacturer include two bond wire connections at each device terminal, with lengths of 325am for the gate, 240m for the drain and 178pm for the source. The manufacturer S-parameter measurements were made by mounting the chip HEMT in a 150/m-deep trench and bonding it to the microstrip lines of a planar circuit. Placing the HEMT in the trench allowed the bond pads to be closer to the transmission line. In the monopulse receiver, fabricating a trench in silicon is difficult, so the transistor is attached directly to the surface of the wafer. Care must be then taken to include the added bond wire inductance resulting from the additional lengths caused by surface-mounting the transistor, especially in the connections to the source terminal. An illustration of this is shown in Figure E.2 of Appendix E. Furthermore, a safe margin for error in the transmission line lengths connected to the source of the HEMT must be included.

49 S11 r I0 45' }Output Matching circuiLl S E LI G X 400 30, 400 400 40n 400 0 80, Oa, O.8pF 3cr 3TJ varactors CbT caT TCI Cb: overlay capacitory 500OOpF Output Silicon (Drain) r -----.- _,___ — - I Overlay BI | Capacitor [ ~~~~~30' i Pad'-iOutput Matchin I I —-- - - 5g g Polyamide! CircuitI Circuit 4~~~. Insulator r ~ ~40,um 15,mNE32100 40/m _ 150/m.t' [- *<T Zzt 80' (Source) - 3~~~~~~~~~~5' MSV- 34064- C 11 ~~~^ ~ ~ ~ ~~1 r I 0.6pF Varactor Varactor Gate Bias Varactor Bias Bias Figure 3.4: Circuit diagram and layout of the VCO design. The CPW transmission line sections all have an impedance of 40Q.

The need for this safety margin becomes apparent during simulation, because the resonance in 1/S11 can be made to pass through the origin of the Smith chart, causing a 1800 change in the phase of 1/S1l and possibly preventing the conditions in equation 3.1 from being satisfied (Fig. 3.6). Furthermore, empirical results from oscillators built at 4GHz and 11GHz show that the oscillator must be designed to operate at a frequency which is approximately 10% higher than the desired frequency of operation. Work has not been performed to determine the cause of this phenomenon (which is referred to here as "frequency compression"), but intuition suggests that it is directly related to the gain compression which takes place when the oscillator signal reaches a steady-state amplitude. Figure 3.5 shows the results of the design process, plotting both 1/S11 and r on the same Smith chart from 22GHz to 26GHz. An output matching section, illustrated in Fig. 3.4, is inserted between the reflection amplifier and a 50Q load in order to ensure that the conditions specified in equations 3.1-3.2, and the stability criterion illustrated in Figure 3.2 are met. Although the output match is designed to drive a 50Q load, the oscillation conditions and first stability criterion are satisfied for load impedances ranging from 20Q to 70Q. The unavailability of a nonlinear model for the HEMT, in conjunction with time limitations, prevented a full simulation of the design in order test the second stability condition which minimizes the AM/PM noise of the oscillator. Also gain compression behavior can be included in a future design which utilizes an approximate nonlinear model of the transistor, with harmonic balance analysis available in commercial software such as Touchstone/Libra, by EEsof [42]. Two varactors are attached to the CPW line connected to the gate of the transistor and allow the frequency of oscillation to be adjusted. The varactors, the MSV-34064-C11 chip varactors manufactured by Metelics [44], have a Cjo of 0.6pF

51 \ \ \a Start: 123GHz I Stop: /25.5GHz Figure 3.5: 1CS,, and output match S-parameters of the VCO design. Arrows indiof the two gate varactors in parallel. A\, \ Start: /23GHz Stop: 25.5GHz fo sc= 2~,. 7-3'GH,, / /' -' — ".:'/

and a minimum reverse-bias junction capacitance of about 0.2pF. Two varactors are used, one on each ground plane of the transmission line to maintain the symmetry of the CPW line to avoid exciting the even transmission line mode. The bond wire used to attach the varactors to the gate CPW each have an estimated length of 350ym, with a calculated inductance of O.llnH. This is then used with the small-signal analysis to predict an operating frequency range from 24.9GHz to 25.3GHz. When the 10% reduction due to frequency compression is taken into account, this translates to an operating range of 22.4GHz to 22.8GHz. In order to bias the gate and varactors, the transmission line connected to the gate of the HEMT is terminated in an overlay bypass capacitor, which acts as a shortcircuit at 23GHz. Experimental evidence indicates that at least 500pF of overlay capacitance is needed in order to suppress any low-frequency oscillations. Each overlay capacitor consists of a gold pad (forming one terminal of the capacitor, which is the ground plane in this terminal) on which an insulating polyimide is deposited (discussed on page 37 in Section 2.6). A second gold pad is evaporated on the insulator and constitutes the second capacitor terminal. The insulating polyimide has a thickness of 1.2ym, determined from measuring the surface of the chip using a DekTak surface profiler [45]. Additional 0lpF chip capacitors are attached to the gate to protect it from static discharge. Measurements on a prototype VCO based on the NE32100 indicate that the circuit produced stable oscillation at 23.3GHz with an output power of 0.5-lmW. The output spectrum produced by the VCO is shown in Figure 3.7, and exhibits a phase noise which is norr- il for a free-running oscillator circuit consisting of planar transmission lines [46]. ie VCO can be phase-locked in order to obtain better performance. The oscillator can be used in the receiver even without phase-locking,

53 ATTEN 10 OdB MKR -27.00dBm RL OdBm 10 OdB/ 23.3326GHz MKR -27.0 dBm CENTER 23.3326GHz SPAN 10.OMHz *RBW 1 00kHz *VBW 30kHz SWP 90ms Figure 3.7: 23GHz VCO output spectrum. since the phase noise is minuscule (even when multiplied by a factor of four due to subharmonic mixer operation) compared to the wide bandwidth of available IF processing elements. The tuning data of the VCO is shown in Figure 3.8 as a function of varactor bias. The power at each point was not optimized by changing the transistor bias, and is within 3dB of 1mW. The measurements indicate that the output frequency is linear with varactor bias with a tuning range of 250MHz. This is a bit smaller than the 400MHz tuning range predicted by small-signal analysis. 3.2 LO Amplifier Circuit Design and Measurement Based on the analysis presented in Section 2.5 of Chapter II, the four mixers located at the apexes of the four antennas require a total LO power level of greater than 20mW (13dBm), and more is needed if circuit losses are taken into account. Therefore the output of the VCO must be amplified. First, an estimate for circuit

54 23.5 ITI. II i I, O0.0 23.4 - S 0 > Xe/ -- Relative Power U 23.3 -3.0 t 23.2:...Frequency 23.1 L5.... I' -6.0 231 5 -10 -5 0 Varactor Bias (V) Figure 3.8: 23GHz VCO tuning data. losses must be obtained in order to estimate the requirements of the LO amplifier chain. This is most easily done by utilizing on-wafer probing techniques and a calibration technique known as "through-reflect-line," or TRL calibration. A description of how the TRL works is found in [47] and [48]. The TRL calibration techniques utilizes calibration standards which consist of three well-characterized, simple planar transmission line elements. This allows the effects of transitions from the coaxial lines of the instrument to the planar circuits to be de-embedded. The TRL calibration technique results in a calibration which spans three frequency octaves. Thus a calibration is said to have a center frequency. The TRL standards consist of a 50Q transmission line 1A long at the center frequency, a highly reflective (but not necessarily ideal) -J1, and a delay line 4A long at the center frequency. Measurements (described in 148]) are then performed on these standards (analogous to a standard coaxial network analyzer calibration), and error corrections are calculated in order

to remove the discontinuities of the transitions to the planar circuit. The test circuits were fabricated using 0.45,um-thick gold evaporated on a 360tmthick silicon substrate with resistivity of 2000 - 3000Q-cm and a relative dielectric constant of 11.7. The dimensions of the CPW lines were determined using Touchstone/Linecalc and included the finite thickness of the substrate (360pm). This resulted in CPW line dimensions marked on Figure 3.9. On-wafer probing utilizing TRL calibration routines with a center frequency of 23GHz, and assuming a guided wavelength of 5200[ym, was used to make the measurements described here. The probe tips used in the measurement have a pitch of 250t1m (distance of center probe contact (signal) to the outer probe contacts (ground)) and are manufactured by Cascade [49]. These were utilized with a probe station manufactured by Alessi [50] and connected via coaxial cables to a HP8510B vector network analyzer with measurement capabilities to 40GHz. Measurements were found to be extremely repeatable, with a signal-to-noise ratio greater than 40dB. The dimensions of the through, reflect, and delay lines are shown in Figure 3.9. On-wafer measurements were performed on a hybrid-mounted DC-isolation beamlead capacitor and 50Q sections of CPW transmission-line. The capacitors (Metelics MBC50-O10B12 [44]) are needed to allow separate biasing of the oscillator and amplifier stages. Losses in the blocking capacitor were measured to be about 0.7dB. This included the return loss from a slight mismatch between the capacitor and the 50Q transmission line, due to the width of the capacitor leads. Taking this mismatch into account, the loss in the beam lead capacitor is about 0.4dB. In the actual receiver design, the capacitors are used in 40Q transmission line circuits to eliminate this mismatch loss. The 50Q CPW transmission line losses from 20-26GHz were measured to be about 0.5dB per guided wavelength (5200Aum) (Fig. 3.10), or about

56 Through (5800,) Delay (7100u) Reflect(2900,L) Capacitor test line 418O m 220 40 350 g r360,un 200 50 400 140 80 500! 60 125 600 T Figure 3.9: Dimensions of the lines used the TRL calibration set and capacitor test fixture. 1dB per centimeter. Measurements performed at 23GHz by other researchers found a comparable loss of 1.3dB for a 50Q CPW lines with w = 75im and g = 56[m. fabricated on a GaAs substrate with 0.5pm gold [51]. Assuming a three-stage amplifier design, the estimate for the circuit losses in the system is 1.5dB for the three DC-blocking capacitors, and 4.5dB losses in the planar transmission lines. This results in a total estimated loss of 6dB. Therefore the oscillator (which delivers about 1mW, or OdBm at 23GHz) requires an amplifier with a gain of 20dB to deliver 5-6mW of available LO power to each subharmonic mixer. The LO amplification circuits are designed around a single-transistor amplifier whose circuit diagram and CPW layout are shown in Figure 3.11. The transistor is

57 0.0 I I T I I T I I I A I T T T I i'-A A LA -0.2 -AAA1 1/4X transmission line -0.4 0 -0.6 - -0.8. [ -l300 10pF beam lead 1. 2 0 ~ 22 24 26 Frequency (GHz) Figure 3.10: Measured losses for CPW line and DC-blocking capacitors a K-band medium-power GaAs MESFET (Fujitsu FLRO16XV [53]) which is hybrid mounted to the circuit using 0.7mil gold bond wires. A quarter-wave bias choke with a center frequency of 23GHz is used throughout the LO amplifier circuit to provide gate and drain DC biases for the transistors without affecting its RF performance. The equivalent circuit for the choke is shown in Figure 3.11, and results in an impedance ZCh in parallel with the input transmission line of the amplifier. Calculations show that the value of Zch exceeds 200Q at frequencies of 20-26GHz, and on-wafer measurements show that the choke has a negligible effect on the RF performance of the circuit at these frequencies. The 6pF capacitor improves the bandwidth of the choke, and consists of two overlay capacitors (similar to those described on page 52) lying on either side of the CPW. Wire-bond crossovers (marked on the layout) equalize the CPW ground planes, and are essential in suppressing the even modes excited by the T-junctions, which form the input and output matching circuits.

58 Port 2 i q Drain bypass 04 capacitance 80,m 65s (102cimn) 80gm Amplifier Layout - FLR16XV Gate bypass R 201 (470Am) -150m 140,m capacice Bias choke' |6)B 90' 90' 70 Tuo0'3 Circuit. | Port I Fgr (1301: yut o aor a sT-Junct ion c Silicon ---- Polyimide outline E Overlay capacitor 2 -I~ I I I I~ eOutput and Amplifier Input 4Z 4Z Bias Equivalent and 35' 354 Choke Circuit Choke DC Bias Bias Choke Open To Equivalent Stub Ampfier Circuit. Zch Figure3.11: Layout of a single-transistor amplifier shown with T-junction corrections, and the equivalent circuit.

On-wafer measurements of the single-transistor amplifier yielded very good agreement with small-signal simulations (Fig. 3.12), for the bias conditions specified oni the manufacturer S-parameter data sheets. The S-parameter data for the transistor includes the parasitic inductance for 250m 0.7mil gold bond wires attached to the gate and drain of the FET. The source of the transistor is connected to the goldplated back-side of the GaAs chip, which is silver-epoxied to the CPW ground plane. Since the height of the FLR016XV is only 60/1m, a bond wire length of 250,um was a good estimate for the actual bond wire lengths used in the receiver fabrication and thus no additional bond wire inductance was added during the small-signal simulation. For the amplifier S-parameter measurements shown, VDS is set at 7V with VGS set to zero, resulting in IDS = 60mA. The S-parameters of the FET vary widely depending on the drain bias voltage, allowing the first stage to be biased for gain, and later stages to be biased for power. The close agreement between measurement and theory in Figure 3.12 are obtained after small adjustments have been made to the transmission line lengths used in the input and output matching networks in the small-signal calculations. The lengths (measured from the center of the T-junction to the stub end) of the CPW matching stubs were adjusted empirically to account for the capacitive end effects of the stub and reduced path length in the bends of the T-junction (Fig. 3.11). They also include corrections for the curvature of the lines at the T-junction, which have a radius of 80[m (about 0.015Ag). The curvature was added to reduce the effects of higher-order modes on the T-junction's behavior since very short transmission line sections connect the GaAs FET. Research by Simons et al. has been performed to experimentally characterize discontinuities in CPW circuits of specific geometries [54, 55]. However, a general method for quickly analyzing a wide number of CPW

VUU09 saI'A0- SA'L saA as pasulq zIos.s -uua~ AX910O.d aq~ uo paseq lag.[lduI~ aes-aiu.rs ~e Io} sla~aturud-S:ZI'~ an.,_['.. / X/ I' /:, --,,,.. \ IS Zsz do " z~OOa:l1ulSJ //',! \ / t x / / — ~/\ I~~~~' gg \,.. ~' ~', <',, // 09

geometries is only now being developed [56]. Characterization of the T-junction is performed empirically, resulting in corrections which include +100 added to the electrical length of the stubs, and a +60 length added to the electrical length of the lines feeding the junction, when determining the distance from the center of the Tjunction to the end of the lines. These corrections compensate for the reduced path length of the bends in the T-junction. A model for curved coplanar waveguide bends can be found in work recently developed by Soghomonian et al. [57]. The single-transistor amplifier is used in a three-stage amplifier chain in order to obtain adequate LO gain (Fig. 3.13). The chain begins with a balanced amplifier, which results in better than 10dB return loss throughout the 21-25GHz range using small-signal simulations (Fig. 3.14). The balanced amplifier approach also ensures that the entire amplifier chain is unconditionally stable. Sections of 40Q transmission line between amplifier stages are used to match each stage, and to allow the hybridmounted DC blocking beam-lead capacitors to be inserted. The input and output 3dB coupler design is a standard 900 hybrid 3-dB coupler design [58]. The balanced amplifier design is tolerant to impedance and transmission line length errors in the hybrid coupler, and extensive modeling of the coupler was not performed. The balanced amplifier performance is also tolerant to mismatches at the termination of the isolation ports in the 90~ coupler, and maintains a return loss which is less than -10dB for a range of terminations, from 30Q to 70Q. Two 1 x 1mm 10OQ chip resistors [59] were used in parallel to terminate the isolated coupler ports. The FLR016XV has a 1dB compression point of 20dBm (100mW) at 18GHz, which is graphically extrapolated from published data to about 18dBm (60mW) at 23GHz. This means that a single device is individually able to supply the required power to the four subharmonic mixers. Therefore two additional stages consisting

62 From VCO Input (Port 1 ) 3dB 90' Isolated Port 3dB 90' Hybrid r 400 (50n Load) Coupler 90' 500 Gate Bias #1 90, Balanced Amp ~ ~ ~~40fi DC Blocking las #2 ia Capacitors 170' Balanced i Ampliftier 1~~~~~~~~~~~~~~~ Balanced Amplifier i Silicon Second Metal Layer - |First Metal Layer —,,7 =:!~~~~~ MDrain Bias Stage 2 Isolated Port DC Blocking' (500 Load) Capacitor jBfiB Da is Load) ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~~~~~~~Drain Bias and Output / s il~~~~~~~~~l ~~(to Mixers) Drain Bias for (to Mixers) Balanced Amplifier'3 _ | |Stage 2 Stage 3 4ate 400 40 [ Bias stage 2 100' 15' Gate Bias Stage 3 Figure 3.13: 23GHz LO 3-stage amplifier circuit layout.

63 25,, / " L I T' I I I T 15 IS111 5 - - - IS221z --- IS221-C IS2112 m5 - \ J / \ / x/ - -2;s'o 22"""-""7I,,.........II _2..D22 24 26 Frequency(GHz) Figure 3.14: The modeled small-signal performance of the 3-stage LO amplifier. of single-transistor amplifiers are used to boost the gain of the design. For simplicity, each stage is identical to the one illustrated in Figure 3.11. Each stage is then connected to 40f sections of CPW transmission line, which allow DC blocking capacitors to be attached while simultaneously acting to impedance match each stage. The LO amplifier chain is designed to maximize the frequency range in which the gain is above 20dB. This criterion was traded off with obtaining optimum gain at a particular frequency, so that the receiver could be used across a wider range of LO frequencies. The small-signal analysis for the three-stage amplifier (Fig. 3.14) exhibits a gain of greater than 20dB from 21GHz to 24GHz. The output of the final stage is then fed to the subharmonic mixers through the CPW T-junction and the CPW-to-slotline transitions (described on page 34 of Chapter II). Figure 3.15 depicts the different wafers which make up the receiver's construction, and the fabricated active wafer is shown in Figure 3.16. The NE32100 oscillator HEMT is attached to the wafer using a five-minute insulating epoxy [60].

The other hybrid-mounted components, including the beam-lead capacitors and MA/COM40422 mixer diodes are bonded to the wafer using silver epoxy [61]. This silver epoxy has been used with no measureable performance degradation, in receiver designs operating at frequencies as high as 340GHz [21]. The devices were handled using modified vacuum tweezers and conventional mechanical tweezers. 3.3 Diode Characterization In order to fully characterize the receiver performance, the diode parameters were first measured for each of the mixers. The mixers are numbered as indicated on Figure 3.16. The series resistance, ideality factor and built-in potential are easily obtained via DC I-V measurements, and are extracted using a curve-fitting technique [62]. An accurate measurement of the junction and parasitic capacitances of the diode pair require more difficult RF measurements to be performed. The antiparallel diode DC-IV characteristics were measured for each diode pair in forward and reverse-bias. The DC measurements for each mixer diode pair (Table 3.1) show that the diodes are well-matched, and the series resistance is several ohms higher than the value of 2-4Q specified by the manufacturer and used during the subharmonic mixer design. Next, RF measurements were performed on the diode to verify the manufacturer's specification for the junction and parasitic capacitances (Cjo and Cp) of the diode pair. These parameters can be estimated accurately at millimeter-wave frequencies by using the diode with a well-characterized antenna whose directivity, RF impedance and losses were determined separately in previous measurements, and measuring the diode's video responsivity at RF frequencies as a function of diode bias current. The measurements are then compared with calculated responsivities based on a diode model which includes Cjo and Cp. This procedure was applied to the MA/COM 40422

crq CD' CA OC CD ) C fC IF, O~~~~~O 0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~6 00( CA p CD Y ( C C LQF''

66 23GHz VCQ| Figue 394 1Integrated Monopulse mon- oucking Receiver gz LO 2x2 94GHz npiifier 1 Integra e -rae ~ hain | Subh txers IF #2 v i 5m II IF [ I A} Figure3.16' The fabricated active wafer of the 94GHz planar subharmonic monopulse receiver.

67 Channel 1 2 3 4 rs (Q) 7,7 6,6 8,9 6,6 n 1.11,1.08 1.09,1.10 1.05,1.09 1.08,1.05 &bi (V) 0.76, 0.75 0.77,0.69 0.74,0.68 0.75,0.77 Is (fA) 14, 15 12, 32 60,50 17,10 Table 3.1: DC diode parameters extracted using I-V curve fitting, listed in pairs for each diode. antiparallel diode pair at 94GHz by forward-biasing one of the diodes. The diode's subharmonic performance was then measured, using the same antenna configuration. A diagram of the experimental setup is shown in Figure 3.17. The antenna used in the experiments is a planar, self complementary log-periodic antenna mounted on a extended hemispherical lens with a 13.6mm diameter, machined from high-resistivity silicon (10,0OOOQ-cm). The log-periodic antenna is located 2200ym behind the center of the hemispherical portion of the lens (Fig. 3.18). This antenna structure has been extensively measured by Kormanyos et al. [15] and used to accurately characterize diode parameters at frequencies above 200GHz. The log-periodic antenna has a = 0.707 (the logarithmic periodicity of the antenna teeth) and 7 = 0.5 (self-complimentary), with a minimum tooth radius of 601um and a maximum tooth radius of 2mm. This makes the antenna very wide-band and periodic at each frequency octave and ensures nominal operation from 26GHz up to 300GHz. The antenna impedance is 741Q, which is determined by dividing the antenna impedance in free space by the mean of the dielectric constants of air and sillcon:

68 -w he 2.22mm 13.7nun as RF power i (a) Log-periodic Antenna with MA/COM 40422 Antiparallel Schottky Diodes Absorber - e.22nm ~ X) 13.7mm _ - LO power LO+RF 1or Beamsplitter RF Poweri (b) Figure 3.17: Experimental setup for RF diode measurements. Video detection (a), and subharmonic mixer measurements (b).

69 /;~~ Dc/IF diode pair 3.3mm Figure 3.18: Dimensions of the planar self-complimentary log-periodic antenna and substrate lens. 189F Zant = (3.4) where 189Q is the impedance of any self-complementary antenna in free space [63]. At 94GHz, the measured directivity D of the antenna is 114 (20.5dB), determined from a full two-dimensional pattern measurement of the front and back radiation patterns, including cross-polarization patterns. The resulting effective aperture is given by A, = -AD = 93mm2. This assumes that the antenna is lossless. However, there are two components of loss due to the lens: absorption loss (eabs), and the mismatch loss at the air-dielectric interface (,ref). These are accurately estimated to be to 0.4dB and 1.5dB, respectively (see [15] for more details). This reduces the effective aperture to A,, = eabsrefAe = 59mm2. The power available at the logperiodic antenna terminals is then related to the incident power density S by the equation Pan,,t = S A,. The physical aperture of the lens is Aphys = 147mm2, and therefore the aperture efficiency is % = A,,er/Aphys = 40%. The diode is mounted at the apex of the log- periodic, and its DC current-voltage curve is used to calculate the junction resistance Rj as a function of bias current. The estimated diode parameters

70 diode model /8R-a Pi, P n RF Circuit L —--------- - - - - - -- + (measuVLFr RDC ~CC LF Circuit {measured ] I fVLF (junction) Figure 3.19: Diode model used in video detection calculations. Pin, is defined as the incident RF power density (S) multiplied by the physical area of the aperture, Aphys - are then used to determine the video response as a function of bias current using the video detection equations described by Zah [62]. The system video responsivity, defined as the low-frequency voltage across a resistive load RDC divided by the total RF power incident on the antenna's physical (S- Aphys) is broken down into four factors (Figure 3.19): IC leR, =IRL (3.5) where 3 is thoe aperture efficiency i calculated above, and RF is the ratio of the power absorbed by the junction to the power available at the antenna terminals and is calculated below:'YRF I-R + 7.|2 (3.6)

The junction impedance Zj and diode impedance Zd are defined below: Zj -1(3.7) R7-j + jW(Cp + Cj) Zd - j + R, (3.8) and the intrinsic diode responsivity is [64] Rh, - 4nkT (3.9) Finally, the ratio of the video voltage at the junction to the voltage across the DC load is RDC?LF (3.10) 7L /=(RDC + Rs + Rj)2 + (Rs + Rj)2(RDCCDC2W7rfmod)2 where RDC and CDC are the DC load resistance and stray capacitance (resulting from coaxial cables, for example) in parallel with the diode, and fmod is the video modulation frequency (usually below 10KHz). Equation 3.5 is used to predict the detected video voltage as a function of bias current, when a known RF power density is incident on the lens. The predicted and measured video voltages are compared as a function of bias current, and the diode parasitic capacitance (and, to a lesser degree, the junction capacitance) is adjusted until the calculations match the measured data. This procedure was performed at 94GHz using the log-periodic antenna configuration described above, with the MA/COM 40422 antiparallel diode pair. A comparison between calculated and measured responsivity is shown in Figure 3.20. A junction capacitance of 75fF is used with a total parasitic capacitance C, + Cj2 of 85fF, where Cj2 is the reversebias junction capacitance of the second diode. Cj2 is calculated to be 50fF by the

72 040 EO / V- / 20 1 Measured 0 - - Calculated 8.1- 1 10 100 Bias Current (microamps) Figure 3.20: Calculated and measured video responsivities as a function of bias current for the following parasitic values: R = 6Q, Cjo - 75pF, Cp - 35fF, bi = 0.75, n=l.1, and I = 1.2 x 10-14. equation Cj = Cjo//l-vj/vb, [40]. The series resistance of the reverse-biased diode has not been included in this calculations. The calculated value for the parasitic capacitance (Cp) is thus found to be 35fF. Both Cjo and Cp are only slightly higher than manufacturer specifications. The cutoff frequency of the diode is defined as [30] 27r(Cjo + Cp)Rs (3.11) and gives a value of 250GHz for R, = 6M. A cutoff frequency ten times higher than the RF frequency is typically vised in mixer designs to obtain acceptable performance, and this condition clearly do not exist here. Once the diodes were characterized, the subharmonic mixer performance of the diode pair was then measured. The measurements were performed using a 20GHz LO

frequency with an 80.36GHz RF signal, resulting in an IF of 360MHz in a 4th subhlarmonic mixing mode. A 20GHz LO was used because high-power LO sources above 20GHz were not available at the time. Both the LO and RF power were injected quasioptically using the setup shown in Figure 3.17 (b). A imil-thick polyethylene sheet was used as a crude beamsplitter to combine most of the LO power with about 5% of the RF power. The diode equivalent model (determined using the video detection experiment described above) was used to calculate the diode video responsivity at 20GHz and 80.36GHz, resulting in values of 65mV/mW and 870mV/mW, respectively. These values were calculated for the MA/COM 40422 diode in the lens-mounted log-periodic antenna (a 74Q antenna), and were used to predict the available LO and RF power being coupled into the diode during the subharmonic mixing experiment. This method for predicting conversion loss and available LO power is accurate to about ~ldB because of uncertainties in the diode parameters. For example, a variation of ~1Q in a series resistance of 5Q, combined with a change of 10pF in the junction capacitance causes a 25% variation in the calculated responsivity of the diode at 20GHz. Additional uncertainties are introduced by inaccuracies in the DC curve fitting and because the series resistance is a function of frequency due to the skin effect. Therefore, the estimated value for series resistance is known to within ~1Q. The LO power was estimated by reducing the LO power level with a calibrated 20dB attenuator to ensure that the diode was operating in the linear region. A 1KHz square-wave amplitude-modulation was then applied to the LO power, and the 1KHz voltage detected by the diode was measured. This voltage was translated to power available at the antenna terminals using the 20GHz diode responsivity of 870mV/mW. A similar measurement was performed to determine the RF power

74 40,, \,, j i 35I~' ~~~- - HBA ml 30- \- Measurement 30 o 0 > 20 X-, 15 102 3 4 5 6 7 Available LO Power (mW) Figure 3.21: Subharmonic conversion loss at 80.36GHz using a 20GHz LO source, for a MA/COM 40422 antiparallel Schottky diode pair mounted in a self- complementary log-periodic antenna, as a function of available LO power. The diode parameters used in the calculations are the same ones listed in Figure 3.20. The conversion loss is defined on page 74 available at the antenna terminals. During subharmonic mixing, the LO attenuation was removed, and the resulting IF signal was fed into an HP 8562A spectrum analyzer. The mixer conversion loss as a function of the available LO power at the antenna terminals is shown on Figure 3.21, and is defined as the power at the IF port (50Q load) divided by the RF power available at the log-periodic antenna terminals (74Q source). The graph shows that the optimum conversion loss of 16.5dB occurred for an available LO power of 6mW, and matches the optimum power level predicted by harmonic balance analysis. The discrepancy of 2dB in the minimum conversion loss between measurement and theory is due to error in the calculated responsivity of the diode at 80GHz resulting from the uncertainty in the diode parameters.

3.4 Antenna and Mixer Measurements The directivity of the individual integrated horn antennas, with an aperture of 1.3A, was estimated by measuring their far-field patterns. This was done by forwardbiasing one of the diodes in the antiparallel pair and operating the receiver in video detection mode. The patterns are uniform for the four antennas, and exhibit symmetry in the E and H-planes (Fig. 3.22). As expected, the single-horn patterns closely match the patterns of for membrane-based integrated horns. The patterns result in a calculated directivity of 12dB based on previously measured full two-dimensional patterns [21], and agree well with a rough figure of 11.5dB given by the approximate expression for antenna gain published by Balanis [65]: 32400 Dmax E3 0 (3.12) where OE and OH are the half-power beamwidths of the horn antenna patterns. For a lossless antenna driving a matched load, a directivity of 12dB translates to an aperture efficiency of 70%, which corresponds to a 1.5dB loss when coupling to a plane wave (not a lens or reflector). An additional loss of about 1dB due to conductor losses in the horn sidewalls can be expected [66, 21]. An additional losses of at least 1dB is estimated, since the horn sidewalls of the active wafer (wafer #3 as marked on Fig. 2.18 in Chapter 2) were not coated with gold. This results in a total antenna loss of approximately 3.5dB when coupling to a plane wave. The next step in characterizing the receiver is to measure the total conversion loss (Lrt) of the receiver, defined as the measured IF power at 200MHz into a 50Q load, divided by the total 94GHz power incident on the horn aperture (P2nc = S. Aphys8) L~t includes the aperture efficiency losses (1.5dB), losses in horn walls (1.5-2dB),

76 0 C..\. co CIS h /'I6 Degrees Degrees A / _ _ Hd jJ E-planecod =E I C 1I- -I aH-plane H - - H-plane Channel 13 Channel 2 2960 -40 -20 0 20 40 60 260o -40 -20 0 20 40 60 Degrees Degrees Figure 3.22: E and H-plane patterns for the individual horns in the monopulse array. The antennas are numbered as indicated in Figure 3.16 The antennas are numbered as indicated in Figure 3.16

impedance mismatches between the pumped diode and antenna at RF, the intrinsic conversion loss of the mixer, and finally, mismatches between the diode and the IF load. This measurement consists of using a 94GHz Gunn diode source feeding a W-band standard gain horn to transmit a known power density in far-field, where the receiver is located. The power transmitted by the 94GHz horn is monitored with an Anritsu [67] power meter, and the standard gain horn has a gain of 23.1dB at 94GHz which is established from full two-dimensional pattern measurements [69] and extensively used formulae [65]. Therefore, the Gunn source transmits a known power density in the far-field. The total 94GHz power the horn aperture is P =Pt Ah (3.13) where Pt is the transmitted power, Gt is 23.1dB, and Aphys is the physical area of the integrated horn, equal to 16.8mm2. The receiver is placed in the far-field of the 94GHz horn, the 23GHz LO is activated and the IF power is measured. The IF is selected to be 200MHz because of the wide availability of signal processing components at this frequency. The measured total single-sideband conversion loss for the four channels is listed in Table 3.2. The total conversion loss (L~t) lies between 28 and 32dB for channels 2,3 and 4. By using R, = 6Q, Cp = 35fF, Cjo = 75fF and the embedding impedances presented in Chapter II (Fig. 2.14 on page 35) with harmonic balance analysis, the calculated mixer conversion loss is 19.5dB, and is broken down into 18.5dB for the intrinsic mixer loss and 1dB for the RF and IF mismatch. When the estimated horn antenna losses of 3.5dB are added, this results in a total calculated conversion loss (PIF/Pinc) of 23dB. It is clear that about 5-6dB of loss remains unaccounted for. The discrepancy between calculated and measured conversion loss may be due to

Channel 1 2 3 4 Conversion Loss 38.1dB 28.1dB 28.9dB 31.1dB Table 3.2: Measured total conversion loss for the four monopulse receiver channels, defined as the IF power divided by the total power incident on the aperture. the following reasons: 1) An insufficient amount of LO power is being delivered to the antiparallel diodes for optimum conversion loss. 2) A higher series resistance exists in the diode due to skin effects at 94GHz. 3) Antenna losses in the horn structure are higher than expected. 4) Physical differences between the modeled antenna and the fabricated antenna are influencing the impedance of the dipole. 3.4.1 LO Power Investigation In order to see if the mixer performance was being optimized with respect to the LO power, a CPW probe was brought into close proximity with the CPW line from the output of the last amplifier stage, and was used to monitor the LO power as the amplifier bias was varied. This data was then corresponded to the measured subharmonic conversion loss data taken at the same amplifier bias points. The conversion losses for channels 2 and 4 are plotted together (Fig. 3.23). Channels 1 and 3 were not measured. The data shows that insufficient LO power is being provided to the mixers for optimum conversion loss. The poor conversion loss of channel 1 is probably due to uneven distribution of the LO power, most likely caused by nonuniformity and defects in the polyimide used to produce the overlay capacitors at various points

Channel 1 2 3 4 IF mixer impedance 619 409/ 37f9 43Q Table 3.3: Measured IF impedances for the four monopulse receiver channels. in the circuit, such as the LO-IF isolation capacitors, and the RF mixer choke. Harmonic balance analysis predicts an IF impedance of about 30Q for the mixer when it is being driven at the LO power levels which result in optimum conversion loss (Fig. 3.24). Furthermore, the calculated IF impedance rises monotonically when the LO power is reduced, and sharply increases when the LO power drops below 1.4mW. The IF impedances for each channel were measured directly using an HP8720B network analyzer at 200MHz (Table 3.3), and the impedance for channel 1 was found to be 60Q at the maximum available LO power levels. This corresponds to a calculated power level of 1.6mW. The IF impedances of the remaining channels were close to 40Q, which corresponds to a power level of 2.8mW. Thus from these measurements and the analysis in Figures 3.23 and 3.24, it is very likely that insufficient LO power is a primary contributor to the excess measured receiver losses. The sidewall wafer which directly contacts the circuit (wafer 4 as marked on Fig. 2.18) had a very important effect on the operation of the mixer. When the sidewall was in place, it caused the conversion losses of channel #1 and channel #4 to drop by several dB. The problem was isolated by measuring the IF impedance of the mixers when they were being pumped with LO power. The measurements showed that the IF impedance of channel #1 increased dramatically, from 60Q to 250Q with the sidewall wafer in place. It was then surmised that wafer #4 was disturbing the distribution of the LO power to the mixers, since a similar effect on the IF impedance could be observed when the LO power was deliberately decreased. In

80 38 36 \ 34' 32 0 0 30 o,..-..Mixer 2 21 (128 au-os-u Mixer 4.6..5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 Normalized LO Power (dB) Figure 3.23: Conversion loss for two monopulse channels, as a function of relative LO power level (dB below maximum LO power). order to eliminate this problem, high-resistivity silicon was used to fabricate sidewall wafer #4. This completely eliminated the attenuating effect the sidewall wafer had on the LO power. The remaining sidewall wafers had no measurable effect on the distribution of LO power, and the conversion losses listed in Table 3.2 were measured with a high-resistivity back-cavity wafer. 3.4.2 Discussion of Losses in the Antenna It is desirable to determine the exact losses in the antenna, and to determine if the mixer has been optimized for conversion loss. Fully characterizing the antenna losses is a difficult process requiring extensive experimental work. This involves remodeling the fabricated antenna structure at microwave frequencies, and performing 94GHz measurements using a calibrated bolometer detector. Due to time constraints, these experiments were not included in this work. Work done in [70] measured a 1dB loss

81 100,,,,,,,,,,,, 30 - - Calculated Conversion Loss 80- 25 C E 0 \ i ~ Pvt - 0 40 - 15 Calculated IF Impedance 20' 2 4 6 8 LO Power (dBm) Figure 3.24: Calculated IF impedance and conversion loss for the monopulse tracking receiver, as a function of LO power, using harmonic balance analysis. (Rs, 6Q, Cj0 o 75fF,Cp = 35fF,n = 1.1,I, = 1.5 x 10-14, b - 0.75V) resulting when only one of the four sidewalls on the active wafer was not coated with gold. Because all four sidewalls on the active wafer are uncoated, a 1dB loss is most likely a conservative estimate. The 90am-thick substrate itself is not yet thoroughly characterized, but is thin enough that it is not expected to contribute significantly to the RF losses. 3.5 IF Monopulse Processor The IF signals produced by the four channels are phase-coherent, because they are driven by the same LO source. Any amplitude or phase noise produced by the LO source simultaneously affects each of the four mixers. This allows the LO to be free-running (phase-locking is not needed), and also permits the IF signals to be processed using an IF monopulse processor, which produces the normalized difference

82 az From Lo Ampliude Receiver Noise I and Chip 4 Amps 4 Phase 4 (4 channels) T rm [ 2 Amplitude Processing Figure 3.25: Block diagram of the 200MHz IF processor. signals and relative phase information. The block diagram for the IF processor is shown in Figure 3.25, and is a conventional monopulse processing design. The IF processing components are mounted on a printed circuit board with a thickness of 24mils (6.0mm) and a dielectric constant of 4.2 at 200MHz. A 50Q microstrip line on this board has a width of 1mm, which is convenient for connecting the closely spaced pins on the components. A parts list, circuit diagram, and detailed circuit description for the IF processor are found in Appendix D. The monopulse antenna patterns produced by the IF amplitude and phase processing circuits are shown (Figs. 3.26, 3.27). The null in the amplitude pattern can be adjusted to be well below 35dB at a given frequency, and the IF processor has a 25dB null width of about 5MHz centered at 200MHz. The phase pattern data is taken directly from the phase discriminator, and is related to the relative phase between the sum and difference patterns by a trigonometric relationship (see Appendix D). The pattern is antisymmetric about 0~, allowing the circuit to easily determine

83 0 do 1 0 -- ~~~~~/, L_ I I _ -'-i ~~~~~~~ I II / A-oH _I I II'1 1 I - \ I -~ ~ ~'\ I I \I Ij I~~~~~~~'~ -20 /2 46 II I A-r 3 MmatI e- — A-H I I - 3'I ~ ~ II'~ I I I'' 2''''4'O -60 c I — 0 — 0 A-H 4 r ~~Degrees'I FiZ e32: M aue o o usea piu epten t9 G z

84 0.15,,,,, I,I I i 1, I 0.10: 0.05 0 0) -0.00 N 0 5-E I -0.05 0 - E-H -0.15 -0.20 I, I I I I I -60 -40 -20 0 20 40 60 Degrees Figure 3.27: Measured monopulse phase patterns at 94GHz.

the angle of arrival relative to the difference null. Note that the high sidelobe levels are due to the antenna element spacing of 1.3A, which results in grating lobes at. ~350 for the sumi channel. The 1.3A element separation was chosen to provide space for the 23GHz LO generating circuits. In future designs when the LO frequency is moved to 46GHz, the antenna elements can be spaced approximately a wavelength apart. This results in grating lobes which occur at 480, where the horn antennas do not radiate much power. A back-of-the-envelope expression (worked out in Appendix D) for the angular resolution for a monopulse antenna system when the target is on-axis can be written: Obw a=, (3.14) where as is the standard deviation of the angle error, Obw is the 3dB beamwidth of the sum antenna and S/N is the signal-to-noise ratio. Thus for a target on axis, with a signal-to-noise ratio of 20dB, the receiver has an angular resolution, integrating over a single pulse, of about 2~. A detailed treatment of monopulse error due to noise and interference can be found in [71]. The tracking receiver and monopulse processor were successfully combined with a data acquisition computer and motorized gimbal mount to track a 94GHz Gunn source in the elevation and azimuth coordinates. The computer samples the IF processor outputs to determine how far off axis the source is, and drives the motors to move the mount accordingly. The most narrowband element in the processor, the phase shifter, has a bandwidth of about 15MHz, so the source is tuned until the IF lies within the phase shifter's operating range. A laser is used to determine the resolution of the tracking system by shining the beam off a reflective surface attached to the mount. This gives a sensitive indication of the resolution and repeatability of

the tracking. The resolution is limited by the signal-to-noise ratio (equation 3.14). The tracking receiver exhibited a repeatability of about 0.60. This is caused by the fact that the unsophisticated control software only utilizes the sign in the relative phase pattern, and thus the gain in the loop was reduced to prevent oscillations, creating a "dead zone" of about 0.6~. The receiver was then used in conjunction with a 10cm lens, which increased the directivity of the sum pattern to 34dB (Fig. 3.28) when using equation 3.12. The lens F/D ratio was chosen to be 0.9, which matches the 6dB points of the monopulse difference patterns, and the first null in sum pattern. Note that the F/D ratio is selected to match to the array pattern, not the individual horn patterns, because the signals received by each horn antenna are combined in a phase-coherent processor. By increasing the effective area of the monopulse antenna, the lens improves the tracking resolution of the system by reducing the beamwidth of the system and improving the signal-to-noise ratio considerably. 3.6 Conclusion and Future Work The integrated tracking receiver design presented in this paper represents the first completely integrated millimeter-wave receiver to date and overcomes many of the difficulties associated with designing a planar receiver. It uses subharmonic mixers which can mix efficiently at even multiples of the LO frequency, allowing the LO to be produced and distributed using efficient planar circuits. The subharmonic mixers are integrated directly with the antennas which consist of efficient integrated horn cavities. All fabrication techniques used to integrate the antenna are compatible with standard MMIC fab' cation. Low loss monopulse processing is performed at the IF, using the phase-coherent signals produced by the receiver. These features make the

87: -— (I \ I-H -10-l /\I I I,,, I J 11 Sr.R,-2 /noh\I /l S o njunt I a, II l I Q~ -30 II 2. - I, 0 -40 z i.,..-1 -5 0 5 10 15 30 Figure 3.28: Measured antenna patterns of the monopulse receiver when used in

receiver suitable for mass-production, with applications such as compact, inexpensive automotive, avionics and space collision avoidance systems, and in munitions guidance. The existing receiver design can be improved in three key areas. First, the LO generation can be made much more compact, by using more efficient high-power MMIC chip modules, such as those described in [72]. Utilizing a MMIC design also allows the LO to be moved to 46GHz to improve subharmonic mixing performance. 94GHz subharmonic mixers can operate with conversion efficiencies which approach those of fundamental mixers [24, 15]. Secondly, the diodes used in the design can easily replaced with ones which have much lower parasitics, optimized for subharmonic mixing at 94GHz. Such diodes can be built with existing semiconductor technology, but until now have had little commercial demand. Planar-doped barrier (PDB) [27] diodes can also be used in place of Schottky diodes, significantly reducing the amount of LO power required to drive the mixers efficiently. The antenna and mixer design can easily be optimized for these new devices. Finally, the IF processing components can also be integrated in a very compact form using existing integrated circuit RF processing technology [73], allowing the entire monopulse tracking receiver to be placed on a single chip. The architecture embodied in the receiver design, combined with the low power requirements of PDB diodes and MMIC technology, would also facilitate the fabrication of a large array of antennas and mixers, driven by a single LO source. Such a large array would include integrated IF processing, allowing an i; -.d 7s-edarray receiver with very high gain and beam-steering capabilities to be constructed. Of course, modified antennas optimized for phased-array antenna operation can be used with the same receiver design in order to improve the antenna patterns. The

89 large, planar array could be used not only in phased-array radar applications to be mounted on virtually any surface of an aircraft, but also in adaptive point-to-point communications on land or space-based systems. It would utilize entirely integrated construction to obtain high performance at low cost.

CHAPTER IV An Integrated Millimeter-Wave and THz Power Sensor Characterization of systems at high frequencies requires a reliable method for measuring absolute power. Typically, the efficiency of the separate components of a receiver such as the antenna or mixer must be found, or the output power of amplifiers, multipliers or oscillators must be determined. In the case of characterizing the performance of a receiver, accurate measurements at mm-wave and THz frequencies are most often done by using the "hot-cold load" measurement technique [21]. However, this method cannot be used reliably to determine the various loss components of the system, such as antenna ohmic losses, mixer reflection, and mixer conversion loss. For any direct measurement of power, a well-calibrated power detector is needed. 4.1 A Brief Technology Review Conventional power measurement at frequencies above a few hundred gigahertz (GHz) is expensive, narrowband, and often inaccurate. Such difficulties arise primarily because conventional power meters rely on waveguide-based systems, which are difficult to machine and calibrate as frequencies increase beyond the mm-wave range. These conventional methods employ a thermistor or a diode suspended in a tuned waveguide structure. Examples of this are the Anritsu, Hughes, and HP power

meters [67]. They are calibrated at the factory, limited to a waveguide band, and become very expensive for frequencies above 100GHz. They also become increasingly inaccurate above 200GHz (~2dB), and are simply not available for frequencies above 300GHz. Among the quasi-optical techniques are the Keating power meter based on the photoacoustic-effect [74, 75], the micro-calorimeter [76], and pyroelectric detectors [77]. The term "quasi-optical" means that the power meter measures an incident power density rather than total power generated by a source. lWork done in [78] shows that the Keating power meter is accurate and easily calibrated. However, it uses a resonant input window which must be changed for different frequencies, is currently limited to frequencies below 300GHz and is expensive. The micro-calorimeter has been used accurately at 94GHz but has a very slow response time, is difficult to calibrate, and must be calibrated for different frequencies. Pyroelectric detectors are also difficult to characterize and calibrate, and are mainly used as crude but sensitive far-infrared (FIR) detectors. In fact, there is a lack of simple and accurate power meters at submillimeter-wave frequencies. The quasi-optical power sensor presented here is specifically intended to fill this need. 4.2 Introduction to the Power- Sensor The large area bolometer has been developed as a monolithic quasi-optical power meter optimal for wide-band (9OGHz-3THz) measurements at millimeter and submillimeter wavelengths (Fig. 4.1) [69]. The bolometer sensor is constructed using a thin, 1000A-thick layer of bismuth. When the sensor absorbs power, its temperature rises, causing a change in its resistance. This, in turn causes a change in the voltage across the bolometer when a DC bias current is applied. In order to calibrate the sensor, one needs to determine the change in resistance of the bolometer when a

known power density is incident on the bolometer. This is discussed in detail under Section 4.4 entitled "Modeling and Calibration". The bismuth bolometer is integrated on a 1.2fm-thick dielectric membrane. This membrane is essential to the operation of the power sensor for two important reasons: 1. It results in a low thermal conductance path between the bismuth bolometer and the supporting silicon wafer, thereby improving the responsivity of the detector. 2. Like the bismuth bolometer, its thickness is much less than a wavelength at millimeter-wave frequencies. This greatly simplifies the electromagnetic analysis-of the structure discussed in Section 4.4.1. This analysis and the resulting transmission line models show that sensor yields a frequency-independent response across a very wide range of frequencies. The low frequency cutoff is given by the size of the bolometer; that is, when diffraction becomes severe and the transmission line and thermal equivalence models break down. The experimental results indicate that this happens when the bolometer is approximately 1.5A0 long. The high-frequency cutoff is given by the electronic properties of thin-film bismuth which occurs in the infrared and UV range [79]. Also, a lim-membrane with a dielectric constant of 4.5 introduces a 3% error at 10THz. However, one could easily integrate the bolometers on 3000A-thick membranes, thereby pushing the cutoff frequency to 30THz. 4.3 Fabrication A 3-layer SiO2/Si3N4/SiO2 structure with respective thicknesses of 5000A/3000A /4000A is deposited on both sides of a (100) silicon wafer. This combination yields a

Incident Radiation DC Contact ~ I~ 1 1 I ~ / ~Bismuth bolometer Membrane 4 to 7mm Silicon Substrate Absorber Figure 4.1: A monolithic wideband quasi-optical power sensor. 1.2/im-thick dielectric layer in tension. The layer must be in tension to yield flat and rigid self-supporting membranes. These are fabricated in two steps. First, an opening in the silicon-nitride layers is defined on the back of the wafer, and the silicon is etched until a transparent membrane appears. In this case, the shape of the surrounding silicon cavity is not important, and any isotropic or anisotropic silicon etchant can be used [80]. Next, the bolometer and contacts are defined on the top side of the wafer. Bismuth is chosen as the detector material for its high resistivity, and high temperature coefficient of resistance [81]. The surface resistance of the detector film should be around 100-200Q for optimal coupling to the incident plane wave, and this is achieved with a 400-700A evaporated bismuth layer. The contacts are evaporated silver 500-1000A thick. One can also use NiCr or Au as bolometer materials, but the evaporated layers are only 30-50A thick and do not yield repeatable performance [82]. The bismuth bolometer is compatible with standard IC fabrication techniques, and it is possible to integrate it with the bias and detection electronics to result in a fully integrated-circuit power meter.

4.4 Modeling and Calibration To be able to use the quasi-optical power meter, one needs to first determine the relationship between the power density incident on the bolometer and the absorbed power in the bolometer. This requires that an electromagnetic model be developed. Once this is done, the responsivity of the bolometer, i7, must then be determined. i is defined as the bolometer's voltage response to a known amount of absorbed power under constant bias conditions, and thus has units of volts per watt. Calibration of the bolometer is the process of finding these two pieces of information. The large size (in wavelengths) of the bolometer, combined with its very small thickness relative to a wavelength, plays an essential role in greatly simplifying the calibration procedure. 4.4.1 Determining an Electromagnetic Model The first part of calibration is simplified by the electromagnetically simple structure of the power sensor. First, operation of the sensor is confined to frequencies where the wavelength is somewhat less than the width of the bolometer. Since it is possible to fabricate bolometers with dimensions larger than 5x5mm, a realistic lower frequency limit, fjl, is 80GHz. The reason for applying this limit is because an electromagnetic plane wave at or above fj1 incident normal to the surface of the sensor will excite currents which are well approximated by a uniform amplitude distribution. This is because the bolometer consists of lossy, resistive material and unlike metallic structures, does not exhibit large current peaks along its edges [83]. Work in [79] shows that the RF impedance of such thin-film materials is equivalent to their DC sheet resistance. Moreover, the moderate variations in the current near the edge contributes negligibly to the overall power absorbed by the bolometer, which is dominated by the uniform currents flowing in most of the resistive sheet. Quan

titative calculations and theory for the scattering of EM waves incident on resistive surfaces support this intuition [84], and millimeter-wave measurements presented in this chapter ind(icate that a bolometer with a minimum square area of 1.5A0-square results in measurements which are in excellent agreement with theory. In addition to its large relative area, the bolometer is suspended on the thin dielectric membrane, which has an electromagnetic thickness of less than Ad/50 at frequencies below 3THz. This allows us to safely neglect the membrane and model the bolometer as a resistive sheet suspended in free space when operating below this upper limit, f,,. Experimental evidence presented in Section 4.7 titled "THz Laser and Spectroscopy Measurements" agrees well with this assumption. Taken together, the large area and thin structure of the bolometer relative to a wavelength permit the use a simple transmission line model to describe the operation of the power meter. The resistive sheet is modeled as a lumped resistor with a value equal to its DC' sheet resistance [79]. The free space existing on either side of the resistive sheet is modeled by a 377Q transmission line (Fig. 4.2). A more complicated model takes into account the angle of incidence of the radiation in the E and H-planes, by varying the value of the impedance of the 3772 transmission line (described in Section 4.5). The model easily gives a simple relationship between the incident power and the power absorbed by the bolometer. Its validity is tested at millimeter waves using several experiments in this chapter. 4.4.2 Finding the Bolometer Responsivity Using a LF Network Because of its large dimensions, the bolometer has a correspondingly large time thermal constant (around 200-500 milliseconds), and responds only to the average absorbed power. Therefore it can be calibrated with signals at frequencies above

i/ To Incident Matched Radiation Load 377[ Rbo1 377Q ~, Figure 42: The equivalent transmission line circuit for the power sensor for a normally incident plane wave. a few hundred 1KHz using a relatively simple low-frequency (LF) circuit. The electromagnetic maodel presented in the preceding subsection indicates that the uniform currents flowing in the bismuth bolometer due to a normally incident RF plane wave at millimieter-wave frequencies do not differ greatly from those excited by a LF signal. Thus the RF and LF absorbed power will produce equivalent thermal distributions across the bolometer, and the responsivity measured using LE signals will be identical to the responsivity at RF frequencies. Thermal modeling on a mass-flow sensor-transducer with a similar geometry indicate that the temperature is nearly constant on the bolometer surface and falls off rapidly to room temperature at the membrane-silicon interface [85]. Thermal modeling can also predict the absolute responsivity of the bolometer without the need for a LF calibration technique. However, because the temperature coefficient of resistance associated with thin-film bismuth is variable and dependent on deposition conditions and thickness, the responsivity is more practically determined by the LF calibration procedure. Essential to the low frequency calibration is fact that the bolometer has a slow thermal response, and will only detect the average applied power. The bolometer absorbs power, heats up and changes its esistangce. There are two ways to calibrate the bolometer at low frequencies; by using either a purely DC method or an ampli

97 Vbias Amplitude-Modulated Function Generator HP3312A Rbia HPF R bol LPF PAR- 124A Lock-in Amplifier Figure 4.3: The low-frequency (0.5 to 5MHz) calibration network tude modulated (AM) AC current. The DC method is simple to use and has been investigated by Rutledge et al. [86]. It consists of applying an incremental DC power to the bolometer, and measuring the change in resistance using a 4-wire measurement technique and a six-digit multimeter. This technique is accurate for resistance changes of 0.05Q or higher, which translates into a minimum detectable absorbed power of about 0.5mW. The AM calibration technique consists of applying to the bolometer a LF (0.55MHz) signal at frequency fLF (Fig. 4.3), modulated at a much lower frequency fnod (lOHz-lKHz). First, consider a signal at fLF with constant amplitude, turned on at t=0 and applied to a bolometer which has been resting at room temperature (Fig. 4.4). The bolometer is a resistive element, and absorbs the constant AC power being applied. As the power is absorbed by the bolometer, it heats up until it reaches a steady state: its resistance decreases by a constant amount proportional to the average amplitude of the signal, but not its instantaneous value which is changing at

t=O t-1 second I I I LF Power (a) L I 1/Frnod Bolometer I Resistance (b) Figure 4.4: A signal in the LF spectrum turned on at t=O and applied to the bolometer (a). The bolometer's thermal response (b). a frequency fLF and is much faster than the thermal time constant of the bolometer. Thus the bolometer is only responding to the average AC power of the signal, not its instantaneous amplitude. When the LF signal is turned off, the bolometer cools down and its resistance rises back to its room-temperature state. Now consider a LF signal being turned on and off (square-wave 100% amplitude modulation) at a rate of fmod, lOHz-lKHz. Because fmod is much lower frequency than fLF, the bolometer heats up and cools down at a frequency fmod and when biased, will produce a voltage signal at fmod. This is exactly the same way it will function in the millimeter and submillimeter-wave region when the incident radiation at 100GHz to 3THz is chopped (100% square-wave amplitude modulated) at fmod. The input and output AM signals are filtered using high quality passive LC filters (Fig. 4.5), and the square wave bolometer response is detected using a lock-in amplifier. The HPF is very important since it prevents spurious, synchronous lowfrequency modulation signals produced by the signal generator from reaching the lock-in amplifier. The LPF merely prevents the carrier at fLF from saturating the

sensitive lock-in amplifier. Since fLF is several orders of magnitude larger than f,,d, the filter designs are simple. The LPF, for example, contains one value of inductance and one value of capacitance, and exhibits resonances at its cutoff frequency of 10KHz, far from either fmod and fLF. Care must be taken to include the 100Q source resistance of the bolometer when designing the filters. The values shown in Figure 4.5 are general guidelines and must be adjusted to account for self-resonances due to the large inductor values. The response as f,,mod is varied from 10Hz to 1KHz is measured using the LF calibration network and compared to the same response found using a millimeterwave system (Fig. 4.6). In this case, a 140GHz source is AM modulated at fmod, and the bolometer response is measured. The output signal drops as l/f in both the LF and 140GHz measurements, indicating a single pole thermal response (Fig. 4.7), and showing that indeed, the thermal response for RF and LF frequencies are the same. The output is also a linear function of the applied LF and RF power and of the bias voltage. Measurements performed on a 4x4mm bolometer yield a responsivity of around 70mV/W at fmod = 100Hz with a bolometer sheet resistance Rbol = 105Q and a bias current of 5mA. When the area of the bolometer is increased to 7x7mm, the responsivity decreases to about 8.6mV/W for Rbol = 99Q and a bias current of 7.7mA. The AM technique is more complicated and requires expensive equipment such as a lock-in amplifier, but it eliminates DC drift problems associated with convection, infrared radiation and thin-film bismuth properties. Also, because it is based on homodyne detection using a lock-in amplifier, it allows extremely sensitive measurements of absorbed power to be made. For example, a 50nW absorbed power can be easily detected with a three-second integration time.

100 R. (5Q) c c c I__. _ I I_ _._'_ (-_ I 000loon) C=0.47IH R*,.1 Lock-in (vloon) L L L r. Amplifier Vbe. C T CT CT I 1OKI Figure 4.5: The high-pass and low-pass filters used in the LF calibration network. Absorber 1 OdB Coupler Multiplier IvSource. Bolometer Power Freq. Meter Meter Figure 4.6: The millimeter-wave quasi-optical measurement set-up.

101 ~1 ~ ~ ~! T T I' I I I I T Bolometer Response versus Modulation Frequency C) 11 a) 0.1 o o o-s-e-. 550KHz 100 1000 4.5 Millimeter-wave Measurements Additional experimental evidence is required to fully justify the simple transmission line model used in analyzing the bolometer structure. The quasi-optical response of the sensor was measured at 90GHz, 140GHz and 240GHz on 7x7mm and 4x4mm bolometers. The bolometer was sandwiched between thick styrofoam layers to reduce the very low frequency drift due to air convection and infrared-radiation. Also, an absorber was used behind the bolometer to minimize reflections. We found that a badly designed or misaligned absorber can affect the measurement by as much as ~25%. A good absorber design would consist of a large absorber-lined pyramidal cavity placed behind the sensor. Figure 4.8 shows that bolometer is polarization independent for normal incidence. This indicates that the silver DC bias contacts do not affect the bolometer response, and that the bolometer is indeed behaving as an electrically large absorbing, resistive sheet. The transmission line model is used to calculate the bolometer response as a

102 function of angle of incidence [87]. For an electromagnetic plane wave incident at an angle 0 on a bolometer with a normalized sheet resistance r with respect to the free-space impedance of 377Q, the load impedance (Fig. 4.9) is given by r Cos 0 Zload = for E- plane incidence, and (4.1) r + cos 0 Zload r for H - plane incidence. (4.2) rcos0 + i The power absorbed by the bolometer is therefore: Pabs = Cos5t (1 - p2) Zload (4.3) where 0 is the angle from normal incidence, cos 0 accounts for the reduction in the effective area of the bolometer, and p is the reflection coefficient (in a 377Q system) of the load as defined in Figure 4.9. The absorption peaks at 50% for a sheet resistance of 189Q (half of the free-space impedance) at normal incidence (0 = 0). For this case, the pattern is independent of the polarization and is given by: 2 cos2 0 Pabs (1 + os0)2 (4.4) (1 + coS n)2 The 140GHz H-plane scan for a 7x7mm bolometer with a surface resistance of 98Q is presented in Figure 4.10. The theoretical and experimental results differ for incidence angles greater than 700 due to the small size of the bolometer. At these angles, the projected length is less than 1.5MA, and the TL model begins to break down. The transmission line equivalent circuit is also checked by measuring the bolometer response at 140GHz and 240GHz with a sliding ground plane behind the silicon substrate (Fig. 4.11). The theoretical power absorbed by the bolometer is given

103 1.20 I 0 QD (12 C4 0 0 o~s r~ Independence of Bolometer Response With Respect to Polarization.~ 0.40 (+/- 3% variation) 0.00. 30 -10 10 30 50 70 90 Angle of Polarization Figure 4.8: Bolometer response at 140GHz as a function of polarization angle for normal incidence. by (1 - Sf1), where S11 is calculated using the TL equivalent circuit, using a 377Q normalizing impedance. A small loss was assumed into the TL model to account for the plane-wave diffraction between the bolometer and the ground-plane. The measured minima occur at a half-wavelength period, and the measured frequency using the mirror data agree exactly with the waveguide frequency meter. The resonant deviation between theory and experiment at 140GHz is most likely caused by interactions between the cavity (formed by the etched silicon behind the sensor) and the mirror backshort. When measurements were performed at 240GHz, the incident radiation was transmitted through membrane from the substrate side of the sensor. This eliminated the cavity effects and allowed the backshort to be positioned closer to the bolometer. Absolute power measurements were done on 4 independent 7x7mm bolometers at 90, 140 and 240GHz. The millimeter-wave plane wave was generated by waveguide

104 E-plane incidence H-plane incidence 1 1 cos~ r cosO cosO r cosO r _ r' Z load Z load Figure 4.9: The equivalent transmission line circuit for E and H-plane incidence. 1.2.............T.........,...,.T...,....V..',,; 1.0 0.8 ~ o0.6. 0.4 o 0.2 - * *Bolometer Response, 980/square - UM Theory, 98~f/square 80 -60 -40 -20 0 20 40 60 80 Angle of Incidence Figure 4.10: H-plane pattern at 140GHz for a 7x7mm bolometer with a sheet resistance of 98Q.

105 1.00 o 0.80 d - Incident Radiation ~ — 3770 R 377 f 0.60 bol o Sl ( 0.40 *> @< to 6.d co F bl r ~T1 Backshort W0.20 Q)= 0.20Incident Transmission Line Theory =:: Backshort Measurement Power (140GHz) - _ l260.... 6.70 7.20 7.70.20 Backshort position d (mm)!.00.. II......... o 0.80 w 0.60 0 0 M 0.40 Backshort 0.20: Transmission Line Theory Incident Backahort Measurement Power (240GCHz) -1 oq2. 11.60 12.00 12.40 Backshort position d (mm) Figure 4.11: Relative absorbed power versus mirror position for a 4x4mm bolometer at 140GHz (top) and 240GHz (bottom).

106 Bolometer Anritsu R(Q) 90GHz 90GHz 140GHz 240GHz 98Q 14.5 15.3 25.3 10.6 98Q 14.7 15.3 23.2 10.4 98Q 15.5 15.3 24.1 10.6 102fQ 14.3 15.3 24.5 10.6 Table4.1: Absolute measured power density of an incident plane wave using four different bolometers at 90GHz, 140GHz and 20GHz. The results at 90GHz agree well with the Anritsu power meter measurements. Gunn-diode sources feeding doublers or triplers (when necessary), and a standard gain horn. The millimeter-wave sources were AM modulated at 100Hz. The bolometers were placed well jnto the far-field of the transmitting horn. The power measurements were done by the method of substitution, that is, first the bolometer output resulting from the millimeter-wave radiation is recorded. The millimeter-wave signal is then removed, and a carrier at 500KHz is AM modulated at 100Hz and applied until the same output voltage is achieved. It is not necessary to de-embed the filters or the lock-in amplifier responses, since the millimeter-wave power absorbed is determined from the amplitude of the amplitude of the 500KHz modulated AC voltage across the bolometer, which is directly measured using a calibrated oscilloscope. The absolute plane-wave power density is then calculated from the surface area of the bolometer and the TL equivalent model. The results, summarized in Table 4.5, indicate that an accuracy of ~5% is achievable over a wide frequency range. This accuracy includes the alignment of bolometers to the incident plane wave, the placement of the back absorber, the voltage readings on the lock-in amplifier and across the bolometers, and connector repeatability.

107 The results are also compared to a calibrated waveguide power meter at 90GlHz. In this case, a standard gain horn and an Anritsu power meter are placed at the same position as the bolometer. The receiving horn gain is independently calibrated to ~2% using standard-gain horn tables [88], two-dimensional pattern measurements for directivity calculations, and the Friis transmission formula using two identical horns [63]. The power density is then calculated from the calibrated antenna gain and the Anritsu power meter reading. The four independent bolometers yielded power density readings which differed by ~3% and agreed to within 5% compared to the Anritsu meter. 4.6 Absolute Output Power of a Millimeter-wave Tripler An important application area is the measurement of the absolute output power of triplers above 180GHz. This provides an experimental verification of non-linear analysis programs, Schottky-diode parameters, and the effects of the parasitic capacitance and series resistance on the performance of the device. The output power of a 220-280GHz tripler (Millitech, MUT-03) was measured using a calibrated standardgain horn for WR-03 waveguides, and a quasi-optical large-area bolometer placed in the far-field of the tripler-horn combination using an experimental setup similar to the one shown in Figure 4.6. The bolometer was placed at a distance of 36.8cm from the mouth of the standard gain horn, and a typical power density of 0.22W/m2 was measured. In each case, the bias, the input power at the fundamental frequency, and the input and output tuning screws were optimized for maximum output power. Table 4.6 shows the results of two specific bolometers which yielded the widest discrepancy across the 220-280GHz range. The output power readings still agree with each other to within ~5% over the entire frequency band, and agree well with the

108 Tripler Bolometer 1 Bolometer 2 Specified Output 98Q 102Q Power 222GHz 0.70mW 0.76mW 0.79mW 240GHz 1.7mW 1.9mW 1.84mW 249GHz 1.5mW 1.5mW 0.36mW1 270GHz 2.0mW 2.2mW 2.01mW 279GHz 1.5mW 1.7mW 1.12mW2 Table 4.2: Measured output power of a Millitech MUT-03 tripler. The bolometers with the largest measurement discrepancy over the 220-280GHz range are presented here. Notes- 1 Limited by a maximum available source power of 6mW; 2 Measured at 277GHz. the output powers quoted by Millitech using a custom-made WR-03 Anritsu power meter. Other application areas of the power meter include efficiency measurements of doublers and triplers at submillimeter-wave frequencies, antenna gain and couplingefficiency measurements and the determination of the total output power from terahertz solid-state sources and far-infrared lasers. 4.7 THz Laser and Spectroscopy Measurements The previous sections established the validity of a simple model in analyzing and calibrating the large area bolometer, and determined preliminary performance of the sensor at several frequencies within the millimeter wave spectrum. The remaining sections in this chapter describe measurements to determine the performance and suitability of the same power sensor beyond 300GHz, into the terahertz spectrum. A Fourier Transform Spectrometer (FTS) was used to analyze the power sensor structure by measuring transmission spectra of the bolometers from 0.6THz to 10THz.

109 Also, a linear array of bolometers was fabricated and used to characterize the Gaussian beam produced by a far-infrared (FIR) laser at 802GHz and 2.54THz. Measurement of the behavior of materials at frequencies beyond 300GHz are difficult to do with coherent sources, simply because continuously tunable power generation and accurate measurement capabilities do not exist at these frequencies. The FTS circumvents these difficulties by using a broadband incoherent source such as a mercury vapor lamp, combined with an automated, dynamically aligned Michelson interferometer to make accurate transmission or reflection measurements from 300GHz up to UV wavelengths [89]. However, because they behave much like black bodies, these broadband sources provide very little power at millimeter wave frequencies (which correspond to the lower operating limit of the FTS). Because of this, very sensitive, liquid helium-cooled detectors are required to make measurements below ITHz. The most obvious measurement needed is to first find the transmission characteristic of the thin dielectric membrane supporting the bolometer. The dielectric membrane structure and fabrication are very similar to that described in the first six sections, except that slightly thicker dielectric layers are used in order to increase the strength of the membrane. The membrane is composed of three layers: 7000k thermally grown SiO2, 3500A LPCVD Si3N4, and 4500k LPCVD SiO2. At millimeter-wave frequencies, the membrane is so thin compared to a wavelength that its effect on the electromagnetic behavior of the sensor is negligible. As frequencies approach lTHz, this is no longer the case, and the layers within the membrane must be included in the model. The dielectric layers are modeled as sections of transmission line with impedances equal to 377Q divided by the respective refractive indices, and the 400k to 1500k bismuth bolometer is modeled as a 190f1 to 40f resistance

110 E SiC2 Incident Radiation E SiAN DC Contact 400A Bismuth ____________________ 4500 A — 3500oo A t- 4 to?7mm -(0S Silicon Substrate Incident Radiation 0 150 0 3770fbi 10015 1900 37711 1.080 1.030 1.68' Figure 4.12: Large area bolometer, shown with the transmission line circuit equivalent at 1THz. (equal to its DC sheet resistance) in parallel with the transmission line (Fig. 4.12). The measurements presented here show that the model is valid up to frequencies approaching 3THz. 4.7.1 FTS Measurements Transmission spectra of the bolometer were measured using a Bomem DA8 Fourier Transform Spectrometer [90]. The FTS was configured with a mercury vapor lamp source, a mylar beamsplitter, and a liquid helium cooled bolometer operating at 4.20K. The measurement resolution was 15GHz, which was more than sufficient to resolve the broad spectral features of the bolometer. Measurements contain gaps near 4THz and 7.5THz, which correspond to nulls of the beam splitter. Transmission spectra were obtained for a 7x7mm dielectric membrane with and without the thin-film bismuth, from 600GHz to 1OTHz (4.13). Since the diameter of the focused FTS beam is larger than the membrane, a variable aperture is used to

111 adjust its size. The transmission spectra are then compared with predicted transmission given by the simple transmission line model. The values for E, used in the model are 4 and 6 for Si02 and Si3N4, respectively. The dielectric constants of these thin films are not known at terahertz frequencies, and the values used here are not very different from those obtained using low-frequency measurements. Fig. 4.13 shows transmission measurements of the bare membrane, which agree to within O.ldB with the model up to 7THz. The materials which make up the membrane have never been characterized at such high frequencies, and it is therefore not surprising that higher than expected absorption of the membrane occurs above 7THz. This may be caused by additional species such as hydrogen present in the Si3N4 which were added during the deposition process. Appendix B, entitled "Fabrication" contains details of the deposition process, as well as suggestions for additional experiments. The sample is then placed in an evaporation chamber, where 400A of bismuth were deposited at pressures below 1,T at a rate of about 5A/s. Measurement of the membrane with the bismuth show a transmission of about 24% at low frequencies, and agree with the model to within 0.5dB, up to 2.5THz. The difference between measurement and model increases to 1.8dB near 7THz. This deviation is most likely due to the change in the sheet resistance of thin-film bismuth at these frequencies, and is not investigated here. However, these results clearly show that the power meter can be used effectively at frequencies approaching 3THz, and that the effects of a 1.5pm-thick membrane are negligible up to terahertz frequencies. 4.7.2 FIR Laser Measurements A 20-element linear array of large area bolometers was constructed in order to easily obtain the profile of a Far Infrared (FIR) laser beam at 802GHz and 2.54THz.

112 The measurements were performed at the NASA-Goddard Space Flight Center. Each element of the array consists of a square bismuth bolometer approximately 800A thick, 1.14mm on a side, with a center-to-center spacing of 1.71mm. This allowed the array to be used with accuracy at frequencies above 400GHz, based on the results presented in previous sections. The bolometers were fabricated on a single rectangular dielectric membrane 38mm long and 1.3mm wide. The thermal isolation among the elements, measured using the LF network, was better than 26dB due to the high thermal resistance of the thin membrane. The responsivity of a single element determined in the LF calibration technique was 172mV/W at a bias of 4V and a modulation frequency of 102Hz. The sheet resistance of 51Q was constant to within ~2% throughout the array. Peak power densities being measured were typically more than 30dB above the sensitivity of the bolometer (about 5nW/mm2 incident power density). The profile and peak power density of the laser beam were measured as a function of distance (d) along the axis of propagation. The beam profiles were clearly dominated by the first order Gaussian mode. The values for measured beam waist (w) at 802GHz are shown in Fig. 4.14 for d=150mm, 279mm, and 4457mm. These agree with Gaussian optics calculations to within 4%, assuming a 3.5mm minimum beam waist at the output coupler of the laser (d=0) [91]. Calculations for total power output at various positions along the axis of propagation yield 2.6mW~8%, assuming circular symmetry of the beam and using a Gaussian fit to the measured waists. At 2.54THz, the power densities were about 5 times higher compared to measurements at 802GHz at corresponding array positions. However, because the beam profile was much narrower, power content was only slightly higher, about 3.1mW, with a variation of ~3% from position to position. This value is lower than normally obtained

113 with this laser at 2.54THz, probably because of the difficulties in stabilizing the laser when the measurements were made. Again, using a 3.5mm minimum waist at d=O, measured beam waists (Fig.4.15) agreed with Gaussian-beam calculations to within 2% at d=889mm and 686mm, and 11% at d=381mm. 4.8 Conclusion In conclusion, the simple transmission line model and low frequency calibration used to characterize the large area bolometer has been shown to be valid for frequencies approaching 3THz. The FTS transmission measurements for the bismuth bolometer fabricated on a 1.5tm show that the spectra are flat and agree with the model to within 11% up to 3THz. The beam profile, peak power density, and total power content of a FIR laser was determined at 802GHz and 2.54THz using a linear array of bolometers. The spreading of the beam at these frequencies agreed closely with Gaussian-beam optics calculations. The transmission spectra obtained from the FTS is evidence in favor of the accuracy of the large area bolometer.

114 1.0 Dielectric membrane - 0.8 C 0.6 0.2 400 A Bismuth layer on dielectric membrane 0.4 0 2 4 6 8 10 Frequency (THz) Figure 4.13: FTS transmission spectra for the dielectric membrane with and without the thin film bismuth. Simple transmission line model theory (dashed lines) is shown for comparison. Squares and triangles mark 802GHz and 2.54THz respectively. o000 d=457mm 1.0 a. d=279mm dPo ~ 00 d 150mm mm |mm mm/mm 08 0 457 16.3 5.8 - _1 1279 9.8 19 150 6.2 42 802GHz FIR Laser t 0.6 Beam Profiles 0.4 0 N t - 0.2 ~ 20 -10 0 10 20 Position Across Beam (mm) Figure 4.14: FIR laser beam profile at 802GHz, at several distances from the laser output coupler. P0 is the peak incident power density.

115 1.0 o oo d=889mm.A.. Ad=686mm d w Po0, 0oooo d=381mm mm mm W/mm ~4-4 889 10.8 16 0 0.8 686 8.0 32 381 4.8 88 2.54THz FIR Laser aU 0.6 Beam Profiles > 0.4 0.2 z 0.0 0.20 -1 0 10 20 Position Across Beam (mm) Figure 4.15: FIR laser beam profile at 2.54THz, at several distances from the laser output coupler. P0 is the peak incident power density.

APPENDICES 116

117 APPENDIX A A Sbigle-1Horn 94GHz Integrated Azimuthal Monopulse Antenna An alternative design for a compact 94GHz monopulse integrated antenna has been investigated and developed. This design uses a single horn cavity to produce monopulse patterns in the H-plane only, and is referred to as a single-horn monopulse design. In contrast to the four-horn monopulse presented as the principle work of this thesis, the single-horn monopulse antenna presented in this appendix excites the sum and difference patterns at RF frequencies by coupling to orthogonal modes within the same horn cavity, instead of processing the IF signals from four separate horn receivers. Pattern measurements of microwave models and on the millimeterwave antennas show good agreement with theory and exhibit symmetric patterns with a sharp -30dB null at broadside for the difference mode. Microwave model measurements show input impedances close to 50Q2, with greater than - 25dB isolation between sum and difference antennas across a 10% bandwidth. A.1 Antenna Description The single-horn monopulse is also based on the pyramidal horn cavities etched in silicon. The monopulse system achieves direction finding in a single coordinate (azimuth) using two separate antennas within the same horn cavity (Figs. A.1

118 /!~~~~A /A Figure AX A l H *igure A.1: A mionolithic H-plane monopulse antenna.

119 A.2). The sum antenna consists of a single dipole and is suspended on a dielectric membrane with a thickness much less than a wavelength. This antenna couples primarily to the TElo mode, which produces a broad peak at normal incidence. Similarly, a set of parallel dipoles is suspended within the cavity on a membrane at a different plane. The dual dipoles, which constitute the difference antenna, are connected together by coplanar strip transmission lines which are crossed over at the center to couple primarily to the TE20 mode of the cavity. This mode produces a pattern which contains a sharp null at normal incidence. The detector is integrated at the center of the membrane where the two transmission lines meet. The feeding point of the dipoles for both the sum and difference antennas are approximately a quarter wavelength away from the horn walls. This conveniently allows quarter-wave coplanar-strip (CPS) transmission lines terminated with a very low impedance to be used to extract the DC bias and IF/video signal from the detector without affecting the antenna RF impedance. This is confirmed by measurement of a microwave scale model, which found the effects of the CPS lines on the input impedance to be insignificant. The construction of this monopulse system is achieved using photolithographic techniques identical to those used in the production of conventional integrated circuits. A.2 Fabrication The integrated monopulse antenna has been fabricated, using the same stacked silicon wafer construction described in Chapter II, Section 2.3. These wafers are etched anisotropically to form the walls of the horn cavity. Additional wafers are processed using procedures described in Appendix B in order to form thin dielectric membranes, on which the sum and difference antennas are integrated. The thick

120 and detector | —~ —-- 1.4A 0.42X AF x 51A X Bias Sum Antenna Figure A.2: Side view showing sum and difference antenna planes. The detector of the difference antenna is integrated very close to the center the crossover network. Note that the two antennas are not drawn to scale.

121 nesses of the front wafers are chosen to position the difference antenna at the proper depth within tile horn cavity. The wafers stacked below the difference antenna position the sum antenna at the proper depth. Finally, wafers are stacked below the sum antenna wafer to complete the pyramidal horn cavity. The detectors are deposited on a membrane using conventional photolithographic methods and consist of 4/um-square bismuth microbolometers [81]. Recent work has also resulted in a design which accommodates a hybrid-mounted beam-lead Schottky diode for heterodyne detection [33, 21]. Finally, the individual wafers are assembled by aligning and gluing them together. A.3 Theory The theoretical patterns of the monopulse configuration are easily obtained by modeling the horn cavity by a cascade of very small waveguide steps, and using modematching theory. The dipole antennas within the cavity are modeled by Hertzian dipoles and the resulting fields at the horn mouth are calculated using the detailed analysis presented in [92]. The difference antenna does not couple to the sum antenna due to mode orthogonality. The CPS transmission lines used for IF/video signal extraction and LO injection are neglected in this analysis, which is validated by measurements at microwave and millimeter-wave frequencies. The sum and difference antennas are located O.4A and 0.8A from the apex of the horn cavity. The corresponding cavity cross-sections are 0.6A and 1.1A, and are slightly larger than the waveguide cutoff dimensions for the TEo10 and TE20 modes, respectively. The input impedance and resonant antenna length of the sum antenna are found using the results of full-wave analysis [92]. The input impedance of the difference antenna was determined using microwave modeling, and its value was

122 adjusted by altering the CPS transmission lines and lengths of the asymmetrically coupled dipoles. A.4 Microwave Measurements Two microwave scale models were constructed for identical scaled designs at 1.2GHz and 3.3GHz. Input impedances and coupling between the antennas were first measured on the 1.2GHz model using an HP8720B network analyzer. The difference dipoles were constructed using flat strips of copper with a width of 0.02A and an overall length of 0.47A. The resonant input impedance of this configuration is measured to be 43Q (Fig. A.3(a)). The sum pattern antenna is located at a cavity opening of 0.57A. It consists of a dipole with width 0.02A and a length of 0.37A. The measured resonant impedance is 53Q, also shown in Figure A.3a. The sum and difference antennas were then measured together in the horn cavity. This had no effect on the input impedance measurements, and agrees with the high isolation measured between the two antennas (Fig. A.3(b)). A 3.3GHz microwave model of the monopulse was constructed for pattern measurement. This model included a metal ground plane which surrounded the horn mouth and extended several wavelengths in each direction. The sharp null of the difference pattern, below -30dB, and the broad peak of the sum pattern are as expected from theoretical analysis (Fig. A.4). Comparison of the calculated with measured patterns shows close agreement. The slightly lower left lobe of the difference pattern is a result of parasitics in the crossover network. The sum pattern agrees very well with previously developed theory and measurements [93]. Cross polarization in the E, H, and 450 is below -20dB and is not shown.

123 /rr Iat "`,,,,, = (a) ~30 C. 1 -40 _,,,,,,,,, ~, ~....-,~ / 1.1 12 1.3 1.4 Frequency (GHz) (b) Figure A.3:% (a) Measured 1.2GHz input impedances. (b) Isolation between. and A

124 -10./ a\ 1 ~ ~ meas.\ I\ t ^ Theory -= 0 i 0 9 A.5 Millimeter-wave Measurements A monopulse antenna was constructed with a design frequency of 92GHz which is ric sidelobes throughout the 90:94GHz range (Fig. A.5). The measured difference patterns exhibit good agreement with the theoretical patterns until 45~. Beyond at the center of the tuning range of the Gunn diode source used in the measurements. patterns exhibit good agreement with the theoretical patterns until 45,. Beyond that, the more rapid decrease in the measurements is caused by aperture blockage and are about 1 dB higher than expected due to aperture blockage. The sum antenna

125 /, (/ / I I v:X 0 I-2 A patterns. produces an E-plane pattern which is very similar to the H-plane pattern shown in Figure A.5. The slight asymmetry in the difference pattern is a result of parasitics from the crossover network. Cross polarization patterns were measured below -20dB for both sum and difference channels and are not shown. A.6 A Receiver Implementation The basic antenna design was then incorporated into a planar receiver design. Several modifications to the single-horn monopulse antenna were investigated after its performance was verified at millimeter-wave frequencies. The modifications improve the gain of the antenna and make it more suitable for use in a receiver. First, the layout of the sum and difference antennas was altered to accommodate a hybridmounted Schottky diode (Fig. A.6. This is inecessary because currently no methods exist to integrate semiconductor devices on thin dielectric membranes. For the differ

126 ence antenna, the center portion of the coplanar strip transmission line connecting the two dipoles was enlarged to form pads to which a beam lead diode could be attached (using silver epoxy 161]). The crossover network was moved to a position adjacent to one dipole. This introduces a small parasitic capacitance adjacent to the apex of the dipole, and thus a similar capacitance without a crossover was introduced near the apex of other dipole in order to maintain symmetry in the design. The sum antenna was widened in order to serve as the bonding pads for the diode, and its length was empirically adjusted to result in a favorable input impedance for the mixer diode. As before, microwave modeling was used to optimize and verify the antenna characteristics prior to fabrication, and showed that these changes did not severely affect the input impedances or patterns (Fig. A.3). At millimeter-wave frequencies, the measured difference patterns exhibit good agreement with the theoretical patterns with a null below -30dB. Cross polarization patterns were measured below -20dB for both sum and difference channels. The second modification made to the antenna was to attach a machined extension to the mouth of the horn in order to increase the directivity of the patterns. This extension introduces a change in both the flare angle and cross section at the integrated horn aperture. This excites two modes within the machined section and improves the gain while preserving the symmetry of the patterns, thereby improving coupling efficiency to a lens or reflector. A detailed analysis of this "dual mode" horn can be found in [94]. A machined horn extension for operation at millimeter-wave frequencies was fabricated and attached to the integrated horn antenna. The difference pattern measurements are shown in Fig. A.7. The null depth was measured at 92GHz to be below -26dB. Cross polarization remains below -25dB. The difference antenna generated similar patterns with sharp nulls below -23dB from 84GHz to

127 High Resistivity Silicmn Substrate C —--— c-ita-n —-.e Crossover 1.x Figure A.6: Modified difference antenna. The enlarged center pad allows a hybridmounted Schottky diode to be attached to the CPS line. The sum antenna did not- require modification. 94GHz, and appeared to be well matched to the diode throughout this range. A.7 LO and Mixer Design The difference antenna design was then incorporated in a planar subharmonic receiver design, shown in Figure A.8. The receiver design follows the procedures presented in Chapters II and III of this thesis. The 23GHz local oscillator is based on a hybrid-mounted GaAs FET in a common-gate reflection amplifier circuit using CPW transmission lines. This eliminates the need for via holes, which would be necessary in a microstrip circuit. Design of the coupler is based on work done in [95] and allows the oscillator to be easily monitored and phase-locked. The commonly used CPW-to- slotline transition is presented in [36] and has been verified using microwave modeling with this design. The 23GHz LO is not radiated by the dipole pair since the antenna impedance is very reactive at this frequency, which is below

128 3.56X' I I'' I. -20 _, I II I ] -; — -- A~ Without extennidn (Me..) _ - -A With Extension IMea&.)[ Machined (Theory) 0 _ ////////RgIntegrated -60 -20 20,,,,,,, 60,,,,,,,,,,, Degrees Figure A.7: Measured A pattern with horn extension at 92GHz, compared with theory. The difference pattern without the machined section is also shown for comparison. the cutoff of the horn cavity. An antiparallel Schottky diode pair was not readily available when the circuit was fabricated. Instead, a single diode, the DMK2784 manufactured by Alpha Industries [96] was mounted in the circuit and used to mix in 4th- harmonic mode. The diode has the following parameters: R- = 4/, Cjo = 30fF, Cp = 15fF, n = 1.07, Is = 1 x 10-16 Aand bi = 0.8V. The power delivered to the diode was estimated to be 2mW. This was done by activating the 23GHz LO, measuring the DC voltage developed across the diode under biased and unbiased conditions. A maximum DC voltage of 0.91V was measured across the diode when the diode was unbiased. This voltage dropped to 0.050V when a 50Q load was placed across the diode. These two voltages were corresponded to values produced by harmonic balance analysis of the diode under the same conditions, in order to arrive at an estimate of 2mW for the available LO power [30]. Conversion loss of the mixer in 4th-harmonic mixing mode was then

129 Modified Difference Antenna CPW-Slotline Transition Drain C+ L U ~~~~~~~~~~~~~~~~~~~Bias Antip alel Diode pair:10dB Coupler 23GHz Oscillator ~aractor Varactor Gate Bias Figure A.8: Layout of 94GHz sub-subharmonic integrated monopulse receiver, difference channel.

130 calculated to be about - 30dB, and agrees well with the measured conversion loss of -32dB~1.5dB at 94GHz. The conversion loss used here is defined as the measured IF power in a 50Q load divided by the power available at the dipole antenna terminals. The measured antenna directivity is used to deembed the antenna from the quasioptically measured conversion loss measurements. With adequate LO power, the harmonic balance analysis described in Chapter II predicts a conversion loss of -15dB when the single diode is replaced with a commercially available antiparallel diode pair (MA/COM MA40422). A.8 Discussion It is worthwhile to compare the single-horn and four-horn approaches. The singlehorn design obtains the monopulse patterns for a single plane at RF, by coupling to orthogonal modes in a single horn. This simplifies IF processing and eliminates the need to calibrate the IF processor. This is because the difference pattern null obtained in the single-horn design is inherent in the mode orthogonality of the antenna and consequently is more stable. However, to achieve phase coherence between the sum and difference channels, the two oscillators located on the sum and difference wafers must be locked together. Though this may be technically easy, it requires additional off-chip circuitry and complicates the design. Furthermore, it is difficult to squeeze an additional antenna into the horn in order to obtain the elevation difference pattern, limiting the single-horn design to tracking in one coordinate only. These difficulties justify pursuing the four-horn design. The single-horn design is elegant, and can be used in applications where tracking in only one coordinate is needed.

131 APPENDIX B Fabrication of Thin Dielectric Memrnbranes on a Silicon Substrate B.1 Determining Suitable Layer Thicknesses The dielectric membrane fabrication processes described in this appendix has been widely used to form the basis for a number of antenna and power sensor structures [68, 69, 21]. Their physical ruggedness and robustness allows them to be compatible with photolithography, photoresist processes, metallization, and hybrid components. Planar millimeter-wave antennas and circuits can then be fabricated on and supported by the dielectric membranes. The process described here is applicable only to silicon substrates, and requires two types of high-temperature furnace thin-films to be grown or deposited on the substrate. First, thermal SiO2 is grown in a standard oxidation furnace. A low-pressure chemical vapor deposition (LPCVD) furnace is then used for the deposition of an intermediate layer of Si3N4, and a top layer of SiO2. In general, when thin-film materials are grown or deposited on a substrate at a very high temperature, tensile and compressive stresses between materials build up as the sample is cooled to room temperature. This may be partly due of their different molecular structures, and also because the thermal coefficients of expansion for the

132 different materials are different from each other and from that of the substrate. TIhe different dielectrics used in the membrane (Si3N4 and SiO2) are necessary in order to result in a small amount of net tensile stress within the membrane. This keeps the membrane flat and allow it to be self-supporting, while preventing too much tension from rupturing the membrane. A general procedure for adjusting the tensile or compressive stress produced by a film is to deposit or grow the film on one side of the wafer only [82]. The curvature of the wafer is then carefully measured optically or with a surface profiling instrument (DekTak [45]). Since the Young's modulus and thickness of the silicon substrate is known (15 - 23 x 106 psi), the stress produced by the material can then be calculated, and the thermal coefficient of expansion for the material can be found if needed. Additional layers can then added, while the substrate curvature-is monitored in order to obtain an estimate of the contributions to the stress by each layer. Once these parameters are found, the thicknesses of layers can then be adjusted in order to change the sign or magnitude of the total stress of the 3-layer membrane. A practical method for doing this is to use the uppermost layer to determine the sign of the stress (compression or tension), and then to gradually reduce its thickness using etching until the desired stress is achieved. It is known from previous work that SiO2 and Si3N4 have thermal coefficients of expansion which are smaller and larger than that for silicon, respectively ([97]). In addition to this, these materials have very high mechanical strengths (with typical compressive strengths greater than 2 x 105 psi), and are commonly used insulators in IC fabrication. Initial choices for the thicknesses of the layers of a trilayer structure placed the membrane under compressive stress. The silicon wafer profile is shown in Figure B.1 (a) after 20 seconds of buffered HF etching. The top layer of oxide was then etched for an additional 20 seconds (Fig. B.1 (b)) to obtain a small amount of tensile stress.

133 The resultant thickness ratios of 5:3:4 for SiO2 (bottom layer): Si3N4: SiO2 (top layer) were then used in subsequent membrane fabrication. The equation for the residual stress in the membrane as a function of wafer deflection h on a wafer of radius r is given [23]: E, h d2 = 3(1 - v) 2 dB.1) where Es is the Young's modulus of the substrate, v is the Poisson ratio of the substrate, d, is the substrate thickness, and df is the film thickness. (1-,) is about 1.8 x 105 Nmm-2 for silicon, and the substrate thickness is 390p. For Fig. B.1 (a), this gives a residual compressive stress of - 0.15Nmm-2 for silicon, and a tensile +0.58Nmm-2 for (b). B.2 Thermal SiO2 Wafers are first subjected to a prefurnace cleaning detailed in Table B.1, in order to prevent contamination of wafers and furnace tube during the high-temperature oxide-growing process. The thermal oxides used in the membranes are grown at close to one ATM pressure. There are two types of thermally grown SiO2: wet and dry. The dry oxide grows at a much slower rate but yields a denser film. The wet process yields faster growth rates and more porous films. The membranes use a standard dry-wet-dry thermal oxide process which is part of standard CMOS processes. This means that thin layers of dense dry oxide layers are grown before and after a thick layer of wet oxide. A furnace temperature of 1100~C is typically used. Dry oxide is grown with an oxygen flow rate of 3L/min. For wet oxide, flow rates of 1.7 liters/minute for 02, and 2.5 liters/minute for H2 are burned to form steam which, at high temperatures, reacts with the silicon substrate to form SiO2. A 7000k-thick

134 1400 1200 0 L 800 400 U 600'4-'04 0 2 4 6 8 10 12 14 16 18 20 22 Position across wafer (rnm) (a) -500 -1000 -1500 C/ -2000 0 U -2500 -3500 -'.. -4000 0 2 4 6 8 10 12 14 16 18 20 22 Position across wafer (mrm) (b) Figure B.1: Substrate surface profiles for initial thicknesses of 5000A/3000A/7500A. After 20 seconds BHF etching (resulting in thickness of 4670A/3000A/7500A) (a), and 40 seconds BHF etch (resulting in thicknesses of 4330A/3000A/7500A) (b).

135 H202: NH30H: H20 1:1:5: 10-20 minutes at 90'C H20 quench 2 minutes HF room temperature, 30 seconds H20 quench 2 minutes H202: HCI: H20 4:4:25 10-20 minutes at 90~C H20 quench 5 minutes H20 rinse with N2 bubbling Wait for resistivity to exceed 13Mf-cm Table B.1: Prefurnace Clean layer typically takes about three hours to grow. Wafers are placed in and removed from the furnace at a temperature of 8000C and ramped to and from oxidizing temperature at a rate of about 5~ per minute. B.3 LPCVD Si3N4 Once the thermal furnace processing is completed, the wafers are then directly transferred to the LPCVD furnace, where Si3N4 and SiO2 are deposited. Deposition rates are typically between 50 and 80A per minute. The deposition parameters are listed in Table B.2. B.4 Comments The membranes are formed by defining a square opening in the membrane layers on the backside of the wafer, and etching the silicon using a solution of ethylene diamine pyrocatechol (EDP). The solution is heated to a temperature of about 100~, and the wafer is etched at a rate of about iltm per minute, until the substrate within the square opening has been entirely etched and only the membrane remains.

136 LPCVD Si02 tilt zone temperatures: 9100C, 9200C, 930 C N2 (dilute) 290sccm N20 120sccm DCS(Dichlorosilane) 60sccm LPCVD Si3N4 tilt zone temperatures: 8100C, 820~C, 830~C NH3 160sccm DCS 40sccm Table B.2: LPCVD Si02 and Si3N4 Deposition Parameters. Since EDP does not etch the membrane materials at a significant rate, the layer thicknesses remain close to their initial values, and the structure is not weakened. Membranes with a 5x5mm area and a thicknesses of 5000A/3000A/4000A had a yield of about 50%. This was significantly improved by increasing the layer thicknesses proportionally by 30%, and 7x7mm membranes have been fabricated with yields of 80-90%. If the Si3N4 layer is thicker than 4000A, the tensile stress may cause fissures to form in the membrane -during etching.

137 Membranes have also been successfully formed using an anisotropic silicon etch,onsisting of a solution of potassium hydroxide (KOH) with a concentration of 50g (OH: H20 at a temperature of 60~. This solution has an etch rate of about 0.5/1 per ninute, and is much less toxic and more easily prepared compared to EDP, and thus t is better suited for safe use in research environments. However, it etches Si02 at rate of about 7TA per minute and completely removes the top layer of oxide before he membrane is formed. The increase in tensile stress combined with the reduction a the thickness of the membrane results in a reduced yield to about 50% for 7x7mm rmembranes. With additional characterization, future membrane designs may omit he top layer of oxide in order to improve membrane yield with KOH etching.

138 APPENDIX C Harmonic Balance Analysis Supplemental The reflection algorithm was developed by Kerr et al. [24] as a way of modeling the way the nonlinear circuit is turned on. For the sake of simplicity, this appendix illustrates how the method works for one linear port and one nonlinear port, though a multiple-diode mixer can be analyzed with this method as well [30]. The reflection algorithm separates the linear and nonlinear circuits using a fictitious, lossless transmission line with a characteristic impedancd Z, and a length which is equal to integer multiples of a wavelength at the single-tone excitation frequency, or fundamental frequency (Fig. C.1). All frequencies existing in the circuit are multiples of the fundamental frequency, since the excitation voltage VLO enforces a periodic boundary condition on the circuit. Thus, the fictitious transmission line has no effect on the solution to the circuit. The value of Z, can be arbitrarily chosen, but is usually chosen to be equal to the characteristic impedance of the system in order to avoid numerical instabilities. The single-tone source is then activated and applied to the transmission line. The voltage wave v, propagates to the nonlinear device, and its voltage is applied across the device. Kirchoff's voltage law is applied, and solutions are found for the voltages across the junctions (vjm(t) for the mth junction) of the devices. This is done by applying the equations describing the current in the

139 Fr — 1, rFictitious transmission line ~I ~ Linear I subcircuit Zc ~~~~I I Vr I O L _ _ _ _ _ _ _ _ __ V_ i — nx Figure C.1: The equivalent circuit used in the reflection algorithm. device. A nonlinear differential equation is usually involved, since the device consists of nonlinear resistances, susceptances and reactances. The resulting current in the device is different from the incident current propagating in the transmission line and therefore a reflected wave v,,is triggered on the transmission line. The voltage of this reflected wave is written: Vr(t) = vi(t) - id(t)Zo (C.1) where id(t) is the current in the device. The reflected waveform contains harmonics of the single tone, and is transformed into the frequency domain. These harmonics propagates along the transmission line back to the linear circuit, where they reflect off the embedding impedance of the single-tone source. The embedding impedance of the source is in general different at each frequency. These source reflections are then combined with the signal of the source and sent along the transmission line to the nonlinear device. The process is repeated until the waveform across the device converges to a stable solution. The current and voltage across the device are used to calculate its large-signal impedance, while the junction voltages vjm are used to determine the time-varying junction impedances. Once the time-varying impedances

140 are obtained, mixer performance is determined via a small-signal analysis which is briefly outlined here: the Fourier components of the time-varying diode impedance can be used to write an Ohm's law-type relationship relating the current and voltage components at different frequencies. This can be reduced to a 2 x 2 conversion matrix involving only the RF and IF frequency components to find the conversion loss of the mixer. This analysis is thoroughly covered in section 4.3.3 of [30]. A thorough analysis of subharmonic diode mixers based on antiparallel Schottky diodes has been performed by Kerr [24]. This analysis is based on the assumption that the diodes are identical and looking at identical embedding impedances. Largesignal analysis is performed for a single diode placed in an embedding network with impedances which are doubled at the odd LO harmonics, and zero at even harmonics. This is justified because the LO source sets a boundary condition on the diodes with a period of one LO cycle (Fig. C.2). This means that at time to, diode 1 is subjected to identical electrical conditions compared to the conditions seen by diode 2 at time t180, except reversed in polarity. Thus, the currents produced by diode 1 and diode 2 cancel at even harmonics of the LO, and add at odd harmonics of the LO. Once the large-signal analysis is complete for a single diode, the resultant small-signal conversion matrix is doubled or zeroed at the appropriate elements to find the conversion matrix for the diode pair. The contribution of this thesis to subharmonic mixer analysis is to treat the two back-to-back Schottky diodes as a single nonlinear device. This is done by writing the nonlinear differential equation to include both diodes, each with its own parameters and parasitics. This results in two coupled second-order differential equations in two variables (vjl and vj2), which is solved numerically during the large-signal analysis. Once this is completed, the conversion matrix for the diode pair can be easily calcu

141 to VLO diode 2 diode 1 t 180 Figure C.2: Illustration of the symmetry used in Kerr's analysis. lated and used in the small-signal analysis. This analysis includes the effects of the parasitic inductance, capacitance and series resistance in the large- signal analysis. The reason for pursuing this alternative method for analyzing the diode pair is to first, verify the results of calculations using Kerr's method. Secondly, this method has the capability of looking at the effects of an asymmetric diode-pair. The diode equations are developed below, with voltages, currents and circuit elements defined in Figure C.3. Notice that the parasitic capacitance, which exists across both diodes, can be lumped together with the linear embedding impedances to simplify the nonlinear analysis. Each diode in the circuit contains two reactive elements (the series inductance L,, and the junction capacitance cjm for the mth diode). This results in two coupled second-order differential equations describing the junction voltages. First write the equation for voltages in loop 1, which includes the source representing the incident voltage wave vi: vjl + (ijl + cjlv')R81 + L,l(ij' + i'1) + (ijl + ij2 + i + ic2)ze = 2vi (C.2) A similar equation exists for loop 2:

142 Vj2 + (ij2 +- Cjlv)R L 2 + 2i2) + (ij2 + ijl + ic2 + icl)Zc = 2vi (C.3) where ijm and ic are the current flowing in the junction resistance, and junction capacitance for diode m, respectively. In order to obtain equations which only contain the junction voltages, write the currents in the junction resistor and capacitor of each diode in terms of the junction voltage, and find their derivatives. ij, io(eV l-/(nlkT/q) 1) (C.4) j -- nkT(ijl- io)vjl kTj (C.5) nkT (i71 nkTZjl Vl Zi1 = C1vjl + cjlvjl (C.6) vj=icl - 3 (Vjl)2 + CjlVjl (C.7) (C.8) where io is the reverse saturation current. Substituting equation C.8 into equation C.3 allows the following expression to be written for v": v2'l = C L ijlLs, +ijl(Rs,+Zc)+glLs,(vV.)2-2vi+vjl-+vjl (Rs +Zc)+Zc(ijz2+C~v2)] V1 - C1L 31 (C.9) where ijl and ijl are not expanded, for compactness. Equation C.9 expresses vj'l entirely in terms of the junction voltages and their derivatives. To solve for the junction voltage waveforms numerically at discretized time intervals, an initial guess for vjl, vj2, v21 and vj2 are needed. They are used to find vi1 and vj'2 using equation C.9. These are then used to find the junction voltages for the next increment in time. This process is carried out over several LO cycles, until the transient response in the

143 Loop 1 Zc Loop 2 Vi + FigureC.3: Diagram of the nonlinear circuit analysis for antiparallel Schottky diodes. solution has died out. Because the equation is second-order, adaptive time intervals are used to ensure accuracy and stability of the solution. The major disadvantage of this technique is clearly evident: it requires a larger number of computations in order to perform the large-signal analysis. For example, a typical unoptimized largesignal calculation running on a 25MHz 80486-based personal computer requires 1 minute to complete. When analyzing a single diode, such as in Kerr's analysis, the series inductance, like the parasitic capacitance, can be combined with the linear embedding impedances so that only a first-order nonlinear differential equation needs to be solved. This reduces the time of computation for a single diode to about 4 seconds on the same computer system. Typical waveforms calculated at 23GHz for the total current in a diode pair (using the diode parameters listed in Table 2.1 on page 31) are shown in Figure C.4. Results of a comparison between the symmetric method and this method show that the RMS error between the current waveforms is about 0.1%, and corresponds to agreement within about 10% for small-signal mixer analysis and 5% for LO impedances. This small difference in calculations is

144 0.01 QJ 0; -0.01 Kerr's method o0 - - Thesis -0.03 0.0 L 40 80 120 t (A.U.) Figure C.4: Typical waveforms for the total current in an antiparallel Schottky diode pair being driven by a single-tone large-signal excitation. insignificant, especially since measurement techniques at millimeter-wave frequencies have yet to obtain comparable accuracy. In conclusion, this appendix presents an alternative method for finding the performance of a subharmonic mixer based on antiparallel Schottky diodes. It has been used to successfully verify the results of previously developed analysis which assumed identical diodes and embedding impedances. Though it requires more computational power, it can be used to solve for asymmetric diode pairs, and successfully verifies the results produced by the Kerr analysis.

145 APPENDIX D IF Processing and Monopulse Accuracy The IF processor is constructed using commercially available 200MHz signal processing components purchased from suppliers such as Mini-Circuits [100]. The hybrid combiners and couplers typically have bandwidths of a few hundred MHz, and cost around $20 for small quantities. Blank, copper-clad printed circuit board (PCB) with a dielectric constant of 2.45 is available in thicknesses with imil (25.4jim) increments [101]. A 24mil thickness was selected so that the impedance of a lmm microstrip line is 50Q [35]. Holes are drilled in the PC-board, and components can be mounted on both sides to reduce the space needed. In a commercial design, the entire IF processing board can be integrated on a single chip, as is currently being done with global positioning system (GPS) processors [73]. The 94GHz monopulse receiver is mounted on a separate, smaller receiver PCB which contains two stages of low-noise IF amplifiers (Fig. D.1). This allows the receiver to be mounted in a mechanical gimbal mount for pattern measurements and tracking. The IF amplifier chain for each channel has a noise figure of 1.7dB and a gain of 45dB. The amplified IF signals are then sent via 50Q flexible coaxial lines (RG-174) to the main IF monopulse processing board. The IF signals from the receiver PCB must be amplitude and phase-trimmed because the four channels

146 5V 1 OV Cdc 200nH IF6 3! (from Cdc Cdc receiver) Cdc= 10OOpF 4700 Cdc Figure D.1: Low-noise 200MHz IF amplifier circuit having a noise figure of 1.7dB and a gain of 45dB. of the receiver each have slightly different mixer conversion losses and IF gains. Commercially available phase shifters were not available, so one was designed using lumped elements following the design described in [102] (Fig. D.2). The constructed circuit gives a phase trimming range of about 55~ when mechanically tuned variable 8pF capacitors are used. The values for the lumped element inductors and capacitors are obtained by experimentally adjusting the calculated values to compensate for the parasitic capacitance associated with the lumped element inductors [103]. The trimmed IF signals then enter the monopulse processor (Fig. D.3). This produces the elevation and azimuth difference signals, and the sum signal. The magnitude of these signals as a function of angle of incidence of a 94GHz plane wave are shown in Figure 3.26 of Chapter III. D.1 Amplitude Processing Logarithmic video amplifiers detect the monopulse difference and sum IF signals. The amplifiers have a dynamic range of 60dB, and produce a DC voltage which is linearly proportional to the log of the magnitude of the IF voltage detected, with

147 TFAS- 1 1Kf Cb=38pF __ Cv= 1-12pF - Cv Ca=6.6pF Lv= 15nH Ls=39nH Lr Lt=27nH Figure D.2: Amplitude and phase trim circuits for a single channel. The TFAS-1 is a matched variable attenuator. a specified proportionality constant A=6mV/dB and a maximum deviation from linearity of ldB (Fig. D.4). These DC voltages are then sent to two differential amplifiers, which produce two normalized difference voltages: V,i = KA(log(A,Il)- log()) = KA(log (D. 1) VaZ = KA(log(A,,) - log(Y)) KA(log( (D.2) where K is the amplifier gain and is selected to be 10 by setting the ratio 20R,/Rg. This results in a maximum voltage of 3.6 volts for a 60dB difference between the difference and sum signals. In practice. the output voltage will exceed 3.6V because the actual dynamic range of the amplifiers is greater than 60dB. Vei and Vaz are sampled by a DT2801 data acquisition board [104], which is controlled by an IBMcompatible personal computer. These voltages give the computer information about the target's proximity to the azimuth and elevation difference pattern nulls, as long as the target is within the peaks of the difference antenna pattern main lobes (+20~). Notice that an ambiguity remains about which side of boresight the target is located.

148 Cd ciao ~ —------- -10@ 1-arm3~~...... 1 I I I. =_ouel otrr CIL3 i E Mi M,. U..I IA O - z W, LPS-10S pe: d>-I: I Ull OUT Figure D.3: The IF monopulse processor, which processes the signals from the four channels after amplification (Fig. D.1) and amplitude and phasetrimming (Fig. D.2). ['~ an-I~~~~~~~~~~,...... 3~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~l~llm 33.i~~~~~O~ OUT1

149 The boresights of each of the antennas in the monopulse receiver are parallel, and thus phase processing is required to resolve this ambiguity. D.2 Phase Processor The phase processor consists of two phase discriminators (PD), corresponding to the two coordinates (elevation and azimuth). The circuit of a single PD is shown in Figure D.5. The PD takes two input signals, A sin(a + jwt) and B sin(3 + jwt) and produces two pairs of outputs, which are square-law detected (Fig. D.6) and RC-filtered, leaving only the DC (video) terms: V1 A2 + B2 + AB cos(a-3 ) (D.3) V2 = A2 + B2- AB cos(a- ) (D.4) V3= A2 + B2 + AB sin(a- /i) (D.5) V4= A2 + B2 - AB sin(ac- -) (D.6) These voltage pairs are then sent to differential amplifiers, which output the even term Ka 2AB cos(ca - ) and the odd term K,*- 2ABsin(ca - ). A similar set of voltages is produced for the elevation coordinate. Only the odd term is needed to locate the target. However, it is not normalized, and cannot be logarithmically detected, as in the case of the amplitude processor signals. Therefore in a commercial system, the phase discriminator must be used with automatic gain control in order to produce voltages within the operating range of the digital sampling circuits. The IF processor sets K4 to be large enough that the differential amplifier saturates when ar -1 differs from O by a small amount when the incident RF power density is above an arbitrary level. Thus the phase processor

150 12t Rs 14 18 IN+ C)~~~~~~~~~1 1,~~~~~~1 Output Z Vin i Rg PM AMP-01 -- 2S / R J Vout= G Vin _ f<~ If G- 20Rs Rg O~~~~~~~:OOK+,OOK! -12V 12V _ 4 17 44 g Tech 19 IF * CLA3-200/50A Vout 1.2.27,28,38 39.43'45 Figure D.4: Circuit diagrams for a differential amplifier (above) and an IF logarithmic amplifier (below).

151 K 2AB sin(a-iB) Odd output Differential Amplifier V3 V4 Square-law I detectors I 0' 90' Anaren 1-H0262-3 PSCJ-2-1 IN ISO PSCQ-2-250 IN 0' 0' IN 5 A sin(a) B sin(/\ 180'6 6 90' o' o. 90. IN ISO Anaren 1-H0262-3 ~I[. ______ Square-law I Square-law! detectors I V1 V2 Differential Amplifier Even output K 2AB cos(a-B) Figure D.5: Circuit diagram for the phase discriminator.

152 4:1 1000pF Input 3 6800 To differential Coicraft TIF Lamplifier Metelics Coilcraft MSS2005S-E26 WB1040PC Zero-bias diode Figure D.6: Square-law detector with a sensitivity of 5600mV/mW at 200MHz. is used to give the polarity of the amplitude information. The data shown in Figure 3.27 of Chapter III is taken directly from the unsaturated output of the differential amplifier for the odd term, and is 4D = K6 - 2AB sin (a - /), where A and B are the voltage amplitudes of the sum and difference signals respectively, and a - d is the phase difference between the sum and difference signals. D.3 Monopulse Error Due to Noise A simple analysis gives a rough estimate for the resolution limit of a monopulse antenna system operating in the presence of noise, for a target on boresight. The normalized rnonopulse voltage slope km on boresight can be approximated by the reciprocal of 3dB half-width of the monopulse difference pattern, km = 2/bw, (Fig. D.7). Thus a noise voltage v, in the presence of a signal voltage v, introduces an angle error signal vn/(km. v). It is evident that v,/v, is simply the voltage signal-to-noise ratio (VSNR). The power signal-to-noise ratio (PSNR) is the square of the VSNR, so the angular error introduced by the noise voltage, Utheta, is equal to 1h/(Em PSR) = Ob-/(2IPI 2Vi). A more complete analysis, which is outside of the research presented in this thesis, has been performed by Sherman [71].

153 2 /I \'I i/ / 4 Component List difference voltage pattern. D.4 Component List

154 Active Components Part Number Supplier NE32100 Low-noise HEMT NEC [39] UPC1658 Low-noise VHF/UHF silicon amplifier UPC1677 Medium-power silicon amplifier FLRO16XV Medium-power K-band GaAs FET Fujitsu [53] PMI AMP-01 Instrumentation amplifier Precision Monolitic Inc. [105] CLA3-200/50 Detection log video amplifier Log Tech, Inc. [106] Table D.1: Active components list.

155 Passive Components Part Number Supplier TFAS-1 variable attenuator Mini-Circuits [100] LPS-109 power splitter PMT-1 0-180~ power combiner PSCQ-2-250 0-90 power combiner PSCJ-2-1 0-180~ power combiner 1-H0262-3 0-90~ hybrid combiner Anaren Microwave, Inc. [107] WB104OPC RF 4:1 Transformer Coilcraft [103] 1008CS-150XJBB chip inductor 1008CS-390XJBB chip inductor 1008CS-270XJBB chip inductor 1008CS-102XJBB chip inductor PCC100CN chip capacitor Digi-Key Corporation [108] PCC39OCN chip capacitor PCC060CN chip capacitor PCC102CN chip capacitor MBC50-10B12 beam lead capacitors Metelics [44] MSV-34064-C11 chip varactors MSS2005S-E26 zero-bias detector diode MSTF-3-s-N-100-01 chip resistor Mini-Systems, Inc. [59] Table D.2: Passive components list.

156 APPENDIX E Bond Wire Characterization The equations provided here are obtained from a journal article in the trade magazine Microwaves & RF and give an estimate for the inductance and series resistance of a bond wire valid.up to 40GHz or higher, as long as the length of the bond does not exceed 0.1A 198]. The equations assume cylindrical bond wires with diameter d. They account for the effects of skin depth, as well as finite, varying distance over a ground plane. The inductance for a straight bond wire of length I in free space is L -=!i{ln[(21/d)+ (1 + 21/d)2)'/2] + d/21-(1 + (d/21)2)12 + ISq } (E.l) s 0.25tanh(4d./d) (E.2) = ds f (E.3) where d,, p and Sl, are the skin depth, resistivity and relative magnetic permeability of the bond wire material, respectively, and f is the frequency in Hertz. The presence of a ground plane (Fig. E.1) reduces the net inductance of the bond wire, and can be written as Lg = L- M(2h) (E.4)

157 d d5 G///ro und Plane'//////////// s=2h Image Figure E.1: Bond wire over a ground plane. s = 2h is the center-to- center separation between the bond wire and its image. where h is the height of the bond wire over the ground plane and M(2h) is the mutual inductance due to the ground-plane image of the wire. The mutual inductance for a pair of straight bond wires carrying equal currents, with a center-to center separation of s is written as [99] M(s) =- 2 {(ln[(l/s) + (1 + I/s)2)1/2] + s/- (1 + (s/1)2)1/2} (E.5) 27r where M(s) is added if the currents in the wires are in the same direction, and subtracted if they are in opposite directions. To account for a finite-resistance ground plane, h can be replaced by h' = h + 4.6d,, since most of the RF current is confined within 4.6 skin depths of the surface. In general, the bond wire is not parallel to the surface, so h is replaced by ha,, the average height of the wire over the surface, which is found by simple integration. Similarly, two parallel wires carrying current in the same direction with a centerto-center distance s, located a distance h above a ground plane have a total inductance of Lg2 = [L- M(2h) + M(s)]/2 (E.6)

158 where +M(s) is the mutual inductance of the two wires, and -M(2h) is the mutual inductance due to the ground plane images. Finally, equations for the DC and RF resistance of the bond wire at DC and RF can be written: Rdc- d2 (E.7) Rs = FpRdccosh[0.041(d/ds)2], d/ds < 3.394 (E.8) Rs = FpRdc(0.25d/d, + 0.2654), d/ds > 3.394 (E.9) Fp = 1.0 + 0.8478e-0.9435s/d, sd > 1. (E.10) where Fp accounts for the change in current distribution in the wire caused by magnetic flux of the wires, and is more pronounced when ds << d. R, is added in series with the net inductance calculated using the previous expressions. The S-parameters measured by the manufacturer include two bond wire connections at each device terminal, with lengths of 3251am for the gate, 240um for the drain and 178pm for the source. An additional 100/um is added to the wire lengths to account for the height of the transistor, which is surface-mounted instead of lodged in a trench (Fig. E.2). The calculated parasitic inductance for a 0.7 mil gold bond wire used to attach the NE32100 HEMT used in the VCO is calculated by assuming two bond wires with center-to-center separation of 125pum, and using d = 20ym, p = 2.4410-8Q-cm, 1 = 100Lm, and an average height above a ground plane of 50pum. The resulting additional inductance of the two wires is calculated to be 0.03nH. The inductance of the bond wires connecting the gate of the varactor to the gate transmission line also needs to be accounted for. A length of 350,um with an average height of 50/um above a ground plane was used to calculate a value of O.llnH and a negligible series resistance of 0.261Q, at 23GHz.

159 320 (a)' ---!r1o00 (b) | 150 11 Figure E.2: Diagram illustrating the dimensions used in the VCO bond wire calculations. Manufacturer's test fixture with HEMT residing in a 150pm-deep trench (above). Surface-mounted HEMT used in the receiver (below).

BIBLIOGRAPHY

161 BIBLIOGRAPHY 1] Stiglitz M. R. and Blanchard, C., "A Laser Radar Experiment," Microwave Journal, pp.57-60, September 1990. 2] Nettleton, J. E., Rohde, R. S. and Buser, R. G., "Active/Passive Infrared Imaging," Proceedings, Fifth Conference on Coherent Laser Radar: Technology and Application, SPIE-The International Society for Optical Engineering, 5-9 June, 1989. 3] Hu, B. B., Zhang, X., Froberg, N. M., "THz Radiation from a Photoconducting Antenna Array," IEEE J. of Quantum Electronics, Oct. 1992, pp. 2291-301. L] Webb, A. R., "Point Source Location using a Millimeter-wave Focal Plane Array Radar," IEE Proc. Part F, Radar and Signal Processing, Dec. 1991, pp. 581-91. i] Goldsmith, P. F., Novak, G., Erickson, N. R., "A 15 Element Focal Plane Array for 100GHz," IEEE Microwave Theory Tech., Jan. 1992, pp. 1-11. i] Lipsky, S. E., Microwave Passive Direction Finding, John Wiley & Sons, Inc., 1987. i] Biernson, G., Optimal Radar Tracking Systems, John Wiley & Sons, Inc., 1990. ] Lee, K. M., Ruey-shi, C. "Design and Analysis of a Multimode Feed Horn for a Monopulse Feed," IEEE Transactions on Antennas and Propagation, pp. 171181, Feb. 1988. i] Newman, J., Jackson, C. M., "A low-cost Ka-band microstrip patch monopulse antenna," Microwave Journal, pp. 125-126, July, 1987. )] Jackson C. M., and J. Newman, "Low Cost, Ka-Band Microstrip Patch Monopulse Antenna," Microwave Journal, pp. 125-131, July 1987..] Watson C. K. and Ringer K., "Feed Network Design for Airborne Monopulse Slot Array Antennas," Microwave Journal, pp. 129-145, June 1988. ] Marsarweh, S. K., Sherer, T. N., Yngvesson, K. S., Gingras, R. L., Drubin, C., Cardiasmenos, A. G., and Wolverton, J. "Millimeter-Wave Slot Ring Mixer Array Receiver Technology,' IEEE 1992 Microwave and Millimeter-Wave Monolithic Circuits Symposium Digest of Papers, June. 1992.

162 [13] Rutledge, D. B., Neikirk, D. P., and Kasilingam, D. P., "Integrated Circuit Antennas," in Infrared and Millimeter-Waves, vol. 10, pp.l-90, K. J. Button, Ed. New York: Academic Press, 1983. [14] Katehi, P. B., Alexopoulos, N. G., "On the Effect of Substrate Thickness and Permittivity on Printed Circuit Dipole Properties," IEEE Trans. Antennas Prop., pp. 34-38, Sept. 1983. [15] Kormanyos B. K., Ostdiek, P. H., Bishop, W. L., Crowe, W. T., and Rebeiz, G. M., "A Planar Wideband 80-200GHz Subharmonic Receiver," Accepted for publication in the October 1993 IEEE-MTT Special Issue on Quasi-Optical Techniques, Sept., 1992. t16] Zah, C. E., Rutledge, D. B., "A polystyrene for matching a silicon lens at millimeter wavelengths," Int. J. Infrared Millimeter Waves, vol. 6, pp. 909-918, Sept. 1985. t17] Kerr, A. R., Siegel, P. H., and Mattauch, R. J., "A simple quasioptical mixer for 100-120 GHz," IEEE-MTT Int. Microwave Symp. Digest, pp. 96-98, 1977. [18] Zmuidzinas, J., "Quasi-optical slot antenna SIS mixers," IEEE Trans. Microwave Theory Tech., vol. 40, pp. 1797-1804, Sept. 1992. [19] Rogers, R. L., Neikirk, D. P., "Radiation properties of slot and dipole elements on layered substrates," Int. J. Infrared and Millimeter Waves, vol. 10, pp. 697728, 1989. [20] Wentworth, S. M., Rogers, R. L., Hetson, J. G., Neikirk, D. P., and Itoh, T., "Millimeter Wave Twin Slot Antennas on Layerd Substrates," Int. J. Infrared and Millimeter Waves, vol. 11, pp. 111- 131, 1990. [21] Ali-Ahmad, W. Y., Bishop, W. L., Crowe, T. W., Rebeiz, G. M., "An 86-106GHz Quasi-Integrated Low Noise Schottky Receiver," to be published in the IEEE Trans. Microwave Theory Tech., April 1993. t223 Ekstr6m, H., Gearhart, S., Acharya, P. R., Rebeiz, G. M., Kollberg, E. L., Jacobsson, S. "348-GHz Endfire Slotline Antennas on Thin Dielecdtric Membranes," IEEE Microwave and Guided Wave Letters, pp. 357-358, Sept., 1992. [23] Rebeiz, G. M., Monolithic Millimeter-Wave Two- Dimensional Horn Imaging Arrays,pp. 2-23, California Institute of Technology, Pasadena, California, 1988. [24] Kerr, A. R., " Noise and Lose in Balanced and Subharmonically Pumped Mixers: Part I-Theory," and "Part II-Experiment," IEEE Trans. Microwave Theory Tech., vol. MTT-27, no. 12, pp. 938-950, December 1979. [25] McMaster, T. F., Schneider, M. V. and Snell, W. W., "Millimeter-wave receivers with subharmonic pump," IEEE Trans. Microwave Theory Tech., vol. MTT-24, pp. 948-952, Dec., 1976.

163 [26] Sze, S. lM., High-Speed Semiconductor Devices, pp. 576-577, John Wiley & Sons, Inc., 1990. [27] Strohm, K. M., Schaffler, F., Guttich, U., "A D-Band Suharmonic Mixer with Silicon Planar Doped Barrier Diodes," IEEE Trans. Microwave Theory Tech., pp. 366-8, Feb. 1991. [28] Mattauch, R. J., Roser, H. P., and Crowe, T. W., "GaAs Schottky Diodes for THz Mixing Applications," Proceedings of the IEEE, pp.1827-41, Nov. 1992. t29] Maas, S. A., Nonlinear Microwave Circuits, Artech House, Inc., 1993. [30] Maas, S. A., Microwave Mixers, Artech House, Inc., 1993. [31] Held, D. N. and Kerr, A. R., "Conversion Loss and Noise of Microwave and Millimeter-wave Mixers: Part I-Theory," and "Part II- Experiment," IEEE Trans. Microwave Theory Tech., vol. MTT-26, pp. 49-61, Feb. 1978. [32] Gearhart, S. S., Ling, C. C. and Rebeiz G. M., "Integrated 222GHz cornerreflector antennas," Microwave and Optical Tech. Lett., 5 Jan. 1991 [33] Ling, C. C. and Rebeiz, G. M., "Advances in 94GHz Integrated Monopulse Antennas," accepted for the 1992 Joint Symposia, Chicago, Illinois, July 1992. [34] Eleftheriades, G. V., Analysis and Design of Integrated-Circuit Horn Antennas for Millimeter and Submillimeter- Wave Applications, The University of Michigan, Ann Arbor, Michigan, 1993. t35] Gupta, K. C., Garg R., Bahl, I. J., Microstrip Lines and Slotlines, Artech House, Inc., Norwood, MA, 1979. [36] Cahana, D., "A New Coplanar Waveguide/Slotline Double- Balanced Mixer," IEEE Microwave Theory Tech. Symposium Digest, pp. 967-70, 1989. [37] Grammer, W., Yngvesson, K. S.,"Coplanar Waveguide Transitions to Slotline: Design and Microprobe Characterization," to be published in IEEE Trans. Microwave Theory Tech., Special Issue on "Modeling and Design of Coplanar MMICs," September, 1993. [38] Selectilux HTR 3 Photosensitive Polyimide, Ciba-Geigy Corporation, 16033 South Twelfth Place, Phoenix, AZ 85044, (602) 460-1311. t39] California Eastern Laboratories, 6823 Hobson Valley Drive, Suite 203, Woodridge, IL 60517, (708) 241-3040. [40] Sze, S. M., Physics of Semiconductor Devices,, pp. 585- 590, John Wiley & Sons, Inc., 1981. [41] Boyles, J. W., UThe Oscillator as a Reflection Amplifier: an Intuitive Approache to Oscillator Design," Microwave Journal, pp. 83-98, June, 1986.

164 [42] EEsof, Inc, 5601 Lindero Canyon Rd., Westlake Village, CA 91362-4020. [43] Wedge, S. W., Compton, R., Rutledge, D., Puff: Computer Aided Design for Microwave Integrated Circuits, Version 2.0, copyright 1991, Puff Distribution, Electrical Engineering M/S 116-81, California Institute of Technology, Pasadena, CA 91125. [44] MBC50-10B12 beam lead capacitors, MSV-34064-Cll chip varactors, Metelics, 975 Stewart Ave., Sunnyvale, CA 94086, (408) 737-8181. [45] DekTak Surface Profile Measurement System, Sloan Technology Division, Veeco Instruments, Inc., 3350 Scott Blvd. Suite 3902, Santa Clara, CA 95054, (408) 982-0600. [46] Adar, A., Ramachandran, R., "An HBT MMIC Wideband VCO," IEEE Microwave Theory Tech. Symposium Digest, pp. 247-250, 1991. [47] "Engen, G. F., Hoer, C. A., " Thru-Reflect-Line: An Improved Technique for Calibrating the Dual Six-Port Automatic Network Analyzer," IEEE Trans. Microwave Theory and Techniques, December 1979. [48] "Network Analysis: Applying the HP 8510 TRL calibration for non-coaxial measurements," Product Note 8510-8, Hewlett Packard Corporation, [49] 305K-200 Microprobes, Cascade Microtech, Inc., 14255 Brigadoon Court, Beaverton, OR 97005, (503) 626-6023. [50] REL-4300 Probe Station, Alessi Inc., 35 Parker St., Irvine, CA 92718, (714) 830-0660. [51] Haydl, W. H., Kitazawa, T., Braunstein, J., Bosch, R., Schlechtweg, M., " Millimeterwave Coplanar Transmission Lines on Gallium Arsenide, Indium Phosphide and Quartz with Finite Metalization Thickness," IEEE Microwave Theory Tech. Symposium Digest, pp. 691-694, 1991. [52] Becker, J. S., Jager, D., "Electrical Properties of Coplanar Transmission Lines on Lossless and Lossy Substrates," Electronic Letters, pp.88-91, 1 Feb. 1979. [53] Fujitsu Microelectronics Inc., Microwave and Optoelectronics Division, 7001 North Locust St., Suite 231, Kansas City, MO 64118, (816) 436-9404. [54] Omar, A. A., Chow, Y. L., Roy, L., and Stubbs, M. G.,"Effects of Air-Bridges and Mitering on Coplanar Waveguide 90~ Bends: Theory and Experiment," IEEE Microwave Theory Tech. Symposium Digest, pp. 823-824, 1993. [55] Simons, R. N., Ponchak, G. E., Martzaklis, K. S., and Romanofsky, R. R., "Experimental Investigations on Channelized Coplanar Waveguide," NASA Technical Memorandum 102494, April 1990.

165 [56] Ponchak, G. E., Katehi, L. P. B., Dib, N, I., "A Theoretical and Experimental Study of Coplanar Waveguide Shunt Stubs," IEEE Microwave Theory Tech., pp. 38-44, January 1993. [57] Soghomonian, M. S., and Robertson, I. D., "Microwave Modeling o fCurved Coplanar Waveguides," International Journal of Microwave and MillimeterWave Computer-Aided Engineering, pp. 126-133, vol. 3, 1993. [58] Pozar, D. M., Microwave Engineering, pp. 412-15, Addison-Wesley Publishing Company, Inc., 1990. [59] MSTF-3-S-N-100-01 nichrome-silicon 100Q1% tol. chip resistor, Mini-Systems, Inc., Thin Film/Thick Film Division, 20 David Road, North Attleboro, MA 02761, (508) 695-0203. [60] QuickCure 5 Epoxy, Rider's Hobby Shop, 115 West Liberty Street, Ann Arbor, MI 48104. [61] Epotek H20E, Epoxy Technology, Inc., 14 Fortune Drive, Billerica, MA 01821. [62] Zah, C., "Millimeter-wave Monolithic Schottky Diode Imaging Arrays," California Institute of Technology, Pasadena, 1986. [63] Elliott R. S., Antenna Theory and Design, Prentice Hall, New York, 1981. [64] Kreisler, A., Pyee, M., and Redon, M., "Parameters Influencing for Infrared Videodetection with Submicron-Size Schottky Diodes," In. J. of Infrared and Millimeter Waves, vol. 5, No. 4, pp. 55999-584, 1984. [65] Balanis, C., Antenna Theory: Analysis and Design, pp. 33, John Wiley & Sons, New York, 1982. [66] Guo, Y., Lee, K., Stimson P., Potter K. A., and Rutledge D. B., "Aperture efficiency of integrated-circuit horn antennas," Microwave and Optical Tech. Lett., vol. 4, pp. 6-9, Jan. 1991. [67] Anritsu America Inc., 15 Thornton Rd, Oakland, New Jersey, 07436; HewlettPackard Co., Test & Measurement Catalog, 3200 Hillview Ave., Palo Alto, CA 94304; Hughes Millimeter-Wave Products, P.O.Box 2940, Torrance, CA 905092940. [68] Rebeiz, G. M., Kasilingam D. P., Stimson P. A.,Guo Y. andRutledge D. B. "Monolithic Millimeter-Wave Two-Dimensional Horn Imaging Arrays," IEEE Trans. Antennas Propag., vol. AP-28, Sept 1990. [69] Ling, C. C. and Rebeiz, G. M., " A Wide-Band Monolithic Quasi-Optical Power Meter for Millimeter- and Submillimeter-Wave Applications," IEEE Transactions on Microwave Theory Tech., pp. 1257-1261, Aug., 1991.

166 [70] Ali-Ahmad, W. Y., Millimeter and Submillimeter-wave Integrated Horn Antenna Schottky Receivers, The University of Michigan, Ann Arbor, Michigan, 1993. [71] Sherman, S. M., Monopulse Principles and Techniques, Artech House, MA, 1984. [72] Dow, G. S., Tan, K., Ton, N., Abell, J., Siddiqui, M., Gorospe, B. Streit, D., Liu, P., Sholly, M., "Ka-Band High Efficiency 1W Power Amplifier," IEEE Microwave Theory Tech. Symposium Digest, pp. 579-581, 1991. [73] Herman, R. M., Chao, A., Clinton, H. M., Pulver, J. R., "An Integrated GPS Receiver with Synthesizer and Downconversion Functions," IEEE Microwave Theory Tech. Symposium Digest, pp. 883-886, 1991. [74] Moss, D. G., Birsch, J. R., Adamson, D. A., Lunt, B., Hodgetts, T., Wallace, A., "Comparison Between Free Space and In-Waveguide Power Measurement Standards at 94GHz," Electronics Letters, pp. 1134-1137, June, 1991. t75] Thomas Keating Ltd., Billingshurst, West Sussex, England. [76] Yamamura, K., Inoue, T., and Yokoshima I., " A Power Measurement Standard for W-Band a:t 94GHz," IEEE Transactions on Instrumentation and Measurement, pp. 227-279, June, 1988. [77] Ghandi, K., Mandelis, A., Peralta, X. B., "Optical power monitor using a thinfilm pyroelectric bimorph," Review of Scientific Intruments, pp. 1038-43, Mar. 1990. [78] Foot, A., Hodges, B. and Dyson, C., "Calibration of Power and Energy Meters for the FIR/Near MMW Spectral Regions," Int. J. of Infrared and Millimeter Waves, vol. 5, No. 4, pp. 773-782, 1981. t79] Block, W. H. and Gaddy, O. L., "Thin metal-film room- temperature IR bolometers with nanosecond response times," IEEE J. Quantum Electron., vol. QE-9, No. 11, 1973. [80] Peterson, K. E., "Silicon as a mechanical material," Proc. IEEE, vol. 70, pp. 420-457, 1982. [81] Neikirk D. P.,Lam W. W., and Rutledge D. B., "Far-Infrared Microbolometer Detectors," Int. J. of Infrared and Millimeter Waves, Vol. 5, pp. 245-276, 1984. [82] Maissel, L. I., and R. Glang, Handbook of thin film Technology, pp., McGraw Hill, New York, 1970. [83] T. B. A. Senior, K. Sarabandi, and F. T. Ulaby, "Measuring and modeling the backscattering cross section of a leaf," Radio Science, Vol. 22, pp. 1109-1116, Nov. 1987.

167 [84] Natzke, J., Volakis, J. L., "Diffraction by Parallel Resistive Half Plane Structures," accepted for publication in IEEE Trans. Antennas Prop. [85] E. Yoon and K. D. Wise, "A dielectrically-supported multi- element mass flow sensor," Proc. IEDM-88, San Francisco, Dec. 1988. [86] Lee K. A., Guo Y., Stimpson P. A., Potter K. A. and Rutledge D. B., "Thin-film power-density meter for millimeter wavelengths," Submitted for publication to the IEEE- Trans. Antennas Propag.. [87] Ramo, S., Whinnery J. R., and Van Duzer, T., Fields and Waves in Communication Electronics Second Edition, ch. 6, Wiley and Sons, 1984. [88] Johnson R. C., Jasik, H. Antenna Engineering Handbook, Chapter 15, Mc-Graw Hill, New York, 1984. [89] Johnston, S., Fourier Transform Infrared: A constantly evolving technology, New York, E. Horwood, 1991. [90] Bomem, Inc., 450 Ave. St.-Jean-Baptiste, Quebec, PQ G2E 5S5 Canada (418) 877-2944 [91] Goldsmith, P. F., "Quasi-Optical Techniques at Millimeter and Submillimeter Wavelengths, " Infrared and Millimeter Waves, Vol.6, pp. 277-343 [92] Eleftheriades, G. V., Ali-Ahmad, W. A., Katehi, L. P. and Rebeiz, G. M., " Millimeter-wave integrated-horn antennas, Part I: Theory, " IEEE Trans. Antennas Propag., November 1991. [93] Eleftheriades, G. V., Ali-Ahmad, W. A., Katehi, L. P. and Rebeiz, G. M., Millimeter-wave integrated-horn antennas, Part II: Experiment, " IEEE Trans. Antennas Propag., November, 1991. [94] Eleftheriades, G. V., Ali-Ahmad, W. A. and Rebeiz, G. M., "A 20dB QuasiIntegrated Horn Antenna," IEEE Microwave and Guided-Wave Letters, February, 1992. [95] Wen, C. P., "Coplanar-waveguide Directional Couplers," IEEE Trans. Microwave Theory Tech., June, 1970 [96] Alpha Industries, Inc., Semiconductor Division, 20 Sylvan Road, Woburn, MA 01801, (617) 935-5150. [97] Maissel, L. I., and R. Glang, Handbook of thin film Technology, pp.29-30, chapter 6, McGraw Hill, New York, 1970. [98] March, S. M., "Simple Equations Characterize Bond Wires," Microwaves & RF, pp. 105-110, Nov. 1991.

168 [99] Rosa, E. B., Grover, F. W., "Formulas and Tables for the Calculation of Mutual and Self Inductance," Bulletin of the Bureau of Standards, Vol. 8, No. 1, January, 1912. t100] Mini-Circuits, P.O. Box 350166, Brooklyn, NY 11235-0003 (718) 934-4500. [101] NVF Co., Multilayer Products Div., Kennett Square, PA 19348. [102] Daughters, T. E., "Lumped Elements Improve Phase-Shifter Design," Microwaves & RF, pp. 89-100, Jan. 1992. [103] Coilcraft, 1102 Silver Lake Rd., Cary, IL 60013, (708) 639-6400. [104] DT2801 general purpose data aquisition board, Data Translation, Inc., 100 Locke Drive, Marlboro, MA 01752-1192, (617) 481-3700. [105] Precision Monolithics, Inc., 1500 Space Park Drive, Santa Clara, CA 95054, (408) 727-9722. [106] Log Tech, Inc., 3529 Old Conejo Rd., Suite 123, Newbury Park, CA 91320, (805) 499-6462. [107] Anaren Microwave Inc., 6635 Kirkville Road, P.O. Box 178, E. Syracuse, NY 13057, (315) 432-8930. 1108] Digi-Key Corporation, 701 Brooks Ave., South, P.O. Box 677, Thief River Falls, MN 56701-0677, (800) 344-4539.