Bandwidth-Enhanced Circularly Polarized mm-Wave Antenna With On-Chip Ground Plane

This article presents a design methodology and a realization of an antenna radiating circular polarization (CP) for Frequency Modulated Continuous Wave (FMCW) radar at mm-wave frequencies on-chip, which combines antennas of different resonance frequencies to increase the operation bandwidth and allow for high-resolution radar. The antenna consists of four dipoles with an on-chip ground plane operating at two resonance frequencies combined with a matching and feeding network, which enables both frequency selectivity in advantage for the resonant antenna and radiation of circular polarization. The dipole arms are based on shorted ${\lambda }{/4}$ resonators, which are enhanced with series capacitances for increased radiation efficiency. A method for the broadband characterization of CP antennas and the measurement results of the designed CP antenna are presented. It is shown that the antenna covers a bandwidth between 220 and 260 GHz, indicating the feasibility of both the measurement method and the antenna concept.

resonant and thus inherently band-limited, which also relates to the bandwidth of the generated CP [7]. Although backsideradiating chip antennas with high-permittivity lenses can easily achieve a wider bandwidth and higher efficiency [8], [9], the use of on-chip ground planes for frontside-radiating antennas eases the packaging process, since neither flip-chip bonding nor lens positioning is necessary [10]. Alternatively, the use of off-chip ground planes [11] or thick substrates [12], [13] usually allows for a wideband CP operation. However, for higher frequencies, substrate waves in the thick substrate [5] need to be suppressed or to become insignificant due to the use of a high-gain lens. Furthermore, the high gain associated with the utilized lens finally limits the applicability in large antenna arrays in combination with beam-steering [14], since the principle of pattern multiplication ultimately reduces the array performance off broadside. With the utilized CP, each reflection will change the polarization from left-handed (LH-) to right-handed CP (RHCP) and vice versa. Thus, the detection of simple targets demands different polarized Tx and Rx antennas or a quadrature hybrid coupler, which feeds two orthogonal polarizations, as in [8]. The antenna introduced here utilizes capacitive discontinuities in quarter-wavelength grounded dipole arms to increase the radiation efficiency [15], [16]. Furthermore, the full antenna design process is presented, and characterization of the antenna's polarization up to 280 GHz is carried out utilizing a rotating rectangular waveguide and maintaining a phase-correct antenna measurement.
This article is organized as follows. Section II will present a generic antenna design to maximize the utilization of the available chip area. Furthermore, the differential feeding network concept assuring frequency selectivity and correct feeding phases for the CP generation is introduced. Based on the single component, Section III shows the total antenna assembly and simulated current density for physical insight. To measure the radiated polarization, Section IV describes the utilized measurement method for polarizationswitchable phase-stable mm-wave measurements at each far-field sampling point. Finally, Section V presents the measured characteristics of the antenna fabricated in the Infineons B11HFC silicon-germanium technology and compares these with full-wave simulation results. frequency-selective feeding network, where the antenna elements utilize most of the available chip area by design (see Fig. 7). In detail, each antenna element is a shorted λ/4 resonator, such that a pair of two opposing elements build a λ/2 dipole with shorted ends. Each of these elements will be referred to as dipole arms to emphasize their final function. Here, the central feeding eases the feeding network design, as will be discussed later (see Section II-B). Since both conductor and dielectric losses usually allow for only moderate radiation efficiencies, special attention is paid to increasing the radiation efficiency by inserting a capacitive discontinuity [15], [16]. Fig. 1(a) and (b) shows the single antenna element geometry for both utilized elements. Within this publication, these elements will be referred to as A and B. Although omitted in most of the drawings presented here for reasons of clarity, the whole antenna is located over a ground plane on the lowermost on-chip layer with ≈ 9 μm distance and surrounded by SiO 2 .  Table I. The rather unusual shape of the antennas allows for a densely packed arrangement on-chip to fully utilize the available chip area, as will be shown in Section III. The short at the end of the antenna allows for a much shorter realization of a λ/4 dipole arm [17, p. 7-3] at the expense of higher losses compared with a λ/2 resonator with an open termination. Especially due to the multilayered via structures used to realize the short and the close proximity of the ground plane, large dissipative conductor losses are expected. To overcome this drawback, a radiating capacitive discontinuity is introduced into the antenna increasing the radiation efficiency to a suitable level. A thorough discussion on the effects of the discontinuity on the bandwidth and efficiency is presented in [15]. Additionally, a full explanation of the radiation mechanism can be found in [18] and [19].

A. Antenna Element Design
Based on the equivalent circuit (EC) from [16], an EC model for the dipole arms is depicted in Fig. 2. Since both dipole arms A and B share the same principle of operation, their general EC is identical, and only the EC elements differ due to the different antenna geometries. Starting from the antenna's feeding point at the open end, which is modeled by a capacitance C End and a radiation conductance G End according to [18] and [20], a tapered microstrip section is followed by a capacitive discontinuity. After the discontinuity, a widened microstrip section is terminated by a short circuit to ground. At resonance, the high impedance at the feeding point will result in low current and thus comparably smaller losses in the narrow part of the tapered section. Along the antenna length, the current increases toward the short, only affected by the discontinuity as discussed in [16]. The discontinuities are controlled by the overlap of the top aluminum and the underlying copper layer [see Fig. 1(c)], and are described by the capacitance C MIM , a dissipation conductance G Diss accounting for ohmic and dielectric losses, and a radiation conductance G Rad to model the radiation of the gap. They yield both an increased radiation efficiency and a further degree of freedom in the antenna design, where the antenna length can be partly compensated by the capacitance, enabling frequency-tuning without modifying the antenna length. The short circuit is described by the series inductance L Short , the dissipation resistance R Diss accounting for mainly ohmic losses, and the radiation resistance R Rad . Here, the radiation resistance of a dipole in a grounded dielectric according to [21, p. 42] can be used for R Rad , and the microstrip line is well described by characteristic impedance [22] and conductor losses [23] by the incremental inductance rule. All remaining EC elements are  extracted as general functions of frequency between 220 and 300 GHz according to the method from [16]. Table I gives the EC elements evaluated at the operation frequencies of the respective antennas. Both antenna elements A and B are designed not only to geometrically match each other and the available chip area but also to operate at different frequencies f A and f B in combination with the feeding and matching network. Fig. 3(a) and (b) depicts the input impedance for both antennas, which indicates the λ/4 resonances, calculated by an FDTD solver and with a circuit simulation of the presented ECs. Given the simplicity of the ECs, the agreement to the full-wave simulation is very good, where only a small frequency shift can be observed. Here, it should be emphasized that the dipole arms A and B will be detuned by the matching network to obtain the operation at f A and f B . Furthermore, Fig. 3(c) shows the simulated radiation efficiency for both antenna elements, intended to operate at f A = 230 GHz and f B = 250 GHz for A and B, respectively. Here, the agreement between FDTD result and EC modeling is very good. Finally, the EC description enables physical insight into the radiation mechanism of the antenna, as depicted in Fig. 4. Here, the contributions to radiation and dissipation losses are visualized for different parts of the dipole arms at their respective operation frequencies. These antenna parts are the open-ended feeding point, TL sections of each ≈ 50 μm length, the capacitive gap, and the short circuit at the via. The power contributions are separated into dissipation and radiation for each element to allow for interpretation of its use in the overall antenna. The main radiation contributions result from the capacitive gap and the short circuit, and only a very small fraction is radiated by the open circuit. The losses are mainly caused by the via and the wide TL sections, which is reasoned by the current distribution depicted as well. Since both arms are operated close to their λ/4 resonance, the current peaks at the antenna end, resulting in the largest loss contributions. For comparison, both the current distributions calculated by the circuit simulation and the FDTD method are given, showing a very good agreement as well. The minimum current calculated by the FDTD at the capacitive gap results from the current being calculated only on the uppermost aluminum layer, neglecting currents in the MIM layers. The input impedance and the corresponding matching conditions will be highlighted in the next paragraph.

B. Feeding Network Design
Within the operation bandwidth of the antenna elements, the feeding network should enable a good matching condition for each four of the eight elements by its frequency selectivity Fig. 5. Schematic of the feeding and matching network including each four antennas A and B, their respective input impedances Z A and Z B , their impedance after transformation Z A and Z B , and the total input impedance Z 1 corresponding to a differential feeding at 1 and 1 . The lengths and impedances (Z C ) of the TLs are chosen to obtain both matching and phase-correct feeding for a circular polarization at 250 GHz. and result in a phase shift to ideally radiate circularly polarized waves. A schematic of the desired feeding network topology is depicted in Fig. 5. At the center frequency of 250 GHz, the aforementioned antenna elements A and B should be fed with a phase according to their geometrical position to obtain a circular polarization, which will unavoidably result in a suboptimal operation at other frequencies. Starting at poles 1 and 1 , the antenna is designed for a differential feeding or a corresponding differential receiver. From each pole, two branches of ≈320 μm length are separated by a λ/4 transformer enabling both a balancing of the power per branch and a 90 • phase shift necessary for the CP generation. Each branch then splits into two transmission lines of 90 • and 135 • electrical length, resulting in both a phase-correct feeding and the desired frequency selectivity for the antenna elements A and B. Fig. 6 shows the input admittance and impedance levels at the different reference points given in Fig. 5. As it can be seen by comparing Figs. 3(a) and (b) and 6(a), the frequency selectivity results from the impedance transformation, where the high impedance of the antenna at resonance is transformed into a high admittance, which then accepts most of the power when connected in parallel ( Y A + Y B ). The total input impedance Z 1 can be finally matched by choosing a suitable characteristic impedance of the quarter-wavelength transformer at the poles 1 and 1 [see Fig. 6(b)]. It should be emphasized that this feeding network enables a relatively wide Fig. 6. Input admittance and impedance at different reference planes in the feeding network (see Fig. 5). (a) Input admittances of both antennas A and B, which are connected in parallel, yielding a frequency selectivity toward the larger admittance. (b) Total input impedance corresponding to a differential feeding of the antenna. To reduce the coupling to the antenna elements, the feeding network is located on a separate metal layer and connected by vias to the individual feeding points. For the characterization of the fabricated antenna, both poles 1 and 1 are connected to a balun and a single-ended GSG-probing pad. The balun is furthermore fabricated in a back-to-back structure with identical line length, which allows measuring the balun and probing-pad losses.
The surface current density along the antenna is simulated using the full-wave FDTD solver Empire XPU and evaluated at 230 and 250 GHz. Fig. 8 shows the instantaneous current densities on the different metal layers projected into one plane with a dynamic range of 30 dB. The normalization is carried out globally for all phase angles, enabling a comparison of feeding amplitudes between different time instances. To indicate the phase-correct feeding, the field is evaluated at two time instances corresponding to a 90 • phase delay. Whereas the feeding of the dipole arms is well separated and comparable in amplitude for 250 GHz, at 230 GHz, the unintended feeding of the vertical and horizontal dipole arms [see Fig. 8(b)] and an amplitude difference of ≈ 3 dB between the different time instances can be clearly seen, presumably resulting in a more elliptical polarization. This is a direct consequence of the feeding network design, which is optimally matched and phase-correctly fed at 250 GHz due to the λ/4 transformers. At the edges of the operation bandwidth, not only the matching but also the feeding phase are only suboptimal due to the decreased electrical length of the feeding lines (see Fig. 5), and as a consequence, the axial ratio degrades.

IV. CP CHARACTERIZATION AT mm-WAVES
For the antenna radiation pattern measurement, a spherical wafer antenna measurement system [24] is used. To minimize the chip area overhead, the probing pads are located close to the antenna. Consequently, the on-wafer probe is covered with thin-layer graphite absorbers to suppress reflections. Furthermore, only a part of the radiation pattern can be observed due to the shadowing of the probe [25], [26]. The cut planes and the coordinate system utilized for antenna characterization are depicted in Fig. 9(a).
To characterize the polarization, namely differentiate between LHCP and RHCP, phase-correct measurements of ϕand θ -polarization are used to calculate the RHCP and LHCP components according to [27, pp. 150], where the additional (2) 1/2 maintains the norm The phase-correct measurement of mm-waves is a difficult task, and consequently, different methods to characterize the purity of a CP have been presented and discussed in the literature. If a receiving linear polarized antenna is used, the gradual rotation of this antenna can be used to accurately measure the elliptical polarization either by many sampling points [28] or by estimating the ellipse based on few samples [25], reducing the necessity of phase-correct measurements. However, at least four gain-correct measurements need to be carried out compared with two gain-and phase-correct measurements, presumably doubling the necessary measurement time. Alternatively, different reflectors can be used for a time-gated characterization of the input reflection coefficient of the antenna [29], which can ultimately be used to characterize LHCP and RHCP, but the reflectors need to be relatively large to overcome the two-way path loss (∝ r −4 ), rendering this method less suitable for radiation pattern measurement. Furthermore, CP reference antennas, which inherently receive LHCP or RHCP and thus do not require a phase-correct measurement of field components, could be used. Here, the typical waveguide-based solutions only provide a band-limited acceptable axial ratio [30], [31], reducing the applicability to narrow-band characterizations. Finally, simply rotating the receiving waveguide antenna seems a very practical and suitable solution, although it needs to be assured that no phase errors due to displacement [32] or cable bend [33] occur. This also rules out the use of a simple waveguide twist [34] for phase-correct measurements, since the mechanical process of reconnecting the flanges might introduce a receiver displacement in the order of the wavelength. Although the measurement setup utilized has been shown to characterize the radiation pattern with good phase stability, the effects of cable bends and temperature drift should be counteracted by the measurement procedure. Here, the receiving antenna is rotated at each measurement position to measure both polarizations in close succession, without any cable movement and only a neglectable temperature drift. A rotary joint would ideally fit this purpose if it were sufficiently phase-stable while rotated and manufacturable for the WR03 band [35]. Since a geometrical rotation of the receiving antenna by 90 • would be sufficient, two rectangular waveguide flanges are rotated around their longitudinal axis by ±45 • . The approach used here is similar to the approach in [36] except for the iris between the waveguide sections, which drastically increases the mechanical fabrication effort to improve input matching. Since the azimuthal misalignment of the waveguides is similar for both measurements, e.g., +45 • for θ polarization and −45 • for ϕ polarization, both amplitude and phase of the transmission characteristic should be highly comparable in the first place. To ensure the radial alignment during the rotation and axial alignment of the flanges, a cylindrical polylactide (PLA) collet is used. Fig. 10(a) and (b) shows technical drawings of the collet including the cross-sectional cuts highlighting the waveguide orientation, and Fig. 10(c)-(e) shows the photographs of the assembled horn in the rotating collet. The transmission parameter S 21 and the corresponding uncertainty in phase and amplitude resulting from this fixture can be evaluated by measuring a well-known polarization for a number of repetitions. As in the CP measurement later, the measurement data of the single repetitions will be only compared with the data acquired in the next and former measurement to reduce temperature drift. As a reference, a horn antenna polarized in the x-direction (see Fig. 11) will be characterized with the measurement system setup to ϕ = 45 • and θ = 0 • , enabling a highly symmetrical characterization when measuring ϕand θ -polarization. Due to the symmetry, the measured transmission parameters should include the same free-space path loss, a polarization loss due to the misalignment of 45 • for both the cases, and the actual loss of the transition. Accordingly, also the phase measured for both polarizations should be comparable (since mechanical differences could only occur due to axial misalignment of the rotation stage or imperfections in the flanges) and, more important, reproducible with low uncertainty. Fig. 12 shows the measured phase and amplitude difference between ϕand θ -polarization for 70 measurements. Comparing the transmission parameters for subsequent θ and φ measurements, the standard deviation of the resulting phase difference can be calculated and amounts to a maximum σ max = 5.9 • and a mean σ mean = 2.2 • , indicating a high phase-stability associated with the rotation of the receiving antenna. Concluding this section, the presented approach of using a misaligned waveguide to measure two orthogonal polarizations with high phase and amplitude repeatability works well, even for the very high frequencies utilized here.

V. mm-WAVE ANTENNA MEASUREMENT
The characterization of the on-chip antenna by RHCP and LHCP is based on (1), where the θ and φ components at each measurement point are obtained utilizing the rotating waveguide from Section IV. The gain of the antenna under test (AUT) is accessed by comparing the transmission parameter S 21 with a well-known horn antenna and additionally compensating for the losses of the mm-wave probe, as in [37], The probe losses can be accessed by determination of the error terms of the mm-wave probe using on-wafer calibration standards (open, short, and load), and the losses of the balun on-chip are accessed by measuring the balun in a back-toback configuration. The manufactured balun and the losses and phase associated with both baluns back-to-back are depicted in Fig. 13. Here, the balun itself is based on a rat-race coupler in the microstrip technique and uses an on-chip poly resistance for decoupling of the balanced ports. Furthermore, the input reflection coefficient and impedance of the antenna with the balun at the pads can be measured and compared with simulation results, as shown in Fig. 14. The pads consist of two ground pads connected to the on-chip ground (M1) and to the signal pad via a resonant loop to improve the transition from the CPW mode of the probe to the on-chip microstrip line. The corresponding full-wave simulations are carried out with the FDTD solver Empire XPU. Here, the agreement between simulation and measurement is acceptable, since especially the imaginary part of the input impedance is measured to be more inductive. It should be noted that this measurement cannot access the input impedance of the antenna at the feeding points 1 and 1 (see Fig. 7), since a back-to-back measurement does not replace a full calibration. However, if antenna and balun are well matched, the losses within a single balun should be comparable, regardless of it being connected to the antenna or another balun.
For the characterization of the radiation pattern, the receiver is rotated around the AUT on a spherical coordinate system according to Fig. 9(a). Here, two cut planes are measured, and the ϕ = 0 • -cut only covers a range of 0 • ≤ θ ≤ 90 • , since smaller angles usually suffer from the shadowing of the probe. Furthermore, the ϕ = 90 • -cut is limited by −90 • ≤ θ ≤ 50 • to hinder a collision with the microscope [24].  Since the orthogonal components E ϕ and E θ of each measurement position are measured in direct subsequence, longterm phase drift due to temperature variations does not affect the accuracy of the measurement. This is indicated by the very good agreement of measured and simulated RHCP gain, depicted in Fig. 15(a) and (b), where only small undulations of the measurement indicate multipath contributions from reflections inside the measurement domain, most probably at the mm-wave probe [26], [37]. To avoid these contributions, the largest parts of the probe are covered with thin mm-wave absorbers, but the very sensitive probe tips and the metal parts nearby cannot be covered. The cut planes of the LHCP [see Fig. 15(c) and (d)] indicate large deviations between measured and simulated gains. It is supposed that the uncovered part of the probe, which is very close to or even in the near-field of the antenna, might convert parts of the RHCP into LHCP by reflection. This contributes to the measured LH component and might even exceed the LH component radiated by the antenna itself, which is supposedly the reason for the large deviation. Due to the very small distance between antenna and probe tip, the LHCP radiation from the antenna and LHCP radiation reflected by the probe cannot be separated by timegating. To show that the large undesired LHCP is caused by the probe tip, the probe tip and a part of the absorber covering it have been modeled in EMPIRE XPU. The modeling is based on the micrographs of the probe tip at different focal planes in  front and top views, and the results of the FDTD simulation for the LHCP gain at 250 GHz are shown in Fig. 16. Here, good agreement, especially considering the complex probe model [see Fig. 16(c)], between simulation and measurement is achieved, which confirms that the larger LHCP component of the measurement is an artifact of the probe.
Finally, the realized gain in the broadside direction over the frequency indicates a very good agreement between simulated and measured RHCP and LHCP, although a small frequency shift of ≈ 5 GHz is observable for both polarizations [see Fig. 17(a)]. The suppression of the LHCP far below 250 GHz will not be sufficient for most mobile systems, since the resulting polarization is strongly elliptical and thus dependent on the antenna orientation, and thus Tx and Rx antennas require an identical orientation on the final frontend. This can also be recognized by the axial ratio [see Fig. 17(b)], which indicates a dominant elliptical polarization.
Finally, the antenna aperture can be increased to increase the overall gain in the broadside direction, e.g., by a dielectric  lens. The lens utilized here is a hyper-hemispherical lens of 800 μm height and 1000 μm diameter of polytetrafluorethylene (PTFE). Due to the low permittivity of PTFE and the utilized series capacitances in the antenna, only a low frequency-downshift and practically no deviation from the desired operation are expected. Theoretically, larger lenses would enable a larger aperture and a proportionally larger gain, but the probing pads need to remain accessible for the probe tips, limiting the maximum useable lens size. A micrograph of the PTFE lens on-chip and a photograph of the probe with the lens are depicted in Fig. 18. The measured realized gain with lens is depicted in Fig. 19, again in a very good agreement for the dominant RHCP over frequency, and again with unexpected large contributions of the undesired LHCP. It should be noted that the lens can only improve both RHCP and LHCP, and thus, the axial ratio cannot be improved by this measure. Over the whole frequency band between 220 and 260 GHz, the realized RHCP gain is extraordinarily constant given the resonant structure the antenna is based on. Again, the suppression of the LHCP is very weak at the edges of the frequency band in both simulation and measurement, which has already been explained by the feeding network designed for optimal CP operation at 250 GHz.

VI. CONCLUSION
This article demonstrates a design approach for achieving an increased operation bandwidth for on-chip antennas. The main concept combines a feeding and matching network, which ensures frequency selectivity toward the suitable antenna and maintains a circular polarization at the design frequency. The feeding network operates suboptimally with increasing difference between operation and design frequency, resulting mainly in elliptically polarized radiation. Furthermore, the dipole arms fed by the feeding network are geometrically tailored for their application, maximizing the utilization of the available chip area. By introducing a simple rotating collet for changing the polarization of the receiving horn, the CP radiation pattern has been measured and compared with simulation with excellent agreement. For the desired radar application, this antenna needs to be mirrored to achieve a complementary LHCP antenna, which could be used with the presented antenna as a Tx and Rx antenna pair. Since the operation bandwidth of the antenna is strongly limited by the realized elliptical polarization, and not by the antenna mismatch, the feeding network could be optimized toward fulfilling the ideal CP phase shift at the operation frequencies of the individual dipoles, instead of the center frequency, on the cost of the matching and frequency selectivity. Only then, the antennas could be arbitrarily oriented within a Tx-Rx system without significant signal loss.