# Heterogeneously integrated flexible microwave amplifiers on a cellulose nanofibril substrate

Jun 19, 2020

### Strategies of fMIC fabrication

Figure 1 presents schematic illustrations of two manufacturing processes that can be used for fabricating bendable MICs: flexible MMIC and fMIC. In the chip thinning down approach21,31 as shown in Fig. 1a, MMICs were built on III-V semiconductor epitaxial wafers that were thinned down to flexible form. The total number of MMICs that can be fabricated out of one epitaxial wafer using this approach is about (slightly smaller than) the area ratio between a wafer area (Awafer) and a circuit area (Acircuit), which is similar to the turnout of previously demonstrated flexible electronics approaches32,33,34,35. For silicon-integrated circuits with a high density of transistors and for Si-substrate MMICs22,23, chip thin-down represents the most convenient approach to producing bendable chips and favorable cost advantages36,37,38, particularly considering the very low cost of Si substrates. However, for (majority of) MMIC chips built on epitaxial substrates (e.g., GaAs, InP, Sapphire, SiC, etc.), as the majority of the chip area is occupied by area-consuming passive components (e.g., inductors, capacitors, and transmission lines, etc.), the chip thin-down approach is no longer cost-effective due to the high cost of the epitaxial substrates.

The cost-effective approach to fabricate fMICs demonstrated herein, as shown in Fig. 1b, is to heterogeneously integrate membrane transistors39,40 transferred from a dense transistor-packed source epitaxial substrate, which have maximized the use of expensive source substrate, with passive components on an inexpensive, microwave-compatible destination substrate, thus avoiding the occupation of the source substrate by large-area passives. It is noted that the fabrication procedures and, therefore, the processing cost of both the active transistors and passive components are identical in the two approaches illustrated in Fig. 1. However, the substantial increase in the number of active transistors fabricated on the source substrate, now determined by the area ratio between Awafer and the area of the transistors (Atransistors), leads to the same number increase in fabricated fMIC (~13 times of Awafer/Acircuit as in Fig. 1a, detailed in Supplementary Note 2). It is also noted that the cost of the substrates of fMIC shown in Fig. 1b (CNF) is nearly negligible in comparison with that used in Fig. 1a. The only extra fabrication procedure used in Fig. 1b was transfer printing. As the transfer-printing techniques and the relevant tools have already been adopted for other mass-production applications41,42 and proven to be a low-cost addition in the applications, they also represent minimal cost additions in handling active transistors and passives upon adopting the present demonstration for massive production. As a result, the fMIC approach shown in Fig. 1b suggests substantial cost advantages.

Figure 2a, b show photographs of a flexible microwave amplifier circuit on a transparent CNF substrate fabricated using the approach illustrated in Fig. 1b. The fMIC amplifier consists of AlGaN/GaN HEMT, which is readily available from large-size AlGaN/GaN-on-Si wafer43 and can be released in membrane form13. Figure 2c shows the circuit diagram of the flexible microwave amplifier. The designed operating frequency range of the amplifier is 5–6 GHz, which can address many commercial applications44. Due to the relatively long wavelength of electromagnetic (EM) wave within the operating frequency range, lumped spiral inductors and metal–insulator–metal (MIM) capacitors, instead of lengthy (at this frequency) transmission lines, were used to build compact passive impedance matching networks. Figure 2d shows the layer structure view and integration sequence of the amplifier circuit fabrication.

### Flexible AlGaN/GaN HEMT

Before constructing a flexible microwave amplifier on CNF substrates, we first fabricated and characterized membrane AlGaN/GaN HEMTs on the same CNF substrates. The purpose of the fabrication and characterizations is to extract the scattering (S-) parameters of the HEMT under the condition that the HEMT is implemented in the microwave amplifier. Supplementary Figs. 1 and 2 schematically illustrate the detailed fabrication process of membrane HEMTs on CNF substrates. The rationale for the choice of gate length and gate width of the HEMTs is explained in Supplementary Note 3. An array of HEMTs with a gate length of 300 nm (see Supplementary Figs. 3 and 4a) and a gate width of 90 μm was fabricated on an epitaxial AlGaN/GaN-on-Si wafer using conventional HEMT fabrication techniques. After completion of fabrications on the original rigid Si substrate, the HEMT array was diced into discrete HEMTs, each having an area of about 500 µm × 500 µm in size (see Supplementary Fig. 4b). It is noted that the HEMT size/area is still mainly dominated by the coplanar ground–signal–ground (G-S-G) probing pads needed for on-chip radio frequency (RF) characterizations. After dicing, an individual HEMT (or multiple HEMTs as needed) was picked up with a polydimethylsiloxane (PDMS) stamp after coating the HEMT with SiO2 by plasma-enhanced chemical vapor deposition (PECVD). The Si-handling substrate of the HEMT was then fully etched away using XeF2 etching (Supplementary Fig. 4c), leaving a membrane HEMT of about 3.5 μm-thick finished. The membrane HEMT on top of the PDMS stamp was then transfer printed to CNF substrates pre-coated with adhesion layers with the assistance of an MJB-3 contact aligner (Supplementary Fig. 5), of which the operation principle is the same as the commercial tools41,42. The images of HEMT sitting on CNF substrates are shown in Supplementary Fig. 6.

Direct current (DC) characteristics and S-parameters of the HEMT sitting on the original Si substrate and on CNF substrates were characterized, and the results are shown in Supplementary Figs. 7 and 8, respectively. Both drain current and peak transconductance were reduced after HEMT transfer to CNF substrates due to the increased self-heating effects in the device, which resulted from the low thermal conductivity of CNF substrate (~1 Wm−1 K−1)45 compared with that of Si substrate (~156 Wm−1 K−1)46. The results shown here are consistent with other reports13,47,48. Improving the thermal conductivity of flexible substrates49,50 is expected to improve the DC characteristics. Considering the low thermal conductivity of CNF and to avoid excessive heating, the HEMT was biased at a drain voltage (VDS) of only 10 V, despite HEMTs’ ability to sustain a much higher voltage (30 V)51.

The values of current gain (|H21|2), Mason’s unilateral power gain (U), and maximum available gain (MAG)/maximum stable gain (MSG) of the HEMTs were extracted from the measured S-parameters. The extracted values were plotted in Fig. 3d–f as a function of measurement frequency. Without de-embedding parasitic effects, the highest cut-off frequencies (fT) of ~34.1, ~37.9, and ~37.0 GHz, and the highest maximum oscillation frequencies (fmax) of ~69.5, ~87.8, and ~75.0 GHz were obtained for HEMTs on Si, SU8/CNF, and polyimide (PI)/SU8/CNF substrates, respectively (Supplementary Table 2)52. The gate bias (VGS) values where the highest fT and fmax were obtained were −1, −1.5, and −1 V, respectively, on the three different substrates. The RF performances of the HEMT on the CNF substrates are comparable to reported similar HEMTs on other substrates (see Supplementary Note 5 and Supplementary Table 3).

According to Eq. (1)53, it was expected that the HEMT on CNF substrate would have lower fT than that on Si substrate due to the reduced peak transconductance (gm) on CNF. However, both fT and fmax values on CNF substrates are higher than that on Si substrate. The reasons for the unexpected RF behavior were analyzed based on three-dimentional EM simulations using CST Microwave Studio® (Supplementary Fig. 9a, b). The CNF substrates have lower dielectric constant (εr) of ~2.654 than Si (~8.9). The simulations show that the parasitic capacitance values (CPAR in Eq. (1)) caused by the metal probing pads of the HEMT on CNF substrates were significantly reduced in comparison with that on Si substrates, which is consistent with other reports13, due to the lower εr of CNF (see Supplementary Fig. 9c–e and Supplementary Table 4). As no de-embedding of the parasitics were performed for the HEMTs, the parasitic effects were included in the measured (raw) HEMT S-parameters. The reduced parasitic effects of the HEMTs on CNF substrate have over-compensated for the adverse effects caused by the lower thermal conductivity of CNF substrate than Si on fT. As a result, higher fT values were obtained for the HEMTs on CNF than on Si. The increase in fmax on CNF is primarily due to the increase in fT and partly to the reduced gate-to-drain capacitance (CGD), which is also a consequence of the lower dielectric constant of CNF than Si as indicated in (2)53. As the HEMT on SU8/CNF does not have extra PI or oxide layers in comparison with PI/SU8/CNF as shown in Fig. 3b, c, the former has less extrinsic parasitic capacitances between electrodes and thus higher fmax values than the latter.

$$f_{mathrm{{T}}} = g_{mathrm{{m}}}/left[ {2pi left( {C_{mathrm{{G}}} + C_{mathrm{{PAR}}}} right)} right]$$

(1)

$$f_{{mathrm{max}}} approx sqrt {f_{mathrm{{T}}}{mathrm{/}}left( {{mathrm{8}}pi R_{mathrm{{G}}}C_{{mathrm{GD}}}} right)}$$

(2)

The HEMT on PI/SU8/CNF, which was used in amplifier circuit, was further tested under different mechanical bending conditions with bending radii (R) of 38.5 mm and 28.5 mm to study the influence of mechanical bending on the microwave performance of flexible HEMT (Fig. 3g). It was observed that the MAG of HEMT decreased very slightly in the interested frequency range as the bending radius became smaller (see the inset of Fig. 3g). The slight degradation of the small-signal power gain was due to the effects of external strain on the two-dimensional electron gas in the AlGaN/GaN heterostructure through piezoelectric effect and surface states, as revealed previously55,56. Correspondingly, the drain current57 and transconductance58 of the HEMT were slightly reduced, which in turn degrades its microwave performance.

### Design and fabrication of flexible microwave amplifier

The flexible microwave amplifier was designed using the characterized HEMT on PI/SU8/CNF, based on the circuit diagram shown in Fig. 2c. The passive components, two spiral inductors and one capacitor, consists of two layers of metal paths10. A layer of spin-cast PI isolates the two layers of metal path, which not only possesses good mechanical properties but also introduces low parasitic capacitances in the spiral inductors due to its low dielectric constant (~3.5). The optimized input impedance matching network contains a 2.5-turn spiral inductor connected in series with the input port. The optimized output impedance contains a low-pass inductor-capacitor network to suppress high-frequency harmonics, which is composed of a 1.5-turn spiral inductor connected in series with the output port and a MIM capacitor of 30 μm × 40 µm in size shunting the output of HEMT. The MIM capacitor employs a 140 nm-thick SiO2 layer deposited by electron-beam evaporation as its inter-metal dielectric layer. Figure 3h shows the circuit layout for circuit simulations, which combines S-parameters of the passive impedance matching networks obtained from the EM simulations and the measured S-parameters of the HEMT on PI/SU8/CNF. The simulation results shown in Fig. 3i indicate that the flexible microwave amplifier can achieve a peak small-signal gain of 6.69 dB at 5.3 GHz.

The amplifier fabrication process flow is shown in Fig. 4a. An extended schematic illustration of the process flow is shown in Supplementary Fig. 10. The amplifier fabrication began with picking up a rigid HEMT from a diced array of HEMTs (Supplementary Fig. 4b). The Si substrate of the HEMT was removed with XeF2 etching, leaving a membrane-form HEMT. The HEMT was then transfer printed to a temporary Si-handling substrate coated with a spin-cast PI layer and polymethylmethacrylate (PMMA) layer. The PMMA layer acts as a sacrificial layer and the PI as an adhesive layer. The input and output impedance matching networks were then fabricated on the Si temporary substrate. To minimize microwave propagation loss in the passive components, a thick metal stack (Ti/Cu/Ti/Au: 10/900/10/100 nm) was used for both layers of the metal paths of inductors, capacitor, and interconnects. A stack of Ti/SiO2/Ti/Au (10/140/10/100 nm) was deposited in one step to form the MIM dielectric layer and the top electrode layer of the MIM capacitors for a precise control of the capacitance value of the MIM capacitor8. Via holes through an insulation layer (PI: ~1.5 μm thick) were formed by reactive ion etching (RIE) and an interconnected metal was formed to connect the two layers of the metal paths, completing the fabrication of the amplifier on the temporary Si substrate. The PMMA sacrificial layer was then dissolved in hot acetone to release the amplifier (Si substrate can be reused) and the amplifier was transfer printed to a SU8-coasted CNF substrate to complete the entire fabrication process. As most of the above fabrication procedures were carried out on a rigid Si substrate, all the conventional MMIC fabrication techniques can be directly adopted here.

The uniqueness of the fMIC fabrication is that area-consuming passive components in the amplifier were integrated on an inexpensive CNF substrate and not on an epitaxial substrate (Fig. 1b). The fabrication method is drastically different from thinning down III–V-based rigid MMIC chip to obtain bendable MMIC21,31 (Fig. 1a). The fMIC method makes much greater use of epitaxial substrate materials and thus significantly reduces material cost.

As manufacturing microwave transistors requires high-cost fabrication processes such as electron-beam lithography, ohmic contact metal, and high-temperature annealing, the increase in active transistors’ density in the proposed approach leads to substantial reduction of processing cost per transistors as well, which is also the most important reason to employ larger semiconductor wafers in the semiconductor industry59. After transfer-printing membrane HEMT on temporary Si substrate, we fabricated the rest of the passive components using conventional low-cost photolithography systems with large feature sizes (e.g., at level of microns). The combined yet sensible use of high-cost processes for fabricating high density of transistors on epitaxial semiconductor wafer and low-cost processes for fabricating large-area passive components on temporary substrate at different fMIC fabrication stages enable a cheaper and faster fabrication process than conventional flexible MMIC approaches (see Supplementary Note 2 for detailed discussion). The conventional fabrication process and the reduced use of epitaxial materials together render the fabrication cost-effective. Only a very small piece of native epitaxial substrate was used for the HEMT fabrication, which has maximized the use of the epitaxial material.

Figure 4b–d show optical images of the amplifier before dissolving the PMMA sacrificial layer (on temporary Si substrate), after being picked up on PDMS stamp and sitting on CNF substrate during on-wafer microwave measurements, respectively. Figure 4e shows a microscopic image of the HEMT integrated in the amplifier. The flexible amplifier has a compact size of 1.4 mm × 2.4 mm, which is comparable to a rigid AlGaN/GaN HEMT-based MMIC chip60. As shown in Fig. 4d, 92.6% of the fMIC amplifier area was occupied by passive components and only the remaining 7.4% of the area was used by the HEMT, for which the G-S-G pads of the HEMT still account for the majority HEMT area (see Supplementary Fig. 4).

### Small-signal characterizations of flexible microwave amplifier

The small-signal microwave performances of the amplifier were characterized on flat (Fig. 5a) and bent surfaces with bending radii of R = 38.5 mm (Fig. 5d) and R = 28.5 mm (Fig. 5g). Figure 5b, e, h present the measured small-signal gain as a function of frequency under flat and bent radii of 38.5 mm and 28.5 mm, respectively. To measure the small-signal gain values, the HEMT in the amplifier was biased at 10 V for VDS and the VGS was swept from −0.8 V to 0.2 V with a 0.1 V step. As the VGS was increased, the small-signal gain values increased until peak-gain values were reached at 5.51, 5.37, and 5.29 dB at VGS of 0.2 V under the three mechanical conditions. It was noted that the peak-gain frequency under the three conditions also shifted to lower frequencies (5.62, 5.60, and 5.58 GHz, respectively) as the bias was increased. To clearly illustrate the trend of the peak-gain shift, Fig. 5c, f, i plot the extracted peak-gain values and the peak-gain frequencies from Fig. 5b, e, h, respectively, as a function of VGS. As the VGS increased from −0.8 V to 0.2 V, the monitored drain current of the HEMT in the fMIC also increased (Supplementary Fig. 11) and so did the transconductance of the HEMT (Supplementary Fig. 7b), which led to the increase in the RF gain values of the HEMT (Eq. (1)) and the small-signal power gain of the amplifier. Shifting of the peak-gain frequency was also observed. The maximum small-signal gain curves obtained at VGS of 0.2 V under flat and bent conditions were compared in Fig. 5j. Overall, the small-signal gain changes were minimal due to bending, showing the mechanical bendability and robustness of the amplifier against bending. Supplementary Table 5 summarizes simulated and measured gain values of the amplifier at the interested frequencies and peak-gain frequency shifting values.

The slight degradation of the small-signal power gain of the amplifier and the observed shifting of the peak-gain frequencies when the amplifier was bent were a result of the degraded RF performance of the HEMT in the amplifier (see Fig. 3g), as well as by the slight impedance mismatch between the input/output matching networks and HEMT in the amplifier circuit. The input and output impedance values of the HEMT could be slightly changed as its bias was changed. Previous studies6 have also shown that inductance value of spiral inductors and capacitance value of the MIM capacitor can be altered under certain degrees of mechanical deformation. These changes led to the shifting of impedance matching in the amplifier and thus gain degradation, which was qualitatively verified by simulations (see Supplementary Note 6, Supplementary Table 6, and Supplementary Figs. 12 and 13 for details). According to the reported trend55,56, it is expected that as the bending radius of the amplifier is further increased, the performance of the HEMT and therefore the amplifier performance will continue to degrade until a mechanical failure occurs when the limit of the tensile strain of GaN is reached.

### Large-signal characterizations of flexible microwave amplifier

Although the amplifier circuit was designed using the small-signal S-parameters of the HEMT, large-signal microwave characteristics of the amplifier circuit were characterized to estimate the RF power handling capabilities of the fMIC for continuous-wave mode operation. Supplementary Fig. 14 shows the measurement setup for large-signal microwave characterization of the amplifier. The microwave loss in cables, bias tee, and connectors in the measurement setup was calibrated. A microwave signal generator was used to sweep the input power level at 5.5 GHz, the designed center frequency. The initial bias of the HEMT in the amplifier was fixed at VGS of 0.2 V and VDS of 10 V. Figure 5k (flat) plots the output microwave power (POUT), the large-signal power gain (Gain), and the power-added efficiency (PAE) of the amplifier as a function of input microwave power (PIN). Figure 5k shows that the amplifier can achieve a peak PAE of 4.7% at PIN of 5.76 dBm without bending. The corresponding POUT at the peak PAE was 9.75 dBm (9.44 mW) and the concurrent gain was 3.99 dB with a gain compression of 2.25 dB from the small-signal gain of 6.24 dB at 5.5 GHz. The POUT at −1 dB gain compression point (POUT | −1 dB) was 7.76 dBm (5.97 mW) and POUT | −3 dB was 10.3 dBm (i.e., 10.64 ± 0.19 mW). To our knowledge, this is the first report of a flexible microwave amplifier that can output microwave power larger than 10 mW beyond 5 GHz. Supplementary Table 7 compares the various parameters of the amplifier with that of a SiGe BiCMOS power amplifier on thinned Si substrate23, illustrating the various performance advantages of the HEMT fMIC amplifier.

The large-signal microwave performance of the amplifier under mechanical bending (R = 38.5 mm, Fig. 5d) was further characterized and the results are shown in Fig. 5l. A peak PAE of 4.04% was measured along with POUT of 9.35 dBm (8.61 mW) and gain of 3.59 dB. The POUT | −1 dB and POUT | −3 dB were 7.98 dBm (6.28 mW) and 9.85 dBm (9.66 mW), respectively. The slightly larger POUT | −1 dB under bending condition compared with the flat condition was due to a larger PIN. These results indicated that the amplifier is mechanically reliable for microwave power amplification.

Furthermore, based on the measured parameters shown above, the temperature distribution of the HEMT in the amplifier was analyzed using simulations. The detailed simulations are shown in Supplementary Fig. 15. The highest temperature of the HEMT experienced during large-signal operation was estimated to be around 184 °C. According to the reported thermal reliability of GaN HEMTs (Supplementary Table 1), the amplifier is also considered thermally reliable.

### Disposal of flexible microwave amplifier

Although rigid-chip- and the rigid-board-based MICs are not easily disposed of and prone to electronic waste, flexible CNF substrate-based electronics have been shown to be biodegradable10,61, a significant benefit over other flexible host substrates such as polyethylene terephthalate and PI. In the demonstrated amplifier, a simple calculation showed that the CNF substrate accounted for ~92% weight of the amplifier. As an alternative approach to biodegradation via fungi10 to the disposal of the CNF-based fMIC circuit, disposition via incineration was attempted. As shown in Fig. 6, the CNF-based amplifier burst into flames instantly when ignited by a candle and turned into ashes after ~7 s, which is similar to that of other wood-based devices62,63. Although a detailed analysis was unable to be performed during incineration, it is expected that the major composition of the gas produced was carbon dioxide considering the highest possible temperature (>1100 K) of a candle flame64. The easy disposal of the amplifier indicates the green feature of fMIC approach, which could be environmentally significant if such fMIC is massively produced in the future.