Toward a flexible and adaptive wireless hub by embedding power amplifier thinned silicon chip and antenna in a polymer foil

Abstract A flexible and adaptive energy-efficient high-speed wireless hub is developed in polymer foil as a Hybrid System-in-Foil (HySiF) using Chip-Film Patch (CFP) technology. In this matter, the SiGe BiCMOS silicon chips (2.39 × 1.65 mm2) are thinned down to 45 μm and are embedded face-up inside a two-polymer CFP carrier. The active pads of the embedded silicon chips inside foil are extended to the surface of the foil to interconnect to the antenna on the foil. The integrated hybrid system has a signal transmission at 5–6 GHz frequency band. The overall thickness of the system is below 100 μm and its bendability is down to 4 mm radius of curvature. The designed and fabricated PA silicon chips operate at 50 mA with a 1.5 V supply voltage. Therefore, in addition to the high lateral thermal resistance of the thinned chip, self-heating loop inside polymer due to the low thermal conductivity of the embedding polymer raises the system temperature. Consequently, the thermal behavior and RF performance of the PA chip under different conditions are investigated. Moreover, the antenna with the required carrier frequency is simulated, fabricated, and measured on top of the polymer foil as a stand-alone system in the flexible CFP.


Introduction
Radio-frequency identification (RFID) applications are countless. However, the combination of specifications such as long-range, high communication data-rates, lower power consumption, low cost, and flexibility of the device is restricted [1,2]. In this matter, the flexible and adaptive energy-efficient high-speed wireless hub is a research topic with the criterion of a "long-range, environment agnostic wireless reading range" means that the device must be small, thin, and mechanically flexible in order to conform to arbitrary surfaces and to accommodate to mechanical cycling.
One major problem affecting the performance of an RFID tag is the variation of feed-point impedance when the antenna is attached on to media with different dielectric permittivity. It is proposed to switch between different feed points of the antenna to reduce feed impedance mismatch. As a first step, a single feed point antenna and a silicon chip are fabricated in CFP foil. The idea is to extend the number of antenna feed points to fabricate adaptable antenna for various substrates and corresponding permittivities. The alternative solution is to fabricate a single antenna with a single feed point for each specific environment, which makes a large area and costly system [3]. Additionally, the carrier frequency should be in the GHz range. Therefore, the high-performance and the low-parasitic transistor is needed, which can be fabricated using silicon technology.
The high-power thin silicon chips are integrated in the CFP technology in combination with the high-frequency antenna fabricated on top of the polymer foil. The conventional two-polymer face-up CFP technology [4] should be optimized to meet the specifications. The schematic illustration of a HySiF is shown in Fig. 1.
In Section "Two-polymer CFP technology", the CFP process flow for embedding and interconnecting ultra-thin chips in the two-polymer CFP is described. In addition, we briefly elaborate on the process challenge of misalignment and rotation, which occurs in the chip placement step and illustrate the adaptive layout as a solution to this issue. The thermal characterization of the embedded power amplifier (PA) chip in the polymer foil, the corresponding self-heating, and the cooling methods are described in Section "Thermal characterization". The electrical characterization of the integrated PA chip inside polymer foil and the antenna on the CFP polymer foil are summarized in Section "Electrical characterization". Finally, conclusions are presented in Section "Conclusion".

Two-polymer CFP technology
The two-polymer CFP technology is based on embedding ultrathin silicon chips, face-up, inside two types of polymer layers. In this technology, a combination of Benzocyclobutene (BCB 4000 series) and polymer (PI) with the trade name durimide are employed. BCB as an embedding polymer is a negative photoresist with high planarization capability [8]. In addition, BCB polymer with its low loss tangent (0.008-0.002 from 1 MHz to 10 GHz) and low dielectric constant (ε r = 2.55) is a suitable substrate to fabricate a large area antenna [5]. BCB acts as a barrier between the embedded lossy silicon chip and connected antenna. However, BCB cannot be used as a single polymer in the CFP technology due to its stiff nature after curing. Therefore, BCB in combination with durimide as a reinforcing polymer has been exploited. Durimide provides the mechanical flexibility of the system-in-foil (SiF), reinforces the layer stack, and shifts the brittle BCB near the neutral line of stress. The two-polymer CFP technology is compatible with CMOS processing which enables a fine-pitch interconnectivity using photo-lithography and IC-production facilities (CMOS backend process) [6]. The schematic cross-section of the two-polymer ultra-thin chip embedding in the CFP technology is illustrated in Fig. 2.
As shown in Fig. 2, the silicon chip is embedded face-up in the pre-defined cavities in the polymer foil substrate using BCB as permanent glue. The primary purpose of this process is to bring the device side and polymer foil to the same level in order to perform further via opening, metallization, and metal patterning. In the face-up CFP technology, the precise and identical thickness of the embedded chips inside polymer foil plays an important role in avoiding extra unwanted topographies, ease of planarization, and further metallization. The schematic illustration of the high-frequency two-polymer CFP technique is depicted in Fig. 3.
During the fabrication process (see in Fig. 3), a Si 3 N 4 layer (1.5 μm) is deposited as a stress release layer [7] on the temporary silicon carrier. Next, the wafer is coated with an adhesion lowering layer. Afterward, BCB and durimide are deposited and cured consecutively. After defining a cavity in the lithographic step and plasma reactive ion etching (RIE), the PA chip is embedded face-up inside the cavity using BCB as a permanent adhesive layer. Afterwards, the BCB layer is deposited and covers the surface of the embedded chip. The thickness of the BCB layer should be thick enough to planarize the surface of the chip and to make the smooth transition from the edges of the embedded chip to the peripheral polymer. In addition, the thickness of the top polymer is symmetrical to the bottom polymer to keep the embedded chip in the neutral line of stress. Then, the micro-vias are opened using metal as a hard mask and RIE with the gas mixture (CF 4 /O 2 -80:20%). Once the vias are opened, a metal interconnection layer (AlSiCu) is sputtered and structured. At the end of the process, the integrated system-in-foil is released from the temporary silicon carrier.
In this work, the high-speed silicon chips are fabricated using 0.25 μm BiCMOS (IHP process SGB25 V) technology for energy-efficient RF circuits operating at 5-6 GHz frequency band. To embed these chips in the polymer foil, the single chip back-thinning and DC characteristic measurements of CMOS and HBTs, before and after back-thinning, are performed and the low difference is observed. From earlier works, an initial chip thickness down to 20 μm was employed. In this work, the minimum chip thickness down to 45 μm ± 10% is achieved. This difference in comparison to 20 μm is due to internal stress on the chip, which was caused by the metallization, dielectric layers, and the chip layout. Therefore, the process flow including the thickness of polymer layers and via opening technology is adapted to the accomplished chip thickness [9].
During chips placement step in the pre-defined cavities on the polymer foil, the limited accuracy of the chip placement tool and an unwanted movement of the glued chip on top of the foil in the curing step lead to an offset [10]. The undesirable misalignment and rotation error for each chip, especially for multiple chip integration and interconnection, have detrimental consequences for the metal interconnects, thus calling for a revised layout routing of metal interconnects with far larger process bias. This, however, contradicts the CFP concept. The already published adaptive layout technique using a maskless laser writer is proposed as a solution [10].
With the help of the adaptive layout technology, the initial interconnect layout between the embedded chips in the polymer   (1) wafer pre-surface treatment, (2) spin coating the base polymer, curing, and cavity opening, (3) face-up PA chip embedding using a permanent glue, (4) top polymer coating, (5) via opening using lithography, (6) metallization, patterning, and antenna fabrication, and finally (7) releasing system-in-foil from the substrate [16].
foil is adopted based on the calculated rotation and misalignment of the embedded chips in the polymer foil using direct laser writer lithography tool [11,17].

Thermal characterization
Thermal management of ultra-thin silicon chips with a total thickness around 45 μm in comparison to the bulk silicon chips with a thickness of around 400 μm is challenging. With reducing the chip thickness, high lateral thermal resistance is implied on the chip's surface in comparison to the bulk silicon chip [12]. The large temperature gradient on the surface of the chip leads to the thermal mismatch and reliability problems [13].
In addition, the maximum surface temperature of a flexible polyimide substrate is higher because of a poor thermal conductivity of the polyimide substrate which causes further self-heating on the chip [9]. Here, the thermal behavior of the silicon chip including three power amplifier (PA) units is investigated. The chip size is 2.39 × 1.65 mm 2 and consists of three differential power amplifiers (DPA) as a thermal source with a maximum 72 mW power and four on-chip temperature-sensing diodes near the thermal sources.

Temperature-sensing elements
The temperature distribution on the chip's surface is investigated using on-chip thermal sensors. Four bipolar temperature-sensing diodes are designed and fabricated close to the power amplifier units on the chip to evaluate the temperature distribution at different locations on the chip's surface (see in Fig. 4). It is to be noted that, in the case of transistors, the absolute temperature can be potentially measured after appropriate calibration of the elements. In the measurement step, to avoid self-heating of the device caused by the temperature sensor, the current in the transistor is kept at a minimum value (10 μA).
The calibration curve for the junction voltage as a function of temperature is approximated by equation 1 and is defined by the equation 2 and equation 3 for the silicon chips before and after back thinning (≃ 45 μm), respectively [14].
V B is the forward voltage of the diode, V B0 is the sensor's voltage at 0°C, and β is the temperature coefficient of the sensor.

Measurement results
The    In all cases, the chips are measured on the thermal chuck for the V-T calibration curve. The thermal chuck acts as a thermal heat spreader for the above device under test (DUT). Therefore, prior to the temperature measurements, each DUT is isolated from the thermal chuck with a low thermal conductive foil. The temperature measurement is done for four on-chip diodes (D 1 − D 4 ) while the PA near second diode (D 2 ) is powered up (T 2 ) and after reaching a stable temperature is switched off (T 1 ) (see in Fig. 4).
The final measurement results are shown in Fig. 5. After chip back thinning, D 1 and D 2 temperature diodes near the thermal source experience higher temperature in comparison to the thick silicon chip due to the larger lateral resistance. In addition, temperature gradient all over the surface for the thick chip is less than the thin chip because of higher vertical thermal resistance. However, the increase in temperature due to the self-heating phenomena is minimized using a heat spreader underneath the embedded thin chip.

Cooling methods
To increase the long-time reliability of the embedded power chip inside polymer foil, beside power optimization and thermal management at the circuit level, the thermal management at the technology level is considered. Here, two concepts at the technology level are proposed; (I) substructuring the thin chip to introduce a physical thermal barrier and (II) adding heat spreader structures underneath the embedded chip.

Substructuring the thin chip
The substructuring process is performed using a test chip including daisy chain structures. The process flow for substructuring the silicon chip after embedding in the polymer foil is illustrated in Fig. 6. As shown in Fig. 6, the test chip includes four islands of daisy chains. The size and thickness of the ultra-thin test chip is 4.6 × 4.6 mm 2 and 20 μm, respectively.
The SEM cross-section of the structured test chip is shown in Fig. 7. The concept of the substructuring thin chip inside the foil is proven using test chip with a single metallization layer. However, the process flow becomes more complex depending on the number of metallizations, layout, size, and thickness of the ultra-thin chip. The substructuring can enhance the bendability of the system in comparison with a heat spreader.

Heat spreader
The heat spreading mechanism is performed for temperature uniformization below PA chip. In addition, the heat spreader underneath the embedded chip inside the polymer foil minimizes the self-heating effect caused by the low thermal conductivity of the     9. Demonstration of the embedded PA chip with the backside heat spreader in the CFP polymer foil. The integrated system-in-foil is released from the temporary silicon carrier and is bent over a cylinder with approximately 4 mm radius of curvature [16].

International Journal of Microwave and Wireless Technologies
polymer. The fabricated heat spreader is chosen based on, (I) available technology for CMOS compatibility of the process flow, (II) the required flexibility of the final system,(III) size and thickness of the embedded chip, (IV) effect of the heat spreader on the above fabricated antenna, and finally (V) the overall cost of the process. The process flow to embed the PA chip with the backside heat spreader inside CFP using 1.5 μm AlSiCu is illustrated in Fig. 8. The heat spreader is twice the size of the embedded chip.
Firstly, a 10 μm polymer layer is spin coated and followed by sputtering 1.5 μm AlSiCu layer thickness. The heat spreader is then defined using lithographic step and metal etching. Then, a 5 μm polymer is spin coated and a cavity is defined by lithography and polymer plasma RIE. The thinned PA chip is placed and fixed in the cavity using 1 μm BCB (3000 series) as a permanent adhesive layer. After curing the BCB, the chip is covered with polymer and vias are opened using a hard mask and plasma etching. The vias are metalized and simultaneously an electrical fan-out is provided. After final metallization, the polymer on the backside of the heat spreader can be removed by plasma etching the backside of the released foil from the carrier. At the end of the process, the embedded chip inside polymer is released (see in Fig. 9) and further electrical measurements are performed. Despite the presence of the heat spreader, the bendability of the system is proven.

Electrical characterization
As described above, the thin PA chip is embedded inside the foil and connected to the flexible antenna on foil, showing mechanical flexibility (see in Fig. 10). RF performance of the amplifier and the antenna in the polymer foil are investigated separately.

3-path PA chip in the CFP
High-speed and cost-driven 95 GHz-fmax 0.25 μm-SiGe BiCMOS HBT technology is chosen, to keep the power consumption low within the required frequency band and avoid technology constraints. The simplified schematic view of the DPA is illustrated in Fig. 11. The fully integrated isolated NMOS switches with the topology of series/shunt configuration are intended to connect the PA array to a multi-feed antenna. All three amplifiers are identical to each other and a single path of a PA-array is investigated. RF performance of the amplifier has been evaluated with respect to the thickness of the substrate and the backside material in the previous study [14]. In this work, the embedded PA chip inside the CFP is measured and plotted in Fig. 12. The center frequency and gain are changing depending on the substrate thickness and the material below the IC. However, a minor S-parameters degradation after embedding the PA in the CFP foil (with/without heat spreader) is observed because of the broadband matching at the output.
The measured center frequency on the thin silicon with backside metallization is shifted about 300 MHz toward higher frequencies compared to the thick silicon due to the image eddy currents within the conducting material. The inductors and transformer on thin Si chip (thickness of 45 μm) with the backside conducting material have slightly lower inductance due to the ground image mirror currents within the ground plane and accordingly degrading quality factor. In this study, conducting materials are a vacuum chuck and a heat spreader. Image of passive devices within the conducting material contributes a negative mutual inductance because the current flow in the return path is opposite direction. On the other hand, there is not much difference on the dielectric material (PCB) and foil, since they have a much higher resistance than the silicon substrate.
This measurement result matches to the simulated frequency shift of 200 MHz with a thin silicon substrate on the perfect electrical conductor (PEC). To improve the accuracy of the post-layout simulation and investigate the impact of the substrate thickness, all passive components except capacitors are extracted in one view and the image of the passive devices in the ground plane is considered by using a full-wave electromagnetic simulator (ADS-Momentum EM simulator).
In addition, according to the large signal measurements on the vacuum chuck, the PA output power at 5.5 GHz is 9 dBm before thinning and 8.1 dBm after thinning for an input power of − 1.3 dBm [14]. At the same input power and frequency, the embedded PA chip inside the CFP has 8.9 dBm output power without a heat spreader and 8.2 dBm output power with a heat spreader. The simulated output power at the 1 dB compression point is 10.8 dBm with a PAE of 15% at 5.5 GHz. The RF performance of the PA is measured by using a VNA with 0 dBm maximum output power. Hence, PA saturation is prevented.

Antenna characterization in the CFP
The fabricated antenna with the carrier frequency at 6 GHz on top of the polymer foil is measured as a stand-alone system. Therefore, each antenna is measured after releasing, straight,  and on bent status with different radii of curvature (R). The measurements are performed using a vector network analyzer (VNA) (see in Fig. 13).

Antenna measurement result:
In this part, the antenna measurement setup in straight (see in Fig. 13(a)) and bent profile (see in Fig. 13(b)) is described. In order to decrease the measurement errors caused by metallic chuck in the probe station, a Rohacell chuck (ε r ≈ 1) with a thickness of 20 mm is produced and the antenna is placed on it. As shown in Fig. 13(a), the antenna placed parallel to the Rohacell cylinder and the end of the feed lines is bent close to the GSGSG pads. In this way, the antenna functionality in the straight profile is measured. In addition, to measure the antenna in bent profile, the prototype is bent perpendicular to the antenna's main axis (see in Fig. 13(b)).
The simulated and measured result of the antenna input reflection magnitude is depicted in Fig. 14. There is a good agreement between measurement and simulation result of the antenna and the small shift in measured resonance frequency could be due to uncertainty in material parameters (thickness and electrical properties) of the substrate and metal layers used in the simulation. In Fig. 14, the simulated reflection coefficient of the antenna for straight and bent (R = 16 mm) profile versus the measured reflection coefficient of the antenna is shown. The bent antenna has measured the reflection coefficient of − 15 dB at the resonance frequency.
The thickness of the antenna is currently limited to 2 μm AlSiCu to follow CFP CMOS compatible fabrication technology. Unlike the feed impedance measurements with a wafer probe, the measurements of antenna radiation characteristics in straight and bent state require reliable connectors on the polymer foil. Then, balun circuit and cables can be connected for measurements in an anechoic chamber. For that means, a glued connection on thin AlSiCu was not successful. As a possible alternative, the use of solderable Silver for fabrication of the antenna on the polymer foil is suggested.

Conclusion
The thin silicon chip can be embedded and interconnected with a large-area antenna by using the high-frequency compatible twopolymer Chip-Film Patch technology. With the help of adaptive layout technique, proper design, and fabrication of heat spreader underneath the embedded chip, a PA chip with 75 mW power can be integrated in a CFP foil component while ensuring bendability and adequate power dissipation. The hybrid system including PA chip and on-foil antenna can operate at 6 GHz. As a part of the future work, a multiple feed antenna can be proposed.With a clever choice of the feed position on a multiple feed dipole antenna, the worst-case impedance mismatch can be minimized while the antenna is placed on the various environments with diverse permittivities. Additionally, the performance of the antenna is going to be compared based on the use of different fabricating materials (AlSiCu versus Silver). Ultimately, the complete system including multiple feed antenna and transmitter in the flexible CFP will be demonstrated.