MM-Wave Microstrip and Novel Slot Antennas on Low Cost Large Area Panel MCM-D Substrates a Feasibility and

d Performance Study
Janusz Grzyb, Didier Cottet, Gerhard Trster, Electronics Lab, ETH Zurich, Gloriastrasse 35, 8092 Zurich, SWITZERLAND Phone: +41 1 632 79 21, Fax: +41 1 632 12 10, e-mail:[email protected]

Abstract Recently, there was a significant growth in wireless telecommunication applications such as Cellurar and Mobile phones, Personal Communiacation System, Wireless Local Area Network, Pagers, Global Positioning Satellite communication, etc. In future, telecommunication applications will shift more and more to the use of higher frequncies. Multichip Modules (MCMs) and MCM technologies are very close to micro- and milimeter-wave systems like transmitter/receivr (T/R) for radars and communication modules for wireless infrastructure links. There are various stringent requirements for MCMs in this frequency range but one of the most critical requirements is the ability of high RF performance together with high yield and low cost high volume production. Integration of RF circuits and elements with digital and analog circuits on the same substrate is essential to reduce the overall cost and physical dimensions of the whole system. The integrated antenna T/R is a very interesting alternative for applications where compact design, low cost, and high volume are important factors. Integration of the antenna in such a module is very difficult because the technological requirements driven by its performance characteristics (radiation efficiency, bandwidth) are opposite to these of non-radiating elements, where the radiation effect is not desirable. This issue is especially imortnat for lowprofile build-up MCM-D technologies where the problem of radiation efficency of microstrip antennas is extremely diffcult to solve. This paper will discuss the EM based modelling and practical design of microstrip and novel slot antennas in the mm-wave frequency range (3085GHz) on low cost MCM-D substrates. The critical parameters of milimeter design such as losses, bandwidth, and radiation pattern will be discussed. To perform this feasibility study a Finite Element Method (FEM) is used as a simualtion tool and HP8510XF Vector Network Analyzer (VNA) as a measurement equipment. I Introduction In the last years there were a number of efforts to develop mm-wave communication, automotive millimeter-wave radar technology and sensors. Processing and packaging of all T/R module types are the challenge for future system design. Very compact RF and antenna arrangements are usually required in these modules. At the same time they must be suited for lowcost and high reliability mass production. With respect to mmwave front-end, this leads to great challenges for both transmitter/receiver circuitry and antenna. A traditional packaging technology with substrates on carriers and integration into milled metallic boxes with complex
0-7803-7038-4/01/$10.00 (C)2001 IEEE

geometries, hermetical sealing and ceramic coaxial feedthrougs cannot be the answer for high volume production. Other important driving factors are performance and weight. Development from a traditional sub-unit module concept to common substrate or package base module architecture can be reached in two evolution steps: Step 1. replacing sub-units with multifunction MMICs [1]. This means reduced GaAs area, less MMICs, higher chip integration and modular chipsets for different kinds of T/R modules Step 2. replacing sub-unit substrates by a common multi-layer substrates. Advantages are: reduced substrate area, one substrate with matching structures, lower number of interfaces (lower loss). It calls for a RF/DC multi-layer technology with the ability to integrate passive components such as inductors, capacitors, resistors, couplers, power dividers, filters and antenna fabrication. MCM technology is best suited for such applications. Different RF multi-layer technologies are available. From the cost reduction point of view an MCM-D technology on laminate carriers for high-speed digital circuits with the option of Large Area Panel Processing (LAP) of the substrates is interesting as the ultimate solution. So far there are only few publications on the integration of mm-wave antennas, especially into low-profile build-up technologies such as MCM-D [2][3]. In the next sections the design and implementation of mm-wave microstrip and slot antennas integrated in a low cost large area panel processing MCM-D technology will be introduced. Their design process and realization will be discussed using full-wave electromagnetic Finite Element Method simulations and network analyzer measurement results. II LAP Technology A new process for low cost MCM-D substrates was developed within the EU Esprit project LAP (Low Cost Large Area Panel Processing of MCM-D substrates and packages) [4]. The cost target for this thin film technology is 1US$1/inch2. To reach this target the project consortium has increased the panel production size from todays 4x 4and 5x 5 panels up to 12x 12 and 24x 24 panels. II.A Processes and materials The LAP technology is a four layer thinfilm process. Its build-up cross-section is shown on Fig.1. The detailed description of all available process materials and options together with their tolerances being under investigation in the project can be found in [4][5].

2001 Electronic Components and Technology Conference

Fig.1 LAP technology process build-up From the variety of possible combinations we focus on two combinations for which the RF test structures based on these materials have been manufactured and partly measured. 1. BCB (dielectric) (5m/layer) (r=2.7, tan=0.002 at 10GHz) on Rogers 4003 (r=3.38, tan=0.0027 at 10GHz) (base) 2. PBO (dielectric) (8m/layer) (r=2.9, tan=0.011 at 1MHz) on Al2O3 (base) As basic interconnect metal composite layer of electroplated or deposited copper (primary conductor) is used: 1.with BCB Ti/W (80nm)/Cu(2m)/Ni (0.5m) plus Au (1m) on top layer 2. with PBO Ti (80nm)/Cu(2m) plus Ni (2m) and Au (1m) on top layer The actual Degree of Planarization level (DOP) [6] of BCB and PBO, after cross-sectioned our test vehicles was in the range of 60% for both dielectrics. III. Analysis, design and measurements III.A Simulation Technique Finite Element Method has been chosen as a simulation technique to calculate the full 3D EM fields with tetrahedron as a basic mesh element. It allows exact simulation of all types of losses, especially ohmic ones. The proper treatment of these losses is extremely important for correct physical simulation results, as will be shown in the following paragraph. Metal structures in most of the simulations are typically treated as flat 2D elements with the assumption of Perfect Electrical Conductor (PEC) in order to reduce the computational resources requirements. We treat the metallizations in our simulations as real cooper conductor 3D objects in order to include all the possible coupling effects with highest accuracy. The effect of ohmic losses is treated by the surface resistance formulation, which in the mm-wave range is a reasonable approximation. The reason for that is that the skin effect in metal layers with the supported thickness of about 2m in this frequency range is well developed (skin depth is 7-10 times lower that the metal thickness). Such a strict approach is justifiable by our experience, namely high ohmic losses caused by frequency range of interest and very strong local fields resulting from very small cross-sections of the structures being simulated. As we observe the PEC assumption can lead to dramatic change of the simulation results especially for radiating elements. For example the PEC assumption in our directly fed rectanular patch antenna designed at 83 GHz caused the change of the antenna input impedance at resonance from 50 (copper simulation) to 22. This means rejection of the design.

The computation of the radiation power and radiation patterns, including cross-polarization level, demands special care. In order to increase the accuracy of the simulation results of these very important parameters one needs to make denser mesh than that normally used in case of non-radiating structures. In practice this means that one needs to seed initially denser mesh on the absorbing boundary and include some more iterations in the process of the solution convergence after the input impedance characteristics had already stabilized with relatively high accuracy. Maximum change in S matrix elements of approximately 0.01 between two last iterations usually means good stability of the antenna input impedance. Simulation of structures which radiate more easily (wideband resonance behaviour) such as slot antennas are more difficult to converge and demand more iterations and therefore consume more memory and time. For the computation of these radiating problems we use SGI Origin 2000 with a superscalar dataflow architecture, 8 processors R10000 (250MHz) and 8GB RAM with a peak performance rate 800MIPS per processor. As the software code is not parallel we use only one processor per task. The amount of memory required depends mainly on the metal structure complexity, the size in relation to the wavelength range of interest and the Boundary Condition (BC) type. For radiating structures as antennas the memory requirements and simulation time increase up to 2-3 times in comparison to the non-radiating boundary condidtion simulation. Typical computational resources needed for the simulation of our antenna problems with the high accuracy discussed earlier are as follows: 1. microstrip antenna at 80GHz - 360390MB of memory, 911min/ frequency point 2. slot antenna at 40GHz and 80GHz 600850MB of memory, 40-60min/ frequency point. This large difference in the computation time and memory required between microstrip and slot antennas is a result of fields computation necessity in the thick lossy substrate (800m) and in the air space underneath in the slot antenna configurations. On the contrary, in order to achieve high accuracy of the field pattern in the microstrip antenna model we do not need to accommodate this substrate into the model at all, because of the large ground plane between the radiating patch and the substrate. Additionally, simulation of the microstrip antennas allows to use the H symmetry plane and cut the model by half. To determine the excitation field pattern at the input port, the 2D solver calculates the natural field patterns (modes) that can exist inside the transmission structure with the same crosssection as the port, together with transmission parametres such as Z0 and complex propagation constant . This solver is very fast and computationally efficient without quasi-static limits. The solution of the simple 2D problem such as microstrip or coplanar line or coupled lines configuration takes only a few seconds on a SUN ULTRA 10. III.B Measurements All the measurements have been made on a HP8510XF Vector Network Analyzer (VNA) with waveguide output and

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

air coplanar waveguide microprobes up to 115GHz with LineLine-Refelct-Match (LLRM) and Thru-Reflect-Line (TRL) calibration standard substrate from Cascade Microtech and then on-wafer with a ground-signal-ground (GSG) pad configuration for on-wafer probing. III.D Microstrip antennas Up to date there are few publications on the integration of the mm-wave antennas into MCM-D technology. The authors in [2] used modified MCM-D process with the BCB layer thickness of 72m and the designed microstrip patch was at 35GHz. Three basic microstrip antenna types [7] in the frequency range of 80GHz have been designed: directly fed rectangular patch, electromagnetically coupled patch and circularly polarized patch with 45 angle patch slot, as at the moment the actual design aim at validation of this technolgy for design of the distributed elements in mm-wave frequency range. The layout photograph of the manufactured test vehicle with the microstrip filters, Lange couplers and antennas and the geometrical parameters of the antennas are shown in Fig.2 and Table1. Detailed design flow and analysis of different design considerations such as surface wave excitation, losses, dielectric and metal profiles, complex metal build-up and their relation to the process setup and tolerances is thoroughly analyzed in [5].

locus of the patch. The price for this low build-up is a large reduction of the antenna bandwidth and radiation efficiency. This happens because of a dramatic increase in Q value of the antenna and the ohmic losses which are the result of very strong local fields (thickness of 16m).

Table1. Parameter list of microstrip antennas The radiation efficiency decreases from 80% simulated with PEC assumption (reference value) down to 30%. For BCB based substrate with thickness of 40m the efficiency could be still held in the range of 70-80% for this frequency range. This is a very important conclusion for the design of microstrip antennas on MCM-D substrates. Radiation efficency for PBO based antennas cannot be made considerably higher even with thicker build-up as the contribution of dielectric losses is too large. It is a known fact that losses increase antenna bandwidth. This may appear to be an advantage for patches with very low bandwidth, but the antenna efficiency, as one can see, is reduced drastically. Based on EM simulations, the bandwidth (defined for 7dB) for the same patch geometry simulated with PEC assumption was 0.9GHz, while with the copper conductor it was 2.5GHz. The bandwidth defined only by the frequency range over which the input match is acceptable is not the only constraint. Equaly important are the radiation pattern characteristics as minor lobes and cross-polarization for linearly polarized antennas or axial ratio and isolation between right-hand and left-hand circular polarization for circularly polarized antennas. For directly fed rectangular patch it is not easy to achieve cross-polarization level below 20dB in the frequency range considerd, especially in the principal H-plane. This is due to a strong inductive offset of the input impedance at the feed position. Another aspect of this inductive offset is the frequncy mismatch between the location of the minimum insertion loss and the resonant frequency (defined as the point where the real part of the input impedance reaches its maximum) of the antenna. This is especially important for antennas with low bandwidth such as microstrip low-profile antennas. This discrepancy leads to the situation where the field pattern and antenna parameters (such as beam area and directivity) are strongly deteriorated for a good match at the input. Using of electromagnetically coupled patches one is able to correct these detrimental effects. Let us consider this effect in detail. Fig.3 shows the input impedance characteristics of the rectangular directly fed patch antenna resonating at 82.7GHz with the following dimensions: width 2000m, length - 1000m, feed inset - 500m. The inductive behaviour of the input reactance at the resonance and its unsymmetry around the resonance is easy visible. Achieved maximum cross-polarization level in the H-plane is -15dB for

Fig.2. Test vehicle with antennas The total thickness of the antenna in this case is only 1618m. This is very far from optimum in connection with bandwidth and radiation efficiency. For optimum bandwidth as large values of the dielectric thickness as feasible under the two constraints, mentioned below, should be used. The first concerns the surface wave excitation and imposes the limit on the thickness of 700m in case of our process specifications. This means that one has only TM0 surface mode propagating with a small value of field component. Low values of the patch thickness and dielectric constant help to reduce the magnitude of this mode to an acceptably low level. The second constraint is that the patch stops resonating after thickness crosses a certain threshold value: 428m in our case. This is the effect of the entirely inductive input impedance

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

this antenna. This is caused partly by this effect and partly by longer total length of the non-radiating edges (longer than /2 because of the feed inset effect).The and components of the field pattern characteristics in the H plane are shown on Fig.4.

Fig.4. Field pattern in the H plane of the directly fed patch

Fig.3 Input impedance of the directly fed rectangular patch In comparison with the predecessor, directly fed patch antenna, the EMC patch antennas exhibit reduced spurious radiation form the feed network. Additionally, using of EMC patches we are able to broaden operational bandwidth (defined as a frequency range over which the return loss is smaller or equal to specified value). Nevertheless, we should be very carefull with this broadening design if we take into consideration the purity of the cross-polarization. Otherwise, we can have very good input match but at the frequency point which does not correspond to the proper resonant frequency with high purity of cross-polarization. Many authors neglect that effect considering only input impedance match. This leads to an improper design. The following example explains this problem. The bahavior of the input impedance of the EMC patch antenna is a function of a number of parameters such as patch width, the patch-feed overlap, and the dielectric constants of two layers between the patch and the conducting ground. In general, there is an infinite number of solutions for the conjugate match of the antenna input impedance with the feed line. This match can be realized by adjusting any of the two parameters, either the patch width or the patch-feed overlap, or the two parameters together. From the maximum bandwidth point of view it is desirable to make the patch wider as the Q factor of the antenna decreases with the patch width increase. The directivity of the radiating system also increase with the wider patch. Irrespective of the patch width, the locus of the input impedance forms a double resonance loop which crosses the real axis of the Smith chart at two different points. With the increase of the patch width, the area of the resonant loop diminishes. As the patch width increases, the loop shifts into the capacitive region which is what we wanted to show.

a)

b) Fig. 5 a). Return loss b) input impedance of the improper matched EMC patch antenna The antenna dimensions in this case are the following: length 1100m, width 2300m, feed inset under the patch 400m. The two figures Fig.5a and Fig.5b explain this effect. At the frequency point (79.5GHz) where the reactance of the input impedance should cross the 0 value (real axis of the Smith chart) there is a capacitive region mentioned above. The best match (-35dB) occurs at the lower frequency (78.5GHz) at which the fields are already much higher cross-polarized. As a result in the whole band (77.7GHz80.3GHz for 7dB definition), though wide, we have very high cross-polarization

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

level in the range of -12dB -3dB with 3dB at the frequency of the best input match (Fig. 6). For properly designed EM coupled patch we have achieved maximum cross-polarization level in the H-plane of 25dB in the whole band. The dimensions of this antenna design are as follows: length 1100m, width 1800m, feed inset under the patch 400m. The input impedance behavior and the field pattern in the H plane of this design are shown on the Fig.7 for comparizon. Another important aspect of the microstrip antenna designs in this low profile MCM-D build-up is the difficulty in the prediction of the input impedance behavior in relation to the patch width. Normally (this means low losses assumption or neglecting the losses at all), the increase of the patch width leads to the input impedance decrease. In our design conditions this is not the case. One has to include the effect of high ohmic losses (strong fields in the patch ) on the input impedance value.

a)

b) Fig.6 Field pattern at 78GHz in the H plane of the improper EMC coupled patch In result there are two contributors to the input impedance value, namely radiation resistance and equivalent loss resistance. The lossresistance represents the losses in the dielectric and ohmic losses in metal (especially in our case). These two factors change the input impedance value in the opposite directions with respect to the change of the patch width. So it is not easy to state a priori if the increase of the patch width decreases or increses the input impedance value. This means that the use of a 3D simulation technique, which includes simulations of high ohmic lossses properly is a must and any of the approximate analysis techniques cannot be used. The most important parameters of the designed antennas are collected below and they are as follows for: BCB and PBO based antennas, respectively: BCB: fr=82.9-83.15GHz (on the whole panel), B=2.4GHz (-7dB), Directivity=9.9dB, Gain=4.15dB, S11=-24-32dB shown on Fig. 8 (on the whole panel) PBO: fr=79GHz (on the whole panel), B=2.9GHz (-7dB), Directivity=9.9dB, Gain=4.7dB, S11=-24 -27dB (on the whole panel) Fig.7 a) Field pattern at 79GHz in the H plane b) input impedance of the properly coupled EMC patch The changes in the resonant frequency for BCB antennas are mainly accounted for by planarity nonidealities of Rogers 4003 laminate. For PBO antennas fr is very stable as a result of very planar ceramic carrier. The large difference between directivity and gain is caused by losses of the antennas. Considering all the possible process tolerances we can conclude that designs of the antennas are repeatable on lowcost low-profile MCM-D large panel substrates. For the supported metal layer width and dielectric thickness tolerances [5] maximum decrease in input impedance match is 4dB5dB, maximum deviation of resonant frequency 0.2GHz. Tight thickness control of the metal layer is not critical because of the low skin depth at 80GHz.

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

Fig.8. S11 of the rectangular patch on BCB III.E Slot antennas The basic drawbacks of above considered microstrip antennas implemented on a very thin MCM-D substrate are very high ohmic losses and small bandwidth. None of the known methods to enhance the bandwidth [8] can be used in this case without further deterioration of the radiation efficiency. The solution seems to be slot antennas. The geometry of the slots fed by coplanar waveguides is suitable for monolithic integration and they dont require holes for mounting active devices as microstrips do. One of the main issues with CPW-fed slots is to provide an easy impedance matching to the CPW line. So far, several input impedance tuning techniques of slot antennas based on a change of slot dimensions, coupling mechanism or both have been reported in the literature. Among them we can quote the onewavelength center- or offset-fed slots, the half-wavelength capacitively- and inductively fed and multiple folded slot [9]. The sizes of antennas remain almost unchanged. The idea of a slot loop antenna directly fed by a CPW line, proposed in [10] allows reducing the size of an antenna. The basic idea is to make the electrical or magnetic current flow along the perimeter of a patch, instead of the diameter or one side of the patch. This antenna is supposed to resonate when its slot perimeter is about half-wavelength. Although the shape of this slot was considered, it was only in the frequency range of 1GHz and only experimental and tough EM analysis has been performed. Therefore many of the antenna features were not discovered. A deep EM study of this basic antenna shape has led to many modifications and primarily feeding techniques. Some of the geometries are shown on Fig.9. With simple direct feeding technique of the basic configuration shown on Fig.9a a good match at the antenna input without any matching networks is not possible at 1st resonance (perimeter is /2) because of the extremely low input impedance (25) in the mm-wave range, namely 3080GHz. Using of the matching impedance transformers at the inut of the antenna leads to a higher area consumption (antenna is no more compact) and can decreaseof the impedance bandwidth and worsen the cross-polarization level. Fig.9 a) basic configuration b) with CPW feed (1st) c) with CPW feed (2nd) d) with microstrip feed At the 2nd resonance (perimeter is ) impedance is much higher as it is a parallel type resonance. Additionally, directivity of the antenna is higher because the two slots on the opposite sides of the antenna (taking the feed location as the reference) radiate in phase (constructive interference) leading to compacter radiation in the E plane. First, let us investigate the properties and behavior of this basic direct fed configuration in the mm-wave frequency range. The shape of the antenna leads to many degrees of freedom to tune the input impedance, bandwidth and directivity by changing the slot widths (both radiating and noradiating), length and relative position between two radiating slots. Making the radiating slots wider decreases the input impedance value. Wider non-radiating slots, on the contrary, increase the input impedance. The two non-radiating slots can be treated as a coplanar waveguide line which tronsforms the impedance of the rear slot to the input. If the slots become wider, the characteristic impedance of this equivalent CPW line becomes higher and transforms the radiation impedance of the rear slot as higher (CPW characteristic impedance is much higher than radiation impedance of the radiating slots). With the constant perimeter value (it determines the resonant frequency) we can change the ratio of the radiating front and rear slot length (W antenna width) to the non-radiating slot length (L length of the antenna) W/L in order to tune the input impedance value. Every value of the W/L ratio influences other important parameters of the antenna such as bandwidth, directivity and field pattern cross-polarization value. Making the ratio W/L larger produces a higher Q factor with narrower bandwidth but higher directivity of the pattern and much lower cross-polarization value. On the other hand, a smaller value of W/L ratio allows to produce very high bandwidth, but at the cost of directivity and cross-polarization level. Primary antenna design is in the frequency range of 3545GHz. To illustrate above mentioned relations between geometries and antenna parameters we compare two designs with direct feeding technique. The geometries are as follows:

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

1.W=2000m, L=1000m, widths of the slots: rear=300m, front=20m, non-radiating slots=20m 2. W=2900m, L=100m, widths of the slots: rear=80m, front=20m, non-radiating slots=20m Both designs are laid out on M4 layer and with substrate (Rogers 4003) thickness of 800m. We decided to keep the width of the front radiating slot to the processing minimum (20m in our case ) in this direct feed configuration. This is necessary in order to prevent the locus of the input impedance from becoming entirely inductive and increasing the cross-polarization level and mismatch between the frequency ranges of a good input match and a proper radiation field pattern. This effect applies especially to antennas with smaller W/L ratio. The, return loss characteristics, input impedance and radiation patterns at resonant frequencies for both antennas are shown on Fig.10, Fig.11.and Fig.12.

that effect. More important, the distortion of the field pattern (low side lobe effects) are much lower, because the two radiating slots are closer to each other and are narrower. This means that the radiation is more coherent and less sensitive to the presence of the substrate beneath. This effect is much more pronounced for higher passband frequencies located away from the resonance. The reason for that is the dielectric substrate on one of the sides of the slot surface. Such a behavior can be explained by the fact that the effective thickness of the dielectric substrate is comparable with the wavelength of the resonating antenna slot perimeter (/8/6). In the frequance range of up to 40GHz it is still reasonable to use the 800m thickness of the Rogers 4003 carrier but mainly at the cost of directivity reduction. We achieved the directivity of 6dB and avoided these side lobes excitation in the whole band in the similar antenna configuration as configuration 1 with the substrate thickness of 200m.

a)

a)

b) Fig.10. Return loss of the slot antenna in a) configuration 1 b) configuration 2 As one can see, the impedance bandwidth for the antenna in configuration 1 is considerably lower (12% for 10dB) than that in configuration 2 (26% for 10dB). This is the price for higher directivity of 4.35dB instead of 3.9dB (for configuration 2). In both cases achieved radiation efficency is 85%, which is very high in comparison with the earlier analyzed microstrip antennas. Cross-polarization level in the configuration 1 is also considerably lower because of the shorter non-radiating slots, which are the main contributors to

b) Fig.11. Input impedance of the slot antennas in a) configuration 1 b) configuration 2 For the applications in the range of 6080GHz the use of thinner substrates is a must (200m Rogers 4003 substrate is supported), otherwise the resonative behaviour of the antenna is no more observed. The radiation pattern in the E plane is low sensitive to the substrate presence and almost undistored. Cross-polarization level in the E plane is not an issue and can be kept below 30dB.

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

The configurations with the electromagnetically coupled feeds shown in Fig.9a, 9b, 9c based on the inherent multi-layer build-up of MCM-D technology allow to overcome some of the above mentioned drawbacks by proper coupling of the feed and radiating slot. Using electromagnetic coupling it is possible to move the inductive reactance of the input impedance at the resonance towards more capacitive even with wider front slot and eliminate the mismatch between the frequency ranges of good input match and low crosspolarization level. This procedure allows to tune the real part of the input impedance simultanously. Another advantage of this multilayer build-up is the ability to achieve compacter geometry of the antenna as we can make the non-radiating slots narrower (using only one layer of dielectric between two metal layers). Narrower non-radiating slots mean the transformation of the radiation impedance of the rear slot into lower value at the input of the antenna. This allows to decrease the width of the slots and improve the radiation characteristics, assuming the same required input impedance.

antennas we conclude that the microstrip feeding technique is not the proper choice. Additionally, the inherently high losses of microstrip lines implemented in the low-profile MCM-D technology will further decrease the efficency of the whole radiating system. Using the coplanar waveguide feeding technique from Fig.9b, 9c we can raise the efficiency up to 83% in comparison with microstrip feed and keep the bandwidth and polarization purity on the same level. Conclusions A general full-wave FEM was used to determine the high frequency characteristics of the components in a multi-layer build-up. Measurements or electromagnetic simulation results demonstrate that microwave and mm-wave antennas may be integrated in MCM-D substrates. The basic drawbacks of the microstrip antennas implemented on a very thin MCM-D substrate are low radiation efficiency and small bandwidth. We indicated slot antennas as an alternative solution made their performance study in the frequency range of 3080GHz based on EM simulation. Achievable bandwidth on a Rogers 4003 carrier is up to 25%, radiation efficiency 86%. The most difficult problem to cope with is the distortion of the radiation pattern and degradation of the directivity because of too thick substrate carrier beneath the slot antenna. Acknowledgments The Swiss Federal Office of Education (BBW), under project 97.0286, partially funded this work. Special thanks to Hansruedi Benedickter from Microwave Lab at ETH Zurich for the measurements support. References [1] A.Schfer, L.Verweyen, T.Grave, G.Bck, M.Schlechtweg, Low Cost GaAs pHEMT MMICs for Millimeter-Wave Sensor Applications, IEEE Trans. MTT, Vol.46, No.12, 1998 [2] U.Dahlgren, J.Svedin, H.Johansson, O.Hagel, H.Zitath, Ch.Karlsson, An Integrated Millimeter Wave BCB Patch Antenna HEMT Receiver in IEEE MTT-S Digest., pp.661-664, 1994, [3] E.Soliman, S.Brebels, E.Beyne, Circularly Polarised Aperture Antenna fed by CPW and Built in MCM-D Technloogy, Electronics Letters, Vol.35, No.4, 1999 [4] Esprit Project 26261: LAP, Low cost large area panel processing of MCM-D substrates and packages, http://www.ife.ee.ethz.ch/mcm/lap [5] J.Grzyb, D.Cottet, G.Trster, MM-Wave Integrated Antennas on Low Cost MCM-D Substrates in 9th Top. Meet. EPEP 2000, Oct.23-25, USA [6] P.Chiniwalla, Multilayer Planarization of Polymer Dielectrics in Proc Int. Conf. On HDI and MCMs., pp.52-57, April 6-9, 1999, Denver, USA [7] J.R. James, P.S. Hall Microstrip Antenna: Theory and Design, Peter Peregrinus Ltd., 1981 [8] K.F.Lee, W.Cheng, Advances in Microstrip and Printed Antennas, John Willey & Sons, Inc., 1997

a)

b) Fig.12. Radiation pattern in the H plane of the slot antennas in a) configuration 1 b) configuration 2 The EM coupled configuration with the microstrip feed with the same W/L ratio as in configuration 1(direct CPW feed) allows to achieve good cross-polarization below 22dB but at the cost of bandwidth and radiation efficiency decrease (eff=74% and B=18%). As we focus on the high efficiency

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference

[9] S.Sierra-Garcia, J.Laurin, Study of a CPW Inductively Coupled Slot Antenna, IEEE Trans. Antenna and Propag., Vol.47, No.1, Jan. 1999 [10] M.Cai, P.S.Kooi,M.S.Leong, A Compact Slot Loop Antenna, Micr. and Opt. Techn. Letters, pp.292294,Vol.6, No.5, April. 1993

0-7803-7038-4/01/$10.00 (C)2001 IEEE

2001 Electronic Components and Technology Conference