Chip-To-Chip Switched Beam 60-Ghz Circular Patch Planar Antenna

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X 1

Chip-to-Chip Switched Beam 60-GHz Circular Patch Planar Antenna Array and Pattern Considerations Prabhat Baniya, Student Member, IEEE, Aimeric Bisognin, Member, IEEE, Kathleen L. Melde, Fellow, IEEE, and Cyril Luxey, Fellow, IEEE

Abstract—A 60-GHz switched beam chip-to-chip antenna ar- grows as the square of the interconnect length. Longer lines ray is introduced. The array consists of four center fed circular consume more power due to higher resistance and capacitance patch elements with side vias in a 2 × 2 grid arrangement associated with them [3]. The increased interconnect delay forming a planar array. The array is designed to fit on a typical multicore chip for reconfigurable interchip wireless com- and power consumption limit the number of cores that can munication. The array main beam is switched by changing the be placed on a chip. One way to circumvent this limitation is inter-element phase shifts in the azimuth plane. The switching to extend the NoC paradigm to connect individual multicore of the main beam is analyzed and verified through full wave chips wirelessly to form an even larger multiprocessing unit. simulation. The design presented is an improvement over a The NoC techniques are used to provide interchip commu- previous design of a two-element antenna array. The Friis transmission equation with polarization components taken into nication wirelessly in addition to intercore communication, account is used to model the interchip wireless link. To verify the and thus realize a multicore multichip (MCMC) computing model, a transmission coefficient measurement is made between unit [4]. Fig. 1 illustrates a MCMC unit with nine chips in a a pair of the two-element arrays separated by a 10 mm distance. 3 × 3 arrangement, each chip with eight cores. The cores on a Both simulated and measured radiation patterns of the two- chip communicate through short high speed wired links. Long element array are presented for use in the Friis equation to calculate the transmission coefficients. Full wave simulation of distance communication between chips is done wirelessly via the array pair is also performed. The calculated results obtained routers, eliminating power consuming long wired connections. from the Friis model agree well with both the measured and full The interconnect topology shown in Fig. 1 is called the hybrid wave simulated results. The Friis model is used to calculate both NoC (HyNoC) since it uses a combination of wired and signal and interference levels. wireless links [5]. Millimeter wave (mmW) antenna arrays on Index Terms—Antenna-in-package (AiP), chip-to-chip an- the chip routers are used to provide reconfigurable wireless tenna, circular patch array, Friis transmission equation, hybrid interchip communication. The antennas communicate through network-on-chip (HyNoC), multicore multichip (MCMC), recon- radiation in the air above the chips at near the speed of light figurable, switched beam antenna, wireless link budget, 60 GHz antenna. [3]. Although path loss due to spherical spreading is inherent in the wireless links, the nearly lossless air above the chips does not contribute to power dissipation and heating within the I.INTRODUCTION chips. The MCMC computing architecture has the potential IRED interconnects between cores on a chip increase to readily incorporate and efficiently link hundreds of cores W in complexity with core count. Dense physical inter- together to build a high performance computing (HPC) system. connections cannot be modeled as a predictable delay chan- The MCMC approach also allows the flexibility to fabricate nel due to hard-to-estimate parasitics associated with closely the antennas separately from the computing chips [6]. spaced lines [1]. To address data and synchronization errors, the network-on-chip (NoC) techniques can be used to provide The antenna arrays are designed for 60 GHz operation so intercore communication and ensure functionally correct op- that they can be easily integrated with the existing CMOS eration of the chip. Wired interconnects, however, have to be transceivers compliant with the WiGig IEEE 802.11ad stan- made longer as more and more cores are placed on a larger dard. WiGig offers unlicensed frequency operation around chip [2]. This is an issue because the wired interconnect delay 60 GHz with a broad bandwidth of 2.16 GHz and high throughput of up to 7 Gbps per channel. A 60 GHz wireless Manuscript received June 5, 2017; revised October 16, 2017; accepted Jan- uary 15, 2018. This work was supported by the National Science Foundation link based on the standard can provide multi-Gbps data rate under Grant ECCS-1027703. needed for interchip communication, with the proper design of P. Baniya and K. L. Melde are with the Department of Electrical and the transceiver [7] and antenna components [8]. Even higher Computer Engineering, University of Arizona, Tucson, AZ 85721 USA (e- mail: [email protected]; [email protected]). throughput is achievable with multichannel operation [9]. A A. Bisognin was with the EpOC Laboratory, Universite´ Nice Sophia switched beam antenna array that operates at 60 GHz is Antipolis, 06560 Valbonne, France. He is now with Qualcomm France RFFE, presented for reconfigurable wireless routing. Fig. 1 shows 06560 Valbonne, France (e-mail: [email protected]). C. Luxey is with the EpOC Laboratory, Universite´ Nice Sophia Antipolis, the switched beam array, a four-element circular patch array 06560 Valbonne, France (e-mail: [email protected]). in a 2 × 2 grid arrangement, on the chip routers. The arrays 0000–0000/00$00.00 c 2018 IEEE 2 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

to-chip communication in the horizontal plane. The 60 GHz Core Core Core Core Core Core Core Core Core antenna arrays surveyed in [14] only have broadside beam A B C scanning capability. Recently, an end-ﬁre scanning array using Core Router Core Core Router Core Core Router Core magneto-electric (ME) dipole elements has been demonstrated in [15] but the angular coverage is limited to 180◦. A planar Core Core Core Core Core Core array of ME dipoles is presented in [16] for 360◦ broadside Core Core Core scanning. The switched beam antenna array presented here has

Core Core Core ◦ Core Core Core Core Core Core end-ﬁre scanning capability with 360 angular coverage. This paper is organized as follows. Section II describes the D E F

Core Router Core Core Router Core Core Router Core synthesis of a general isotropic 2 × 2 planar array. Section III presents the detailed structure and simulation results of a 60

Core Core Core Core Core Core GHz center fed circular patch with four side vias. Section IV Core Core Core provides the simulation results of a 60 GHz switched beam × Core Core Core 2 2 circular patch planar array for reconﬁgurable interchip Core Core Core Core Core Core wireless communication. Section V presents radiation pattern G H I measurement and simulation results of a previous generation Core Router Core Core Router Core Core Router Core two-element antenna array on a patterned ground plane. Mea- sured and simulated transmission curves between the two- Core Core Core Core Core Core

Core Core Core element array pair are also presented with the Friis model to understand the wireless link between chips and see how y N Switchable main beam antenna orientation, side lobe levels (SLLs) and polarization W E x 2 × 2 Circular patch planar array can affect interchip communication. This provides validation z S of the Friis model for use with the simulation results of the Fig. 1. 3 × 3 MCMC architecture with HyNoC interconnection using pattern switched beam circular patch array. Section VI summarizes switchable mmW antenna arrays. the paper along with the importance of the work.

II.ANTENNA ARRAY FOR INTERCHIP WIRELESS can be thought of as the hub of interchip communication. COMMUNICATION Specifically, we consider the wireless links between adjacent For the chip-to-chip communication scenario depicted in routers in the diagonal directions, namely E–A, E–C, E–G, Fig. 1, planar arrays are particularly well suited since they and E–I using the switched beam antenna array. Because of can provide a full 360◦ scan of the main beam [17]. This the wireless interconnects in the HyNoC topology, single hop ensures that each chip is able to communicate with its adjacent communication in the diagonal directions is made possible. A neighbors in all eight directions: north, south, east, west and mesh topology wired connection between the chips would re- the four diagonals, provided the array has enough elements quire two hops to communicate in diagonal directions [10]. In to scan the main beam laterally in 45◦ steps (in the azimuth MCMC parallel processing, multiple chips share the workload plane). A linear array is not suited since it is not capable of requiring frequent transfer of data between them. The intercon- achieving a 360◦ scan [17]. In this section, we consider a nect topology must provide fast communication between chips planar array for end-fire scanning of main beam only along in order to achieve high computational speed. Furthermore, the four diagonal directions in 90◦ steps. pattern adaptation can provide real time reconfiguration of the data paths as broken links can be bypassed by switching the array main beam. The design of the antenna array thus requires A. Array Factor a careful and thorough consideration. A four-element planar array in a 2 × 2 grid arrangement can In addition to broad bandwidth and switchable main beam, be used to achieve 360◦ lateral scan in 90◦ steps. A 2 × 2 the 60 GHz antennas for interchip communication have other grid arrangement of four isotropic elements a1, a2, a3, and challenging requirements. These include CMOS compatibility, a4 forming a planar array is shown in Fig. 2. In the far- small footprint, and low power dissipation for use on a typical field, the normalized array factor AFn of the planar array with multicore chip [11]. Moreover, in the MCMC architecture, uniform amplitude excitation and ignoring coupling between since the antenna arrays all reside in the same azimuth plane the elements [18] is given by (horizontal xy plane in Fig. 1), they must be capable of scan- k d sin θ cos φ + β AF = cos 0 x x ning their main beams in the lateral (end-fire) directions (the n 2 xy plane which contains the antenna) to communicate with one k d sin θ sin φ + β another. This is different from pattern reconfiguration in mmW × cos 0 y y (1) WLANs and Wi-Fi networks where beam scanning is done 2 mostly in the broadside, above the antenna plane [12], [13]. where k0 = 2π/λ0 is the free space wave number at the Therefore, the MCMC systems posit the unique requirement wavelength λ0, dx and dy are the inter-element separations of 360◦ end-fire scanning on the antenna arrays. Wireless in the x and y directions in the azimuth plane, θ and φ are interconnects based on such arrays enable reconfigurable chip- the elevation and azimuth observation angles, and βx and βy BANIYA et al.: CHIP-TO-CHIP SWITCHED BEAM 60-GHz 3

The array factor by itself is derived using isotropic elements y and does not consider the directional characteristics of the an- Far-field tenna elements. The pattern multiplication in (4) incorporates observations the directional characteristics of the antenna elements into the total ﬁeld through the element ﬁeld. The inter-element separations, dx and dy, in the azimuth plane are set so that the main beams in the four diagonal directions are obtained with 90◦ inter-element phase shifts. That is, |βx| = |βy| = π/2 (6)

d which can be achieved by setting, a1 a2 dx = dy = d = 0.3535λ0 (7) ϕ0 d x Since d < λ0/2, grating lobes are avoided as well [18]. The 90◦ value is chosen as it can be easily obtained at the ◦ a3 a4 output of the quadrature (90 ) hybrid couplers. This would be helpful in designing feed network based on these couplers [19]. With (6) and (7) satisfied, βx and βy must cycle through four different combinations of 90◦ values listed in Table I, Fig. 2. Planar array of four isotropic antenna elements in a 2 × 2 grid arrangement in the azimuth plane. obtained for four different values of main beam angles φ0 along the diagonal directions. TABLE I Ideally, for the interchip wireless communication illustrated INTER-ELEMENT PHASE SHIFTS REQUIREDFOR MAIN BEAM in Fig. 1, it is desirable to have the maximum of both FORMATION IN THE AZIMUTH PLANE the element field and array factor lie in the azimuth plane ◦ φ β β (θ = 90 ) i.e., end-fire maximum condition. From (4), it is 0 √ x √ y +45◦ −k d/ 2 −90◦ −k d/ 2 −90◦ easy to see that this would maximize the total far-zone field 0 √ 0 √ +135◦ +k d/ 2 +90◦ −k d/ 2 −90◦ in the azimuth plane along φ0 directions provided (2) and (3) 0 √ 0 √ −135◦ +k d/ 2 +90◦ +k d/ 2 +90◦ are satisfied and therefore maximize interchip transmission. 0 √ 0 √ ◦ ◦ ◦ However, printed antenna elements that have end-fire radiation −45 −k0d/ 2 −90 +k0d/ 2 +90 (in the azimuth plane) are hard to realize in the presence of an unavoidable ground plane [20]. Vias can be incorporated into are the inter-element phase shifts in the x and y directions, the antenna element to realize a monopole-like pattern and respectively. The inter-element phase shifts required to have improve lateral radiation. However, larger ground planes are the main beam of the array sweep along the four diagonal required to move the element pattern maximum towards the ◦ ◦ ◦ ◦ directions i.e., φ0 = 45 , −45 , 135 , and −135 in the azimuth plane [18]. Due to the limited chip size, the ground ◦ azimuth plane (θ0 = 90 i.e., end-fire condition for a planar planes cannot be made arbitrarily large. Thus, there is a trade- array) are given by off between the antenna performance and geometry. In summary, the selection of antenna element is dictated by β = β = β = −k d cos φ (2) x a2,1 a4,3 0 x 0 the need to maximize lateral radiation (in the azimuth plane).

βy = βa2,4 = βa1,3 = −k0dy sin φ0 (3) Microstrip antennas such as the rectangular and circular patch which have their pattern maximum in the vertical plane are not Substituting the four different values for φ in (2) and (3) 0 optimized for lateral transmission. Furthermore, to get good with d = d = d yields four different combinations of β x y x impedance matching, these antennas must be probe fed off and β which are listed in Table I. y center. This results in azimuthal asymmetry in the element radiation pattern. The asymmetric probe radiation can result B. Main Beam Formation in unwanted side lobes [18]. For the phase shifts given in (2) and (3), only the array factor is maximized in the azimuth plane (i.e., end-fire) along III.CENTER FED CIRCULAR PATCH WITH SIDE VIAS φ . The element pattern has to be considered as well. The total 0 In light of these issues, we propose a center fed circular far-zone electric field of the array E~ is the product of the total patch with symmetrically placed side vias around the center field of a single element E~ and the array factor AF [18], single as the antenna element. Such circular patches have been used expressed as at low frequencies to reduce mutual coupling due to surface ~ ~ Etotal(θ, φ) = Esingle(θ, φ) × AF(θ, φ) (4) waves [21] and improve pattern smoothness [22]. Four side vias around the center feed serve to improve impedance match- The element far-field can be decomposed into vertical E θ ing. The location of side vias must be optimized to improve and horizontal E polarization components as follows. φ matching since the center feed via cannot be moved to change ~ Esingle(θ, φ) = Eθaˆθ + Eφaˆφ (5) the input impedance. The vias maintain azimuthal symmetry 4 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

G (W = 7.3 mm) G (W = 14 mm) G (W ) G (W ) θ g θ g ϕ g = 7.3 mm ϕ g = 14 mm θ ϕ

(a) (b) Fig. 4. Simulated gain patterns (dB) of the center fed circular patch (with four side vias) at 60 GHz showing both Gθ and Gφ gain components for two different ground plane sizes. (a) Vertical plane pattern (φ = 0◦). (b) Horizontal plane pattern (θ = 90◦).

make it resonate at 60 GHz. The length and the width of the ground plane are each set to Wg = 7.3 mm. The side view of (a) the antenna structure with the stackup of the layers is shown in Fig. 3(c). There are two conductor layers: antenna and ground plane. The layers are separated from each other by a RO4003C laminate. The laminate thickness h is 0.2 mm. The thickness t of printed copper is 35 µm for both the antenna and ground planes.

B. Radiation Pattern Full wave simulation of the center fed circular patch with four side vias was performed in HFSS. The simulated gain (b) patterns of the patch at 60 GHz in the vertical (elevation) and horizontal (azimuth) plane are shown in Fig. 4. Both vertical Gθ and horizontal Gφ gain components are shown. From the vertical plane pattern shown in Fig. 4(a), we can see that the pattern maximum has tilted away from vertical (θ = 90◦). Moreover, the horizontal plane pattern shown in Fig. 4(b) has (c) azimuthal symmetry within 1 dB and can be thought of as isotropic in that plane. It is also important to note from the Fig. 3. HFSS 3D model of center fed circular patch with four side vias. (a) Top view. (b) Perspective view showing vertical and horizontal planes. patterns that the patch radiates mostly in Gθ polarization. The (c) Stackup view. vertical pattern in Fig. 4(a) has no Gφ component above the minimum plot scale value of −20 dB. The horizontal pattern in Fig. 4(b) shows that Gθ polarization is at least 19 dB in the element pattern if their distances from the patch center higher than Gφ polarization in the diagonal directions. The are kept the same. In addition, the side vias improve lateral highest Gφ polarization is 9 dB below Gθ polarization for radiation because they cause the pattern maximum to move Wg = 7.3 mm. away from vertical and towards the horizontal plane. The ground plane size was increased to 14 mm to see its effect on the radiation pattern of the circular patch. In Fig. 4(a), A. Structure of Circular Patch with Side Vias we can see that the increase in ground plane size has further The high frequency structural simulator (HFSS) 3D model tilted the pattern maximum away from the vertical while of center fed circular patch with four side vias is shown in producing more ripples in the pattern. The ripples are due to Fig. 3. Rogers RO4003C is used as the antenna substrate [23], edge diffraction, the nature of which depends on the ground a low loss hydrocarbon ceramic with dielectric constant of 3.55 plane size [18]. In the horizontal plane pattern of Fig. 4(b), the and loss tangent of 0.003. The diameter of the circular patch Gθ component has increased slightly while the Gφ component a is 1.8 mm. The radial locations of the vias b deﬁned from has reduced. Circular ground planes can be used to further the patch center to the via centers are 0.61 mm each. The reduce the Gφ polarization levels [24]. However, the antennas diameters of the center feed via and four side vias af are 0.15 are expected to be mounted on chips that have square or mm each. These are optimized values for the patch element to rectangular geometry. A ground plane of similar shape is easier BANIYA et al.: CHIP-TO-CHIP SWITCHED BEAM 60-GHz 5

Gϕ Gθ Gθ + Gϕ ϕ ϕ

(a) (b)

ϕ ϕ Fig. 5. HFSS 3D model of 2 × 2 circular patch planar array overlaid with simulated 3D gain pattern (dB) at 60 GHz. The main beam is pointed at ◦ φ0 = +135 . to integrate and also emulates the antenna performance better in the scenario.

IV. CIRCULAR PATCH PLANAR ARRAY The isotropic elements in Fig. 2 are replaced by the circular patches to form a 2 × 2 circular patch planar array. The patch elements are excited with equal amplitude and have the (c) (d) same inter-element separation and phase shifts as the isotropic Fig. 6. Simulated gain patterns (dB) of the 2 × 2 circular patch planar array at 60 GHz in the horizontal plane (θ = 90◦) showing switching of elements they are replacing. ◦ ◦ the main beam (for Wg = 7.3 mm). (a) φ0 = −45 . (b) φ0 = +45 . ◦ ◦ (c) φ0 = −135 . (d) φ0 = +135 . A. Array Structure The detailed 3D structure of the 2 × 2 circular patch planar array is modeled in HFSS which is shown in Fig. 5 along with the overlay of simulated 3D gain pattern (dB) at switching of the main beam can be seen in the horizontal 60 GHz. The inter-element separations, dx and dy are both gain patterns shown in Fig. 6, corresponding to four different fixed at d = 1.86 mm using (7) at wavelength λ0 = 5 mm, corresponding to 60 GHz center frequency. The size of the combinations of βx and βy values in Table I. ground plane W for the array is kept at 7.3 mm (i.e., same g The gain patterns show both vertical G and horizontal G as that for the patch element). Due to mutual coupling, the θ φ gain components along with their total (G + G ). The sum patch elements in the array do not resonate at 60 GHz when θ φ represents the total gain in the polarization matched case. Both a and b are set to the values given in Fig. 3. The parameter f gain components contribute to the total transmission coeffi- b was optimized to shift the resonance frequency of the patch cient when the TX and RX antennas are polarization matched, elements in the array back to 60 GHz. The optimized value as will be shown using the Friis equation in Section V-B. The of b is 0.67 mm. The main beam of the array is pointed at horizontal patterns represent the horizontal cut of the full 3D φ = +135◦ in Fig. 5 as a result of realizing β = +90◦ 0 x patterns. For example, Fig. 6(d) is simply the horizontal plane and β = −90◦ in the excitation of the simulation and this y cut of the 3D gain pattern shown in Fig. 5. The array has matches with what is expected from Table I. It verifies that a peak gain of 4.5 dBi at 60 GHz in the horizontal plane. the array is working as designed. The array radiates mostly in Gθ polarization (at least 24 dB higher than Gφ polarization) along the main beam directions B. Switching of Main Beam as shown in the radiation patterns of Fig. 6. This is to be The circular patch elements are uniformly excited with expected since the patch element itself radiates mostly in Gθ inter-element phase shifts given in Table I to scan the main polarization (See Fig. 4). The highest Gφ polarization of the beam of the array in the four diagonal directions. Each array is 7 dB below the peak gain. Increasing the size of the combination of phase shifts produces a main beam in one of ground plane to 14 mm increased the peak gain of the array four diagonal directions. Each element can be excited with a to 5.75 dBi while reducing the highest Gφ polarization to 12 different absolute phase in HFSS. By setting the appropriate dB below the peak gain. This is again expected since a similar absolute phase to each element in the excitation, the required trend was observed in the radiation pattern of the isolated patch phase difference between elements can be easily realized. The element (See Fig. 4) when the ground plane size was increased. 6 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

TABLE II SENSITIVITYOF ARRAY PERFORMANCETO CHANGESIN ANTENNA PARAMETERS

Array Performance Parameter a |S11| BW Deviation f0 (at 60 GHz) (|S11|≤−10 dB) +0.03 mm 63.1 GHz −4.9 dB N/A ∆af −0.03 mm 56.8 GHz −9.2 dB N/A +0.03 mm 60.9 GHz −14 dB 4.5 GHz ∆b −0.03 mm 58.7 GHz −11.1 dB 3 GHz

∆Wg +6.7 mm 60.2 GHz −29.6 dB 4.9 GHz a Deviations are changes from the original parameter values: af = 0.15 mm, b = 0.67 mm and Wg = 7.3 mm.

1 Fig. 7. Reﬂection coefﬁcient magnitude (dB) of the center fed circular patch Antenna Plane elements in the 2 × 2 array. RO4003C h Ground Plane

C. Reflection Coefficient RO4450F h Feed Layer A center fed circular patch by itself has poor return loss C4 bumps C4 bumps because the electric field value vanishes at the center [18]. The side vias change the boundary condition at the patch center CMOS chip and allow much better return loss with a center feed [21], [22]. Since the patches are so close to one another, near-field Fig. 8. Illustrative side view of the antenna structure showing antenna, ground coupling between them also affects their input impedance that and feed layers in AiP implementation. The antenna module is shown as is inherently taken into account in the full wave simulation. mounted on a CMOS chip using C4 bumps. In the presence of mutual coupling, the optimized value of b = 0.67 mm makes the patch elements in the array resonate lower loss and thus better signal-to-noise ratio (SNR) over an at 60 GHz. Fig. 7 shows the optimized reflection coefficient (in antenna-on-chip (AoC) implementation in lossy silicon [26]. solid blue) of the circular patch elements in the array, obtained Before the antenna can be packaged with the chip, the feed from simulation. An impedance bandwidth of 5.6 GHz is network that attains the required inter-element phase shifts achieved, using the |S | ≤ −10 dB criterion, around the 60 11 must be realized on a separate layer below the ground plane GHz when a = 0.15 mm, b = 0.67 mm and W = 7.3 mm. f g as depicted in Fig. 8. The design of feed network is beyond Fig. 7 also shows how the reflection coefficient of the the scope of this work. The antenna module can be mounted circular patch elements in the array changes with variations onto the CMOS chip by using flip chip (C4) bumps [25], [27] in the parameters a , b and W . Only one parameter is f g to connect the feed layer to the chip. The solid ground plane varied at a time while the other two are kept at their original helps to minimize interference by blocking antenna radiation values given in Fig. 3. The resonance frequency (f ), reflection 0 into the feed and the CMOS circuits underneath. coefficient magnitude (|S11|) and impedance bandwidth (BW) are most sensitive to the changes in a . Due to a large f E. Antenna Fabrication Considerations shift in the resonance frequency, the |S11| at 60 GHz has gone above −10 dB and the BW cannot be defined for Rogers RO4003C laminate was chosen as the antenna substrate because it has low dielectric loss, and unlike polyte- ∆af = 0.03 mm. Therefore, this parameter should have tight tolerances for fabrication. The performance of the array trafluoroethylene (PTFE) based laminates, it does not require has moderate sensitivity to the changes in b, whereas little any additional chemical wetting preparation process for plated- through hole (PTH) via formation. Since each patch element sensitivity to the increase in Wg. The larger ground plane has multiple vias, the use of RO4003C laminates can greatly did help to improve (reduce) the |S11| at 60 GHz but with a slight reduction in BW. The performance changes have been simplify the fabrication of the antenna array. Moreover, as summarized in Table II. shown in Fig. 8, RO4003C laminates can be stacked and bonded with RO4450F bondplys following standard printed circuit board (PCB) fabrication techniques to realize a cost D. Antenna Packaging Considerations effective multilayer structure [23]. The array achieved a simulated radiation efficiency of 97% at 60 GHz. This high efficiency is attributed mainly to the use V. LINK BUDGETWITH POLARIZATION FOR WIRELESS of the low loss RO4003C substrate. Since the substrate used INTERCHIP COMMUNICATION is not silicon, an antenna-in-package (AiP) implementation of The Friis transmission equation can be used to model the the proposed array is intended for integration with the CMOS wireless link between chip-to-chip antennas provided the far- chip. The AiP solution is recommended [25] because it offers field criterion is met. BANIYA et al.: CHIP-TO-CHIP SWITCHED BEAM 60-GHz 7

A. Far-Field Criterion that the angle dependent gains are also functions of the The Friis equation is valid if the antennas are in the far- pattern configuration selected in Fig. 6. The received power field regions of one another. For an antenna array, the far-field Pr represents the signal power S when the RX antenna is criterion [18] is satisfied at a distance R from the center of receiving from an intended TX and interference power I when the array if the RX antenna is receiving from an unintended TX. 2D2 Accurate analytical expressions for the reflection coefficient R ≥ (8) and the gain of the antenna arrays considered in the paper λ0 do not exist. Instead, simulated and measured values must where D is the largest dimension of the array. For broadband be used in (9). One might wonder why the Friis equation antennas, the criterion must be satisfied for all wavelengths is useful if the simulated values must be obtained and used. (frequencies) within the antenna bandwidth for accurate mod- The reason is that the Friis model only requires a full wave eling of the wireless link. For the 2 × 2 circular patch planar simulation of a single antenna array and the results can be used array shown in Fig. 5, the largest dimension is the diagonal to model the wireless link between any two far-field separated length of the array i.e., D = 4.5 mm and the far-field criterion identical antenna arrays in any orientation. This offers much is satisfied for R ≥ 7.78 mm at 60 GHz. The criterion is more flexibility and is much less time consuming than the most restrictive for the highest frequency fmax (corresponding full wave simulation involving two antenna arrays in specific to the smallest wavelength λmin) considered. If the criterion is orientations separated by a large far-field distance [28]. satisfied at fmax, it will be satisfied at any frequency lower than In order to see how polarization affects transmission, the fmax i.e., f ≤ fmax (corresponding to λ0 ≥ λmin in (8)). Since gain vectors of the TX and RX antennas are decomposed into the highest measurement frequency is fmax = 67 GHz for the vertical Gθ and the horizontal Gφ polarization components this work, the far-field criterion is satisfied at all frequencies as follows. f ≤ 67 GHz if the antenna arrays are separated by R ≥ 9 mm. ~ This far-field criterion is easily satisfied for typical chip-to- GTX (f, θt, φt) = Gθt (f, θt, φt)âθ + Gφt (f, θt, φt)âφ (10) chip distances at few tens of millimeters [11]. For the two- G~ (f, θ , φ ) = G (f, θ , φ )â + G (f, θ , φ )â (11) element antenna array over an emulated artificial magnetic RX r r θr r r θ φr r r φ conductor (AMC) ground in [28], D = 3 mm and the far- The gain components Gθ and Gφ are related to their field criterion is easily satisfied at distance R = 10 mm for respective electric field components Eθ and Eφ of the antennas all frequencies up to fmax = 67 GHz. The Friis equation as follows [18]. cannot accurately model the wireless link in the near-field 2π 2 region because it does not take into account the field behavior Gθ,φ = |Eθ,φ| (12) associated with the near-field. Furthermore, the near-fields of η0Pt the antenna arrays are perturbed if they are in those regions where η0 = 377 Ω is the intrinsic impedance of free space. 2 of one another. Perturbation of the near-field of an array can Each reflection loss term (1 − |S11(f)| ) in (9) can be cause undesirable and unpredictable changes in the reflection combined into a gain term Gθ,φ and the product is called the coefficient, impedance bandwidth and radiation pattern (i.e., realized gain Gθ,φ(rlzd), expressed in equation form as far-field) of the array. 2 Gθ,φ(rlzd) = (1 − |S11(f)| )Gθ,φ (13) B. Frequency Dependent Friis Equation with Polarization The ‘+’ sign is chosen in (9) if both the TX and RX anten- To account for frequency variations, the frequency depen- nas have their vertical and horizontal electric field components dent reflection coefficient, and the antenna gains are used in in the same direction. The electric field vectors of the TX the Friis equation as follows [28]. and RX are aligned and thus are polarization matched. For example, assume the TX antenna has its vertical component 2 2 Pr 2 λ0 in the +θ direction and the horizontal component in the +φ = 1 − |S11(f)| Pt 4πR direction i.e., (+Eθt , +Eφt ). If the RX antenna has the same polarization direction as the TX i.e., (+E , +E ), then ‘+’ × G~ (f, θ , φ ) · G~ (f, θ , φ ) θr φr TX t t RX r r sign is chosen in (9). The case in which the RX antenna p has (−Eθr , −Eφr ) polarization components can be shown to 2 Gθt Gθr Gφt Gφr (9) be equivalent to the case of (+Eθr , +Eφr ) and ‘+’ sign is where Pr and Pt are the received and transmitted (input) used in (9). In this case, the TX and RX field vectors are powers respectively, R is the distance between the anten- just anti-aligned but are still polarization matched. The ‘−’ nas, λ0 is the free space wavelength, S11(f) represents the sign is chosen in (9) if either the vertical or horizontal (but reflection coefficient of the identical TX and RX antennas, not both) component of the TX antenna is in the opposite and G~ TX (f, θt, φt) and G~ RX (f, θr, φr) are the gain vectors direction to the respective vertical or horizontal component of the TX and RX antennas, respectively, along the line of of the RX antenna. The electric field vectors of the TX and sight. G~ TX (f, θt, φt) and G~ RX (f, θr, φr) can be same or RX are orthogonal and thus are not polarization matched. For different even for identical TX and RX antennas depending example, if the RX antenna has (−Eθr , +Eφr ) polarization on their orientation (θt, φt) and (θr, φr) with respect to the components, then ‘−’ sign is chosen in (9). The case in which line of sight. The switchable nature of the antennas means the RX antenna has (+Eθr , −Eφr ) polarization components 8 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

TABLE III (Simulated) (Simulated) SIGN DETERMINATION IN (9) FORTHE TX AND RXANTENNAS BASED G? G? (Measured) G3 G3 (Measured) ONTHE DIRECTIONOF THEIR POLARIZATION COMPONENTS ϕ ϕ aa 0 0 a RX (+E ,+E )(+E ,−E )(−E ,+E )(−E ,−E ) -30 30 -30 30 TX aa θr φr θr φr θr φr θr φr 0 0 -5 -5 (+Eθ ,+Eφ ) + − − + -60 60 -60 60 t t -10 -10 (+E ,−E ) − + + − θt φt -15 -15 (−E ,+E ) − + + − θt φt -90 90 -90 90

(−Eθt ,−Eφt ) + − − +

-120 120 -120 120

-150 150 -150 150 180 180 (a) (b)

ϕ ϕ 0 0 -30 30 -30 30 (a) (b) (c) 0 0 Fig. 9. Two-element antenna array over a patterned ground plane. (a) Fab- -5 -5 ricated PCB prototype. (b) Radiation pattern measurement setup in a V-band -60 -10 60 -60 -10 60 anechoic chamber. (c) HFSS 3D model. -15 -15 -90 90 -90 90 can be shown to be equivalent to the case of (−Eθr , +Eφr ) and ‘−’ sign is used in (9). When patterns are switched, the -120 120 -120 120 directions of field components can change. The changes in the -150 150 -150 150 polarization direction must be taken into account by choosing 180 180 the proper sign in (9). Table III considers all the combinations (c) (d) of polarization component directions between the TX and RX Fig. 10. Measured and simulated realized gain patterns (dB) of the two- antennas. element antenna array at 60 GHz in the horizontal plane (θ = 90◦) showing Let us consider a polarization matched case in the both Gθ and Gφ gain components. (a) 57 GHz. (b) 60 GHz. (c) 63 GHz. MCMC scenario of Fig. 1. All the antenna arrays in (d) 66 GHz. Fig. 1 are identical and have the same orientation. If the main beams of any two arrays are pointed at one an- layer facing down and the feed layer facing up as shown other (consider e.g., E and C), then Gθ = Gθ and t r in Figs. 9(a) and 9(b). The foam emulates air since it has Gφt = Gφr with the ‘+’ sign in (9). Using (10) and (11) in (9) with the aforementioned substitutions, the fac- a dielectric constant close to one, while providing a stable platform and causes only small perturbation in the near-field of tor G~ (f, θ , φ ) · G~ (f, θ , φ ) + 2pG G G G TX t t RX r r θt θr φt φr the array. This ensures that the platform has very little effect on simplifies to (G + G )2 in (9), and thus both gain compo- θt φt the radiation pattern measurement [29]. Realized gain patterns nents add to the total transmission. of the two-element array were measured at 57, 60, 63, and 66 GHz in the horizontal plane and are shown in Fig. 10 along C. Signal and Interference with the simulated patterns. In MCMC computing, several pairs of chips are expected to Now consider that the 2 × 2 circular patch planar arrays communicate concurrently in order to achieve a high degree of on the routers E, F, H and I in Fig. 1 are replaced by the parallelism. Side lobes in the radiation pattern of an antenna two-element antenna arrays in different orientations as shown pair can interfere with neighboring antenna pairs that are also in Fig. 11 along with their radiation patterns. The measured communicating. Polarization components of the radiated fields and simulated gain patterns of Fig. 10(b) were overlaid on the must be properly considered to estimate interference levels in routers in Fig. 11 to show how the change in pattern orientation addition to the signal levels. follows the change in antenna orientation. The antenna arrays In order to understand the effect of antenna radiation in the on the routers E and F are oriented so that their main beams MCMC system, the two-element antenna array of [28] is fur- point at one another enabling maximum transmission between ther studied. Note that this is a previous generation array to the them and are referred to as the communicating pair. Imagine switched beam circular patch array discussed in Section IV. at the same time, the antenna arrays on the routers H and The fabricated PCB prototype, pattern measurement setup and I are turned on. Radiation from the antenna arrays on the HFSS 3D model of the array are shown in Fig. 9. routers H and I in the direction of the routers E and F The measurement was made in a V-band anechoic chamber contribute to interference and vice versa. The objective is using a 150 µm pitch Ground-Signal-Ground (GSG) probe to maximize radiation between the antenna arrays in the on the coplanar waveguide (CPW) feed line of the array. The intended direction e.g., between E–F (communicating pair) antenna module is placed on a foam platform with the antenna while minimizing radiation in other directions e.g., between BANIYA et al.: CHIP-TO-CHIP SWITCHED BEAM 60-GHz 9

0 Measured Core Core Simulated

Core Core Core Core -10 Friis (Measured gain) | (dB)

21 Friis (Simulated gain) E F S -20 Core Router Core Core Router Core

-30

Core Core Core Core -40 Core Core

-50

Core Core Transmission coefficient | Core Core Core Core -60 56 57 58 59 60 61 62 63 64 65 66 67 Frequency (GHz)

Core Router Core Core Router Core Fig. 13. Measured, simulated and calculated transmission coefﬁcient between H I the pair of two-element antenna arrays of Fig. 12.

Core Core Core Core

Core Core tenna pairs if the radiation pattern is known. By simply taking the appropriate gain values in the direction of interest, the link N y signal and interference power levels can be calculated using W E Two-element antenna array (9). The frequency dependent and angle dependent antenna with radiation pattern x gains are used in the Friis model in (9) to account for signal S z variations due to frequency and beam squint. This was done for the communicating pair E–F. The transmission coefficient Fig. 11. 2 × 2 MCMC unit showing two-element antenna arrays on routers curves are calculated using the measured and simulated gain overlaid with the 60 GHz radiation pattern taken from Fig. 10(b). ◦ values such as the ones shown in Fig. 10, at φt = φr = 180 for the TX and the RX array. The calculated results are in good agreement with the measured and the full wave simulated results as shown in Fig. 13. To account for the interference levels, the transmission measurements should also be made between all antenna pairs that are not intended for communication. Such transmission measurements are different because the antennas are not aligned for maximum transmission. The measurement requires actually laying out the antenna pair in the specific orientation and making the measurement. Instead, (9) can be used to calculate the interference levels by taking the appropriate gain values of the interfering antenna pair in the direction of interest. This was done for the interfering pair E–H in Fig. 11 Fig. 12. A pair of two-element antenna arrays separated by 10 mm aligned for maximum transmission to mimic the communicating pair E–F of Fig. 11. by using the antenna gain values, both measured and simulated ◦ In the inset, fabricated prototype of the pair with the GSG probes on the CPW shown in Fig. 10 for the given four frequencies, at φt = 180 feed lines is shown. ◦ for the TX array and at φr = 90 for the RX array. The HFSS 3D model that emulates the interfering pair is shown E–H and E–I (interfering pairs). in Fig. 14. The calculated results are in good agreement with To understand the communication scenario between the transmission coefficient obtained from full wave simulation as routers, the transmission coefficient was measured and verified shown in Fig. 15. with the Friis model in [28] for a pair of two-element array The Friis model also makes it easier to calculate the signal- ◦ on a patterned ground plane that are aligned for maximum to-interference ratio (SIR). In the horizontal plane (θ = 90 ), transmission, which is shown in Fig. 12. The pair mimics the the SIR at the RX router E due to signal from the TX router F ◦ ◦ communicating pair E–F in Fig. 11. The measurement was (φt = 180 , φr = 180 ) and interference from the TX router H ◦ ◦ made on a fabricated prototype of the pair using 150 µm pitch (φt = 180 , φr = 90 ) can be derived from (9) and expressed GSG probes on the CPW feed lines, which is shown in the as follows. ◦ ◦ ! inset of Fig. 12. The result of the measurement is shown in G~ TX (f, φt = 180 ) · G~ RX (f, φr = 180 ) Fig. 13. p 2 Gθ Gθ Gφ Gφ The transmission measurements are not necessary if the SIR = t r t r (14) ~ ◦ ~ ◦ ! radiation pattern is measured in the antenna plane. The Friis GRX (f, φt = 180 ) · GRX (f, φr = 90 ) equation can be used to estimate transmission between the an- p 2 Gθt Gθr Gφt Gφr 10 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

TABLE IV ARRAY PERFORMANCE COMPARISON AT 60 GHz

Antenna Type Peak Gain Maximum SLL Pattern Single Circular Fixed Beam −1 dBi −1.3 dBi Patch (Omnidirectional) Two-Element Fixed Beam 4 −5 Array dBi dBi (Directional) 2 × 2 Circular Switched Beam 4.5 dBi −0.7 dBi Patch Array (Directional)

array has a higher horizontal gain than the patch element and Fig. 14. A pair of two-element antenna arrays separated by 10 mm, one of them rotated by 90◦ to mimic the interfering pair E–H of Fig. 11. the two-element array. Therefore it can provide higher SNR on the channel and maintain the same signal level at a larger communication distance. This is evident from (9). At the same 0 time, the maximum SLL of the switched beam array is also Simulated Friis (Measured gain) higher. The antenna performances are summarized in Table IV. -10 Friis (Simulated gain) | (dB) The SLLs should be minimized (e.g., through null formation) 21

S in order to reduce unwanted radiation (interference) towards -20 unintended receivers. The feed network for the array should be able to achieve simultaneous beam and null formation to -30 direct power only in the intended directions [30]. This will undoubtedly increase the feed network and system complexity -40 for implementation and integration, and the associated costs for the switched beam array but the beneﬁts offered are -50 paramount. Transmission coefficient |

-60 56 57 58 59 60 61 62 63 64 65 66 67 Frequency (GHz) VI.CONCLUSION The design and simulation of a 2 × 2 circular patch Fig. 15. Simulated and calculated transmission coefficient between the pair of two-element antenna arrays of Fig. 14. The transmission contributes to planar array capable of switching its main beam for recon- interference. figurable interchip communication in the MCMC architecture is presented. The antenna array allows a chip to dynamically communicate to its diagonal neighbors reducing hop count. A Using (14) with the simulated gain values shown in simplified systematic approach is followed in the design of Fig. 10(b) at the given φt and φr angles, results in the the array. The mutual coupling between antenna elements is calculated SIR = 12.4 dB. This is 4.3 dB off from the ignored as it greatly simplifies the determination of the inter- simulated SIR = 16.7 dB obtained by taking the difference element separations and phase shifts required for switching of of transmission values at 60 GHz from the simulated curves main beam. A feed network with control switches that can in Figs. 13 and 15. realize and cycle through the required inter-element phase The two-element array discussed in this section serves to shifts should be designed and integrated with the array in demonstrate and validate the Friis model in the interchip the future before the array can be packaged with the CMOS communication scenario. It, however, has a fixed radiation transceivers and chips. The work demonstrates for the first pattern like that of the patch element and cannot be used for time a compact and pattern switchable 60 GHz antenna array reconfiguring the interchip link. Still, the verification of the that can be placed on the multicore chips to provide seamless Friis model using this previous generation array sets precedent interconnection and thus realize a readily scalable massively for the switched beam circular patch array. This provides multicore computing unit. This can have significant impact confidence in using the Friis model for the switched beam on large scale scientific computing by breaking performance array with the added feature of reconfiguration. Finally, it is barriers in parallel processing. Reconfigurable wireless links also desirable to have a constant link budget in the different can also improve system reliability by rerouting signals around scan directions e.g., between routers E–A, E–C, E–G and E– broken links. I in Fig. 1 when the switched beam circular patch arrays A frequency dependent Friis transmission equation is con- are used, so that same throughput can be expected between sidered for modeling the interchip wireless link. The often all communicating pairs. The array presents identical patterns ignored effects of polarization are taken into account in the in all the scan directions as shown in Fig. 6. Thus it can wireless link budget and discussed in detail in this work. Pairs maintain a constant link budget between the communicating of the two-element antenna arrays are studied for the purpose router pairs as the patterns are switched from one router to of verifying the Friis model. The far field criterion is checked another. In addition to reconfigurability, the switched beam and shown to satisfy through calculation at typical interchip BANIYA et al.: CHIP-TO-CHIP SWITCHED BEAM 60-GHz 11

distances for all frequencies within the antenna bandwidth. [22] D. Jackson, J. Williams, A. K. Bhattacharyya, R. L. Smith, S. J. The Friis model is then used to calculate the transmission Buchheit, and S. Long, “Microstrip patch designs that do not excite surface waves,” IEEE Trans. Antennas Propag., vol. 41, no. 8, pp. 1026– coefficients using both the simulated and measured antenna 1037, Aug. 1993. gain values. The calculated results obtained from the Friis [23] Rogers RO4003 Laminates Data Sheet, Rogers Corp., Chandler, AZ, equation agree well with the results obtained from full wave USA. [Online]. Available: http://www.rogerscorp.com [24] J. Liu, Q. Xue, H. Wong, H. W. Lai, and Y. Long, “Design and analysis simulation and measurement, thus validating the use of Friis of a low-profile and broadband microstrip monopolar patch antenna,” model. Further, the interference between antenna arrays due IEEE Trans. Antennas Propag., vol. 61, no. 1, pp. 11–18, Jan. 2013. to radiation in unwanted direction is calculated using the Friis [25] D. G. Kam, D. Liu, A. Natarajan, S. K. Reynolds, and B. A. Floyd, “Organic packages with embedded phased-array antennas for 60-GHz model and verified through simulation. wireless chipsets,” IEEE Trans. Compon. Packag. Manuf. Technol., vol. 1, no. 11, pp. 1806–1814, Nov. 2011. [26] S.-S. Hsu, K.-C. Wei, C.-Y. Hsu, and H. Ru-Chuang, “A 60-GHz REFERENCES millimeter-wave CPW-fed Yagi antenna fabricated by using 0.18-µm CMOS technology,” IEEE Electron Device Lett., vol. 29, no. 6, pp. [1] L. Benini and G. De Micheli, “Networks on chips: A new SoC 625–627, Jun. 2008. paradigm,” Computer, vol. 35, no. 1, pp. 70–78, Jan. 2002. [27] W. Hong, K.-h. Baek, and A. Goudelev, “Grid assembly-free 60-GHz [2] N. H. E. Weste and D. M. Harris, CMOS VLSI Design: A Circuits and antenna module embedded in FR-4 transceiver carrier board,” IEEE Systems Perspective, 4th ed. Boston, MA, USA: Addison-Wesley, 2011. Trans. Antennas Propag., vol. 61, no. 4, pp. 1573–1580, Apr. 2013. [3] A. Karkar, T. Mak, K.-F. Tong, and A. Yakovlev, “A survey of emerging [28] P. Baniya, A. Bisognin, K. L. Melde, and C. Luxey, “Impact of gain and interconnects for on-chip efficient multicast and broadcast in many- polarization in the design of reconfigurable chip-to-chip antennas,” in cores,” IEEE Circuits Syst. Mag., vol. 16, no. 1, pp. 58–72, 2016. Proc. 10th Eur. Conf. Antennas Propag. (EuCAP), Davos, Switzerland, [4] H.-H. Yeh, N. Hiramatsu, and K. L. Melde, “The design of broadband Apr. 2016, pp. 1–4. 60 GHz AMC antenna in multi-chip RF data transmission,” IEEE Trans. [29] D. Titz, F. Ferrero, and C. Luxey, “Development of a millimeter-wave Antennas Propag., vol. 61, no. 4, pp. 1623–1630, Apr. 2013. measurement setup and dedicated techniques to characterize the match- [5] L. P. Carloni, P. Pande, and Y. Xie, “Networks-on-chip in emerging ing and radiation performance of probe-fed antennas [measurements interconnect paradigms: Advantages and challenges,” in Proc. 2009 3rd corner],” IEEE Antennas Propag. Mag., vol. 54, no. 4, pp. 188–203, ACM/IEEE Int. Symp. Networks-on-Chip (NoC), San Diego, CA, USA, Aug. 2012. May 2009, pp. 93–102. [30] P. Baniya, S. U. Rehman, and J. W. Wallace, “Efficient optimization [6] H.-H. Yeh and K. L. Melde, “Development of 60-GHz wireless inter- of reconfigurable parasitic antenna arrays using geometrical considera- connects for interchip data transmission,” IEEE Trans. Compon. Packag. tions,” in Proc. 5th Eur. Conf. Antennas Propag. (EuCAP), Rome, Italy, Manuf. Technol., vol. 3, no. 11, pp. 1946–1952, Nov. 2013. Apr. 2011, pp. 270–274. [7] C. W. Byeon, C. H. Yoon, and C. S. Park, “A 67-mW 10.7-Gb/s 60- GHz OOK CMOS transceiver for short-range wireless communications,” IEEE Trans. Microw. Theory Tech., vol. 61, no. 9, pp. 3391–3401, Sep. 2013. [8] S. Liao and Q. Xue, “Dual polarized planar aperture antenna on LTCC for 60-GHz antenna-in-package applications,” IEEE Trans. Antennas Propag., vol. 65, no. 1, pp. 63–70, Jan. 2017. [9] K. Okada et al., “Full four-channel 6.3-Gb/s 60-GHz CMOS transceiver with low-power analog and digital baseband circuitry,” IEEE J. Solid- State Circuits, vol. 48, no. 1, pp. 46–65, Jan. 2013. [10] H. Zhou, “High performance computing architecture with security,” Ph.D. dissertation, Dept. Elect. Comput. Eng., Univ. Arizona, Tucson, AZ, USA, May 2015. [11] J. Warnock et al., “Circuit and physical design implementation of the microprocessor chip for the zEnterprise system,” IEEE J. Solid-State Circuits, vol. 47, no. 1, pp. 151–163, Jan. 2012. [12] A. Artemenko, A. Maltsev, A. Mozharovskiy, A. Sevastyanov, V. Ssorin, and R. Maslennikov, “Millimeter-wave electronically steerable integrated lens antennas for WLAN/WPAN applications,” IEEE Trans. Antennas Propag., vol. 61, no. 4, pp. 1665–1671, Apr. 2013. [13] D. Liu, J. A. Akkermans, H.-C. Chen, and B. Floyd, “Packages with integrated 60-GHz aperture-coupled patch antennas,” IEEE Trans. An- tennas Propag., vol. 59, no. 10, pp. 3607–3616, Oct. 2011. Prabhat Baniya (S’11) was born in Kathmandu, [14] D. Liu and Y. Zhang, “Integration of array antennas in chip package for Nepal, in 1985. He received the B.E. degree in com- 60-GHz radios,” Proc. IEEE, vol. 100, no. 7, pp. 2364–2371, Jul. 2012. puter engineering from Pulchowk Campus, Institute [15] Y. Li and K.-M. Luk, “A multibeam end-fire magnetoelectric dipole of Engineering, Tribhuvan University, Kathmandu, antenna array for millimeter-wave applications,” IEEE Trans. Antennas in 2008, and the M.S. degree in electrical engi- Propag., vol. 64, no. 7, pp. 2894–2904, Jul. 2016. neering from the University of Massachusetts Dart- [16] ——, “60-GHz dual-polarized two-dimensional switch-beam wideband mouth, North Dartmouth, MA, USA, in 2012. He antenna array of aperture-coupled magneto-electric dipoles,” IEEE is currently pursuing the Ph.D. degree in electrical Trans. Antennas Propag., vol. 64, no. 2, pp. 554–563, Feb. 2016. engineering with the University of Arizona, Tucson, [17] P. Ioannides and C. A. Balanis, “Uniform circular arrays for smart AZ, USA. antennas,” IEEE Antennas Propag. Mag., vol. 47, no. 4, pp. 192–206, From 2008 to 2010, he was a Part-Time Lecturer Aug. 2005. with Pulchowk Campus, Kathmandu. From 2010 to 2012, he was a Teaching [18] C. A. Balanis, Antenna Theory: Analysis and Design, 3rd ed. Hoboken, Assistant with the Department of Electrical and Computer Engineering, NJ, USA: Wiley, 2005. University of Massachusetts Dartmouth. From 2012 to 2015, he was a [19] C.-H. Tseng, C.-J. Chen, and T.-H. Chu, “A low-cost 60-GHz switched- Teaching Assistant and a Research Assistant with the Department of Electrical beam patch antenna array with Butler matrix network,” IEEE Antennas and Computer Engineering, University of Arizona, where he has been a Wireless Propag. Lett., vol. 7, pp. 432–435, 2008. Graduate Research Associate since 2015. In 2015, he joined Corning Optical [20] Z. N. Chen, D. Liu, H. Nakano, X. Qing, and T. Zwick, Handbook of Communications RF LLC, Glendale, AZ, USA, as a Microwave Engineering Antenna Technologies. Singapore: Springer Nature, 2016, vol. 1. Intern. He joined MEGGITT, Securaplane Technologies Inc., Oro Valley, [21] M. A. Khayat, J. T. Williams, D. R. Jackson, and S. A. Long, “Mutual AZ, USA, as a RF Designer, in 2016. His current research interests include coupling between reduced surface-wave microstrip antennas,” IEEE reconfigurable antennas, antenna array design, active microwave circuits, Trans. Antennas Propag., vol. 48, no. 10, pp. 1581–1593, Oct. 2000. meta-materials, electromagnetic modeling, and energy harvesting. 12 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. XX, XXX 201X

Aimeric Bisognin (S’13–M’15) was born in Cyril Luxey (M’03–SM’07–F’16) was born in Nice, Toulouse, France, in 1989. He received the engineer- France, in 1971. He received the master’s degree ing degree in electronics from Polytech’Nice Sophia, (Hons.) (DEA) and the Ph.D. degree (Hons) in elec- Sophia Antipolis, France, in 2012, the M.S. degree in trical engineering from the University Nice-Sophia telecommunications from EDSTIC, Sophia Antipo- Antipolis, Nice, in 1996 and 1999, respectively. lis, France, in 2012, and the Ph.D. degree (Hons) in He was involved in printed leaky-wave antennas, electronics engineering from the University of Nice quasi-optical mixers, and retrodirective transponders. Sophia-Antipolis, Nice, France, in 2015. From 2000 to 2002, he was with the, Mobile Phone He was with the Electronique pour Objets Com- Division, Alcatel, Colombes, France, where he was municants Laboratory, STMicroelectronics, Crolles, involved in the design and integration of internal France, where he was a Post-Doctoral Researcher. antennas for commercial mobile phones. In 2003, He has authored or coauthored over eight publications in journals and 19 he was an Associate Professor with the Polytechnic School, University publications in international conferences. His current research interests include Nice Sophia-Antipolis. Since 2009, he has been a Full Professor with the millimeter-wave communications, especially in the field of the design and IUT Reseaux´ et Tel´ ecoms´ in Sophia-Antipolis, Valbonne, France. He is measurement of antenna in package, lens, and reflector antennas for the 60-, the Co-Head of the Electronique pour Objets Communicants Laboratory, 80-, and 120-GHz frequency bands. where he is doing his research. In 2010, he was appointed as a Junior Member of the Institut Universitaire de France for five years. Also, he collaborates with Berkeley Wireless Research Center, Berkeley, CA, USA, and Stanford University, Stanford, CA, USA, on millimeter (mm)-wave front- end transceivers at mm-wave frequencies. He also works on electrically small antennas, multiantenna systems for diversity, and MIMO techniques. He has authored or coauthored more than 250 papers in refereed journals, in international and national conferences, and book chapters. He has given more than 15 invited talks. His current research interests include the design and measurement of millimeter-wave antennas, antennas-in-package, plastic lenses, and organic modules for 60-, 120-, and 240-GHz frequency bands. Dr. Luxey and his students received the H.W. Wheeler Award of the IEEE Antennas and Propagation Society (AP-S) for the best application paper of Kathleen L. Melde (S’84–M’95–SM’97–F’12) re- the year 2006. He was a recipient of the University Nice-Sophia Antipolis ceived the B.S. degree from California State Uni- Medal in 2014 and a recipient of the University Coteˆ d’Azur Medal in 2016. versity, Long Beach, CA, USA, the M.S. degree He was also the co-recipient of the Jack Kilby Award 2013 of the ISSCC from California State University, Northridge, CA, conference, and the Best Paper Award of the EuCAP 2007 conference, the USA, and the Ph.D. degree from the University of Best Paper Award of the International Workshop on Antenna Technology California, Los Angeles, CA, USA, all in electrical (iWAT2009), the Best Paper Award at LAPC 2012, the Best Student Paper engineering. Award at LAPC 2013 (third place), the Best Paper Award of the ICEAA From 1985 to 1996, she was with the Radar 2014 conference, and the Best Paper Award of the innovation contest of the Systems Group, Hughes Electronics, El Segundo, iWEM 2014 conference (second place). He has been the General Chair of CA, USA, where she was with the Electromagnetic the Loughborough Antennas and Propagation Conference 2011, the Award Systems Laboratory and the Solid State Microwave and Grant Chair of EuCAP 2012, the Invited Paper Co-Chair of EuCAP Laboratories of the Radar and Communications Sector. She has made con- 2013, and the TPC Chair of EuCAP 2017 in Paris, France. He is also the tributions to the design and development of antennas and transmit/receive Delegate of the French Research Ministry for the COST IC1102 action VISTA modules for airborne-phased and active arrays. She has extensive experience (Versatile, Integrated, and Signal-aware Technologies for Antennas) within in modeling, fabrication and measurement of the performance of antennas, the ICT Domain. Since 2015, he has been a member of the IEEE AP-S antenna arrays, high-density microwave circuits, and high-speed packaging Education committee. He was an Associate Editor of the IEEE ANTENNAS interconnects. She was a Task Leader for several internal research and AND WIRELESS PROPAGATION LETTERS from 2012 to 2017, a Regular development projects. In 1996, she joined the faculty of the Electrical and Reviewer of the IEEE TRANSACTIONSON ANTENNASAND PROPAGATION, Computer Engineering Department, University of Arizona, Tucson, AZ, USA, the IEEE ANTENNASAND WIRELESS PROPAGATION LETTERS, the IEEE where she is a Professor. Her current projects include tunable RF front TRANSACTIONSON MICROWAVE THEORY AND TECHNIQUES, the IEEE ends for cognitive radio, high-speed electronics packaging, on-chip antennas, MICROWAVE AND WIRELESS COMPONENTS LETTERS, Electronics Letters, and computational photovoltaics. She has authored or coauthored over 90 and Microwaves, Antennas and Propagation journals and several European publications and five U.S. patents. She has been an expert witness and and U.S. conferences in the field of microwave, microelectronics and antennas. consultant in the area of RF circuits and antennas. Her current research interests include applied electromagnetics, antenna theory and design, and microwave circuit design. Dr. Melde is a member of the IEEE Antennas and Propagation Society (AP- S), the IEEE Microwave Theory and Techniques Society, the International Radio Science Union, Eta Kappa Nu, Tau Beta Pi, and Sigma Xi. She was named the University of Arizona, College of Engineering Teaching Fellow in 2012. In 2010, she received the Excellence at the Student Interface Award from the University of Arizona, College of Engineering. She was a recipient of the 2008 IBM Faculty Award. She has been an invited Keynote Speaker on several occasions, such as the California State University Northridge, CA, USA, School of Engineering Commencement and the Conference on the Electrical Performance of Electronic Packages and Systems (EPEPS). From 1999 to 2001, she served on the Administrative Committee for the IEEE AP-S. She was the Co-Chair for the 2012 and 2013 Conferences on the EPEPS. She was on the Organizing Committee for the 2016 and 2017 Antennas and Propagation Symposia. She was an Associate Editor of the IEEETRANSACTIONSON ANTENNASAND PROPAGATION and the IEEE ANTENNASAND WIRELESS PROPAGATION LETTERS.