Fig. 2(a) shows that the proposed AiP comprises an IC, a bonding structure, and an LTCC antenna. To operate as the primary signal source while also verifying performance, the IC is composed of a VCO generating a signal of 100 GHz, designed by implementing a 28-nm CMOS process. The widely used cross-coupled topology was adopted to meet the oscillation condition for generating a stable signal. The 100-GHz signal was extracted from a fundamental oscillation frequency of 50 GHz using a push–push technique, which extracted the output signal from the center tap of the resonant tank inductor. As is customary in general CMOS implementations, Fig. 2(a) shows that the output configuration of the chip includes a ground–signal–ground (GSG) pad for probing measurements and AiP integration. Fig. 2(b) shows the target-matching impedance range of the LTCC antenna based on the bonding inductance and Z_VCO, including the capacitance effect of the pad. It is observed that the 10 dB return-loss matching circle normalized to 50 Ω is different from the impedance trajectory of the antenna. These must be practically matched, considering the packaging effect.

The AiP consists of a 100 GHz antenna bearing the structure of a well-proven 60 GHz LTCC parasitic stacked patch antenna [27] (Fig. 3). The parasitic patches introduced an additional resonance that significantly extended the bandwidth. The signal was directly fed to the driven patch by a straight via, given that signal loss during the via-to-line conversion process was not negligible and the ground vias overlapped at 150 μm intervals to minimize field leakage. The parameters of the antenna design are listed in Table 1.

As shown in Fig. 4(a), the initial matching design exhibited dual impedance bandwidths of 85.9–107.4 GHz (22%) and 114.9–124 GHz (7.6%). The corresponding radiation beam patterns at each resonance frequency are depicted in Fig. 4(b). It is observed that at 100 GHz, the beam maintains a proper direction without noticeable distortion, but the maximum gain is relatively low—6.7 dBi. Furthermore, the surface current distribution illustrated in Fig. 4(c) indicates that electric coupling is dominant at 100 GHz. In such a scenario, unequal coupling strengths between the driven patch and adjacent parasitic patches result in suboptimal power transfer, which limits radiation efficiency and gain, even though the beam shape is preserved. At 120 GHz, as shown in Fig. 4(d), magnetic coupling dominates, and the beam splits symmetrically along the y-axis, leading to pattern degradation and a further reduction in gain, which declines to 5.8 dBi. To resolve these issues, the spacing between the driven and parasitic patches was carefully optimized to balance the coupling intensities. This coupling optimization enabled the achievement of improved gain without beam distortion. The final design achieved a wide impedance bandwidth of 89.2–111.2 GHz (22%) and an enhanced gain of 6.9 dBi at 100 GHz, which are suitable for accurate performance verification. Furthermore, the impedance of the optimized antenna was calculated to be 40.3–j·16 Ω at 100 GHz, which is within the target impedance range.

To couple the VCO to the antenna, a minimum bonding length of 0.05λ₀ or less had to be maintained to ensure a wide bandwidth and to minimize RF signal loss. For ultra-compact and high-density integration, the same height of the cavity with a 200 μm-thick VCO was manufactured and stationed at the bottom of the antenna as an IC attachment (Fig. 5). A part of the ground via adjacent to the signal via was removed to minimize the distance of the cavity-to-signal via. When a cavity structure is added, a minimum module thickness of 600 μm must be ensured for stability. The antenna module size was set to 3 mm × 16 mm, which was increased in a direction independent of the polarization of the antenna for insignificant performance variations. The bottom side of the cavity was plated for the ground connection of the chip, bearing a size of 1.5 mm × 1.5 mm, with a margin to reduce process complexity. The dimensions of the AiP design are listed in Table 2.

Several solutions have been proposed to address the impedance mismatch problem caused by the local bonding area, such as the self-matching method and the reactive impedance surface (RIS) method. In self-matching, a bonding length of 1.5 mm (λ_g/2) would increase the package volume. However, in the LTCC process, metals can be stacked in units of 50 μm. Therefore, up to three layers of RIS were inserted between the directly fed patch in Layer 2 and the ground in Layer 6, without taking recourse to any additional stacking. A RIS consists of a periodic arrangement of square patches with a minimum pattern width of 60 μm and a gap of 50 μm. The parasitic inductances of the wedge bondwires can be compensated for by conjugating the antenna input impedance using a multilayer RIS structure. Fig. 6 depicts a comparison of the antenna performance achieved using the self-matching technique and different numbers of RIS structure layers. Compared to self-matching, increasing the number of RIS layers from 1 to 3 improves the impedance matching bandwidth by 4.8%, 7.5%, 8.8%, and 17.7%, respectively. The simulation results indicate that the three RIS layers improved the impedance bandwidth by a factor of 3.67. Furthermore, the gain beam pattern results indicated that the bandwidth enhancement resulting from the addition of RIS structures was not due to signal loss, with the gain

As shown in Fig. 7(a), the improved impedance bandwidth resulted in a small performance variation in the cavity structure fabrication for the VCO chip insertion and wedge-bonding connection errors. Notably, it is difficult to guarantee precise wedge-bonding dimensions and an even flat cavity using the LTCC process. Therefore, the height of the chip protrusion, wedge-bonding height, and length variations were accounted for in the simulation by varying H_diff, H_wedge, and L_wedge. Considering the 10 dB return loss obtained through optimized bonding, along with the maximum impedance bandwidth, acceptable process tolerances for L_wedge, H_diff, and H_wedge were considered to be ±100 μm, ±200 μm, and ±125 μm, respectively. Notably, the tolerance ranges being sufficiently larger than the process error by several tens of μm confirms that the proposed AiP can be fabricated practically. Fig. 7(b) shows the variation in the realized gain over the acceptable error range. The results indicate a maximum gain of 6.18 dBi and a gain variation of 0.72 dB within the tolerance range at 100 GHz. Furthermore, in Fig. 7(c), the optimized wedge bonding for the AiP is represented as a modulated π-model equivalent circuit [26, 27]. A comparison of the electromagnetic simulation results with those obtained using the equivalent circuit composed of a self-inductance of 77 pH and parasitics shows that the two impedance trajectories on the Smith chart are almost identical. The transmission coefficient and phase shift of wedge bonding, extracted using the equivalent circuit model, are shown in Fig. 7(d). The loss is 1.54 dB at 100 GHz, as the change in loss is 0.86 dB over the 90–110 GHz frequency range.

The unit cells considered in this study comprise stacked patches that are mutually coupled to a grid or a cross-dipole slot, which contributes to improving the polarization isolation of antennas. At sub-wavelengths, the capacitance component of the patch and the inductance component of the grid were observed to be dominant. Notably, the cross aperture of the dipole acts as a parallel LC resonator. Therefore, the unit cell exhibit a bandpass response, with the order of the corresponding filter increasing with the number of metal layers.

Notably, the phase shift value of the unit cell changed rapidly with frequency, thus securing a broader phase modulation range. Fig. 8(a) depicts a unit cell structure designed using the conventional PCB process. Notably, a narrow phase range and high insertion loss of the PCB-based unit cell result from the insertion of bondply material, with the minimum pattern size being 100 μm for both polarizations [30]. In this study, each unit cell has the same thickness of 300 μm (0.1λ₀) and a size of 600 μm × 600 μm (0.2λ₀). The unit cell structure consists of a grid or cross-dipole aperture between one or two stacked patches (Fig. 8(b)–8(d)). Seven metal layers are required to design the lens, with the proposed unit cells arranged based on the thickness of each layer. The S₂₁¹² parameter was neglected because each proposed LTCC unit cell exhibited a high polarization characteristic of 50 dB or more. Furthermore, unit cells with independent phase-shift values for each polarization were required for different beam-steering features with respect to antenna polarization. Therefore, the patch, grid, and cross-dipole slot constituting the unit cell structure had independent lengths along the x- and y-axes. A total of 144 unit cells were employed to achieve a phase-tunable range of 360° in units of 30° for the two linear polarizations. After 72 unit cells were defined, the remainder were automatically determined since the unit cell characteristics were symmetrical along the two orthogonal axes.

As shown in Fig. 9(a) and 9(b), the phase-tunable range when using only three metal unit cells is 180° for each polarization, whereas the unit cells with five metal layers achieve a range of 360° for both types of polarization. Notably, the figures depict the phase-tunable range for only one type of polarization. Table 3 summarizes the design parameters of the unit cells shown in Fig. 9. Furthermore, Fig. 9 shows that when the x-axis polarized wave (X-pol) is incident, it is uniformly distributed over a 360° range in units of 30°, as indicated by the dotted line, whereas the solid line shows that the phase distribution is confined to a specific value for the y-axis polarized incident wave (Y-pol). The inserftion loss based on the frequency of each unit cell is shown in Fig. 9(c) and 9(d). Although unit cells satisfying an insertion loss of less than 3 dB were selected because of high signal attenuation loss in the W-band, most cells exhibited insertion losses less than 1.5 dB at 100 GHz. Notably, when multiple unit cells have the same phase-shift value, a unit cell with less insertion loss is adopted during the final selection process. Moreover, if unit cells having the same 360° phase-tunable range consider only one polarization scenario, a sufficiently low insertion loss condition will be realized.

The first step in the lens design process was to capture the phase of the fields radiating from the AiP. The effect of the AiP radiation field due to the LTCC jig and the top metal of the baseboard was also accounted for in Fig. 10(a). Fig. 10(b) shows the tapering, spill-over, and feed efficiency of the lens at each focal length. As shown in Fig. 10(c), lens placement at short focal lengths, where the amplitude change of the captured field is discrete, should be avoided to achieve a unit cell zone. Therefore, in the proposed metasurface AiP, the lens was placed at a focal length of 10 mm, given its compact design and high efficiency. A total of 784 unit cells were used to configure the LTCC metasurface, with 28 unit cells—0.6 mm in size— arranged along each axis. A final lens f/D value of 0.59 was selected for zone formation, thereby achieving a compact system.

Fig. 11 illustrates the lens design process using the selected unit cells. Given that the captured phase distribution was symmetric to the y-axis, the lens design sequence for half the area is shown in Fig. 11(a). The captured phase was compensated for using a specific constant value to convert the spherical wave into a plane wave. In addition, a specific phase value was determined to satisfy the scenario in which unit cells with low insertion loss exhibited the greatest level of dispersion in the final stage.

The antenna phased array theory involves a phase offset process in which a uniform phase difference must be applied to each array of unit cells located at regular intervals along the y-axis to steer the beam. When the desired beam steering angle is determined, the offset value can be easily calculated from the array factor (Fig. 11(b)). Subsequently, the unit cells corresponding to the phase values at each position can be matched. As shown in Fig. 11(c), if unit cells corresponding to the phase of each position are accurately arranged, the performance of the lens degrades [31]. However, this problem can be solved by zone formation, which refers to the grouping of adjacent unit cells with similar phase values. Another phase distribution (Fig. 11(d)) can be obtained by repeating the same process for other types of polarization. For convenience, the unit cell distribution for the two polarizations are combined into one diagram in Fig. 11(e).

Fig. 12 shows the structural view and photograph of the bottom side of the LTCC antenna, observed under a microscope, depicting its wedge bonding with the VCO. Using the LTCC process, the cavity was machined such that the average distance from the signal through the edge to the cavity, as indicated by the cavity−via parameter, was 57.4 μm. In addition, plating for the electrical connection of the chip was achieved without having to account for rigorous limitation conditions. Since the cavity area was sufficiently wider than the chip area, it bonded to the bottom of the cavity, providing a ground connection. For DC bias bonding, a rectangular pad was fabricated, accounting for the wedge tail. The pad was electrically connected to the circular pad located at the bottom right through a bias line inside the LTCC module, which was then solder ball-bonded to the baseboard and directly attached to the power supply. Fig. 13 shows the three fabricated lenses with steering angles of (0, 0), (30, 30), and (40, −20) for X-pol and Y-pol. The thickness of each lens module for the seven metal layers was 300 μm, while the minimum printed pattern width and gap were 60 and 50 μm, respectively. Fig. 14 depicts the solder-ball bonding between the AiP and the baseboard, with the voltage supplied through a DC connection. Notably, the baseboard was fabricated from a 3 mm × 3 mm FR4 substrate. The metasurface lens was placed at a focal length through the LTCC jig, which aided in reducing the sidelobes caused by baseboard interconnection.

To verify the performance of the LTCC antenna before VCO attachment, the reflection coefficient of the W-band was measured using a FormFactor i110 150 μm pitch GSG probe. A transmission line with a minimum length of 50 μm was attached to the bottom side of the fabricated antenna, surrounded by ground metal spaced 50 μm apart to reduce signal loss. The S₁₁ of the antenna was verified by attaching the flipped antenna onto Styrofoam, which had a dielectric constant identical to that of air and a thickness of 5 mm to satisfy the far-field condition. Return loss was measured through single-port measurement using the W-band extender module of a vector network analyzer. Fig. 15 plots the simulated and measured reflection coefficients, with the measured impedance bandwidth being 23 GHz (83–106 GHz). Notably, the transmitting AiP operated the VCO by applying a DC voltage to generate a signal, which was subsequently radiated through the LTCC antenna using bonding. Measurements were conducted at 100 GHz by tuning the frequency to account for the VCO design conditions. Maintaining a distance of 6 cm from the receiver, a 10–7025 Aerowave WR10 horn antenna was employed as the receiving antenna. The absolute power of the downconverted signal was measured using a Keysight E4448A spectrum analyzer through a Keysight 11970 W harmonic mixer (Fig. 16). The AiP gain amplitude was normalized to a peak gain of 9.2 dBi by calculating the free-space loss, down-conversion loss by the mixer, and cable loss in the W-band. Signals in the 90–110 GHz band generated through the Keysight E8275D signal generator were delivered to the Formfactor i110 100 μm pitch GSG probe, electrically connected via the CPW line, which is nearly lossless. The power was measured with the same mixer and spectrum analyzer used in the AiP setup. Various loss factors, such as those of the mixer, cable, and probe values, over the 90–110 GHz band were calibrated using a power meter to determine the absolute power at the harmonic mixer and signal generator. Fig. 17 presents the measured results of the AiP beam pattern at 100 GHz, obtained through a ±90° rotation using an automatically controlled turntable. The measured peak gain and sidelobe level were 8.86 dBi and 3.24 dB, respectively, indicating good agreement with the simulation results.

Two of the three types of polarization-dependent metasurface lenses were designed to have beam steering angles of ±0° and ±30° for each type of polarization, where the sign indicates the corresponding beam steering direction for X-pol and Y-pol. To verify that the steering angles would differ for each polarization type, we manufactured and measured a lens that bent the peak beam at +40° for X-pol and -25° for Y-pol. Lens gain calculation was easily performed by considering the difference between the received power values in the spectrum analyzer with and without lenses using the measurement setup illustrated in Fig. 16. The results showed that the beamwidth decreased as the gain increased. Notably, owing to the output limit of the VCO, the measurements were performed only at an angle around the peak gain.

The simulation and measurement results for the three metasurface types at 100 GHz are presented in Fig. 18, indicating good agreement. For the lens that enhanced the gain in the boresight direction for both types of polarization, the simulation and measurement gains for X-pol were 18.5 and 17.8 dBi, respectively, while the simulation and measurement gains for Y-pol were 18.9 and 17.7 dBi, respectively. For the lens designed to steer the beam at 30° for both types of polarization, the simulation and measurement gains were correspondingly 17.0 and 16.2 dBi for the X-pol incident wave, and 17.1 and 16.6 dBi for the Y-pol incident wave, respectively. Similarly, the lens designed to steer the beam by +40° and -25° for X-pol and Y-pol operated as intended. The simulation and measurement gains were 16.3 and 16.0 dBi for the X-pol incident wave, and 17.2 and 16.5 dBi for the other types of polarization. Fig. 19 shows the boresight gain of the metasurface AiP with regard to the frequency. Notably, the frequency range of the measurement results was limited by the oscillation frequency range of the fabricated VCO. Fig. 19 shows that the X-pol and Y-pol 3-dB gain bandwidths are 98–105 GHz and 98–104 GHz, respectively, while the maximum aperture efficiencies are 15.2% and 15.1%, respectively.

The typical processes optimized for lens fabrication are PCB and LTCC. Table 4 presents a performance comparison of the proposed lens and state-of-the-art metasurface lenses that operate in W- and D-bands [21–24]. Prior studies on metasurface lenses for the W- and D-bands only considered combinations with standard horn antennas, without accounting for packaging using RF modules. In contrast, this paper proposes a metasurface lens capable of independently steering a beam in dual polarization, while also being integrated with AiP through the same process employed for beam steering. Lens is composed of seven metal layers with a thickness of 300 μm and a phase modulation range of 360°, while also achieving the smallest f/D value of 0.59. Compared to the PCB lens capable of beam steering at an operating frequency of 145 GHz, more than the current metal stack is possible, demonstrating that a sufficient improvement in gain can be achieved for the source antenna using LTCC. Furthermore, compared to the beam-steerable PCB lens operating at 145 GHz [22], the LTCC process employed in the present study can be used to design a small unit cell with a via-less structure using more than twice the number of metal layers. In contrast, although the PCB lens offers the advantages of a short focal length and beam steering, it is limited to one linear polarization (LP). Meanwhile, another beam steering approach proposed in a previous study combined multiple source antennas with a Huygens metasurface lens and then fed the antenna at an arbitrary location [25]. This approach is more efficient than a phased array antenna in a beam steering system, but it requires multiple feed lines. Furthermore, a beam steering angle of ±12° is possible when the feed position is moved 6 mm from the center of the lens. Most lenses reported in previous studies have been designed based on a standard horn antenna in the W- and D-bands, whereas the PCB metasurface operating at 130 GHz was designed based on AiP, achieving a simulation gain of 8.4 dBi [26]. Notably, the unit cell selection tasks considered in this study to ensure independent phase shift values for scenarios with dual LP modes are more stringent than those associated with previously studied unit cell designs, which only serve to increase the boresight gain or steer the beam, with gain enhancements for a single LP. Moreover, since a high insertion loss is inevitable owing to the operating frequency, the proposed unit cell can be selected to satisfy an insertion loss of less than 3 dB, as desired in most dual LP scenarios.