ON-CHIP ELEMENT AND ARRAY DESIGN FOR SHORT RANGE MILLIMETER-WAVE COMMUNICATIONS

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the

Graduate School of The Ohio State University

By

Rudy M. Emrick, B.S. M.S.

*****

The Ohio State University

2007

Dissertation Committee: Approved by

John L. Volakis, Adviser Mohammed Ismail Adviser Chih-Chi Chen Graduate Program in Robert Lee Electrical and Computer Engineering c Copyright by

Rudy M. Emrick

2007 ABSTRACT

Large amounts of worldwide unlicensed spectrum at 60 GHz is currently being considered for high speed solutions. However, a number of challenges remain for this spectrum to be a viable solution for high volume consumer applications. In this dissertation we look more closely at requirements for indoor antenna connectivity with particular focus on the signal to noise ratio needed to overcome fading in multi- path channels. A new analytical channel model, including multipath scattering, is proposed and adapted to determine antenna requirements. These requirements are then used to develop realistic signal to noise ratios for Silicon-based Frequency

(RF) front ends. This dissertation considers three candidate antennas that show promise for compact on-chip implementation. Given their small size and possible losses at millimeter wave frequencies, we also focus on antenna efficiency for practical metalizations on Silicon. Therefore, relevant material properties are examined to determine the most accurate parameters to be used in the computational models.

It is concluded that arrays of the candidate antennas with spatial power combining must be employed, but are still small enough for on-chip realization. The proposed antenna array that meets performance requirements is as little as 7x7mm2, making it about 1/3 of the target maximum size of 25x25mm2, required to enable integration as part of a portable consumer devices.

ii ACKNOWLEDGMENTS

Many thanks to my wife Rita and two sons, Sam and Josh. Without their patience, understanding and assistance, my completion of this degree would not have been possible. I would also like to thank my advisor John Volakis, who was extremely helpful and understanding in my being a non-traditional student. I also greatly appreciate the help of George Simpson and Bob Neidhard from the Air Force Research

Lab for acquiring probes and taking the measurements which are included as part of this work. In addition, I could not have succeeded without the generosity and solid support from Motorola and my manager Vida Ilderem.

iii VITA

1991 ...... B.S. Electrical Engineering, Michigan Technological University 1994 ...... M.S. Electrical Engineering, Ohio State University 2005-present ...... Graduate Student, Ohio State Univer- sity

PUBLICATIONS

Research Publications

R.M. Emrick and J.L. Volakis “Antenna Requirements for Short Range High Speed Wireless Systems Operating at Millimeter-Wave Frequencies.”. IEEE International Microwave Symposium Digest, pp.974–977, 2006.

R.M. Emrick and J.L. Volakis “Millimeter-Wave and Terahertz Antennas”. Antenna Engineering Handbook, McGraw-Hill, Chapter 23, 2007.

D. M. Ah Yo and R.M. Emrick “Frequency Bands for Military and Commercial Applications.”. Antenna Engineering Handbook, McGraw-Hill, Chapter 2, 2007.

R.M. Emrick and J.L. Volakis “Inductively Loaded Millimeter-Wave Spiral Array on Silicon.”. IEEE Antennas and Propagation Symposium, 2007.

FIELDS OF STUDY

Major Field: Electrical and Computer Engineering

Studies in Electromagentics: Prof. John L. Volakis

iv TABLE OF CONTENTS

Page

Abstract ...... ii

Acknowledgments ...... iii

Vita ...... iv

List of Figures ...... vii

Chapters:

1. Introduction to Millimeter-Wave and Terahertz Antennas ...... 1

1.1 Applications ...... 2 1.1.1 Wireless ...... 3 1.1.2 Radars ...... 5 1.1.3 Imaging ...... 5 1.2 Millimeter-wave Antennas ...... 7 1.2.1 Waveguide Antennas ...... 8 1.3 On-Chip Antennas ...... 14 1.4 Submillimeter-wave and Terahertz Antennas ...... 23 1.5 Chapter Conclusions ...... 25

2. Antenna Requirements for High Speed Wireless Systems Operating at Millimeter-wave Frequencies ...... 29

2.1 Introduction ...... 29 2.2 Fading Multi-Path Channels ...... 32 2.3 Effect of MIMO in Fading Multi-Path Channels ...... 34 2.4 Channel Model Within a Room ...... 38 2.4.1 Analytical Model for a Room ...... 38

v 2.4.2 SNR Calculations ...... 38 2.5 Chapter Conclusions ...... 45

3. Material Properties of Gold and Silicon at High Frequencies and Their Effect on Efficiency for Candidate Antenna Elements ...... 46

3.1 Initial Antenna Element Analysis ...... 46 3.2 Antenna elements used for this analysis ...... 57 3.3 Electrical Properties of Gold ...... 62 3.4 Electrical Properties of Silicon ...... 73 3.4.1 Measured Results for the Spiral Element ...... 79 3.5 Material Property Effects on Efficiency ...... 80 3.6 Chapter Conclusions ...... 84

4. Antenna Array Implementation ...... 85

4.1 overview ...... 85 4.2 Array Analysis ...... 85 4.3 The Array Factor ...... 86 4.4 Array and System Performance ...... 89 4.5 Chapter Conclusions ...... 95 4.5.1 Design Guidelines ...... 97

5. Conclusions and Summary of Contributions ...... 100

5.1 Summary and Conclusions ...... 100

Bibliography ...... 102

vi LIST OF FIGURES

Figure Page

1.1 Millimeter-wave (MMW) spectrum and applications. Bands shown are unlicensed 60 GHz, easily licensed 70 and 80 GHz, 77 GHz automotive radar, unlicened 90 GHz and emerging bands above 100 GHz. . . . . 3

1.2 Signal-to-noise (S/N) ratio at the receiver as a function of separation distance between the and receiver. An antenna with 20-dBi gain was assumed (the horizontal lines show the required S/N ratios for the indicated data rates and configuration) (after R. M. Emrick and J. L. Volakis [3] IEEE 2006)...... 6

1.3 Automotive sensors for advanced safety systems ...... 7

1.4 Horn antennas, connected to a WR15 waveguide, operating at millimeter- wave frequencies. The horn antenna on the left is precision assembled whereas the horn on the right is cast and plated for lower cost. . . . . 10

1.5 Parallel plate slot array to improve manufacturability. Dielectric con- stant of the employed material is 2.17 and an efficiency of 29 percent was achieved (after J. Hirokawa and M. Ando [7] IEEE 1998) . . . . 11

1.6 Waveguide to microstrip antenna coupling using an aperture at the broadwall to feed the microstrip antenna (after D. Pozar [8] IEEE 1996) 11

1.7 Double slot antenna implemented on an LTCC package (after K. Maruhashi et al [9] IEEE 2000) ...... 13

1.8 A multilayer parasitic microstrip antenna array implemented using LTCC (after T. Seki et al [10] IEEE 2005). Measured absolute gain for this antenna is 7.17 dBi...... 13

vii 1.9 Comparison of Substrate Properties for LTCC, FR4, and LCP. Data is shown at 1 MHz for LTCC and 20 GHz for LTCC and LCP. . . . . 14

1.10 High-gain antenna utilizing multiple LCP layers to form a vertical array of spirals over a . The achieved gain is 12.3 dBi with a top surface occupying 1.2 mm, and computations were carried out using the Remcom XFDTD (after R. M. Emrick and J. L. Volakis [3] IEEE 2006)...... 16

1.11 Transmission line losses in dB/mm using thin dielectrics. The SiO2 substrate thickness 3 µm is representative when implementing 50 ohm transmission lines in the top layers of silicon wafer processes. Losses for thicker GaAs substrates are also shown for comparison (use of higher dielectric constants produce similar effects)...... 17

1.12 Approach for reducing losses by increasing the air volume near the structure or components of interest (after C. Nguyen et al [12] IEEE 1998)...... 17

1.13 Two-dimensional array utilizing membranes and air cavities to reduce losses (after G. Rebeiz et al [14] IEEE 1990) ...... 19

1.14 Formation of surface wave and other substrate modes can effect per- formance of on-chip antennas (after N. Alexopoulos et al [15] IEEE 1983)...... 20

1.15 Example layout of an edge-fed element on silicon for compact on-chip antennas. When minimized in size, the antenna element measures only 360 mm 135 mm delivering 3.2-dBi gain and a 3-dB bandwidth of 17%. 21

1.16 Example of a millimeter wave antenna comprised of a half-circle el- ement with a tuning slot. When the size is minimized, the element measures 480 mm 240 mm delivering a gain of 3.8 dBi with a 3-dB bandwidth of 17%...... 22

1.17 Example antenna on a thin membrane integrated with a detector op- erating up to 700 GHz (G. Rebeiz et al [13] IEEE 1987)...... 24

1.18 Log-periodic antenna element with a lens coupled to a hot-electron bolometer for operation at 1-6 THz (after A. Semenov et al [17] IEEE 2007)...... 24

viii 1.19 Micromachined waveguide antenna for 1.6-THz operation (after J. Bowen et al [18] IEEE 2006). A gain of about 13 dBi was achieved...... 26

1.20 Photoconductive antenna to generate and transmit or receive THz sig- nals (L is antenna length) ...... 27

2.1 Wireless Standards Snapshot...... 30

2.2 Worldwide Spectrum Available at 60 GHz. 3 GHz of common world- wide spectrum exists from 59-62 GHz, as highlighted...... 31

2.3 Bit Error Rate as a function of γb for a single input single output system. Bit error rates for K=0, 6 and 12 for BFSK and BPSK are shown...... 35

2.4 MIMO implementation using t transmit and r receive antennas . . . . 36

2.5 Bit Error Rate as a function of the average SNR per bit for a 2x2 MIMO system having L = t·r = 4 MIMO channels (SISO is shown for comparison). Bit error rates for K=0, 6 and 12 with BPSK are shown. 37

2.6 Multipath delay profile for transmit and receive separation of 5m, 15m and 35m which is described by Equation 2.5 at 60 GHz ...... 39

2.7 SNR at the receiver as a function of transmit and receive antenna separation with of 6dBi assumed for both transmit and receive. Required SNR levels are shown for various conditions at 60 GHz 41

2.8 SNR at the receiver as a function of transmit and receive antenna separation with antenna gain of 20dBi assumed for both transmit and receive. Required SNR levels are shown for various conditions at 60 GHz 42

3.1 Mineaturized spiral elements included in this analysis using a) straight arms and b) square meander-line inductive loading ...... 47

3.2 Example layout of antenna approach using LTCC to implement wide- band triangle and Yagi antennas for ”Antenna in Package” approach from [38]. Exploded layout view (a, wideband triangle b) and Yagi c) layouts are shown...... 48

ix 3.3 Input return loss of spiral elements ...... 49

3.4 Block diagram of proposed implementation using compact spiral ele- ments integrated with active silicon circuits ...... 50

3.5 Realized gain of straight and square meander inductive loaded spiral antenna elements. Gain of at least +2dB is required for acceptable system efficiency and performance...... 51

3.6 Scaled version of loaded spiral used for analysis of losses. Substrate λ thickness is 7 at 60 GHz on a high resistivity silicon substrate. Line width of the spiral is ∼ 3.5µm...... 53

3.7 and gain for a scaled loaded spiral element with isolation of loss mechanisms. Directivity along with gain when using gold and PEC are shown...... 54

3.8 Alternative spiral using substantially more metalization. Substrate λ thickness is 4 at 60 GHz on a high resistivity silicon substrate. Line width of the spiral is ∼ 57µm...... 55

3.9 Directivity and gain for a spiral element with much greater metaliza- tion. Line widths are 15x greater than the loaded spiral...... 56

3.10 Dimensions of spiral element on silicon ...... 59

3.11 Dimensions of edge fed bow-tie element on silicon ...... 60

3.12 Dimensions of half circle element on silicon ...... 61

3.13 Gain of the antennas shown using all lossless materials ...... 63

3.14 Efficiency of the antennas using all lossless materials ...... 64

3.15 Antenna pattern for spiral antenna at 55, 60 and 65 GHz...... 65

3.16 Antenna pattern for bow-tie antenna at 55, 60 and 65 GHz...... 66

3.17 Antenna pattern for half circle antenna at 55, 60 and 65 GHz. . . . . 67

x 3.18 Measured data (markers) and extrapolated values (solid) for the imag- inary part of the permittivity of bulk gold as a function of frequency. Extrapolated values were determined by fitting a Drude model known values at low and optical frequencies...... 70

3.19 Measured data (markers) and extrapolated values (solid) for the real part of the permittivity of bulk gold as a function of frequency. Ex- trapolated values were determined by fitting a Drude model to known values at low and optical frequencies...... 71

3.20 Conductivity of bulk gold calculated from an extrapolation of measured properties at optical frequencies...... 72

3.21 Measured data (markers) and extrapolated values (solid) for the real part of the permittivity of intrinsic silicon as a function of frequency. The curve was fit to match high frequency measured data and the know low frequency value...... 75

3.22 Measured data (markers) and extrapolated values (solid) for the imag- inary part of the permittivity of intrinsic silicon. The curve was fit to match high frequency measured data and known low frequency con- ductivity...... 76

3.23 Conductivity of intrinsic silicon calculated from an extrapolation of measured properties at optical frequencies. Low frequency conductivity was used as part of curve fitting for the imaginary part of the permittivity. 77

3.24 Loss Tangent of intrinsic silicon calculated from an extrapolation of measured properties at optical frequencies. Loss tangent contributions from the dopant, intrinsic silicon and their sum are shown ...... 78

3.25 Measured input return loss for the spiral antenna. Thick solid like is the simulated result and is shown with the measured response of 4 fabricated spiral elements ...... 80

3.26 Efficiency of the antennas with gold metallization and lossless silicon 81

3.27 Efficiency of the antennas using gold and 1kohm-cm silicon...... 82

3.28 Efficiency of the antennas using gold and 10ohm-cm silicon...... 83

xi 4.1 Planer rectangular array geometry ...... 87

4.2 Planer linear array geometry ...... 88

4.3 Rectangular array configuration used in this analysis. Element to ele- ment spacing is 1mm for this analysis...... 90

4.4 Effect of increasing array size using half circle element antennas. Op- erating frequency is 60 GHz with element to element spacing of 1mm. Antenna gain here does not include losses associated with the feed network...... 93

4.5 Planer antenna array feed for an 8x8 array of half circle elements. In order to feed each element in phase, substantial line length is needed. In this example, at least 15mm of line length is needed to feed each element, resulting in substantial loss and reduction in efficiency. . . . 94

4.6 Exploded view layout of proposed approach for optimum integration on silicon for millimeter-wave transmit array...... 95

4.7 Effect of increasing array size given that each element is driven by a silicon amplifier providing 10 dBm of output power. PGP is defined in (2.8). Operating Frequency is 60 GHz with element to element spacing of 1mm...... 96

xii CHAPTER 1

INTRODUCTION TO MILLIMETER-WAVE AND TERAHERTZ ANTENNAS

There are a number of definitions describing antennas operating in the millimeter- wave (MMW) or terahertz (THz) bands. Most commonly, antennas operating in fre- quencies whose wavelengths are measured in millimeters (30 GHz to 300 GHz) are referred to as millimeter- wave antennas. On the other hand, the band between 300

GHz and 3 THz is referred to as the submillimeter-wave band, in which the corre- sponding wavelengths are measured in units less than one millimeter. One can say that a fair amount of freedom has been used in the literature when defining the MMW and THz bands. Some describe a device operating at 24 GHz as a millimeter-wave sensor and those operating around 300 GHz as THz devices. A possible cause for these inconsistencies may be the specialized use of MMW and THz devices. Historically, devices operating in the millimeter-wave, submillimeter-wave, and THz regimes have been limited to specialized applications, likely due to their high cost of realization.

Specifically, at these higher frequencies there are a number of challenges, including

(1) the availability of sources operating in these bands, (2) fabrication challenges of the small features required for these devices, (3) maintaining tolerance for the small features to achieve repeatable designs, and (4) fabrication costs. It is certainly

1 well-known that antenna dimensions scale inversely proportional to their frequency.

However, at some point, the features size becomes a fabrication challenge, and tra- ditional printed circuit-board techniques may no longer be applicable. Nevertheless, recent and ongoing advances in a number of fabrication technologies indicate that low-cost solutions at the MMW and THz frequencies should be possible [1]. Specif- ically, silicon transceivers can now support frequencies of 60 GHz and higher. That is, with the capability of traditional silicon devices continuing to trend higher [2], the challenges of obtaining sources and transceiver ICs is effectively being addressed. To some extent, this moves the challenge in realizing MMW and THz devices to other components, including antennas. This chapter is, of course, specifically focused on the antenna design and fabrication for MMW and THz devices.

1.1 Applications

Applications for millimeter-wave and THz frequencies include wireless, radar, and imaging (see Figure 1.1). For antennas, it is often found that the design is application- specific where integration, loss, and gain requirements may vary among applications.

An issue for the designer is also the allowed frequency band of operation. Specifi- cally, the unlicensed spectrum at 60 GHz overlaps an oxygen absorption band that adds an additional 15 dB/km of attenuation. This is certainly a challenge for long communication ranges, but may not be of concern for shorter range applications, as is the case for wireless personal area networks. In fact, this atmospheric attenuation is even desired since it allows for frequency reuse and security of the band by limiting its reach to local areas of operations. From Figure 1.1, imaging applications are typ- ically targeted for bands where there is a local minimum in atmospheric attenuation,

2 ISSUEFORTHEDESIGNERISALSOTHEALLOWEDFREQUENCYBANDOFOPERATION3PECIFICALLY THE UNLICENSEDSPECTRUMAT'(ZOVERLAPSANOXYGENABSORPTIONBANDTHATADDSANADDITIONAL D"KM OF ATTENUATION4HIS IS CERTAINLY A CHALLENGE FOR LONG COMMUNICATION RANGES

 !!

  $

$

                    "#$ &)'52%  -ILLIMETER WAVE--7 SPECTRUMANDAPPLICATIONS

Figure 1.1: Millimeter-wave (MMW) spectrum and applications. Bands shown are unlicensed 60 GHz, easily licensed 70 and 80 GHz, 77 GHz automotive radar, un- licened 90 GHz and emerging bands above 100 GHz.

as is the case with the 94 GHz and 140 GHz frequency bands. This is important, especially for passive imaging, as the desired signals can be extremely low in power and additional atmospheric attenuation can severely impact their reception. Next we consider the three applications areas separately.

1.1.1 Wireless

As wireless devices become more prevalent and the desire for increased data rates continues, there are challenges in the employed lower frequency spectrum (from 824

MHz for the cellular bands up to 2.4 GHz and 5 GHz for the ISM and 802.11 bands.

A solution is to use the large amounts of spectrum in the millimeter-wave frequen- cies. Wireless applications in these bands include wireless personal area networks

(WPAN), point to point for backhaul applications, and the extension of fiber and hybrid fiber-coax systems beyond their current reach. The latter presents an advan- tage over wired networks as it can be cost prohibitive to extend them beyond their

3 current coverage. Specifically, a large portion of unlicensed spectrum is available worldwide around 60 GHz for either WPAN or other applications such as backhaul.

In the United States, easily and cheaply licensed spectrum is now available in two

5 GHz blocks near 75 and 85 GHz. It is, however, believed that spectrum targeted for commercial backhaul applications must be licensed so the operator has some re- course in case of interference. Challenges faced in antenna development for wireless applications may differ significantly. For backhaul applications, the antennas are typically directional and physically large for acceptable performance over 12 km in range under typical rain-fade conditions. However, in the case of WPAN, the chal- lenges range from low cost to sufficient gain for acceptable system-level performance allowing for high data-rate connectivity. To achieve multigigabit data rates at reason- able ranges, especially for non-line-of-sight (NLOS) conditions, it has been recently shown [3] (using simple channel models) that antenna gains on the order of 20 dBi are likely required. Though these antennas will be smaller, an array may be required to achieve such large gains while accounting for expected chipset efficiencies. Figure 1.2 shows signal-to-noise (S/N) ratio as a function of separation distance under different propagation conditions and transmission systems such as single-input single-output

(SISO) and multiple-input multiple-output (MIMO). The propagation channels in- cluded in Figure 1.2 include the usual line-of-sight (LOS) with multipaths and the non-line-of-sight (NLOS), referred to as LOS-blocked in the figure. An observation for Figure 1.2 is that MIMO has a clear advantage in achieving higher data rates.

4 1.1.2 Radars

Millimeter-wave radar sensors are in early deployment for applications such as au- tomotive anticollision systems. Virtually every major automaker now has an optional adaptive cruise control system that may become part of a future automotive safety system for collision warning/avoidance, lane departure warning, blind spot detection, backup, and parking aids. Such systems are already being used for parking assis- tance. Though there is debate on the future pervasiveness of anticollision systems, it is widely expected that these technologies will make driving safer. For example, without active avoidance, perimeter sensors may detect a situation having a high probability of collision and initiate pre-crash safety steps, including pretensioning of seat belts, adjusting brake line pressure to shorten stopping distance, and so on. As displayed in Figure 1.3, there are a wide range of sensors likely to be used; these include millimeter-wave, optical, and ultrasonic sensors. Sensors operating in the 77-

GHz range with a combination of 24-GHz sensors for shorter ranges (see Figure 1.1) are a likely option. Specifically, millimeter-wave frequencies are considered a good choice because they penetrate fog and heavy dust in addition to allowing for high resolution and being small enough so as not to affect vehicle appearance.

1.1.3 Imaging

The area of millimeter-wave and THz imaging has seen significant growth in a number of applications. A popular application is security screening for detection of concealed objects such as ceramic knives hidden beneath clothing [4] (done pas- sively). Another application is the use of passive millimeter-wave imagers for aircraft systems used by pilots to see landing areas under low visibility conditions (fog or

5 Ó·{ #(!04%247%.49 4(2%%

 *") &)



-&(



 - &(

 &(   $")%% () %) +'   "%!

           &') %$),$'$(# ))'$ +' # &)'52%  3IGNAL TO NOISE3. RATIOATTHERECEIVERASAFUNCTIONOFSEPARA TIONDISTANCEBETWEENTHETRANSMITTERANDRECEIVER!NANTENNAWITH D"IGAINWAS FigureASSUMEDTHEHORIZONTALLINESSHOWTHEREQUIRED3.RATIOSFOR 1.2: Signal-to-noise (S/N) ratio at the receiver asTHEINDICATEDDATARATES a function of separation distanceANDCONFIGURATION AFTER2-%MRICKAND*,6OLAKIS between the transmitter and receiver. An antennaÚ)%%%  with 20-dBi gain was assumed (the horizontal lines show the required S/N ratios for the indicated data rates and configuration) (after R. M. Emrick and J. L. Volakis [3] IEEE 2006). INCLUDINGPRETENSIONINGOFSEATBELTS ADJUSTINGBRAKELINEPRESSURETOSHORTENSTOPPINGDIS TANCE ANDSOON!SDISPLAYEDIN&IGURE  THEREAREAWIDERANGEOFSENSORSLIKELYTOBE

6 INCLUDINGPRETENSIONINGOFSEATBELTS ADJUSTINGBRAKELINEPRESSURETOSHORTENSTOPPINGDIS TANCE ANDSOON!SDISPLAYEDIN&IGURE  THEREAREAWIDERANGEOFSENSORSLIKELYTOBE USEDTHESEINCLUDEMILLIMETER WAVE OPTICAL ANDULTRASONICSENSORS3ENSORSOPERATINGINTHE  '(ZRANGEWITHACOMBINATIONOF '(ZSENSORSFORSHORTERRANGESSEE&IGURE  AREALIKELYOPTION3PECIFICALLY MILLIMETER WAVEFREQUENCIESARECONSIDEREDAGOODCHOICE BECAUSETHEYPENETRATEFOGANDHEAVYDUSTINADDITIONTOALLOWINGFORHIGHRESOLUTIONAND BEINGSMALLENOUGHSOASNOTTOAFFECTVEHICLEAPPEARANCE

               

                    

&)'52%  !UTOMOTIVESENSORSFORADVANCEDSAFETYSYSTEMS

Figure 1.3: Automotive sensors for advanced safety systems

sand storms). Such imaging sensors can operate at higher frequencies that extend to

THz and may enable a wide range of systems for detection and diagnosis. A great potential for imaging using MMW and THz exists in the medical field. Applications include tumor recognition, disease diagnosis, recognition of protein structural states, detection/imaging of tooth decay, and so on. [5]

1.2 Millimeter-wave Antennas

As mentioned previously, challenges associated with antennas at high frequencies include fabrication tolerances, efficiency/losses, and cost, though these are all some- what interrelated. As is often the case, it can be very difficult to optimize each of these parameters simultaneously, and a number of approaches have already been pur- sued and developed. In this section, we discuss technologies used for the fabrication of different MMW antennas. We discuss these under the categories of (1) waveg- uide antennas, (2) printed planar antennas, and (3) on-chip antennas, which may be

7 fabricated using micromachining or semiconductor lithography techniques. Here, the

first two approaches will be discussed with the category of on-chip antennas being discussed in a later section.

1.2.1 Waveguide Antennas

An approach to minimizing losses in millimeter-wave antennas is to use waveg- uide antennas. It is well-known that waveguides and waveguide antennas can provide higher performance and lower losses, but they do not typically align well with com- pact or low-cost solutions. Nevertheless, in this context, one can consider waveguide feeds and horns or other waveguide antennas, but hybrid approaches have also been explored. When fabricating such components for MMW applications, a lower-cost approach is to utilize metalized plastics [6], but many question the viability of this approach. Some challenges associated with the metalized plastics include the type of metallization, interfacing with other components, and maintaining electrical continu- ity. Also, the overall cost of this approach and other measures relating to integration bring additional challenges. For example, the design must insure that under temper- ature cycling the joining or movement between dissimilar materials does not degrade metallization and metal contacts. In addition to using metalized plastics to reduce cost, casting, or electroforming, the antenna followed by gold or other plating is a viable alternative. Two millimeter- wave horn antennas are shown in Figure 1.4. The horn on the left is fabricated using precision assembly and brazing/ welding tech- niques, and the horn on the right is fabricated by plating a cast base metal. The electrical performance of the two horns shown in Figure 1.4 is very similar (within one to two tenths of a dB). However, casting (with plating) can significantly reduce

8 fabrication cost from one to several orders of magnitude, depending on the fabrica- tion volume. An example approach to fabricating waveguide antennas for volume manufacturing is to employ parallel plate slot arrays [7], as shown in Figure 1.5. In this case, losses may be greater since a dielectric must be used for filling the waveg- uide. Of course, lower dielectric constants are preferred to avoid further reduction in the metallization dimensions, thus reducing fabrication costs and ohmic losses (issues related to high dielectric constants or thin substrates will be discussed later in more detail). An additional example of a waveguide feed is shown in Figure 1.6. In this example [8], a waveguide is used to feed the microstrip antenna, and this may be a good compromise for certain applications since losses in the MMW feeding networks can be significant. The feed shown in Figure 1.6 is rather standard and employs an aperture in the broadside of the rectangular waveguide positioned under a microstrip antenna. Challenges for an approach like this, as in other cases, are likely to increase as the frequency of operation continues to increase since at higher frequencies, the dimensions of the aperture can become quite small, and placement of the microstrip antenna properly over the aperture can become a dominant challenge. Printed Planar

Antennas As may be expected, there is strong interest in printed planar antennas for millimeter-wave applications. At lower frequencies, these are very easy to fabricate and are compatible with other portions of a transceiver that may be implemented on the same substrate allowing for greater integration. There are a range of substrates used for planar antennas, but there are some that appear to have garnered the greatest interest based on the volume of current work and technical publications. The most commonly used substrates are currently low temperature cofired ceramic (LTCC), liquid crystal polymer (LCP), and silicon. Interest in LTCC and LCP is fairly wide

9 DAPPROACHESHAVEALSOBEENEXPLORED7HENFABRICATINGSUCHCOMPONENTS ONS ALOWER COSTAPPROACHISTOUTILIZEMETALIZEDPLASTICS BUTMANYQUES F THIS APPROACH 3OME CHALLENGES ASSOCIATED WITH THE METALIZED PLASTICS METALLIZATION INTERFACINGWITHOTHERCOMPONENTS ANDMAINTAININGELECTRICAL EOVERALLCOSTOFTHISAPPROACHANDOTHERMEASURESRELATINGTOINTEGRATION ALLENGES&OREXAMPLE NSURE THAT UNDER TEM EJOININGORMOVEMENT R MATERIALS DOES NOT ONANDMETALCONTACTS SINGMETALIZEDPLASTICS STING OR ELECTROFORM OWEDBYGOLDOROTHER ALTERNATIVE4WOMILLI NTENNAS ARE SHOWN IN ORNONTHELEFTISFABRI &)'52%   (ORN ANTENNAS CONNECTED TO A 72 ONASSEMBLYANDBRAZ FigureWAVEGUIDE OPERATINGATMILLIMETER WAVEFREQUENCIES4HE 1.4: Horn antennas, connected to a WR15 waveguide, operating at millimeter- QUES ANDTHEHORNONwaveHORNANTENNAONTHELEFTISPRECISIONASSEMBLEDWHEREAS frequencies. The horn antenna on the left is precision assembled whereas the TED BY PLATING A CASThornTHEHORNONTHERIGHTISCASTANDPLATEDFORLOWERCOST on the right is cast and plated for lower cost.

because their electrical properties are some of the best at millimeter-wave frequencies

and are also well-suited for integration with other portions of a transceiver. Silicon

is certainly of high interest for on-chip antenna integration.

Packages with integrated antennas using LTCC have already been demonstrated,

and there are a number of options available in the type and implementation of printed

planar antennas utilizing LTCC. A double slot antenna [9] on LTCC is shown in Figure

1.7. In this case, the antenna is formed using several layers to produce a multilayer

package delivering a gain of 4 dBi for the double slot antenna. This example represents

good feasibility for antenna integration with the transceiver package. However, having

the antenna and transceiver close to or colocated with the transceiver is not always

preferred. Depending on the application and the transceiver, it may be preferable

10 ÓÎ‡È #(!04%247%.49 4(2%%

    

   

        ERREDTOAVOIDFURTHERREDUCTIONINTHEMETALLIZATION  DIMENSIONS  THUS COSTSANDOHMICLOSSESISSUESRELATEDTOHIGHDIELECTRICCONSTA&)'52%  0ARALLELPLATESLOTARRAYTOIMPROVEMANUFACTURABILITY$IELECNTSORTRIC CONSTANTOFTHEEMPLOYEDMATERIALISANDANEFFICIENCYOFWASACHIEVED FigureAFTER*(IROKAWAAND-!NDO 1.5: Parallel plate slot arrayÚ)%%%  to improve manufacturability. Dielectric constant EDISCUSSEDLATERINMOREDETAIL  of the employed material is 2.17 and an efficiency of 29 percent was achieved (after MPLEOFAWAVEGUIDEFEEDISSHOWNIN&IGURE )NTHISEXAMPLJ. Hirokawa and M. Ando [7] IEEE 1998) E A FEEDTHEMICROSTRIPANTENNA ANDTHISMAYBEAGOODCOMPROMISEB L 4H L I L F F H H H I &I FOR  I I IL NCELOSSESINTHE--7FEEDINGNETWORKSCANBESIGNIFICANT4HEFEED SRATHERSTANDARDANDEMPLOYSAN    DEOFTHERECTANGULARWAVEGUIDE CROSTRIPANTENNA#HALLENGESFOR ASINOTHERCASES ARELIKELYTO     ENCY OF OPERATION CONTINUES TO &)'52%  7AVEGUIDETOMICROSTRIP ERFREQUENCIES THEDIMENSIONSOF Figure 1.6: Waveguide to microstrip antenna coupling using an aperture at the broad- MEQUITESMALL ANDPLACEMENTOF wallANTENNACOUPLINGUSINGANAPERTUREATTHE to feed the microstrip antenna (after D. Pozar [8] IEEE 1996) BROADWALL TO FEED THE MICROSTRIP ANTENNA APROPERLYOVERTHEAPERTURECAN AFTER$0OZARÚ)%%% HALLENGE 11 i˜˜>Ã

THEREISSTRONGINTERESTINPRINTEDPLANARANTENNASFORMILLIMETER TLOWERFREQUENCIES THESEAREVERYEASYTOFABRICATEANDARECOMPAT ONSOFATRANSCEIVERTHATMAYBEIMPLEMENTEDONTHESAMESUBSTRATE INTEGRATION 4HERE ARE A RANGE OF SUBSTRATES USED FOR PLANAR ANTEN to have the antenna located where it can provide the best performance with the rest of the transceiver, for instance, located near its interfaces with the rest of the system. Having the antennas separated from the transceiver is, of course, common at lower frequencies, but at millimeter-wave frequencies interconnect losses between the transceiver and antenna are of concern. In this case, separations of many wavelengths can be expected for a package of a few millimeters in size. Using more available layers for antenna design with LTCCs is also an option, as demonstrated with the multilayer parasitic microstrip antenna array in Seki et al [10] (see Figure 1.8). This antenna was formed using three LTCC layers and incorporates the feed element and two parasitic element layers. It delivered a measured gain of 7.17 dBi. In addition to LTCCs, a substrate material that has been gaining interest is the liquid crystal polymer (LCP) [11], though LCP has been under consideration for much less time than LTCC. LCP has captured interest due to its good electrical characteristics. They are also considered very stable since LCP does not absorb moisture and is therefore near hermetic.

A comparison of LTCC, LCP, and FR4 is shown in Table 23-1. As shown, LCP has a loss tangent similar to LTCC with a lower dielectric constant. The latter may be an advantage at MMW frequencies, but as compared to LTCCs, it offers fewer dielectric constant choices. Using LCP substrates with multiple layers, as may be done with

LTCCs, a number of antenna possibilities exist. One such possibility is shown in

Figure 1.10 where multiple LCP layers have been used to implement a vertical array of spiral antennas to achieve more than 12-dBi gain occupying a top surface area of only 1.2 mm. Additional increases in gain may be achieved by increasing the number of layers and increasing the number of elements. The operation of an array like this

12 -),,)-%4%2 7!6%!.$4%2!(%24:!.4%..!3 Ó·

&)'52%  $OUBLESLOTANTENNAIMPLEMENTEDONAN,4##  FigurePACKAGEAFTER+-ARUHASHIETAL 1.7: Double slot antenna implementedÚ)%%% on an LTCC package (after K. )NADDITIONTO,4##S ASUBSTRATEMATERIALTHATHASBEENGAININGINTERESTISTHELIQUID MaruhashiCRYSTALPOLYMER,#0 et al [9] IEEETHOUGH,#0HASBEENUNDERCONSIDERATIONFORMUCHLESSTIMETHA 2000) N ,4##,#0HASCAPTUREDINTERESTDUETOITSGOODELECTRICALCHARACTERISTICS4HEYAREALSO CONSIDEREDVERYSTABLESINCE,#0DOESNOTABSORBMOISTUREANDISTHEREFORENEARHERMETIC

               

            

&)'52%  !MULTILAYERPARASITICMICROSTRIPANTENNAARRAYIMPLEMENTEDUSING,4##AFTER43EKIETAL Ú)%%% -EASUREDABSOLUTEGAINFORTHISANTENNAISD"I Figure 1.8: A multilayer parasitic microstrip antenna array implemented using LTCC (after T. Seki et al [10] IEEE 2005). Measured absolute gain for this antenna is 7.17 dBi.

13 Ó·n #(!04%247%.49 4(2%%

4!",%  #OMPARISONOF3UBSTRATE0ROPERTIESFOR,4## &2 AND,#0

,4## &2 ,#0 2ELATIVEPERMITTIVITY '(Z -(Z '(Z ,OSSTANGENT    2ELATIVECOST -EDIUM 6ERYLOW ,OW

!COMPARISONOF,4## ,#0 AND&2ISSHOWNIN4ABLE !SSHOWN ,#0HASALOSS FigureTANGENTSIMILARTO,4##WITHALOWERDIELECTRICCONSTANT4HEL 1.9: Comparison of Substrate Properties for LTCC,ATTERMAYBEANADVANTAGEAT FR4, and LCP. Data is shown--7FREQUENCIES BUTASCOMPAREDTO,4##S ITOFFERSFEWERDIEL at 1 MHz for LTCC and 20 GHz for LTCC and LCP.ECTRICCONSTANTCHOICES 5SING,#0SUBSTRATESWITHMULTIPLELAYERS ASMAYBEDONEWITH,4##S ANUMBEROF ANTENNAPOSSIBILITIESEXIST/NESUCHPOSSIBILITYISSHOWNIN&IGURE WHEREMULTIPLE ,#0LAYERSHAVEBEENUSEDTOIMPLEMENTAVERTICALARRAYOFSPIRALANTENNASTOACHIEVEMORE THAN D"IGAINOCCUPYINGATOPSURFACEAREAOFONLYMM!DDITIONALINCREASESINGAIN is similar to a Yagi-Uda antenna where the ground plane acts as the reflector element and the element above is the driver. Additional spiral elements above the driven element act as director elements. A challenge in this approach is feeding the driver element.

1.3 On-Chip Antennas

Since antennas are physically small at high frequencies, overcoming challenges with interconnect losses has forced implementations toward on-chip integration. A number of approaches beyond those that are part of standard packaging techniques

(to minimize dielectric and metal losses) have been explored [12–15]. In order to min- imize losses, methods are often used to maximize the portion of the fields that are in air or to maximize the metallization used to print the transmission lines and antennas.

0.5 However, since the characteristic impedance of transmission line is Z0 ∼ (L/C) , and the capacitance increases with the dielectric constant, as the capacitance increases the transmission line impedance is also lowered. To compensate for the increased capacitance, the width of the microstrip line can be decreased. This does raise the line impedance but also increases ohmic losses associated with the transmission line.

14 Figure 1.11 shows loss contributions due to thin dielectrics for a microstrip trans- mission line using simulations. The figure highlights the effects of ohmic losses and how these can become dominant as frequencies are increased (assuming a substrate with reasonable loss characteristics). The assumed relatively thin SiO2 layer of 3µm is consistent with the upper layers available for the transmission line as part of a silicon process. For reference, losses of 0.2 dB/mm are expected when using gallium arsenide (GaAs) with a commonly used substrate thickness of 100 mm. An approach for reducing losses in structures and components by reducing the substrate dielectric used is shown in Figure 1.12. By creating an air cavity below the structures, both the dielectric and ohmic losses can be reduced. Air cavities can be created to form such structures either by etching or by micromachining away materials from a thick substrate. Alternatively, the desired devices and components can be placed on a thin membrane either before or after being applied to a supporting structure. It is not clear that approaches like this will lend themselves to low-cost, high-volume manufac- turing, though they are clearly of interest for specialized lower-volume applications.

For such approaches to be viable for high volume, they should also be compatible with standard semiconductor processing. Additional issues may relate to the device packaging for long-term reliability as moisture or other contaminants that enter the air cavity can significantly affect performance. A larger scale example of this ap- proach is shown in Figure 1.13. Figure 1.13 shows a two-dimensional array fabricated so the antenna element is suspended on a membrane over a cavity shaped as a horn.

This approach allows for reduction in both dielectric and ohmic losses. The SiO2 substrate thickness 3µm is representative when implementing 50Ω transmission lines

15  P P THETRANSMISSIONLINEIMPEDANCEISALSOLOWERED4OCOMPENSATEFORTHEINCREASEDCAPACITANCE THEWIDTHOFTHEMICROSTRIPLINECANBEDECREASED4HISDOESRAISETHELINEIMPEDANCEBUTALSO

&)'52%  (IGH GAINANTENNAUTILIZINGMULTIPLE,#0LAYERSTOFORMAVERTICALARRAYOFSPIRALSOVER AGROUNDPLANE4HEACHIEVEDGAINISD"IWITHATOPSURFACEOCCUPYINGMM ANDCOMPUTATIONSWERE CARRIEDOUTUSINGTHE2EMCOM8&$4$AFTER2-%MRICKAND*,6OLAKISÚ)%%%  Figure 1.10: High-gain antenna utilizing multiple LCP layers to form a vertical array of spirals over a ground plane. The achieved gain is 12.3 dBi with a top surface occupying 1.2 mm, and computations were carried out using the Remcom XFDTD (after R. M. Emrick and J. L. Volakis [3] IEEE 2006).

in the top layers of silicon wafer processes. Losses for thicker GaAs substrates are also shown for comparison (use of higher dielectric constants produce similar effects).

This may have the effect of reducing dielectric losses in addition to possibly re- ducing ohmic losses due to decreased distributed capacitance. The horns were fab- ricated by anisotropic etching of the silicon in an ethylendediamine- pyrocatechol solution. The etchant used forms pyramidal holes bounded by < 111 > crystal planes in < 100 > silicon. The flare angle of the horn was fixed at 70.6 degrees by the orientation of the crystal planes. This angle was larger than desired to accommodate the employed fabrication method. It is possible that the horn angle could be better controlled if reactive ion etching or ion-beam milling were used. In this case, the membrane was produced by depositing a silicon oxynitride layer on the front wafer

16 -),,)-%4%2 7!6%!.$4%2!(%24:!.4%..!3 Ó·

"!!



  

 & !!  " !! #!   

 !!!"  !!



 "!! 

     #$ % ESIREDDEVICESANDCOMPONENTS&)'52%  CANBEI IPLACED LI LON I DA THIN IMEMBRANE HI DI L I HEITHER 3I/ B Figure 1.11: Transmission line losses in dB/mm using thin dielectrics. The SiO2 NGAPPLIEDTOASUPPORTINGSTRUCTURE)TISNOTCLEARTHATAPPROsubstrate thickness 3 µm is representative when implementingACHESLIKE 50 ohm transmission SELVESTOLOW COST HIGH VOLUMEMANUFACTURING THOUGHTHEYARElines in the top layers of silicon wafer processes. Losses for thicker GaAsCLEARLY substrates are IALIZEDLOWER VOLUMEAPPLICATIONS&ORSUCHAPPROACHESTOBEVIAalso shown for comparison (use of higher dielectric constants produceBLEFOR similar effects). SHOULDALSOBECOM DARD SEMICONDUCTOR     NALISSUESMAYRELATE      KAGING FOR LONG TERM UREOROTHERCONTAMI E AIR CAVITY CAN SIG   ERFORMANCE! LARGER   ISAPPROACHISSHOWN     &IGURE   SHOWS ARRAY FABRICATED SO &)'52%   !PPROACH FOR REDUCING LOSSES BY FigureINCREASING 1.12: Approach THEfor AIR reducing VOLUME losses NEAR by THE increasing STRUCTURE the air OR volume COMPO near the ENT IS SUSPENDED ON  structureNENTSOFINTERESTAFTER#.GUYENETAL or components of interest (after C. Nguyen et al [12]Ú)%%%  IEEE 1998). A CAVITY SHAPED AS 4HIS MAY HAVE THE EFFECT OF REDUCING DIELECTRIC LOSSES ACHALLOWSFORREDUC IN ADDITION TO POSSIBLY REDUCING17 OHMIC LOSSES DUE TO RICANDOHMICLOSSES DECREASEDDISTRIBUTEDCAPACITANCE using chemical vapor deposition. The silicon was then etched away to leave the free- standing membrane. It is noted that for the membrane to be flat and rigid, it must be in tension. An additional potential challenge associated with on-chip antennas is the limitation in dielectric properties and thickness. Silicon substrates have a relative permittivity of about 11.7 but the resistivity of the substrate can vary. Of course, substrates are limited to those available for already established semiconductor pro- cesses, and wafer thicknesses may also be limited to stay with standard processing.

Typical high-volume silicon substrates may have resistivity on the order of 10Ωcm, though higher resistivities of 1kΩcm or more are available but are more expensive and may not be available in large diameters. Of course, substrate thickness and the dielectric constant can effect antenna performance beyond dielectric and ohmic losses.

For example, if the substrate thickness and dielectric constant are not chosen prop- erly, surface waves and other substrate modes can significantly effect efficiency and bandwidth (see Figure 1.14). Conditions that generate surface wave and substrate modes can significantly affect performance so they should be avoided. Antennas implemented at high frequencies tend to be relatively simple geometries and try to maximize metallization to minimize ohmic losses. Figure 1.15 is an example of an edge-fed microstrip antenna that can be implemented in a very small space by trading bandwidth.

This microstrip antenna can be optimized to occupy only 360mm × 135mm and deliver a gain of 3.2 dBi with a corresponding 3-dB bandwidth of 17%. Compact antenna elements such as this can be used to realize higher gain by implementing an array of on-chip elements. Another antenna element that shows promise is illustrated in Figure 1.16. This element is formed from a half circle and incorporates a tuning slot

18   

            

   

     

&)'52%   4WO DIMENSIONAL ARRAY UTILIZING MEMBRANES AND AIR CAVITIES TO REDUCE LOSSES

Figure 1.13: Two-dimensional array utilizing membranes and air cavities to reduce losses (after G. Rebeiz et al [14] IEEE 1990)

that allows this element to be tuned so it can be impedance-matched more easily. This is another example of a broadband element that can be implemented in a relatively small space in trade for bandwidth. Though this element is larger than the previous example, it can be designed to achieve a 3-dB 17% bandwidth gain of 3.8 dBi while occupying only 480mm×240mm. Each of these approaches can be implemented with much greater bandwidth if larger physical sizes are acceptable; each of these example microstrip antenna elements offers a wide range of usefulness and can be tailored for a particular application where size, bandwidth, or gain may be the most important design parameter.

19 PY PROCESSING4YPICALHIGH VOLUMESILICONSUBSTRATESMAY 7 CM THOUGHHIGHERRESISTIVITIESOFK7 CMORMOREARE ANDMAYNOTBEAVAILABLEINLARGEDIAMETERS/FCOURSE ECTRIC CONSTANT CAN EFFECT LECTRICANDOHMICLOSSES     KNESSANDDIELECTRICCON SURFACE WAVES AND OTHER LY EFFECT EFFICIENCY AND #ONDITIONSTHATGENERATE ESCANSIGNIFICANTLYAFFECT VOIDED &)'52%   &ORMATION OF GHFREQUENCIESTENDTOBE Figure 1.14: Formation of surface wave and other substrate modes can effect perfor- D TRY TO MAXIMIZE MET manceSURFACE of on-chip antennas WAVE (after N. ANDAlexopoulos OTHER et al [15] IEEE SUBSTRATE 1983) MODES CAN EFFECT PERFORMANCE SSES&IGURE ISAN OF ON CHIP ANTENNAS AFTER . TRIP ANTENNA THAT CAN BE !LEXOPOULOSETALÚ)%%% CEBYTRADINGBANDWIDTH

20 -),,)-%4%2 7!6%!.$4%2!(%24:!.4%..!3 Ó·££

&)'52%  %XAMPLELAYOUTOFANEDGE FEDELEMENTONSILICON FORCOMPACTON CHIPANTENNAS7HENMINIMIZEDINSIZE THEANTENNA Figure 1.15: Example layout of an edge-fed element on silicon for compact on-chip antennas.ELEMENTMEASURESONLY When minimized in size, theOM antennas elementOMDELIVERING D"IGAIN measures only 360 mm 135 mm deliveringANDA D"BANDWIDTHOF 3.2-dBi gain and a 3-dB bandwidth of 17%.

21 &)'52%   %XAMPLE OF A MILLIMETER WAVE ANTENNA COM FigurePRISEDOFAHALF CIRCLEELEMENTWITHATUNINGSLOT7HENTHESI 1.16: Example of a millimeter wave antenna comprised of a half-circle elementZE withISMINIMIZED THEELEMENTMEASURES a tuning slot. When the size is minimized, theOM elements measuresOMDELIVERING 480 mm 240 mm deliveringAGAINOFD"IWITHA D"BANDWIDTHOF a gain of 3.8 dBi with a 3-dB bandwidth of 17%.

22 1.4 Submillimeter-wave and Terahertz Antennas

Approaches used at submillimeter-wave and THz frequencies are very similar to those already discussed. Fundamentally, the fabrication challenges and losses are the same but more of an issue at THz frequencies. At submillimeter wavelengths, fabrication techniques are limited to micromachining and semiconductor lithography or equivalent due to their smaller dimensions. Interconnects between the antenna and other components such as detectors are also more challenging, leading to higher levels of integration to achieve acceptable performance. An example of a log-periodic antenna fabricated on a thin membrane with an integrated detector operating up to 700 GHz [16] is shown in Figure 1.17. As shown, membranes and air cavities along with an integrated antenna detector were employed to achieve performance.

Antennas operating at frequencies greater than 1 THz have also been demonstrated.

As an example, a log-periodic antenna element with a lens coupled to a hot-electron bolometer was fabricated and shown to operate from 16 THz [17] (see Figure 1.18).

Again, lithography requirements are not achievable with standard printed wiring- board techniques, and integration with other components is required to minimize interconnect losses. As the operational frequency increases, higher levels of integration are required to avoid degradation due to interconnects.

Micromachining has also been used to successfully fabricate THz antennas. A

1.6-THz waveguide antenna fabricated using micromachining [18] is shown in Figure

1.19. Though there is some room for improving the performance of this antenna, micromachining techniques show promise for THz antennas. A gain of about 13 dBi was achieved in this prototype. As in other cases, volume fabrication has not been demonstrated, but it does show promise for specialized applications within the THz

23 PHYREQUIREMENTSARENOTACHIEVABLEWITHSTANDARDPRINTEDWIRING BOARDTECHNIQUES AND INTEGRATIONWITHOTHERCOMPONENTSISREQUIREDTOMINIMIZEINTERCONNECTLOSSES

   

   



&)'52%  %XAMPLEANTENNAONATHINMEMBRANEINTEGRATEDWITHADETECTOROPERATINGUPTO'(Z '2EBEIZETALÚ)%%% !STHEOPERATIONALFREQUENCYINCREASES HIGHERLEVELSOFINTEGRATIONARE REQUIREDTOAVOIDDEGRADATIONDUETOINTERCONNECTS Figure 1.17: Example antenna on a thin membrane integrated with a detector oper- ating up to 700 GHz (G. Rebeiz et al [13] IEEE 1987).

&)'52%  %XAMPLEANTENNAONATHINMEMBRANEINTEGRATEDWITHADETECTOROPERATINGUPTO'(Z '2EBEIZETALÚ)%%% !STHEOPERATIONALFREQUENCYINCREASES HIGHERLEVELSOFINTEGRATIONARE REQUIREDTOAVOIDDEGRADATIONDUETOINTERCONNECTS

&)'52%   ,OG PERIODIC ANTENNA ELEMENT WITH A LENS COUPLED TO A HOT ELECTRON BOLOMETERFOROPERATIONAT 4(ZAFTER!3EMENOVETALÚ)%%% Figure 1.18: Log-periodic antenna element with a lens coupled to a hot-electron bolometer for operation at 1-6 THz (after A. Semenov et al [17] IEEE 2007)

24 frequency range. An alternative to micromachining is standard semiconductor lithog- raphy techniques to create repeatable feature sizes well into the submicron range. In addition to the above, other fabrication approaches have also been explored for THz antennas. One such approach is the photoconductive antenna [19], also known as a photoconductive switch and shown in Figure 1.20. In this case, electrodes are formed on a semiconducting photoconductive film. The film is (in most cases) a III-V semi- conductor such as GaAs. The photoconductive antenna can be considered as a dipole of length L with its resonance shifted due to the refractive index of the semiconductor

(GaAs at THz frequencies has a refractive index of n = 3.4). For operation, a voltage is established across the electrical contacts and the excited carriers are accelerated by the electric field during the optical excitation pulse. This results in a short broadband electromagnetic pulse in the THz region. When used as a receiver, during the optical pulse the excited carriers are accelerated by the electric field component of the THz pulse, which creates a measurable current. The receiver is, as usual, connected to a current amplifier across the terminal electrical contacts.

1.5 Chapter Conclusions

There are a large number of applications that either have been or are expected to leverage the use of millimeter-wave spectrum. Each of these applications has faced some similar and some different challenges associated with the use of this spectrum.

Historically, sources have been difficult to realize and the use to more exotic semicon- ductor materials like GaAs or InP have made realizing hardware expensive. Recent advances in silicon devices has shown that low cost sources using standard silicon are

25 -),,)-%4%2 7!6%!.$4%2!(%24:!.4%..!3 Ó·£Î

&)'52%  -ICROMACHINEDWAVEGUIDE FigureANTENNAFOR 4(ZOPERATIONAFTER*"OWEN 1.19: Micromachined waveguide antenna for 1.6-THz operation (after J. Bowen et al [18]ETAL IEEE  2006).Ú)%%% !GAINOFABOUTD"I A gain of about 13 dBi was achieved. WASACHIEVED -ICROMACHINING HAS ALSO BEEN USED TO SUCCESSFULLY FABRICATE 4(Z ANTENNAS !   4(Z

26 x

Figure 1.20: Photoconductive antenna to generate and transmit or receive THz signals (L is antenna length)

27 possible. This has, to a large extent, pushed the top challenge area to the implementa- tion of antennas so that they can be realized in a reasonable size with efficiencies and gain that allow for sufficient system level performance. Next, we look more closely at the system requirements which drive the antenna specifications for short range high speed wireless applications.

28 CHAPTER 2

ANTENNA REQUIREMENTS FOR HIGH SPEED WIRELESS SYSTEMS OPERATING AT MILLIMETER-WAVE FREQUENCIES

2.1 Introduction

Wireless connectivity trends have tended to track but lag wired solutions. Wired speeds using 100Mbps network interface cards (NIC) on PC’s have been widely used for many years now. Current wireless standards address data rates on the order of tens of megabits per second, with 802.11b at up to 11 Mbps and 802.11a/b at up to 54 Mbps. Standards in early development for 802.11n and 802.15.3a are expected to address data rates into the low 100’s of Mbps. These standards in addition to those well established and others in development are shown in Figure 2.1. The next generation of NIC operates at 1Gbps and is available with new PC’s as part of a

”standard” configuration on the motherboard. It is believed that as consumers grow accustomed to transferring files and utilizing streaming multimedia applications at

1Gbps+, they will also demand the option of doing so wirelessly.

One approach for meeting this demand is to utilize unlicensed spectrum that is available at millimeter-wave frequencies like 60 GHz, where at least 3 GHz of spectrum

29 5000

1000 ? 500

100 802.15.3a (Disbanded) UWB 50 802.11n

10 802.11a/g 802.15.3 802.16 WiMax 5 Canopy 802.11b 1 WiFi 802.16e / 20 Bluetooth UMTS / HSDPA / 0.5 802.15.1 1xEVDO GPRS/EDGE ZigBee 0.1 802.15.4 GSM/TDMA Transmission rate (Mbps) Blackberry

1 10 100 1000 10 000 100 000 Range (meters)

Figure 2.1: Wireless Standards Snapshot.

is available worldwide as shown in Figure 2.2. The propagation characteristics of these frequencies, in combination with the amount of spectrum available, leads many to believe it is best suited for short range high data rate applications [1,20].

In this paper, we look closely at the requirements for antennas to implement short range wireless solutions that utilize this unlicensed spectrum. We will limit this analysis to a single room and look closely at expected system performance for both line-of-sight (LOS) and non-line-of-sight (NLOS) conditions. Operation within a room could be LOS or NLOS depending on the contents of the room and the position of the transmitter and receiver. We will look closely at the expected performance in LOS and NLOS conditions to determine antenna requirements for robust performance of

30

Frequency GHz 54 55 56 57 58 59 60 61 62 63 64 65 66

Japan Unlicensed

Europe Unlicensed

Unlicensed USA

Figure 2.2: Worldwide Spectrum Available at 60 GHz. 3 GHz of common worldwide spectrum exists from 59-62 GHz, as highlighted.

very high speed wireless systems utilizing available spectrum at 60 GHz for operation within a room.

One of the key challenges of utilizing this spectrum is cost. Recent work [2,21] has shown that key building blocks, such as transceivers operating in this band, can be implemented using Si. Thus, to make a realistic assessment of antenna requirments in these types of systems, we will use the expected performance of Si IC’s in this analysis for transmit power and receiver noise figure for fading multi-path environments such as a room.

When looking at the use of high frequency unlicensed spectrum around 60 GHz, the antenna implementation are physically smaller than they would be at lower fre- quencies. The size of these antennas are on the order of what leads to interest in integration of these antennas on chip with circuitry that has been implemented on low cost silicon. Integrating antennas on chip may lead to lower cost solutions in addition to the elimination of interconnect losses and challenges that arise when the antenna is off chip. Arrays of on chip elements can be used to realize high gain.

31 Realizing antennas has some key challenges. First, the relative permittivity of silicon εr ' 11.8 at low frequencies, which does allow for smaller antenna designs may increase ohmic losses if the area of metallization becomes too small. In addition to the relatively high εr, silicon substrates are typically doped. In some cases they may be highly doped resulting in a low frequency resistivity of 10Ω · cm to as high as 1k − 10kΩ · cm, with the lower resistivity substrates being more widely available.

Lower resistivity substrates are also available in larger diameters, which allows for lower cost fabrication.

In looking at antenna implementations at millimeter-wave frequencies does require simulation. Accurate simulation requires an understanding of properties of the mate- rials that will be used when the antennas are fabricated. Relatively large amounts of data are available at low frequencies and optical frequencies. Available information on the electrical properties of the types of materials likely to be used, like gold and silicon, are not readily available for the 60 GHz band. Here, we will look closely at the electrical properties of typical materials in the frequency band of interest and quantify their impact on three candidate antenna elements.

2.2 Fading Multi-Path Channels

Within an indoor area, it is well known that scattering will occur in the room due to features such as furniture and other items located within it. As a result of this scattering, signals will arrive at the receiver through multiple paths within the room, causing fading. In looking more closely at the multi-path environment of a room, there are two general cases. When there is LOS between the transmitter

32 and receiver, the received signal can be described with a Ricean distribution [22–24].

Alternatively, when there is a NLOS condition, the received signal can be described

using a Rayleigh distribution [25,26].

Before considering multiple-input-multiple-output (MIMO) channels, it is appro-

priate to first mention some of the essential aspects of the indoor channel for single-

input-single-output (SISO) settings. As is the case for every channel, the PDF plays

a key role in determining the bit error rate (BER). Thus, it is important to identify

the key parameters in describing the channel, for example, the Ricean distribution

takes the form

h 2 2 i   ( r exp − r +A I Ar ; A ≥ 0, r ≥ 0 p(r) = σ2 2σ2 0 σ2 (2.1) 0 ; r < 0 where r is the received amplitude signal with the mean power of the multi-path signal related to the variance of the distribution, 2σ2. Key parameters are the constant, A, representing the peak amplitude of the dominant signal and it is common to define

A2 the Ricean K-factor as K = 2σ2 . Of course, when the K-factor K=0 (NLOS), (2.1) reduces to the well known Rayleigh distribution.

Having the PDF, one can proceed to obtain the BER. For a fixed attenuation α

(in a time-invariant channel), the error rate for BFSK and BPSK is given by [27]

q P2(γb) = Q( 2gγb) (2.2)

2 where g=0.5 for BFSK, g=1 for BPSK and γb = α Eb/N0 is the received signal to

noise ratio(SNR) per bit.

33 To obtain the error probabilities when α is random you must average (2.2) over

the probability density function of γb defined by (2.1) with a change of variables [27].

Doing so, we obtain the error as

Z ∞ P2 = P2(γb)p(γb)dγb. (2.3) 0

With a change of variables (2.3) can be expressed as a function of K, g, γb (average

2 SNR/bit γb = 2σ Eb/N0) [28]. The results of (2.3) are shown in Figure 2.3. As would

be expected, Figure 2.3 shows that BFSK requires a higher γb than BPSK under the

same conditions. Figure 2.3 also shows that the NLOS case (K=0) requires a higher

γb at the receiver than when there is a dominant LOS component.

2.3 Effect of MIMO in Fading Multi-Path Channels

It is well known that multiple transmit and receive antennas increase capacity and

thoughput in wireless systems. The type of system that may be described as MIMO

is shown in Figure 2.4. In this type of system there are t transmit and r receive

antennas so that there are L = t · r MIMO channels. To quantify the performance of this type of MIMO system, (2.3) must be modified.

Specifically, for L channels we have [22]

Z ∞ Z ∞ Z ∞ L P2M = ... P2({γbl}l=1) 0 0 0 L Y · p(γbl)dγb1dγb2 . . . dγbL (2.4) l=1

where P2M represents the BER for L multipath fading MIMO channels. Figure 2.5

shows that a system with L = 4 MIMO channels will require a significantly lower γb

34 0 10 K=0 BPSK SISO K=0 BFSK SISO -1 10 K=0 BPSK SISO K=6 BFSK SISO K=12 BPSK SISO -2 10 K=12 BFSK SISO

-3 10

-4 10

-5 10 Bit Error Rate Error Bit -6 10

-7 10

-8 10

-9 10

-10 10 0 5 10 15 20 25 30 35 Average SNR/Bit (dB)

Figure 2.3: Bit Error Rate as a function of γb for a single input single output system. Bit error rates for K=0, 6 and 12 for BFSK and BPSK are shown.

35 RX TX1 1

RX TX2 2

RX TX(t-1) (r-1)

RX TXt r

Figure 2.4: MIMO implementation using t transmit and r receive antennas

36 0 10

-5 10 10-6 10-8 -10 10 10-10 10-12

-15 10

-20 10 Bit Error Rate -25 10

-30 10 K=0 BPSK SISO K=0 BPSK 2x2 MIMO K=6 BSPK SISO -35 10 K=6 BPSK 2x2 MIMO K=12 BPSK SISO K=12 BPSK 2x2 MIMO -40 10 0 5 10 15 20 25 30 35 Average SNR/Bit (dB)

Figure 2.5: Bit Error Rate as a function of the average SNR per bit for a 2x2 MIMO system having L = t · r = 4 MIMO channels (SISO is shown for comparison). Bit error rates for K=0, 6 and 12 with BPSK are shown.

than a SISO system under that same conditions. In many consumer wireless systems

a BER of 10−6 is a typical performance level. Figure 2.5 also indicates that in order to achieve this level of performance a 2x2 MIMO system in NLOS operation (K=0) will require a γb of ∼13dB. By comparison, a SISO system of equivalent performance will require γb of ∼7dB with a dominant LOS component (K=12). Though not shown in

Figure 2.5, SISO with NLOS requires γb ∼55dB, which is a rather impractical value

since it is difficult to achieve with available chipsets and even more difficult to achieve

with emerging low cost Si chipsets.

37 2.4 Channel Model Within a Room

In determining a systems performance in a room under the described conditions,

it is useful to first use a simple analytical model to determine the received power.

2.4.1 Analytical Model for a Room

To accurately determine the received power in a room, all geometrical and mate-

rial parameters must be known. These parameters will, of course, vary significantly

when comparing different rooms under different conditions. For the purposes of this

analysis, a simple model has been developed where the received power is expressed

as

M X −jφn Pr = PLOS + ane . (2.5) n=1

2 Pr is comprised of the direct LOS component in which PLOS = PtGT GR/(4πd/λ) and the sum accounts for the multipath components on the assumption of a Rayleigh distribution with random phase between 0 ... 2π. Also, M is the number of multipath rays included in the model.

Figure 2.6 shows the results of the the model described by (2.5) for transmit and receive separations of 5m, 15m and 35m at 60 GHz. This model agrees well with published measurements in this frequency band [29–31] for a single room.

2.4.2 SNR Calculations

Equation 2.5 can be used to extract the SNR. We will use a transmit power of

+10dBm and a receiver NF of 10 dB, a reasonable performance level for Si chipsets operating at these frequencies [2, 21]. For all cases, it was assumed that multipath

38 -55

5m Separation -60 15m Separation 35m Separation -65

-70

-75

-80

-85 Received Power (dBm) Power Received

-90

-95 0 20 40 60 80 100 120 140 160 180 200 Time(ns)

Figure 2.6: Multipath delay profile for transmit and receive separation of 5m, 15m and 35m which is described by Equation 2.5 at 60 GHz

39 rays arrive every 5ns after the LOS ray shows up and continue until the last ray has arrived at 200ns. For the distances used in these calculation, extending the time duration to 400ns has virtually no impact on the results. Calculations were also made assuming the transmit and receive antennas are the same for the cases of 6dBi and

20dBi antenna gain. For all cases, the frequency of operation was 60 GHz and the required γb (for the selected limits) were obtained from the results shown in Figure

2.5. This must be scaled for the full data rate of the system. Having these values for

γb, the required SNR is

PR ( )dB = 10log(Rb) + (γb)dB + LMdB (2.6) N0 where Rb is data rate in bits/sec and LMdB is the desired link margin in dB. Typically,

LMdB = 10dB, which we have done already.

The calculated received SNR for the chosen limits of interest are shown in Figures

2.7 and 2.8. In each case, the LOS signal, of course, dominates the receiver SNR. The

NLOS SNR does show variability due to the random phase of the Rayleigh distributed multipath components. Figures 2.7 and 2.8 include 25 runs of the calculation to capture the SNR ranges. When 6dBi antennas are used it can be seen that for even short distances (5m), the received SNR is not sufficient to support 1.25 Gbps, the required peak data rate for gigabit Ethernet. However, since we cannot increase transmit power or improve the receiver NF (doing so would prevent the use of a

Si chipset) we must increase antenna gain. Thus, let us instead consider transmit and receive antennas having 20dBi gain. Now, with LOS operation, the system will support 1.25 Gbps to ranges beyond 50m. NLOS operation under these conditions is still insufficient even at ranges as short as 5m using 2x2 MIMO. Using MIMO does

40 120

110 2x2 MIMO NLOS@2Gbps 2x2 MIMO [email protected]

SISO [email protected] 100

LOS+Multipath 90

80

70 Signal to Noise Ratio at Receiver(dB)

60 LOS Blocked

50 5 10 15 20 25 30 35 40 45 50 Separation Between Transmitter and Receiver(m)

Figure 2.7: SNR at the receiver as a function of transmit and receive antenna sepa- ration with antenna gain of 6dBi assumed for both transmit and receive. Required SNR levels are shown for various conditions at 60 GHz

41 140 LOS+Multipath

130

2x2 MIMO NLOS@2Gbps

120

110 2x2 MIMO [email protected]

SISO [email protected]

100 Singal to Noise Ratio at Receiver(dB) 90 LOS Blocked

80 5 10 15 20 25 30 35 40 45 50 Separation Between Transmitter and Receiver(m)

Figure 2.8: SNR at the receiver as a function of transmit and receive antenna sepa- ration with antenna gain of 20dBi assumed for both transmit and receive. Required SNR levels are shown for various conditions at 60 GHz

42 help to reduce receiver SNR requirements but it does imply a need for additional antennas, which require space. These results indicate that LOS or near-LOS operation with good antenna gain is required for robust system operation. It can be concluded from the above, that to meet these requirements adaptive arrays with significant gain will be required for system level performance acceptable for wide scale consumer acceptance.

A number of challenges exist in implementing these adaptive arrays. For example, a 20 dBi patch array will be nearly 2x2 inches in size (depending on the implemen- tation) and more than one of these will be required for MIMO implementations.

However, this is too large for hand held applications. Compatibility with consumer electronics will likely require an antenna aperture under 1 inch.

Though Figure 2.8 represents performance using 20 dBi antennas it can be easily seen that in order for NLOS performance to be acceptable at ranges of up to 10m, 25 dBi antenna gain will be required to achieve acceptable performance. This analysis assumes that transmit power is fixed at 10dBm, which is what can be expected from a single silicon amplifier at these frequencies. We know that Pr in (2.5) is comprised of the direct LOS component in which

2 PLOS = PtGT GR/(4πd/λ) (2.7)

To reduce the antenna gain requirements, additional transmit power would be re- quired as can be seen by the numerator in (2.7). For the purpose of this analysis we will refer to the numerator in (2.7) as the Power Gain Product (PGP)

43 P GP = PtGT GR (2.8)

In this case with 25 dBi antennas and 10 dBm of transmit power, the required PGP is 60 dBm. Using this metric, a number of alternative solutions can be evaluated to determine which provide an overall systems solution.

It is possible to increase the transmit power beyond the capability of a single chip if the output of multiple amplifiers is combined. The challenge in this approach is that power combining at millimeter-wave frequencies has proven to be very challeng- ing, often requiring precision machining and waveguide combining to achieve good performance over a reasonable bandwidth [32, 33]. Even with the use of expensive fabrication techniques and materials the number of devices that can be reasonably combined is typically limited to a few and maybe up to 10. These approaches can limit practical implementations that will be required for commercial deployments.

Work has also been done in looking at spatial power combining, though these imple- mentations have also typically been complex and expensive [34,35].

In looking at the trade space between transmit power, power combining, antenna gain and antenna size we propose the use of spatial power combining using compact antenna elements that can be implemented on silicon substrates. This approach enables the power combining of a large number of silicon devices in a small space with broadband performance.

44 2.5 Chapter Conclusions

Antenna requirements for short range wireless systems that use either single an- tennas or MIMO have been examined. In order to asses performance within a room, an analytical model for the multi-path scattering environment within a room was developed. In addition, the required SNR for a range of data rates was also devel- oped, as they relate to the antenna approach used, either single antenna or MIMO.

In looking closely at system requirements, it is clear that antenna gain is required to achieve acceptable system level performance. Antenna gain on the order of 25 dBi is required when using MIMO in order to realize Gbps wireless speeds in at least near

NLOS operation within a room. Challenges associated with realizing these anten- nas, of course, relate to the size, cost and efficiency of these antennas. For practical applications, the antenna size is limited to less than one inch square so that they can be integrated with small devices without impacting the overall form factor and appearance of the device they are integrated with.

45 CHAPTER 3

MATERIAL PROPERTIES OF GOLD AND SILICON AT HIGH FREQUENCIES AND THEIR EFFECT ON EFFICIENCY FOR CANDIDATE ANTENNA ELEMENTS

3.1 Initial Antenna Element Analysis

Two spiral antenna elements were selected for comparison which are shown in

Figure 3.1 which include both straight arm and square meander-line inductive loading, since these two configurations have been shown to be the best candidates for lower frequencies [36]. A 3 dimensional conical coil was shown to be better at improving very low frequency performance but this would be very difficult to fabricate for millimeter- wave frequencies. The dimensions of these spiral elements are small so that a large number of them can be included within a space limit of 25x25mm2, since the diameter of each is only 0.65mm.

The spiral elements shown in Figure 3.1 were simulated using HFSS. All simula- tions used high resistivity silicon substrates of standard 0.5mm thickness. An initial attempt was made to simulate using FDTD tools but these broadband spiral elements did not converge using this method. Others have reported similar issues in analysis using FDTD methods for these types of antennas [37].

46 0.62mm 0.65mm

2 μm (a) Line Width (b)

Figure 3.1: Mineaturized spiral elements included in this analysis using a) straight arms and b) square meander-line inductive loading

Broadband simulations were completed with input return loss results shown in

Figure 3.3. The straight arm spiral does have a similar frequency response but does have more significant ripple in the band of interest. Here we see that the inductive loading of the spiral arms is reducing reflections from the ends of the spiral arms resulting in a much flatter frequency response to frequencies below 50 GHz. This frequency response also provides for good margin so that it will not be sensitive to variations in fabrication in the primary band of interest, 59-62 GHz.

To date, there have been two main approaches to implementing millimeter-wave antennas. Examples of these two approaches can be found in the literature but they can be categorized as integration of the antenna in the chip package [38] and implementation on chip utilizing micro-machining or other techniques to create an effective dielectric constant different than standard silicon. There are pros and cons to each of these two approaches. Integration in the package moves the antenna to

47 ceramic material is LTCC ferro-A6 with a relative permittivity and loss tangent of 5.9 and 0.002, respectively. The antenna is formed by the 10-ȝm thick Ag metallization. As shown, four ceramic layers are observed to form the package. The 1st layer is 0.38 mm thick with a truncated ground plane. The 2nd layer is the radiator layer with a small cavity of 3.8×2×0.285 mm3. The 3rd layer has a cavity of 5×3.2×0.19 mm3 and 4th layer has a cavity of 5×3.8×0.38 mm3. The above three cavity layers form a package cavity to load a chip. It should be noted that the transceiver chip is adhered to the cavity base of the common ground plane. This configuration will contribute to the shielding of the chip from the antenna. In addition, the aperture feeding scheme using via through ground is avoided here to minimize the transmission loss at high frequency band of 60 GHz. Instead the CPW single or differential feeding directly from the chip pads using bond wires is adopted. The CPW pitch is 250 ȝm. Fig. 1 (b) shows the layout of a wideband triangle (WB-triangle) AiP. The size of the whole AiP is 12.5×8×1.235 mm3. The beveled structure is well known for its wideband impedance matching capability. Here, we creatively use this structure to create a slot antenna with L = Ȝ. Fig. 1 (c) shows the layout of a differential Yagi AiP. The size of the whole AiP is 12.5×8.6×1.235 mm3. It consists of one driven element, four director elements on the back of the 1st ceramic layer, and one reflector element 100 ȝm deeper inside the same layer. Its return loss is defined as follows, § Z  Z · RL 20lg d0 (1) ¨ Z Z ¸ ©¹d0 where Zd is the input impedance of the differential driven antenna, calculated using two port Z-parameters as Zd = 2(Z11 í Z21) = 2(Z22 í Z12) and Z0 is 100 ȍ.

Fig. 2 shows the impedance matching performance of the AiPs. For WB-triangle AiP the simulated -10 dB impedance bandwidth is 9.3 GHz from 58 to 67.3 GHz. The measured S11 is lower than -10 dB from 58.3 to 59.5 GHz and from 62.5 to 65 GHz, indicating an acceptable matching to a 50-ȍ source at these frequency bands. As shown, the good agreement between simulated and measured results is observed for the frequency range of 55-59 GHz. The great discrepancy beside this frequency range may be due to the surface waves in LTCC substrate. The surface wave can be suppressed by placing metal strips around the antenna structure [17]. The metal strips will be implemented in the next generation of the AiPs. Fig. 2 also shows the simulated impedance performance of the Yagi AiP. It has a -10 dB impedance bandwidth of 9.4 GHz from 55.9 to 65.3 GHz. Fig. 3 shows the radiation patterns of the AiPs both in H (YZ) and E (XZ) planes at 64 GHz. The measured gain for WB-triangle AiP varies from 5.1 to 7 dBi over the frequency range of 59-64 GHz with a value of 6 dBi at 64 GHz. In simulation, it is also found that the WB-triangle and Yagi AiPs both have efficiency better than 93%.

y Top view z x

L= Ȝ

Explored view Connected to the without chip chip by bond wires CPW-feed antenna

CPW grounded vies

Ground plane

Bottom view Bond wires with chip IBM Chip (a) (b) (c) Fig. 1 60-GHz AiPs: (a) WB-triangle 3D view, (b) WB-triangle layout, and (c) Yagi layout.

Figure 3.2: Example layout of antenna approach using LTCC to implement wideband triangle and Yagi antennas for ”Antenna in Package” approach from [38]. Exploded layout view (a, wideband triangle b) and Yagi c) layouts are shown.

280 another material like low temperature co-fired ceramic (LTCC) which can also create challenges in addition to requiring interconnect between the chip and the antenna at high frequencies. The use of air cavities does allow for high levels of integration on silicon but micro-machining or etching away precision cavities is not well aligned with low cost high volume fabrication techniques.

In order to avoid package interconnect issues and fabrication challenges with cre- ating air cavities in silicon substrates, the proposed approach for this project is a hybrid solution that maximizes integration of millimeter-wave electronics on silicon and avoids difficult high frequency interconnects. Antenna elements can then be fab- ricated to form an array that is flip-chip mounted as shown in Figures 3.4. With this configuration, the array is fabricated on silicon so that all of the high frequency active circuits can be fabricated on the same substrate as the array. Interconnects to the array are done using flip-chip mounting and are needed only for low frequency

48 -2 Square Meander Loading -4 Straight Arms

-6

-8

-10

-12 S11 (dB) -14

-16

-18

-20

-22 0 50 100 150 200 Frequency (GHz)

Figure 3.3: Input return loss of spiral elements

and DC connection to the array. Complementary active and passive circuits can be attached to the standard substrate which may include baseband, DC regulation and other circuits. The standard substrate can be any typical substrate used for printed wiring boards and assemblies to maintain a low cost of implementation. Antenna arrays developed as part of this project will be fixed beam. Electronically steering for broadband applications are expected to be too expensive and difficult to fabricate for low cost high volume applications. To date, for example, phased arrays have not been deployed in any consumer electronics.

49 Silicon Substrate Flip-Chip Connections (DC and High Frequency RF)

Active Circuits Antenna Elements

Package Surface/Standard Substrate

Flip Chip Interconnect Active Circuits for MMW portion of transceiver

Figure 3.4: Block diagram of proposed implementation using compact spiral elements integrated with active silicon circuits

The gain response vs. frequency for each spiral is shown in Figure 3.5. Though the return loss is extremely broadband, the realized gain is not acceptable. In fact, this gain response indicates that the elements shown in Figure 3.1 must be roughly twice the current size to have realized gain that approaches what may be usable, which

I have done for the remaining analysis. Using these element the overall efficiency of each individual element would be fairly low but it will reman fairly constant as the number of elements in the array is increased so this approach favorable when combining more than a few devices. This leads to the question of what the optimum trade is between element gain, size and number of elements that can be fit within a given space. This trade analysis will be completed as part of this research plan.

In order to better evaluate the losses of the loaded spiral, the element was scaled so that it has reasonably good directivity in the frequency band around 60 GHz, as

50 0

-2 Square Meander Loading Straight -4

-6

-8

-10

-12 Realized Gain (dBi) -14

-16

-18

-20 20 40 60 80 100 120 140 160 180 200 Frequency (GHz)

Figure 3.5: Realized gain of straight and square meander inductive loaded spiral antenna elements. Gain of at least +2dB is required for acceptable system efficiency and performance.

51 shown in Figure 3.6. Here the diameter of the spiral is now about 1.1mm using a line

λ width of 3.5µm and a substrate thickness of 7 at 60 GHz. For this analysis dielectric and mismatch losses are assumed to be negligible. In comparing the performance when using gold compared with PEC, it can be seen that the ohmic losses associated with using gold represent about 15 dB of loss as shown in Fig 3.7. In this case the ohmic losses are clearly significant and result in an antenna element that is basically unusable. The use of inductive loading which may be of benefit at lower frequencies appears not to be a good approach for higher frequencies. It is important to note that the conductivity for gold in this analysis was assumed to be its well known low frequency conductivity, which is unlikely to be accurate at millimeter-wave frequen- cies.

In order to reduce the ohmic losses, the dimensions of the metalization will have to be changed. Losses can be reduced by increasing the cross sectional area to lower the overall resistance of the conductor. When increasing the thickness, there will be a limit since all of the current will be carried in the first 3-5 skin depths of the material.

Beyond the thickness, the width of the conductor can be increased. A spiral element which used line widths that are 15 times greater than the loaded spiral in Figure 3.6 is shown in Figure 3.8. The diameter of this spiral is only 0.7mm on a high resistivity

λ silicon substrate that is 4 at 60 GHz.

This initial analysis indicates that loaded spiral antennas are not a good choice

for the applications being considered here. This also leads to questions in two areas.

First, this leads to the question of what material properties to assume for those being

used in this analysis, gold and silicon. This also leads to the question of what the

52 Ohmic losses using gold Directivity (dBi) result in ~15dB of loss

Metal is Gold Gain (dBi) Gain curves assume a perfect impedance match

Gain>Directivity at some freq likely due to convergence and PEC, but they are about the same, PC memory may not allow any more mesh refinement

Scale is Directivity (dBi) Blue

Gain (dBi) Green Diameter is ~1.1mm

Metal is PEC

Figure 3.6:λ/7 Scaled substrate, version ofHigh loaded resistivity spiral used Si for analysis of losses. Substrate thickness is λ at 60 GHz on a high resistivity silicon substrate. Line width of the Line7 width is ~3.5um spiral is ∼ 3.5µm.

53 5

0

Gain Using Gold -5 Gain Using PEC Directivity

-10 Gain or Directivity (dB/dBi)

-15

-20 45 50 55 60 65 70 75 80 85 Frequency (GHz)

Figure 3.7: Directivity and gain for a scaled loaded spiral element with isolation of loss mechanisms. Directivity along with gain when using gold and PEC are shown.

54 Directivity (dBi) Red

Gain (dBi) Green

Metal is Gold

Diameter is ~0.7mm

λ/4 substrate, High resistivity Si Figure 3.8: Alternative spiral using substantially more metalization. Substrate thick- Lineλ width is ~57um ness is 4 at 60 GHz on a high resistivity silicon substrate. Line width of the spiral is ∼ 57µm.

55 5

0

-5 Directivity Gain using Gold -10 Gain or Directivity (dB/dBi)

-15

-20 45 50 55 60 65 70 75 80 85 Frequency (GHz)

Figure 3.9: Directivity and gain for a spiral element with much greater metalization. Line widths are 15x greater than the loaded spiral.

56 best antenna elements to use are. Since the final solution will be an array, it seems unlikely that spiral elements are the best choice, partly due to challenges in feeding them properly which can create challenges in packaging. In order to minimize ohmic losses, elements should be chosen such that the metalization can be maximized in the small area that the element occupies.

3.2 Antenna elements used for this analysis

In quantifying the effects of losses due to materials, three antenna elements were selected. Here, two elements that maximize metalization within their occupied area have been chosen along with a spiral element that also maximizes metalization. The spiral element is included for comparison, though it is not expected to be the element of choice due to challenges in feeding an array of spirals along with challenges in true compact implementations that may be inherent in the spiral. The details of each antenna are shown if Figures 3.10-3.12. Each of these antenna elements was designed so as to minimize their size while achieving a 3dB gain bandwidth covering the frequencies of 55-65 GHz. In each case the feed is assumed to be 50Ω.

The spiral element shown in Figure 3.10 has it’s outer diameter as small as it can be to meet the gain target at the lower end of the band. Spiral antennas were first introduced in the 1950s [39] when it was demonstrated that constant input impedance and circular polarization over a wide range of frequencies can be achieved. Since then much progress has been made in improving the understanding and performance of spiral antennas [40–45]. More recently [36, 46] it has been shown that the use of inductive loading enables the reduction in size of spiral antennas. This previous work has looked at both lumped element and distributed element loading but only at lower

57 frequencies which are below 10 GHz. At much higher frequencies, lumped element

loading is not an option due to the size of the antenna. Distributed element loading

can be implemented using standard semiconductor lithography techniques.

It is well known that in a spiral antenna radiation occurs when the currents from

nearby arms are in phase for constructive radiation which occurs at the radius cor-

responding to a circumference length of c = λ. The lower limit of the operational

frequency is determined by the spiral diameter. The upper frequency limit is deter-

mined by the arm details near the center, which improves with tighter arms near the

center. In this case, the spiral outer dimensions have been chosen so that the lower

end of the frequency band is covered. In this implementation, the outer diameter is

610µm over a ground plane with substrate thickness of 367µm which is about λ/4 in

silicon.

A modified bow-tie antenna is the 2nd antenna element used for this analysis, as

shown in Figure 3.11. Here, in order to have the antenna resonant in the band of

interest, the overall dimensions of the antenna are 360x135µm. A typical bow-tie

antenna will have two triangular shaped sections with their narrowest side at the

center, with the center being the most common feed point. In this case, the bow-tie

element is edge fed with rectangular sections defined to that the element remains

symmetric while allowing for customization of the shape to allow for it to resonate

only in the band of interest. In this case, separation between the antenna and the

ground plane is a little more than λ/4 at 423µm. The bow-tie element essentially appears as a load at the end of the and looks somewhat like a defect at the end of the transmission line that radiates.

58 367μm

610μm

51μm width

Figure 3.10: Dimensions of spiral element on silicon

59 1260x405μm

423μm

Overall 45x135μm 360x135μm

67.5x135μm

45x112μm 45x27μm

Figure 3.11: Dimensions of edge fed bow-tie element on silicon

60 810x1260μm

423μm

31μm width

150μm 239μm

479μm

Figure 3.12: Dimensions of half circle element on silicon

This final antenna used for this analysis is essentially a half circle at the end of the feed with a slot that allows for some amount of tuning so that a good impedance match can be achieved. The separation from the ground plane is the same as was used for the bow-tie element but the half circle element is a little larger overall. The addition of the slot allows for some alteration in where the currents will flow. Section of the antenna may essentially become resonant tuning elements in combination with the antenna. This is analogous to adding a tuning to the antenna or transmission line. This technique has been used with other elements like patches [47].

61 The performance of each was determine by simulation in HFSS using ideal or loss- less materials. The realized gain of each is shown in Figure 3.13. The corresponding radiation patterns for each are shown in Figures 3.15, 3.16 and 3.17. Each of the elements meets the target of having a 3dB bandwidth that covers 55-65 GHz with a minimum gain of 3 dB, though they may not be centered exactly. For the purposes of this analysis, the performance of each is more than adequate. The spiral and half circle antennas have slightly higher gain, 4-5 dB compared with the peak gain of the bow-tie which is 1-2 dB less. This is not really unexpected, though, since the bow-tie antenna is the smallest of the three. The efficiency of each antenna is shown in Figure

3.14. As expected, each has an efficiency of nearly 1, since all lossless materials were used. In some cases, the efficiency does go above 1, but this is due to convergence or round off errors in the simulations which are not uncommon when using all lossless materials. Of course, the actual efficiency can not really exceed 1. Next we will look more closely at material properties to determine how to most accurately simulate the effect of the losses associated with real materials that would be used in the fabrication.

3.3 Electrical Properties of Gold

Material properties for gold have been looked at very closely at very high frequen- cies that fall within wavelengths that would be referred to as optical [48,49] which is the range of 30-300 THz. Based on measurements of material properties at optical frequencies, the complex permittivity can be extracted. For a homogeneous isotropic medium the wave number is given by

62 6 Spiral 5

4

3 Bowtie 2

1 Realized Gain (dB)

0 Half Circle

-1

-2 50 52 54 56 58 60 62 64 66 68 70 Frequency (GHz)

Figure 3.13: Gain of the antennas shown using all lossless materials

63 1.1

spiral 1.05 bowtie

1

0.95

Efficiency Half Circle

0.9

0.85

0.8 50 52 54 56 58 60 62 64 66 68 70 Frequency (GHz)

Figure 3.14: Efficiency of the antennas using all lossless materials

64 60GHz

65GHz

55GHz

Figure 3.15: Antenna pattern for spiral antenna at 55, 60 and 65 GHz.

65 60GHz

65GHz

55GHz

Figure 3.16: Antenna pattern for bow-tie antenna at 55, 60 and 65 GHz.

66 60GHz

65GHz

55GHz

Figure 3.17: Antenna pattern for half circle antenna at 55, 60 and 65 GHz.

67 σ k2 = ω2µε = ω2µ(ε − j ) (3.1) 0 ω where ε = ε0−jσ/ω is the complex permittivity of the medium. At optical frequencies n and α are real and are the standard refraction index and the attenuation constant which have available measured values [49]. These parameters can be used to determine the complex permittivity of a material using the relation

ε = ε0 − jε00 = (α2 − n2) − j2nα. (3.2)

Using this data at optical frequencies, the behavior at lower frequencies was extrapo- lated. In order to most accurately determine the lower frequency values, the behavior of these curves below the region where measured data exists is important. It is known that good conductors like gold will exhibit characteristics of a Drude metal, so that the general shape of the response is known and can be predicted with a Drude model.

The real and imaginary part of the permittivity at lower frequencies are given by [50]

2 0 ωp ε (ω) = 2 2 − P (3.3) ω + ωτ and 2 00 ωpωτ ε (ω) = 2 2 (3.4) ω(ω + ωτ ). In this case the frequency is ω with the other parameters that must be extracted being the plasma frequency ωp relaxation frequency ωτ and the polarization term P . The polarization term is used to account for contributions to the total dielectric function that includes contributions due to conduction electrons and the interband transitions.

Each of these terms are discussed in [50].

68 Since the Drude parameters can be chosen to fit a particular frequency range, a low frequency reference was required to optimize the fit and both high and low frequencies where measured data exists. High frequency ε is given by (3.2) and was used in combination with the well known low frequency conductivity of gold

σ = 4.1x107S/m. In order to relate all of the parameters together in the best way the frequency dependant imaginary part of the permittivity can be related to the conductivity using [51]

00 σ(ω) = ωε0ε (ω). (3.5)

Using (3.3),(3.4) and (3.5) and simultaneously fitting to the known high and low fre- quency values, the Drude parameters can be extracted. In this case the parameters were extracted while defining the angular frequency in eV. These can be easily con- verted to rad/s using the equivalence 1eV = 1.519x1015rad/s. The resulting Drude parameters that provide a good fit are ωp = 9.5eV , ωτ = 30meV and P = 100. The resulting extrapolated curves for ε0 ,ε00 and σ over frequency are shown in Figures 3.18,

3.19 and 3.20.

It is shown that good agreement was achieved for this extracted Drude parameters so that both the real and imaginary part of ε agree well with data from [49] along with a low frequency conductivity of about 4.1x107S/m. It can be concluded then, that the frequency response for each of these parameters in the extrapolated region is accurately given by (3.3) and (3.4).

When looking more closely at the conductivity of gold as a function of frequency as shown in Figure 3.20, it can be seen that its frequency response if very flat at

69 10 10

8 10 ε’’ 6 10 Extrapolated (Solid) 4 10 Measured (markers)

Imag(Epsilon) 2 10

0 10

-2 10 5 10 15 20 10 10 10 10 Frequency (Hz)

Figure 3.18: Measured data (markers) and extrapolated values (solid) for the imag- inary part of the permittivity of bulk gold as a function of frequency. Extrapolated values were determined by fitting a Drude model known values at low and optical frequencies.

70 6 10

5 10 ε’ 4 10

3 Extrapolated 10 (Solid)

2 10 Measured 1 Real(Epsilon) 10 (Markers)

0 10

-1 10

-2 10 5 10 15 20 10 10 10 10 Frequency (Hz)

Figure 3.19: Measured data (markers) and extrapolated values (solid) for the real part of the permittivity of bulk gold as a function of frequency. Extrapolated values were determined by fitting a Drude model to known values at low and optical frequencies.

71 10 10

9 10

8 10

7 10

6 10 Conductivity (S/m) Conductivity

5 10

4 10 5 10 15 20 10 10 10 10 Frequency (Hz)

Figure 3.20: Conductivity of bulk gold calculated from an extrapolation of measured properties at optical frequencies.

72 lower frequencies with the knee being well above the frequency band of interest for this analysis. We can then conclude that the bulk conductivity value for gold that is commonly used at low frequencies is also valid in the millimeter-wave band. This does, however, assume that the gold metallization is smooth enough so that surface roughness does not have an impact, which is the assumption we will make for this analysis.

3.4 Electrical Properties of Silicon

In order to determine the electrical properties of silicon for the band of interest we will separate two contributions to its loss characteristics. When data for silicon substrates is given, such as the resistivity, this is a low frequency measured parameter.

It is well known that silicon substrates are comprised of intrinsic silicon and a dopant.

Intrinsic silicon itself is very difficult to produce but is in general a low loss material.

The substantial conductivities seen in silicon substrate are essentially due to the dopant. For this analysis, we will separate these two contributions and assume that the effective loss tangent for doped silicon can be expressed as

tanδe = tanδsilicon + tanδdopant (3.6) where there are contributions from the dopant and intrinsic silicon. Measurements for silicon equivalent to those for gold exist [52] and can be used in combination with

(3.2) and (3.5) to extrapolate data for the band of interest. The Drude model used for gold is not valid for silicon, since it best describes the frequency behavior of good conductors. Using the measured data from [52] in combination with the known low frequency ε0 = 11.87 and σ = 4−5x10−4S/m [51] for intrinsic silicon, curves can be fit

73 to extrapolate. These real and imaginary parts of the permittivity can be represented

by exponential functions in the region of interest.

ε0 (f) = e1.81x10−45f 2−8.31x10−32+1.28x10−16 (3.7)

ε00 (f) = e7.32x10−31f 2+1.12x10−14−7.71 (3.8) where f is the frequency in Hz.

Plots of the measured and extrapolated data is shown in Figures 3.21, 3.22 and

3.23. The extrapolated curves fit the known optical and low frequency data so we can conclude that the extrapolated data is accurate as long as there are no resonances in these parameters in the extrapolated region. In comparison, gold does have some resonances but these do not occur until you reach very high frequencies that are in the infrared range.

In order to calculate the effective loss tangent tanδe the two contributions defined in (3.6) will be defined as

ε00 (f) tanδ = (3.9) silicon ε0 (f)

σd tanδdopant = . (3.10) ωε0εr

where σd is assumed to be directly from the low frequency resistivity of the silicon

substrate. Each of the equations (3.6), (3.9) and (3.10) are shown in Figure 3.24. At

74 20

Measured 18 ε’ (Markers)

16

Extrapolated (Solid) 14

Real(Epsilon) 12

10

8 8 10 12 14 16 10 10 10 10 10 Frequency (Hz)

Figure 3.21: Measured data (markers) and extrapolated values (solid) for the real part of the permittivity of intrinsic silicon as a function of frequency. The curve was fit to match high frequency measured data and the know low frequency value.

75 2 10

Measured Data (markers) 1 10 ε’’

0 10

-1 10

Imag(Epsilon) -2 10 Extrapolated (solid)

-3 10

-4 10 8 10 12 14 16 10 10 10 10 10 Frequency (Hz)

Figure 3.22: Measured data (markers) and extrapolated values (solid) for the imag- inary part of the permittivity of intrinsic silicon. The curve was fit to match high frequency measured data and known low frequency conductivity.

76 6 10

4 10

2 10

0 10

-2 10 Conductivity (S/m)

-4 10

-6 10 8 10 12 14 16 10 10 10 10 10 Frequency (Hz)

Figure 3.23: Conductivity of intrinsic silicon calculated from an extrapolation of measured properties at optical frequencies. Low frequency conductivity was used as part of curve fitting for the imaginary part of the permittivity.

77 8 10 10Ω-cm

6 10 tanδe = 1kΩ-cm tanδδsilicon+ tan dopant

4 10 2kΩ-cm

2 10 11kΩ-cm

Loss Tangent (microradians) Loss Tangent Intrinsic Si 0 10

-2 10 8 10 12 14 16 10 10 10 10 10 Frequency (Hz)

Figure 3.24: Loss Tangent of intrinsic silicon calculated from an extrapolation of mea- sured properties at optical frequencies. Loss tangent contributions from the dopant, intrinsic silicon and their sum are shown

78 lower frequencies, the contribution from the dopant dominates the loss tangent. There

is a region where tanδdopant and tanδsilicon are of comparable contributions which is dependant on the dopant resistivity for the substrate. At high frequencies, intrinsic silicon begins to dominate the characteristics of the substrate. This analysis indicates that the contribution from the dopant is still the dominant contributor for the material properties in the 60 GHz band. At higher frequencies, the contribution from intrinsic silicon must be taken into account for accurate predictions of performance.

3.4.1 Measured Results for the Spiral Element

The spiral element shown in Figure 3.10 was fabricated using high resistivity silicon and gold metalization. The spiral antenna impedance was measured using the

Agilent V-band extension to an 8510C automatic network analyzer. The wafer was placed directly on the metal chuck of an Alessi wafer probe station. To accommodate the relatively narrow pad spacing on the spirals, a 60µm pitch GSG Pico Probe was acquired from GGB industries and mounted to the probe station. Parameters for the probes were entered into the 8510C and a 1-port Short Open Load calibration was performed using the Cascade Microtech ISS (Impedance Standard Substrate).

Measured results from 4 spiral elements on the same substrate along with pre- dicted performance are shown in Figure 3.25. Good agreement between predicted and measured performance was achieved, though there is some variation in the mea- sured results. Probe pads with 60µm pitch were added to the center of the spiral to

allow for measurement. This small pitch was needed because standard probe pitches

would not fit within the center of the spiral, do to it’s small size. In addition to

parasitics from these pads, the probes used for the measurement had to be in close

79 0

-2

-4

-6

-8

-10 Simulated S11(dB) -12

-14

-16

-18

-20 55 55.5 56 56.5 57 57.5 58 58.5 59 59.5 60 Frequency (GHz)

Figure 3.25: Measured input return loss for the spiral antenna. Thick solid like is the simulated result and is shown with the measured response of 4 fabricated spiral elements

proximity to the spiral itself, while the measurement was being made. Since the probes were very close to the spiral, differences in probe placement can also result in differences.

3.5 Material Property Effects on Efficiency

This analysis has shown that the material properties for gold and silicon that should be used when developing on chip antennas in the 60 GHz band are the low frequency conductivity for gold (assuming the surface is smooth enough so that surface

80 1

Spiral 0.95

0.9

bowtie Half Circle 0.85 Efficiency

0.8

0.75

0.7 50 52 54 56 58 60 62 64 66 68 70 Frequency (GHz)

Figure 3.26: Efficiency of the antennas with gold metallization and lossless silicon

roughness does not significantly impact performance) along with the conductivity

defined by the low frequency resistivity measurement of the silicon substrate. In

order to determine the effects of real gold and silicon properties on the three antennas

discussed earlier, simulations where run using 10Ω · cm and 1kΩ · cm silicon.

The contribution to degradation in efficiency due to the conductivity of gold is shown in Figure 3.26. Comparing these results with those shown in Figure 3.14, it can be seen that the additional loss is minimal and on the order of a few percent.

Since the loss contribution from gold is not significant for these antennas, gold will be included in simulations that also add losses associated with silicon.

81 1 Spiral

0.95 Half Circle

0.9

0.85 Efficiency bowtie 0.8

0.75

0.7 50 52 54 56 58 60 62 64 66 68 70 Frequency (GHz)

Figure 3.27: Efficiency of the antennas using gold and 1kohm-cm silicon.

When adding the resistivity of 1kΩ · cm to silicon, the overall efficiency of the antenna does degrade some but remains in the range of 0.9 to 0.95, which is still reasonably good. This indicates that the losses due to gold and a relatively high resistivity substrate still allow for the implementation of a good performing antenna on chip. The spiral does appear to have slightly better efficiency compared with the other two , but the spiral does not include any losses due to the interconnect or feed. The cases for the bow-tie and half circle each have some loss associated with the feed line in the simulation.

82 0.25

0.2

Spiral

0.15

Efficiency 0.1 Half Circle

0.05 bowtie

0 50 52 54 56 58 60 62 64 66 68 70 Frequency (GHz)

Figure 3.28: Efficiency of the antennas using gold and 10ohm-cm silicon.

As expected, there is clearly a limit to the level of conductivity for the silicon substrate that will allow for a good antenna implementation. When using a relatively conducive substrate with a resistivity of 10Ω · cm, the efficiency drops to 0.2 or less, as shown in Figure 3.28. This drop in efficiency does move the performance to a level that is unlikely to be acceptable in any application. Again, the spiral does appear to have slightly better efficiency but the other two antennas do include some losses associated with the feed line that are not included for the spiral, though they would be in practice.

83 3.6 Chapter Conclusions

Three antenna candidates have been proposed that have bandwidth for the un-

licensed millimeter-wave band. Of these, the spiral had the largest dimensions of

610µm, where the bow-tie had a maximum dimension of 360µm. A careful study

was done to evaluate the conduction and substrate losses. The electrical properties of

both silicon and gold were examined closely to determine the parameters that should

be used in simulated the proposed antenna elements to assess losses and efficiency. In

using the Drude model for gold and curve fitting for silicon, it was determined that

the low frequency conductivity for gold and the substrate resistivity for silicon are

the proper parameters to be used in the analysis. The main conclusion was that the

standard 10Ω · cm substrate leads to losses that yield efficiencies of only 20% but use of 1kΩ · cm or higher silicon does allow for antennas to be realized with efficiencies

that are quite good.

84 CHAPTER 4

ANTENNA ARRAY IMPLEMENTATION

4.1 overview

As an initial analysis in evaluating the array performance, an array factor was used in addition to the element parameters for initial analysis. Initially we will assume that element to element coupling effects for the time being are minimal or negligible.

In looking at the system level performance, there are multiple contributors. Key contributors to the system performance include the antenna element gain, the array configuration used and the maximum output power of each silicon amplifier in the transmitter. Since we have proposed to use spatial power combining in order to keep the overall antenna size in an acceptable range, as the antenna array grows in size, both antenna gain and transmit power will increase.

4.2 Array Analysis

The array configuration for this preliminary analysis is shown in Figure 4.1, where we have set element spacing Dx and Dy to 1mm for a square array, which has the same number of elements in each direction, x and y. As the number of elements and size of the array increases, it is important to note that both the antenna gain

85 increases as well as the transmit power, since here we assume that each element is

driven by a silicon amplifier that is capable of providing 10 dBm of RF power. Other

variations could also be implemented, where a single amplifier could drive more than

one antenna element. The array analysis is limited, for now, to the use of the half

circle with , since its performance considered good for the size of the element used.

The same analysis can be carried out of a wide range of antenna elements and array

configurations.

4.3 The Array Factor

In order to determine the antenna array performance, the effects of using multiple

antenna elements in an array must be quantified. As mentioned, the array configu-

ration considered here is shown in Figure 4.1 but first we will consider the case of a

linear array as shown in Figure 4.2. If we were to assume that we have a linear array

with N elements along the x axis, we know that the array factor can be expressed

as [53] N X j(n−1)(kdx sin θ cos ϕ+βx) AF = I1ne (4.1) n=1 where dx and βx represent the spacing and progressive phase shift between elements along the x-axis, respectively. I1n in (4.1) is the coefficient of excitation of each element. This expression can be expanded to describe a rectangular array, as is our interest here by modification to address elements in each direction as

N " M # X X j(m−1)(kdx sin θ cos ϕ+βx) j(n−1)(kdy sin θ sin ϕ+βy) AF = I1n Im1e e (4.2) n=1 m=1

86 z

dy y dx

x

Figure 4.1: Planer rectangular array geometry

87 z

y dx

x

Figure 4.2: Planer linear array geometry

88 Further, if the amplitude excitation coefficients of the rectangular array in the

x-direction are proportional to those in the y-direction then the amplitude of the

(m, n)th element can be written as

Imn = Im1I1n. (4.3)

If we further assume that the excitation of the array is such that the amplitude is uniform over the array (4.2) can be simplified and written as

M N X j(m−1)(kdx sin θ cos ϕ+βx) X j(n−1)(kdy sin θ sin ϕ+βy) AF = I0 e e (4.4) m=1 n=1 so that we now have an expression for the array factor of a rectangular MxN array which can be used for both rectangular and square arrays.

4.4 Array and System Performance

In order to determine the overall array performance we must apply (4.4) to deter- mine the increase in antenna gain due to the array configuration we have chosen. In applying this to the circular element, the results shown in Figure 4.4 were obtained.

The array configuration is a square array, where we have set element spacing Dx and Dy to 1mm. The array analysis is limited to the use of the half circle element.

The array performance is shown in Figure 4.4. This represents the antenna array gain when using the half circle elements in this configuration as a function of N.

Antenna gain of more than 20 dBi is achieved with an 7x7 array. With the used element size of only 479µm and element to element spacing of 1mm, this array only occupies 7x7mm2 compared to a goal of 25x25mm2. The data shown in Figure 4.4 also indicates that the desired antenna gain of 25 dBi can be achieved with a 12x12

89 … NxN Elements …

… …

Dx Dy

… … … …

Figure 4.3: Rectangular array configuration used in this analysis. Element to element spacing is 1mm for this analysis.

90 array. An array of this size will fit within an area of about 12x12mm2, assuming that the antenna efficiency is high. Any additional losses that result in a reduction in the array efficiency will result in the need for a larger array in order to compensate. It is also important to note that the larger the array the more difficult it is to achieve high efficiency since the feeding structure and potential variations in fabrication will typi- cally increase along with the array size. Beyond the antenna size itself, the efficiency of the array is important in another key system parameter which is the overall sys- tem efficiency. Lower efficiencies mean that more system power is required to operate which is undesirable for portable battery powered electronics, since lower efficiency results in diminished battery life.

In looking more closely at how standard combining techniques may be used to feed a planer array, we will look more closely at feeding an 8x8 array, as shown in Figure

4.5. In order to feed each element with uniform amplitude and phase, a relatively large feed structure is required. Here, we must essentially break the array into subsections that have uniform feeding. It should be further noted that in order to feed the elements in uniform amplitude and phase, the total number of elements in the array must be multiples of 4 elements in each direction. The resulting line length needed to feed each element increases substantially. In this example, the line length from the feed to each element is about 15mm. If we use the expected losses shown in Figure

1.11, which represent the line losses per millimeter on a standard silicon substrate with thin dielectrics, the additional loss in this feed is about 30dB. Of course, this loss will increase as the array size increases. Clearly, this loss makes this type of combining non-practical. These losses can be less if using the type of silicon processing we have proposed but this would also create challenges in implementing the transmission

91 line. If the elements are realized without a ground plane the transmission line would

have to be implemented using co-planer waveguide. Using co-planer waveguide for

this application has challenges associated with placing ground close to the radiating

elements further adding parasitics. In order to realize a compact array with good

efficiency, the use of spatial power combining is likely required. An exploded view of

the low loss combining approach is shown in Figure 4.6. Here, each element of the

array is fed directly from a silicon power amplifier. In doing so, the losses associated

with the feed network are eliminated in addition to the increase in transmit power

due to the power combining of the output of multiple power amplifiers.

In order to capture the effects of spatial power combining, we must also determine

the additional transmit power due to feeding each antenna element with it’s own

silicon power amplifier. Using again, the expected transmit power for a single silicon

power amplifier of 10 dBm, we can multiply to output power of a single amplifier

by the number of elements in the array to determine the total power transmitted.

We are, of course, interested in how this effects the system performance, so we must

take into about the P GP that was defined in (2.8). As previously discussed, in

order to meet acceptable system level performance we require a P GP of at least

60 dBm. By using the P GP we are able to simultaneously take into account both the antenna gain and the transmit power of the system. Results for the analysis using the half circle elements are shown in Figure 4.7, where the realized array gain, combined transmit power and the calculated P GP are shown. These results indicate

that in order to achieve acceptable performance using a single silicon amplifier, a

12x12 array is needed. When using spatial power combining, the required array size

is significantly reduced so that acceptable performance can be achieved with a 6x6

92 30

25

20

… NxN Elements … 15 … … 10 D Realized Array Gain (dB) x Dy

5 … … … …

0 1 2 3 4 5 6 7 8 9 10 11 12 N for the NxN Array

Figure 4.4: Effect of increasing array size using half circle element antennas. Oper- ating frequency is 60 GHz with element to element spacing of 1mm. Antenna gain here does not include losses associated with the feed network.

93 Feed Point

Figure 4.5: Planer antenna array feed for an 8x8 array of half circle elements. In order to feed each element in phase, substantial line length is needed. In this example, at least 15mm of line length is needed to feed each element, resulting in substantial loss and reduction in efficiency.

94 Antenna Elements with Silicon substrate millimeter-wave electronics

Optimum Antenna Elements Baseband, DC, Low frequency electonics Determined as Part of this Project Flip-Chip Interconnect

Standard substrate/package

Figure 4.6: Exploded view layout of proposed approach for optimum integration on silicon for millimeter-wave transmit array.

array and significant extra margin can be realized when using a 7x7 array. With the inter-element spacing used these arrays can be realize in 6x6mm2 or 7x7mm2 respectively. When comparing with the overall size goal of 25x25mm2, this indicates that it is possible to realize a system as described here using antennas well under the size limit goal,

4.5 Chapter Conclusions

The antenna array gain was formulated and it was determined that using the half circle elements the required gain of 25 dBi can be achieved with a 25x25 element array. This does not, however, account for loss in efficiency due to the feed network.

In implementing the feed network, the size of the arrays used are limited in that they must allow for feeding with uniform amplitude and phase. In doing so, substantial

95 90 Calculated PGP 80 PGP Goal 70

60

50 Total Combined TX Power (dBm) 40

30

20

Realized Array Gain, Power PGP or (dB, dBm) 10 Realized Array Gain (dB) 0 1 2 3 4 5 6 7 8 9 10 11 12 N for the NxN Array

Figure 4.7: Effect of increasing array size given that each element is driven by a silicon amplifier providing 10 dBm of output power. PGP is defined in (2.8). Operating Frequency is 60 GHz with element to element spacing of 1mm.

96 line length is added. In the example of and 8x8 array shown, at least 15mm of line length was added between the feed point of the array and each element, adding substantial loss. In looking more closely at spatial power combining, the issues of feed network losses along with limits on the size of the array of uniform feeding are eliminated. In addition, the constraints of transmit power from a single silicon amplifier are eliminated. Here, it has been shown that by using this approach, the required system performance can be achieved, with margin, using a 7x7 array that occupies about 1/3 of the antenna size goal of 25x25mm2.

4.5.1 Design Guidelines

Since one of the key challenges in realizing millimeter-wave antennas is loss or efficiency, it is important to focus on areas that will influence this. First, as is the case at lower frequencies, it is important to select substrate materials that are sufficiently low loss at high frequencies. In selecting substrates it is also important to look closely at the frequencies that the loss characteristics have been characterized at. For example, a polyamide substrate may appear to have low loss characteristics but the loss tangent is typically valid only for low frequencies, maybe up to 10 GHz. At much higher frequencies a substrate like this will no longer be considered low loss.

An example of a substrate that does have good loss characteristics at high frequencies is liquid crystal polymer (LCP). Also part of the substrate selection is the relative permittivity of the substrate. It is important to note that this will have an effect on the ohmic losses, since a higher relative permittivity will typically results in reduced sizes in antennas which can impact ohmic losses.

97 Beyond the dielectric losses, the ohmic losses should also be considered carefully.

When using higher dielectric materials or thin dielectric layers, it is important to realize that the increased capacitance resulting from these essentially reduces the dimensions of the metallized areas, so the substrate selected will also have an impact.

Often, this is a desirable effect since in many applications smaller antennas and passive elements may be preferred since in many applications smaller passives allow for higher levels of integration. This can be especially important at very low frequencies where the physical size of antennas are so large they cannot be integrated in hand held devices when they are a significant percentage of a wavelength in size. However, in reducing the metallization it is also important to realize that doing so will often increase the ohmic losses. When selecting an antenna element at high frequencies, it is important to consider mainly those that allow for maximum metallization in the area that the antenna will occupy. As usual, there are a wide range of possible element types that can be selected but the choices are reduced when applying this constraint.

There are a range of elements that may occupy approximately the same area but will have substantially less metallization in the area and will have substantially higher losses.

The detailed analysis done here, shows clearly that antenna gain will be required to achieve good system performance at millimeter-wave frequencies. Realizing this antenna gain in a reasonable size that is also low cost to fabricate guides efforts in the direction of planer arrays. In fabricating a planer array choices include traditional printed circuit board techniques in addition to semiconductor fabrication, since at millimeter-wave frequencies on-chip antennas become a possibility. In realizing these antennas on silicon, it is important to use high resistivity substrates with refractivities

98 on the order of 1kΩ · cm or more. When designing antennas, the use of the well known low frequency conductivity of the metallization and the rated resistivity of the substrate are sufficient. At much higher frequencies (in the THz range), these values may no longer be valid.

Power combining for large planer arrays can have a significant impact on the efficiency of the antenna. Since uniform amplitude and phase feeding is typically desired, the effective length of the feed line is significantly increased. In order to reduce the losses associated with the feed and to increase the overall transmit power, spatial power combining offers great potential. In doing so, both the transmit power and the efficiency of the antenna can be increased substantially. Due to the high frequency of operation, this type of combining can also be implemented in a small size with very high levels of integration, with a combination of on-chip and system in package approaches.

99 CHAPTER 5

CONCLUSIONS AND SUMMARY OF CONTRIBUTIONS

5.1 Summary and Conclusions

The motivation for this work is to take advantage of the 60GHz unlicensed spec- trum for high speed wireless communications. So, this dissertation provides for the

first time

• Guidelines for antenna design were developed; In this regards, indoor propa-

gation models were adapted, with a new analytical model developed, for the

millimeter-wave spectrum to establish minimum criteria for data rates within

a room under both LOS and NLOS conditions. This model also takes into

account the effects of MIMO on channel capacity.

• Several standard and novel antenna elements were proposed and examined for

performance at millimeter-wave frequencies, with packaging issues taken in to

account.

• Material loss models were development and employed for the analysis and char-

acterization of the antenna elements on 10 Ohm and 1k-Ohm Silicon substrates.

100 • Measurements were provided by AFRL on spiral elements that were fabricated

for use in this band on silicon substrates. Predicted performance agreed well

with measurements, which validates the material loss models that have been

developed.

• Spatial combining arrays were recommended and design guidelines were given

to satisfy the high data rate and distance/coverage requirements.

• It was found that by using this approach, an array of 7x7 elements (making

it about 1/3 of the overall size limit) can deliver a capacity of 1.25 Gigabits

per second at distances greater than 50m in open space and the same capacity

indoors at ranges of up to 10m under NLOS conditions when utilizing MIMO.

101 BIBLIOGRAPHY

[1] R. Emrick, S. Franson, J. Holmes, B. Bosco, and S. Rockwell. Technology for emerging commercial applications at millimeter-wave frequencies. Wireless Com- munications and Applied Computational Electromagnetics, 2005. IEEE/ACES International Conference on, pages 425–429, 2005.

[2] C. H. Doan, S. Emami, D. Sobel, A. M. Niknejad, and R. W. Brodersen. 60 GHz CMOS radio for Gb/s wireless LAN. Radio Frequency Integrated Circuits (RFIC) Symposium, 2004. Digest of Papers. 2004 IEEE, pages 225–228, 2004.

[3] R.M. Emrick and J.L. Volakis. Antenna Requirements for Short Range High Speed Wireless Systems Operating at Millimeter-Wave Frequencies. Microwave Symposium Digest, 2006. IEEE MTT-S International, pages 974–977, 2006.

[4] L. Yujiri, M. Shoucri, and P. Moffa. Passive millimeter wave imaging. Microwave Magazine, IEEE, 4(3):39–50, 2003.

[5] P.H. Siegel. THz applications for outer and inner space. Electromagnetic Com- patibility, 2006. EMC-Zurich 2006. 17th International Zurich Symposium on, pages 1–4.

[6] T.J. Muller, W. Grabherr, B. Adelseck, E. Microwave Factory, and G. Ulm. Surface-mountable metalized plastic waveguide filter suitable for high volume production. Microwave Conference, 2003. 33rd European, 3, 2003.

[7] J. Hirokawa and M. Ando. 40 GHz Parallel Plate Slot Array Fed by Single-layer Waveguide Consisting of Poast in a Dielectric Substrate. IEEE Antennas and Propagation Society International Symposium, 1998.

[8] D.M. Pozar. Aperture coupled waveguide feeds for microstrip antennas and microstrip couplers. IEEE AP-S Int. Symp. Dig, 1:700–703, 1996.

[9] K. Maruhashi, M. Ito, L. Desclos, K. Ikuina, N. Senba, N. Taka- hashi, and K. Ohata. Low-cost 60 GHz-band antenna-integrated transmit- ter/receivermodules utilizing multi-layer low-temperature co-fired ceramictech- nology. Solid-State Circuits Conference, 2000. Digest of Technical Papers. ISSCC. 2000 IEEE International, pages 324–325, 2000.

102 [10] T. Seki, K. Nishikawa, and K. Cho. Multi-layer parasitic microstrip array an- tenna on LTCC substrate for millimeter-wave system-on-package. Microwave Conference, 2003. 33rd European, 3, 2003.

[11] R. Bairavasubramanian, D. Thompson, G. DeJean, G.E. Ponchak, M.M. Tentzeris, and J. Papapolymerou. Development of mm-wave dual-frequency mul- tilayer antenna arrays on liquid crystal polymer (LCP) substrate. IEEE AP-S Int. Antennas and Prop. Symp. Dig, pages 393–396, 2005.

[12] C.T.C. Nguyen, L.P.B Katehi, and G.M. Rebeiz. Micromachined devices for wireless communications. Proceedings of the IEEE, 86(8):1756–1768, 1998.

[13] G.M. Rebeiz, D.P. Kasilingam, Y. Guo, P.A. Stimson, and D.B. Rutledge. Mono- lithic millimeter-wave two-dimensional horn imaging arrays. Antennas and Prop- agation, IEEE Transactions on, 38(9):1473–1482, 1990.

[14] G.M. Rebeiz. Millimeter-wave and terahertz integrated circuit antennas. Pro- ceedings of the IEEE, 80(11):1748–1770, 1992.

[15] N.G. Alexopoulos, P.B. Katehi, and D.B. Rutledge. Substrate Optimization for Integrated Circuit Antennas. Microwave Theory and Techniques, IEEE Trans- actions on, 83(7):550–557, 1983.

[16] G.M. Rebeiz, W.G. Regehr, D.B. Rutledge, R.L. Savage, and N.C. Luhmann. Submillimeter-wave antennas on thin membranes. International Journal of In- frared and Millimeter Waves, 8(10):1249–1255, 1987.

[17] A.D. Semenov, H. Richterr, H.W. H¨ubers, B. G¨unther, A. Smimov, K.S. Il’In, M. Siegel, and J.P. Karamarkovic. Terahertz performance of integrated lens antennas with a hot-electron bolometer. IEEE transactions on microwave theory and techniques, 55(2):239–247, 2007.

[18] J.W. Bowen, S. Hadjiloucas, B.M. Towlson, L.S. Karatzas, S.T.G. Wootton, N.J. Cronin, S.R. Davies, C.E. McIntosh, J.M. Chamberlain, R.E. Miles, et al. Micromachined waveguide antennas for 1.6 THz. Electronics Letters, 42(15):842– 843, 2006.

[19] H. Murakami, S. Ono, A. Quema, G. Diwa, E. Estacio, N. Sarukura, R. Nagasaka, Y. Ichikawa, E. Ohshima, H. Ogino, et al. Zinc oxide single crystal as substrate for photoconductive antenna device generating radiation in the terahertz frequency region. Infrared and Millimeter Waves and 13th International Conference on Terahertz Electronics, 2005. IRMMW-THz 2005. The Joint 30th International Conference on, 2, 2005.

103 [20] B. Bosco, S. Franson, R. Emrick, S. Rockwell, and J. Holmes. A 60 GHz transceiver with multi-gigabit data rate capability. Radio and Wireless Con- ference, 2004 IEEE, pages 135–138, 2004.

[21] S. Reynolds, B. Floyd, U. Pfeiffer, and T. Zwick. 60GHz transceiver circuits in SiGe bipolar technology. Solid-State Circuits Conference, 2004. Digest of Technical Papers. ISSCC. 2004 IEEE International, pages 442–538, 2004.

[22] M.S. Alouini and A. J. Goldsmith. A unified approach for calculating error rates of linearlymodulated signals over generalized fading channels. Communications, IEEE Transactions on, 47(9):1324–1334, 1999.

[23] H. S. Abdel-Ghaffar and S. Pasupathy. Asymptotical performance of M-ary and binary signals overmultipath/multichannel Rayleigh and Rician fading. Commu- nications, IEEE Transactions on, 43(11):2721–2731, 1995.

[24] W. Lindsey. Error probabilities for Rician fading multichannel reception of binary and n-ary signals. Information Theory, IEEE Transactions on, 10(4):339–350, 1964.

[25] T.S. Rappaport. Wireless Communications: Principles and Practice. IEEE Press Piscataway, NJ, USA, 1996.

[26] R. Vaughan and B. Andersen. Channels, Propagation and Antennas for Mobile Communications. Institution of Electrical Engineers, 2003.

[27] J.G. Proakis. Digital Communications. McGraw-Hill, 4th edition, 2001.

[28] C.P. Lim, J.L. Volakis, K. Sertel, R.W. Kindt, and A. Anastasopoulos. Indoor propagation models based on rigorous methods for site-specific multipath envi- ronments. Antennas and Propagation, IEEE Transactions on, 54(6):1718–1725, 2006.

[29] J. Kunisch, E. Zollinger, J. Pamp, and A. Winkelmann. MEDIAN 60 GHz wide- band indoor radio channel measurements and model. IEEE Vehicular Technology Conf., 4:2393–2397, 1999.

[30] J. Purwaha, A. Mank, D. Matic, K. Witrisal, and R. Prasad. Wideband Channel Measurements at 60 GHz in Indoor Environments.

[31] T. Zwick, T.J. Beukema, and H. Nam. Wideband Channel Sounder With Mea- surements and Model for the 60 GHz Indoor Radio Channel. Vehicular Technol- ogy, IEEE Transactions on, 54(4):1266–1277, 2005.

[32] K. Chang and C. Sun. Millimeter-Wave Power-Combining Techniques. Mi- crowave Theory and Techniques, IEEE Transactions on, 83(2):91–107, 1983.

104 [33] Y. Lee, JR East, and LPB Katehi. Micromachined millimeter-wave module for power combining. Microwave Symposium Digest, 2004 IEEE MTT-S Interna- tional, 1, 2004.

[34] S.S. Sarin, C.D. Prasad, and D. Singh. Quasioptical power combining at millimetre waves. Microwaves, Antennas and Propagation, IEE Proceedings-, 147(4):301–303, 2000.

[35] J.A. Benet, A.R. Perkons, S.H. Wong, and A. Zaman. Spatial power combining for millimeter-wave solid state amplifiers. Microwave Symposium Digest, 1993., IEEE MTT-S International, pages 619–622, 1993.

[36] M. Lee, C.C. Chen, and J.L. Volakis. Ultra-Wideband Antenna Miniaturization Using Distributed Lumped Element Loading. IEEE AP-S International Sympo- sium, 2005.

[37] M.N. Afsar, Y. Wang, and R. Cheung. Analysis and measurement of a broadband spiral antenna. Antennas and Propagation Magazine, IEEE, 46(1):59–64, 2004.

[38] Y.P. Zhang, M. Sun, K.M. Chua, L.L. Wai, D. Liu, and B.P. Gaucher. Antenna- in-Package in LTCC for 60-GHz Radio. Antenna Technology: Small and Smart Antennas Metamaterials and Applications, 2007. IWAT’07. International Work- shop on, pages 279–282, 2007.

[39] E.M. Turner. Spiral Slot Antenna, December 2 1958. US Patent 2,863,145.

[40] J. Dyson. The equiangular spiral antenna. Antennas and Propagation, IRE Transactions on, 7(2):181–187, 1959.

[41] J. Dyson, R. Bawer, P. Mayes, and J. Wolf. A Note on the Difference Between Eqiangular and Archimedes Spiral Antennas. Microwave Theory and Techniques, IEEE Transactions on, 7(2):181–187, 1959.

[42] W. Curtis. Spiral antennas. Antennas and Propagation, IEEE Transactions on [legacy, pre-1988], 8(3):298–306, 1960.

[43] V. Rumsey. Frequency independent antennas. IRE International Convention Record, 5, 1957.

[44] J. Kaiser. The Archimedean two-wire spiral antenna. Antennas and Propagation, IEEE Transactions on [legacy, pre-1988], 8(3):312–323, 1960.

[45] R.G. Corzine and J.A. Mosko. Four-arm Spiral Antennas. Artech House, 1990.

[46] D.S. Filipovic and J.L. Volakis. Broadband meanderline slot spiral antenna. Microwaves, Antennas and Propagation, IEE Proceedings-, 149(2):98–105, 2002.

105 [47] Y.L. Chow, Z.N. Chen, K.F. Lee, and K.M. Luk. A design theory on broad- band patch antennas with slot. Antennas and Propagation Society International Symposium, 1998. IEEE, 2, 1998.

[48] E. Topsakal and J.L. Volakis. Frequency selective volumes for optical spatial filters. Antennas and Wireless Propagation Letters, 3, 2004.

[49] E.D. Palik. Handbook of optical constants of solids. Academic Press Handbook Series, New York: Academic Press, 1985, edited by Palik, Edward D., 1985.

[50] I. Pirozhenko, A. Lambrecht, and VB Svetovoy. Sample dependence of the Casimir force. New Journal of Physics, 8(10):238, 2006.

[51] C.A. Balanis. Advanced Engineering Electromagnetics. Wiley, 1989.

[52] M. A. Green and M. J. Keevers. Optical properties of intrinsic silicon at 300K. Progress in Photovoltaics, 3(3):189–192, 1995.

[53] C.A. Balanis. Antenna Theory: Analysis and Design. Wiley, 1982.

106