A) eWLB package architecture

Figure 1(a) shows a schematic diagram for the package architecture that mainly consists of two layers, namely the mold compound and a dielectric layer [Reference Meyer, Ofner, Bradl, Brunnbauer and Hagen9]. The mold compound has a relative permittivity of $\varepsilon _r \approx 3$ and a loss tangent of tan δ ≈ 0.005 at 77 GHz. The dielectric layer has a relative permittivity of $\varepsilon _r \approx 3.6$ and a loss tangent of tan δ ≈ 0.026 measured at 1 GHz. The mold compound has a thickness of 450 µm and the dielectric layer has an overall thickness of about 17 µm. The thickness of the dielectric layer varies based on the location of the chip and the redistribution layer (RDL) as well. In the eWLB packaging concept, the pads of the die in the fan-in area, limited to the size of the die, is mapped to an arbitrary pad pitch in the fan-out area. This redistribution or mapping is realized in the dielectric layer using a thin film copper layer called the RDL shown in Fig. 1(a). Indeed, eWLB packages were fabricated with different sizes such as 8 × 8 mm² and 8 × 12 mm² for our investigation. In addition, a single eWLB package can combine two dies or more that can be interconnected through the RDL. Figure 1(b) shows the MMIC and the antenna system integrated in one eWLB package and hence there is no need for 77 GHz connections outside the package. The antenna system is patterned into the thin film copper layer in the RDL, which can also be used to make a transmission line to interconnect the antenna and the MMIC as indicated in Fig. 1(b). In addition, on top of the PCB carrier and beneath the antenna, there is a copper plate that acts as a shield for the PCB and a reflector for the antenna system. The antenna radiates through the mold compound in the broadside direction.

Fig. 1. SiP concept. (a) An embedded water level ball grid array (eWLB) package and (b) eWLB package and integrating an antenna and a chip.

B) Radiating elements

In eWLB packaging technology, the RDL can be a single or a double layer. Only a single layer RDL is considered in the ongoing design. Utilizing one metalization layer, differential antennas fed with a coplanar stripline (CPS) such as loop antennas, dipole antennas, folded dipoles, Vivaldi antennas, and coplanar patch antennas can be designed. Although, single-ended antennas fed with a coplanar waveguide transmission line (CPW) such as patch antennas and bowtie antennas are attainable, the current designs are limited to differential antennas as the used MMIC has a differential output. Antennas in eWLB package can be backed with a reflector on the PCB and/or directors patterned on top of the mold compound.

In the ongoing investigation of the viability of the eWLB package as an antenna platform, antennas are characterized independently from the chip. In other words, the silicon die is a dummy chip and it is not connected neither to the antenna nor to its DC biasing. It is included to examine its geometrical and electrical characteristics on the antenna performance. Figure 2 depicts the used model where the antenna is fed from an external source. The silicon chip is modeled as a die that has a relative permittivity of $\varepsilon _r=11.9$, a resistivity of ${\rm \varrho }=18\, \Omega \, {\rm cm}$, and an overall dimension of $3.0 \times 3.0 \times 0.4\, {\rm mm}^3 $. In the next section, we focus on two types of antennas that radiate in the broadside direction, namely: the loop antenna and the dipole antenna. Such antennas will radiate in both directions perpendicular to their planes and therefore it is necessary to back each antenna with a reflector on the PCB.

Fig. 2. Passive AiP.

C) Measurement concept

Figure 3 depicts the proposed measurement concept, where an AiP will be soldered on an evaluation board that is interfaced to the measurement equipment. Evaluation boards are designed as an interface between the eWLB package and the measurement device. Since the Si-chip is a dummy chip and the antenna is fed from an external source as depicted in Fig. 2, RF-transitions are developed to transfer the 77 GHz-signal to/from the package (RF-T1) and to/from the measurement equipment (RF-T2) as well. The RF signal is differentially carried in the eWLB package with a CPS transmission line and on the PCB with a conductor-backed coplanar stripline (CB-CPS) transmission line. Both transmission lines are designed with a characteristic impedance of ${\rm Z}_{\rm o}=100\, \Omega $. Figure 4 shows the first transition RF-T1 that interfaces the eWLB package and the PCB. It converts the differential signal carried by a CPS in the package to a differential signal carried by a CB-CPS on the PCB. It utilizes the soldering balls to guide the RF signal. The CPS in the package is optimized to minimize the stray radiation in the end-fire direction while the CB-CPS on the PCB employs a quarter-wavelength transformer for impedance matching. Soldering balls, close to the transition zone, are removed to avoid undesired coupling.

Fig. 3. Measurement concept.

Fig. 4. RFT1 and its scattering parameters.

A row of vias, marked in blue in Fig. 4, is used to ground the landing pads of the balls around the transition. In a complete design of an evaluation board, these vias are extended to ground the antenna reflector mounted on the PCB. Figure 4 shows the simulation results of the developed transition RF-T1 for the differential mode excitation. The transition is well matched over a wide frequency range including the range of interest from 76 to 81 GHz. Quite acceptable performance is observed in this range such as an insertion loss of 0.8 dB at 76.5 GHz. Moreover, 50% of the recorded insertion loss is due the radiation induced by the discontinuity over the transition. Since the transition is symmetric, there is no coupling between the differential mode and the common mode. This is expressed in the simulation result by an isolation better than 40 dB due to the numerical approximation.

The second RF-transition RF-T2 is to interface the evaluation board to the measurement equipment. There are several techniques to develop this transition such as the usage of a balun and a single-ended finline or direct usage of a differential finline [Reference Alhenawy and Schneider6]. In [Reference Alhenawy and Schneider11], our developed differential finline transition has a wideband impedance bandwidth of about 37 GHz and an insertion loss less than 0.5 dB, which allows a precise characterization of the device under test. Finally, a wafer prober station is used for the measurement of the reflection coefficients of the antennas while their radiation patterns are measured in an anechoic chamber.

III. PACKAGE ENVIRONMENT PARAMETRIC STUDY

This section includes the study of the fabrication tolerance on the antenna performance and the optimization of the antenna location within the package. Also, the characterization of the antenna performance over the frequency range of interest is included.

A) Mold compound and soldering balls

The thickness and the size of the mold compound plays a major role in forming the radiation pattern and the efficiency of the antenna. Therefore, the acceptable thickness of the mold is investigated for a fixed package size. Figure 5 shows how the gain of a PLA varies versus the thickness of the mold compound $t_m $ in the broadside direction, i.e., $\theta=0$°, at 76.5 GHz. It also reveals a continuous decrease in the directivity in the broadside direction with almost a constant efficiency. Increasing the thickness of the mold creates a dip in the broadside direction which splits the main lobe into two beams. For example, at a mold thickness $t_m=940\, {\rm \mu m}$, the dip splits the main beam into two beams with maxima at $\theta=\pm 30$°, not shown. The smaller the mold thickness, the better the performance of the antenna, but the minimum thickness is limited by the die height. The fabricated samples have a mold thickness of 450 µm with a tolerance of ±20 µm, which leads to a maximum gain drift of 1 dB. Furthermore, the resonance frequency of the antenna is insensitive to the tolerance of mold compound, results are not shown here.

Fig. 5. Gain versus mold thickness at 76.5 GHz

The soldering balls have a pitch of 500 µm and a conductivity of σ = 7 MS/m. Generally, soldering balls can be utilized for a wide variety of functions such as DC biasing, intermediate frequency (IF) transmission, RF transmission, heat transfer, and mechanical stability of the package. In the current design, two balls are employed in the transition RF-T1 and a grid of balls, distributed on the outer border of the package, are connected through the RDL and also on the PCB level to form a cavity beneath the antenna as shown in Fig. 6. Our designed antennas are based on soldering balls with a height of $h_b $ = 200 µm among other two possible options of 150 and 250 µm. The nominal height value is 200 µm with a soldering tolerance of about ±30 µm. After soldering, the spherical shape of the balls turns into an oval shape with different diameters as depicted in Fig. 6. The effect of the ball height variation on the antenna gain is plotted in Fig. 7 (a). It shows a robust performance with a maximum drift of about 0.5 dB. The soldering tolerance with respect to the diameters of the balls has a negligible effect on the antenna performance. However, changing the height of the balls has remarkable effects on the reflection coefficient of the antenna as shown in Fig. 7(b). It shows that the resonance frequency of the antenna is shifted back and forth by about 3 GHz in response to a tolerance of ±30 µm around the nominal value. Such deviation could limit the usability of the antenna; however, the measurement of the balls' height after soldering for different samples and charges shows an actual tolerance of ±5 µm, which causes an acceptable deviation of about 400 MHz around the desired resonance frequency.

Fig. 6. Soldering ball and evaluation board.

Fig. 7. Effects of the soldering balls' height variation on the antenna performance by 3D simulation. (a) Antenna gain versus soldering balls' height at 76.5 GHz and (b) antenna reflection coefficient versus balls' height.

B) Dielectric and RDL layers

The dielectric layer consists of two layers, namely Diel. I and Diel. II, as shown in Fig. 8(a). The fabrication tolerance of the dielectric layer thicknesses ranges from ±1.0 to ±3.0 µm. In the tolerance analysis, the performance of the PLA is observed versus variation. The PLA is designed at a resonance frequency of 78 GHz. The part of the Diel. II on top of the RDL layer has a fabrication tolerance of ±3 µm which results in a shift in the resonance frequency by about 200 MHz as shown in Fig. 8(a). Figure 8(b) shows that the thickness variation of Diel. I has a negligible effect on the resonance frequency. Figure 8(c) shows a dominant drift in the resonance frequency, in the order of 500 MHz, appears when the thickness of the Diel. II on Diel. I varies by ±1 µm. The thickness of the thin film RDL layer has a fabrication tolerance of ±1.5 µm. On the one hand, decreasing the RDL thickness has a slight effect on the antenna performance, but on the other hand, the increase of the thickness causes a drift in the resonance frequency by about 500 MHz to the left side as shown in Fig. 8(d). The radiation pattern of the PLA is insensitive to the variation of the dielectric and the RDL layer within the tolerance limits. Moreover, the simulation results show that the antenna efficiency is reduced by 8 percentage points due to the loss in this dielectric layer.

Fig. 8. Tolerance analysis for a PLA in an 8 × 8 mm² package using 3D simulation. (a) S ₁₁ versus the thickness of Diel. II on RDL, (b) S ₁₁ versus the thickness of Diel. I, (c) S ₁₁ versus the thickness of Diel. II on Diel. I, and (d) S ₁₁ versus the thickness RDL.

C) Radiation pattern versus frequency

The frequency range of interest for automotive radar sensors is from 76 to 81 GHz. Therefore, the radiation pattern and the effects of the lossy silicon chip on the antenna performance are investigated over this range. Figures 9 and 10 show simulation results for the PLA with and without the dummy Si-chip over the frequency range from 76 to 81 GHz. They show that as the frequency increases, the gain decreases, e.g., the difference between the gain at 76 GHz and at 81 GHz is about 1.5 dB. Figures 9(a) and 9(b) also show a slight change over frequency in the E-plane of the antenna pattern at the package edge, i.e., $\vartheta=90\degr $. Since the Si-chip lies in the H-plane of the antenna, its effects are more pronounced in shaping the pattern in the H-plane compared to the E-plane. Although the Si-chip slightly reduces the antenna efficiency, it sharpens the main beam across its H-plane.

Fig. 9. E-plane pattern versus frequency and effects of dummy Si-chip in 8 × 8 mm² package size. (a, b) Radiation pattern along the E-plane (i.e., yz-plane).

Fig. 10. H-plane pattern versus frequency and effects of dummy Si-chip in 8 × 8 mm² package size. (a, b) Radiation pattern along the H-plane (i.e., xz-plane).

D) Si-chip Location in an eWLB Package

The location of the Si-chip in the package is crucial for the performance of an AiP. Therefore, a parametric study for the distance between the antenna and the dummy chip is carried out.

Figure 11(a) shows the studied configuration. Figure 11(b) shows that changing the distance i between the antenna and the Si-chip leads to a shift in the resonance frequency of about 500 MHz. This shift in the resonance frequency can be compensated in a redesign. The effects of changing the distance between the antenna and the chip is more pronounced on the antenna radiation pattern and its efficiency η as shown in Figs. 12(a) and 12(b). The closer the antenna is to the chip, the lower the gain and the efficiency will be as well. The efficiency is decreased as more energy is coupled to the low resistivity silicon substrate. There are two factors that contribute to the gain reduction: firstly, the energy lost in the chip body, i.e., decreased efficiency. Secondly, the geometrical effect of the chip on the radiation pattern as the main beam becomes more flat in the boresight direction. For instance, when the distance between the chip and the antenna is i = 2.5 mm, the antenna has a gain and an efficiency of 9.3 dBi and 80% at 76.5 GHz, respectively. In comparison, at a distance i = 0.5 mm, the gain and the efficiency are recorded to be 5.5 dBi and 45% at 76.5 GHz, respectively. If the chip were active, the antenna could be placed farther than 2.5 mm and up to 3.5 mm. However, in the current configuration, the distance i is determined to be 2.0 mm to minimize the undesired coupling between the antenna and both the Si-chip and the RF-transition RF-T1.

Fig. 11. I. Parametric study of the antenna location in an 8 × 8 mm² package using 3D simulation. (a) Simulated eWLB package and (b) Reflection coefficient.

Fig. 12. II. Parametric study of the antenna location in an 8 × 8 mm² package using 3D simulation. (a) H-plane radiation pattern at 76.5 GHz and (b) E-plane radiation pattern at 76.5 GHz.

A) Printed loop antenna (PLA)

A loop antenna has two different modes of radiation: it can radiate in either the end-fire or in the broadside direction. A loop antenna is classified as electrically small when its perimeter C is less than one-tenth of the operating wavelength λ. Such loops radiate in the end-fire direction while electrically large loops, where C ≥ λ, radiate in the broadside direction. Although small loops are inefficient radiators, they could be used for inter-package communication. In electrically large loops, the electrical current is no longer uniformly distributed on the perimeter and therefore its main lobe departs the plane of the antenna toward the broadside direction. Indeed, the size of the loop that corresponds to a maximum gain in the broadside direction and its input impedance should be determined in order to design a matching network. Therefore, we investigated the characteristics of the loop by changing its perimeter in an 8 × 8 mm² eWLB package and observing the gain in both the broadside, i.e., ϑ = 0° and the end-fire direction at ϑ = 90°. The planar circular loop has a fixed strip width of a = 187 µm and a variable radius r as shown in Fig. 13(b). The feeding source is applied directly between points 1 and 2. Figure 13(a) shows the simulated gain and directivity, in both end-fire and broadside directions, versus the perimeter of loop normalized to the effective wavelength $C/\lambda _{eff} $ within the package environment. It shows that the maximum gain in the broadside direction is achieved with a loop perimeter of $1.8\lambda _{eff} $. The gain is independent of the strip width as long as it is considerably smaller than the operating wavelength; e.g., the ratio of the strip width to the loop perimeter at $1.8\lambda _{eff} $ is about 4%. Figure 13(b) plots the real and the imaginary parts of impedance between point 1 and point 2. It also shows that the first and the only antiresonance occurs at a loop perimeter of $0.7\lambda _{eff} $. Therefore, the appropriate practical point of operation is the second resonance, i.e., the first series resonance, where the input impedance has a reasonable value for matching and the maximum gain is achieved. The gain as well as the input impedance can be determined by approximating the non-uniform current distribution over the loop by a Fourier series as in a conventional wire loop antenna [Reference Pocklington12]. In [Reference Alhenawy and Schneider6], a single PLA antenna with an inner perimeter of 2.85 mm is developed and measured in an 8 × 8 mm² eWLB package as shown in Fig. 14. The reference plane of the measurement and the simulation is the package edge. The measurement along the H-plane shows strong ripples compared to simulation. It is due to the parasitic radiation from the waveguide port, the CPS transmission line, and the PCB-to-package transition that are not included in the simulation. In addition, in practical applications, these transitions will be excluded and hence their effects. The distance between the antenna and the Si-chip is optimized to minimize the effect of the chip on the antenna efficiency.

Fig. 13. Planar loop antenna in an 8 × 8 mm² eWLB package by 3D simulation at 76.5 GHz. (a) Directivity and gain versus loop perimeter and (b) Input impedance versus loop perimeter.

Fig. 14. Radiation pattern of PLA at 76 GHz [6].

B) Printed dipole antenna (PDA)

The inherent nature of a differential feed suggests the use of a dipole antenna as a good candidate for an AiP concept. A single PDA, supported by a reflector at a distance of 200 µm, was developed in an 8 × 8 mm² eWLB package. Figure 15(a). shows the measured and the simulated radiation pattern of a conventional PDA along the E-plane and the H-plane [Reference Alhenawy and Schneider6]. The measurements of different samples show a deformation in the H-plane pattern at positive ϑ-angles for the PDA. One way to better integrate a PDA in a package and to minimize the coupling in an antenna array configuration is to change the orientation of the dipole wings. Figure 15(b) shows the simulated and measured E-plane and H-plane radiation pattern of a PDA with 90-tilted wings. A lower gain is recorded in the broadside direction with a more pronounced dip in the H-plane. The dip is a result of the propagation through the mold compound. Increasing the mold thickness from 401 to 940 µm results in a dip level difference of 22 dB as shown in Fig. 5 for a PLA.

Fig. 15. Simulation and measurement results of PDA antennas in 8 × 8 mm² eWLB package. (a) PDA at 76 GHz [6] and (b) PDA with 90°-tillt 76 GHz.

In a multichannel system, more than one antenna per package is needed. Therefore, we developed four PDAs in one package as shown in Fig. 16(a). This is a generic configuration that can be used in a wide range of applications such as a 4-channel radar system where the four channels are simultaneously powered on for transmission but receiving from each channel individually. In Fig. 16(a), the antennas are uniformly distant by 2.5 mm in an 8 × 12 mm² eWLB package. The geometrical structures of the antennas are identical. Also, a dummy Si-chip is centered along the width of the package. Figure 16(b) shows the simulation results of the reflection coefficient for each antenna and the coupling between them as well. It shows that all four antennas have almost the same resonance frequency. The drift in the resonance frequency is due to the different electrical length of each antenna, which stems from its position in the package and its interaction with the chip and the other antennas. The isolation between the different channels is always better than 18 dB. The isolation can be further improved by using a PLA or a tilted PDA. Figure 17 shows the radiation pattern when all elements are simultaneously fed with identical amplitudes and identical phases. The gain and the efficiency of the array antenna are 12.0 dBi and 79%, respectively. Figures 18(a) and 18(b) show the radiation pattern for each antenna while the other antennas are switched off, i.e., their ports are matched. It is observed that the H-planes of A 1 and A 4 are identical while their E-planes are mirrored due to their locations. Similarly, this behavior is observed for the inner two PDAs. Their H-planes are identical while their E-planes are mirrored along the xz-plane. The usage of the antenna array configuration and the addition of directors on top of the mold compound can improve the radiation characteristics.

Fig. 16. Antenna array in 8 × 12 mm² eWLB package. (a) Four PDAs in an eWLB package and (b) Refelection coefficients and coupling.

Fig. 17. Radiation pattern of four PDAs.

Fig. 18. Antenna array in an 8 × 12 mm² eWLB package. (a) Radiation pattern of PDA 1 and PDA 4 and (b) radiation pattern of PDA 2 and PDA 3.

C) Gain enhancement

1) Dielectric lens

Using the antenna in package, an improved gain can be obtained with the aid of a dielectric lens that collimates the radiated waves. We used a dielectric lens with a diameter of 55 mm, mounted on top of the eWLB package at a distance d = 25 mm. Figure 19 shows the measured radiation pattern of a PDA antenna with and without the use of a dielectric lens. The usage of the lens increases the gain to 26.5 dBi and forms an HPBW of 4.5°. A tolerance in the distance d between the lens and the antenna by ±1.5 mm leads to a gain diminution less than 0.5 dB. The choice of a proper lens enables the realization of different gains that suit the target application.

Fig. 19. PAD with a dielectric lens at a distance d = 25 mm. (a) E-plane (yz-plane) at 76 GHz and (b) H-plane (xz-plane) at 76 GHz.

2) Stacked directive elements

Increasing the effective aperture of an antenna increases its directivity and gain. This can be achieved by adding more elements in the plane of the antenna or in a vertical fashion. Similar to a Yagi-Uda antenna, directors are patterned on the top of the mold compound of an 8 × 8 mm² eWLB package as shown in Fig. 20(a) The distance between the fed PDA and the director, i.e., the mold thickness, is optimized together with the length and the width of the director to increase the gain of the antenna. The director has a length of $0.45\lambda _e $, where $\lambda _e $ is the effective wavelength and a width of about $0.13\lambda _e $. Hence, the length of the director is 28% smaller than the directly fed PDA. The usage of the director increases the gain by 1.2 dB compared to a PDA in package without directors as shown in Fig. 20(b). Furthermore, to enhance the radiation characteristics in the H-plane of the antenna, a second parasitic element is added along that plane. The antenna has a simulated gain and efficiency of about 9.9 dBi and 71%, respectively.

Fig. 20. An 8×8 mm² eWLB package with parasitic directors. (a) Top view of an eWLB package and (b) radiation pattern.