INTRODUCTION
Wireless communication has become an essential part of our daily lives and revolutionized the way people live and work1⇓⇓-4. It allows us to com- municate with each other regardless of time and distance. As illustrated in Fig. 1, wireless communication has evolved significantly, with the first generation (1G) system providing only low-quality voice calls and the fifth generation (5G) system offering a global internet that connects people and devices, enabling smart homes, cloud computing, autonomous vehi- cles, drones, robots and artificial intelligence. Each generation introduces new services and higher data rates, necessitating new technologies and more electromagnetic spectrum (or bandwidth) to support data commu- nication. In order to maintain backward compatibility with older genera- tions, higher frequencies and more frequency bands of the spectrum have been pursued by international standardization organizations (e.g., 3GPP5, IEEE6), with an emphasizing on improving spectrum efficiency. This has led to increased complexity and difficulty in designing radio frequency (RF) systems, making RF filters a crucial component in RF front-end de- sign. RF filters are electronic devices that allow specific frequency com- ponents of a signal to pass through with low loss while greatly suppressing other frequency components, enhancing the system’s anti-interference ca- pability and signal-to-noise ratio for better quality communications.
Fig. 1 The evolution from 1G to 5G. |
Most filter technologies are based on resonating elements 7⇓⇓⇓-11, such as electromagnetic or acoustic resonators. Filter size is usually proportional to the wavelength of the corresponding physical wave that propagates in the resonator cavity. The advantage of acoustic resonators is that their wavelength (μm level at GHz) is several orders of magnitude smaller than the electromagnetic wavelength (cm level at GHz in air) for the same op- erating frequency, allowing for the miniaturization of acoustic filters into small chips with an area of less than 1 mm2. The most common acoustic filter technology is the surface acoustic wave (SAW) technology which is originated from the invention of interdigital transducers12 (IDT) by R. M. White at the University of California, Berkeley. Before SAW devices were used for direct RF filtering, they played an important role in intermediate frequency (IF) filtering and signal processing in radar, television (TV) and frequency control applications13,14. As wireless communication evolves from 1G to 2G and 3G, the main frequency bands adopted in the standards range from several hundred MHz to 2 GHz. Simultaneously, mainstream photolithography in the early 2000s allowed for fine resolution of a few μm for the IDT electrodes. As a result, SAW resonators achieved a decent quality factor (Q>800) and electromechanical coupling coefficient (k2 > 10%) at RF frequencies. Additionally, due to the relatively simple fabri- cation process and low cost, SAW filters and duplexers have been widely utilized in today’s cellphones, especially for 2G and 3G frequency bands, for example, GSM (900 MHz), PCS (1900 MHz), Band 1 (2100 MHz), Band 5 (850 MHz), etc.
The development of the 4th generation (4G) wireless communication, which is also known as Long Term Evolution (LTE), is primarily driven by the need to support Internet Protocol (IP) based services in a mobile device. Although the voice services remain the fundamental function of a mobile phone, there is an ever-increasing demand for higher data rates for web browsing, video streaming and file transfer, which pushes the peak data rates of mobile systems from Kbit/s for 2G to Mbit/s for 3G and get- ting to Gbit/s for 4G (Fig. 1). The Shannon-Hartley theorem states that higher electromagnetic bandwidth is required by higher data rates to sup- port the information transmission in the presence of noise. Therefore, in addition to the 1 GHz bands that were extensively used in 2G and 3G, the frequency bands (electromagnetic spectrum) from 2 to 3 GHz naturally became the choice for 4G. The 4G LTE standard defined many new bands, such as Band 2 (1.9 GHz), Band 7 (2.6 GHz), Band 40 (2.3 GHz) and Band 41 (2.6 GHz). Consequently, the 2 to 3 GHz range quickly became overcrowded, which was exacerbated by the fact that the unlicensed Wi-Fi standard also operates in the 2.4 GHz band. To isolate the different bands and minimize interference, high Q filtering technology became the bottle- neck for 4G phones. At the time, the SAW technology could not work well above 2 GHz, and as a result, the bulk acoustic wave (BAW) technology was developed to deliver high Q (> 2000) at high frequencies. The BAW technology relies on thin film deposition of piezoelectric aluminum nitride (AlN) on a silicon (Si) substrate. The bulk acoustic wave travels and res- onates in the vertical direction so that the thin film thickness determines the resonant frequency. There are two ways of confining acoustic energy within the thin film piezoelectric layer: one is to use surface microma- chining or microelectromechanical systems (MEMS) process to suspend the AlN plate at a certain gap from the Si substrate, which is named as a film bulk acoustic resonator (FBAR) 15 ; the other is to deposit multiple layers of thin films with different acoustic impedance to serve as a Bragg reflector 16, which reflects most of the acoustic energy back into the piezo- electric AlN plate, known as a solidly mounted resonatr (SMR) 17.
With respect to 5G new radio (NR) 18, two main characteristics are ob-served for the frequency band allocation: high operating frequency (>3 GHz) and wide bandwidth (> 500 MHz). The most common 5G NR bands deployed globally are n77 (3.3 to 4.2 GHz) and n79 (4.4 to 5.0 GHz), with the corresponding bandwidth of 900 MHz and 600 MHz, respectively. These bandwidths are nearly ten times greater than that of a typical 4G band, such as Band 2, which exhibits a downlink/uplink bandwidth of 60 MHz. Acoustic filter technologies, such as SAW and BAW, have been de- veloped and optimized for relatively narrower band applications in 2G to 4G, especially for frequency-division duplexing (FDD) bands in which the downlink and uplink bands exhibit a narrow gap (e.g., only 20 MHz for Band 2). Conventional SAW and BAW resonators based on bulk single- crystal lithium tantalate (LiTaO3 or LT) substrate and thin-film polycrys- talline AlN, do not have sufficient electromechanical coupling coefficient (k2) 19 to support filter design that requires 10 times larger bandwidth. Therefore, in the first phase of 5G deployment, low-temperature co-fired ceramic (LTCC) technology20 has become the mainstream filter technol- ogy which is widely applied in phones. The LTCC technology involves stacking multiple layers of conductor and dielectric to form inductors (L) and capacitors (C), which are the two basic electromagnetic components for wideband filter design. In spite of the fact that LC filter design is not new in the microwave field, the LTCC technology delivers a decent quality factor (inductor Q > 50 @1 GHz) in a tiny footprint (area < 1.6 ×0.8 mm2), making it an attractive option for cellphone applications in which small form factor is a primary concern.
However, in order to meet the increasing need of consumers for even higher data rates in future augmented reality (AR), virtual reality (VR) and metaverse applications21,22, more electromagnetic spectrum and new wireless communication standards are currently being formalized, such as 5.5G and Wi-Fi 6E/723. The high-frequency band (3 to 10 GHz) is once agian becoming increasingly crowded, just as the 2 to 3 GHz band dur- ing 4G development. This new scenario demands high frequency, wide bandwidth and sharp roll-off(high rejection) all in one filtering solution for the next phase of 5G evolution, which drives the invention of next- generation emerging filter technologies: LC-acoustic hybrid 3, XBAR 24 and XBAW 25. Along the way, there are also other variants of advanced fil- ter technologies which were proposed to compete with each other, includ- ing temperature-compensated SAW (TC-SAW) 26,27, incredible high per- formance SAW (IHP-SAW) 28, contour-mode resonator (CMR) 29, cross- sectional mode resonator (XMR) 30 and thin film lithium niobate (LiNbO3 or LN) based Lamb wave resonator 31. Fig. 2 shows the commonly used RF filter types and their suitable frequency range and bandwidth, all of which will be explained and discussed in the current work to present the advantages, disadvantages and future directions of these filter technolo- gies. The following sections are organized according to the timeline when different filter technologies were introduced alongside each generation of wireless communication standard.
Fig. 2 Advanced RF filters for wireless communications. |
2G AND 3G: SAW
In 1965, R. M. White and F. W. Volmer invented the interdigital transduc- ers (IDT) 12, which triggered the widespread application of surface acous- tic wave (SAW) devices. Owing to the advantages of low cost, small size, light weight and high reliability, SAW-based RF filters are widely used in second-generation (2G) and third-generation (3G) mobile communication systems 32⇓-34. In the frequency range up to 1 GHz, SAW filters are almost exclusively employed.
Fig. 3 shows a typical SAW resonator structure, consisting of a bulk piezoelectric substrate with patterned metal electrodes on top. Part of the metal electrodes is designed as interdigital transducers (IDT), while the rest is used as reflection g rating str uctures to confine acoustic wave energy within a resonating cavity. The working principle of SAW is illustrated in Fig. 3 : when an alternating voltage is applied to the IDT, acoustic wave is generated and travels along the surface of the piezoelectric substrate in the directions perpendicular to the fingers 34,35. When the frequency of the applied voltage aligns with the frequency (or wavelength) of the periodic structure of the IDT and reflectors, acoustic wave will be confined be- tween the two reflectors and therefore a resonator (acoustic wave cavity) is formed.
Fig. 3 Typical structure of a SAW resonator. |
The most critical parameter for SAW device design is the center fre- quency, which is determined by the period of the IDT fingers and the acoustic velocity 36 :
$ f_{0}=\frac{V_{S A W}}{\lambda}$
where,λ is the wavelength, given by the periodicity of the IDT and VSAW denotes the acoustic wave velocity. The IDT finger width is usually equal to a quarter wavelength (λ /4), so the interdigital electrode pitch (p) and the grating period are equal to half of the wavelength (λ/2). The generated acoustic wave largely depends on the type of piezoelectric substrate ma- terials, the orientation of the piezoelectric crystal, the materials, thickness and geometry dimensions of the metal electrodes 37.
The development and performance enhancement of SAW devices can be significantly attributed to the evolution of modeling and design the- ories. R. H. Tancrell was the first to propose that each edge of the IDT electrodes could be treated as an independent δ-function source of acous- tic energy 39, so that the IDT as a whole can be regarded as a superposi- tion of periodic independent wave sources, as shown in Fig. 4a. A refer- ence point at the center (x=0) is assumed for summing the distributed delta function contributions and the summation yields the frequency re- sponse of one set of IDTs. The amplitudes of delta sources are assumed to be constant, but the phase contribution changes for each electrode ac- cording to the distance from the reference point. The delta (δ) function model is the simplest model of IDT and has been widely used in SAW de- sign. However, this model cannot account for the effects of reflections in- side the IDT, which can sometimes severely affect the performance of the device.
Fig. 4 a, Delta function model (arrows indicate delta function sources). Electric field directions of b, real distribution, c, “cross field”model. Reprinted with permission from ref. 36. ©2000, Springer. and d, “in-line field”model. Reprinted with permission from ref. 38. ©2014, Institute for Environmental Nanotechnology. |
To address this issue, W. R. Smith proposed equivalent circuit mod- els 40 for IDT. He developed a simplified model for the electric field dis- tribution between IDT based on D. A. Berlincourt’s 41 work about equiv- alent circuits for the length expander bar with parallel and perpendicular electric field, as shown in Fig. 4 b-d, and used Mason’s equivalent circuit model 42 todescribe onepairofIDT fingers. Thedrawbackis thatitcannot account for mechanical wave reflection or energy storage. Another equiv- alent model is based on the Coupling-of-Modes (COM) theory, which takes both surface acoustic waves’ multiple scattering and their genera- tion by IDT electrodes into consideration 43⇓⇓⇓-47. The COM equations were proposed by C. S. Hartmann 48, M. S. Sandler 49 and D. P. Chen 44, and they gave a simple mathematical form to offer analytical expressions for many IDT parameters based on the force and voltage analogy. K. Hashimoto has composed an excellent book on COM theory used in SAW devices 36. To extract COM model parameters, it is necessary to analyze the prop- agation characteristics of surface acoustic wave in the periodic structure and determine the stop band boundary of the dispersion curve. Common extraction methods include the Green’s function 50, effective permittivity for grating 51, the discrete Green function 52 as well as the periodic Green function 53.
With the rapid development of computer technology in recent years, finite/boundary element modeling (FEM/BEM) has emerged as a cru- cial numerical calculation method for the accurate analysis of SAW de- vices 54⇓⇓⇓⇓-59. It can handle arbitrary materials and crystal cuts, diverse elec- trode shapes, and complex composite structures composed of multiple metal and dielectric layers. However, the FEM method typically must take an overwhelmingly large number of degrees of freedom into considera- tion, which results in high memory usage and long computing time 60,61. In 2016, J. Koskela et al. proposed the hierarchical cascading technique (HCT) 60 for fast simulation of SAW devices based on the two-dimensional (2D) finite element method (FEM). In this approach, each unique elec- trode period was modeled and simulated only once, which could drasti- cally reduce memory consumption and simulation time. Recently, D. Sui et al. proposed a universal hierarchical cascading technique (HCT) for 3D periodic modeling, which performed better than other approaches in quasi-periodic cascading 62.
The initial applications of SAW technology were mainly in mili- tary systems, but these devices failed to achieve high volume produc- tion. Around 1975, Philips, Plessey and Siemens started mass produc- tion of SAW intermediate frequency (IF) filters for TV applications. With the emergence of cellular phones and the prevailing heterodyne receiver architecture in 2G 64, SAW IF filters have been used for over a decade in almost every cellular phone. Until the 3G era when fre-quency division duplex (FDD) became mainstream, RF SAW filters working at high frequencies near 1 GHz were developed adopting two different circuit topologies for duplexer implementation, as illustrated in Fig. 5. One such topology is the double mode SAW (DMS) fil- ter 65, which utilizes two identical resonant modes acoustically coupled in the longitudinal direction, allowing impedance transformation and of- fering balun functionality for free. The other topology was the Ladder- type filter 66, which utilizes acoustic resonators to form cascaded series and shunt branches. Fig. 5 shows a PCS SAW duplexer, including a Ladder-type transmit (Tx) filter and a DMS receive (Rx) filter, show- ing ultra-low loss and steep cut-offcharacteristics. The SAW fabrica- tion process is relatively simple and typically involves a single litho- graphically defined metal layer on top of a bulk single-crystal piezoelec- tric substrate (LiTaO3). Due to their low cost and simplicity of the pro- cess, SAW RF filters have been dominating the low band applications (< 2000 MHz) 67.
Fig. 5 a, Schematic of a PCS SAW duplexer including a Ladder-type transmit (Tx) filter and a DMS receive (Rx) filter. b, Frequency response of the SAW duplexer. Reprinted with permission from ref. 63. ©2003, IEEE. |
As discussed above, the evolution of wireless communication has been driving the industry and academia to seek ways for further increas- ing the operating frequency and bandwidth of SAW devices. To achieve higher frequency operation, one approach is to reduce the line width and spacing of the IDT, which are ultimately limited by the resolution of pho- tolithography available from a SAW foundry. Narrower electrodes nor- mally contribute to higher parasitic resistance and lower current or power handling capability, which significantly degrades the resonator or filter performance. Therefore, more and more researchers resort to other ap- proaches: new substrate materials with higher acoustic velocity 68, higher electromechanical coupling or high-order resonant modes 69. Years of ef- fort and innovation have promoted the SAW technology to take several big steps forward in inventing further around the IDT concept with new thin- film materials and structures (thanks to the development of semiconductor and MEMS processes), which greatly enhances the filter performance and will be discussed in the following sections.
4G: TC-SAW AND BAW
TC-SAW
As shown in Fig. 6, when it came to 4G long-term evolu- tion (LTE), more frequency bands were defined in the 2 to 3 GHz range 26, which was compounded by the fact that the mainstream Wi-Fi band (2.4 GHz) had already been there. In one extreme case, the upper edge of Band 40 (2.3 to 2.4 GHz) coincided with the lower edge of the Wi-Fi band (2.401 to 2.483 GHz) with no transition band, making it almost impossible to achieve isolation between the two bands. In order to address this issue, 4G LTE required sharp roll-offin filter technology, resulting in the need for high Q and low temperature coefficient of the resonators 71⇓-73. How- ever, normal SAW filters at that time exhibited relatively high frequency drift over temperature (-35 to -45 ppm/ °C) 74, and they often failed to meet the stringent specifications of 4G standards 75,76. Researchers began exploring ways to compensate for temperature on the IDT structure by developing temperature-compensated SAW (TC-SAW) technology, which soon became the preferred approach for Japanese filter companies that had already invested heavily in SAW manufacturing infrastructure.
Fig. 6 Mobile frequency band allocation in mainland China. Reprinted with permission from ref. 70. ©2015, IEEE. |
The TC-SAW technology evolved in two flavors independently 26,78 : one is to deposit a silicon dioxide (SiO2) thin film layer on IDT to realize temperature compensation. The commonly utilized piezoelectric single- crystal substrate materials for normal SAW exhibit a negative temperature coefficient of frequency (TCF), whereas SiO2 exhibit a positive TCF that can cancel out the negative TCF of the piezoelectric substrate 79 and is readily available from semiconductor processes. Therefore, this approach provides the capability of achieving a TCF close to 0 ppm/ °C. Fig. 7 a plots the frequency characteristics of a TC-SAW filter that shows little drift with the temperature changing from -30 °C to 80 °C. However, since amorphous SiO2 is non-piezoelectric 80, this additional layer would de- grade electromechanical coupling, increasing propagation loss and caus- ing spurious resonant modes. To mitigate the spurious modes in TC-SAW, the Panasonic research team proposed two techniques: selectively remov- ing (or patterning) the top SiO2 film for confining the SAW energy in the active IDT region and suppressing transverse-mode spurious responses 81 ; controlling the cross-sectional shape of the SiO2 /electrode structure to suppress the Rayleigh-mode spurious responses 82, as illustrated in Fig. 8.
Fig. 7 Two options for TC-SAW implementation. a, Deposited SiO2 /IDT/LiTaO3 structure and the corresponding filter frequency characteristics at different temper- atures 73. b, Bonded LiTaO3 /sapphire structure and SAW duplexer frequency characteristics based on different substrates 77]. Reprinted with permission from refs. 73,77. ©2004, IEEE. |
Another temperature compensation approach involves bonding a LiTaO3 wafer to a supporting substrate with a low coefficient of ther- mal expansion 83. The physical mechanism of the bonding approach is to use the bulk supporting substrate to suppress the thermal expansion of LiTaO3, which is normally in a thin film form. Therefore, the support- ing substrate material must have a smaller thermal expansion coefficient (TEC) than LiTaO3 and, if possible, a larger Young’s modulus. Sapphire is an ideal material with a small TEC of about 5 ppm/ °C and a very large Young’s modulus of 470 GPa. The temperature compensation effect of sapphire bonding has been demonstrated and commercialized in the indus- try 77,84, as shown in Fig. 7 b. Silicon is another choice for the supporting substrate 85,86 due to its good thermal conductivity, which is approximately 4 times better than that of sapphire and 16 times better than that of quartz. Excellent heat extraction is critical for Tx filter applications. However, bonding a supporting substrate cannot realize zero TCF, as can be done by the SiO2 approach. Although the utilization of a thinner LiTaO3 layer can help reduce the TCF, it also comes with spurious responses.
Apart from low loss, high rejection and small TCF, high power dura- bility is also desperately required. For SAW devices, submicron-width aluminum (Al) electrodes undergo acoustic migration under alternating stress, leading to device failure at high power 87. Various compositions and structures, such as Al-Cu/Cu/Al-Cu 88, AlMgCu/Ti/AlMgCu/Ti 89 andsin- gle crystalline Al 90 have been explored to mitigate this migration problem. A research team at Tsinghua University proposed new electrode com- positions of Al/Ti/Cu/Ti 91, Al-0.9wt.%Cu(Al-Cu)/Ti/Cu/Ti 92, etc., which greatly improved the SAW filter power durability up to 34.5 dBm. Despite the advancement of SAW and TC-SAW technologies, it is still difficult to deliver high Q and high power durability in the frequency range from 2 to 10 GHz. This is also the reason that researchers (Hewlett-Packard Laboratories and Infineon Technologies) start to look into bulk acoustic wave (BAW) technologies for the commercial solution of sharp roll-offRF filters, as required by the 2 to 3 GHz bands in 4G.
BAW
FBAR
The film bulk acoustic resonator (FBAR) is a widely used bulk acoustic wave (BAW) device, which is known for its high operat- ing frequency, high Q and excellent power durability 93. There are two ways to fabricate an FBAR resonator: membrane-type and airgap-type 94, which corresponds to bulk 15 and surface 95,96 micromachining processes, respectively (Fig. 9). The membrane-type FBAR is realized by etch- ing the supporting substrate (typically Si) from the backside until the etching stops at the bottom electrode, so as to form a suspended elec- trode/piezoelectric/electrode sandwich structure which can freely vibrate in the vertical direction. As for airgap-type FBAR, a sacrificial layer is first formed before depositing the electrode/piezoelectric/electrode thin film layers and later is etched away (released) to form the airgap that al- lows the sandwich structure to vibrate freely.
Fig. 9 Cross-sectional drawing of FBAR device structures: a, membrane-type FBAR and b, airgap-type FBAR. |
In 1980, the first FBAR resonator was proposed by K. M. Lakin et al. 97 with a detailed theoretical derivation. Five years later, in 1985, H. Satoh et al. 95 proposed the first airgap-type FBAR device, which used zinc oxide (ZnO) as the piezoelectric layer material, and metal gold (Au) and Tita- nium (Ti) as the top and bottom electrode materials. Later, C. Vale et al. 98 and R. Ruby et al. 15 reported FBAR devices operating in the GHz range based on ZnO and aluminum nitride (AlN) in 1990 and 1994, respec- tively, both of which were membrane-type FBAR devices. In particular, R. Ruby et al. 15 reported FBAR resonators with Q values over 1000 and resonant frequencies in the range of 1.5 to 7.5 GHz, which made it fea- sible and attractive to commercialize FBAR devices. As a result, FBAR filters and duplexers were successfully commercialized by this team from Hewlett-Packard Laboratories, which later became Agilent, Avago and now Broadcom. The big commercial success also attracted other semi- conductor companies to invest in this MEMS filter technology, including TAIYO YUDEN 99,100, Samsung 101,102 and STMicroelectronics 103,104.
FBAR resonator can be analyzed with a simplified three-layer model, which primarily comprises a piezoelectric thin film sandwiched between upper and bottom electrodes. As shown in Fig. 10 a, applying an alternat- ing voltage to the top and bottom electrodes could generate an electric field (E), electric displacement (D) and polarization (P) within the pol-ing axis of the cr ystallog raphic str ucture. This axis is also known as the c-axis, as illustrated in Fig. 10 b. Due to the reverse piezoelectric effect, mechanical deformation arises along the z direction (c -axis) in the form of strain or an acoustic wave, as shown in Fig. 10 c. The upper surface of the top electrode and the lower surface of the bottom electrode serve as two boundary conditions, where the sandwich structure abuts air. Con- sequently, the acoustic wave encounters a significant acoustic impedance mismatch and gets reflected. In this way, the acoustic wave bounces back and forth between the two surfaces, transforming the structure into a res- onator (acoustic cavity) with the majority of the acoustic wave energy confined within the mechanically vibrating body itself. Fig. 10 d shows a typical frequency response of an FBAR resonator. The electrical charac-teristics of FBAR exhibit infinite pairs of resonant points, and each pair is composed of a series resonant frequency (fs) and a parallel resonant frequency (fp).
Fig. 10 a, Electric displacement and polarization in an FBAR resonator. Reprinted with permission from ref. 105. ©2008, Universitat Autònoma de Barcelona. b, Deformation of the crystal structure when electric field is applied in the c -axis. Reprinted with permission from ref. 106. ©2001, Avagotech. c, The direction of electrical field and acoustic wave propagation in an FBAR resonator. d, Typical frequency response of an FBAR resonator. |
To characterize the electrical properties of piezoelectric bodies, W. P. Mason proposed the traditional piezoelectric theory in 1948 42, which included the Mason equivalent circuit model, as shown in Fig. 11. The Mason model can be utilized to describe not only piezoelectric materials but also non-piezo materials. The multi-layer stacking of the piezoelec- tric thin film and two metal electrode layers could be accounted for by cascading multiple sections of transmission line in the equivalent circuit model. Each section of the transmission line includes two acoustic ports: one on the left and the other on the right, while the section for the piezo- electric layer exhibits an additional electrical port at the bottom, with the transformer representing energy conversion between acoustic and electri- cal domains.
Fig. 11 The Mason equivalent circuit model. Reprinted with permission from ref. 42. ©1948, D. Van Nostrand Company, Inc. |
Although the Mason model could accurately describe the electrical behavior of FBAR, the model contains transformers and negative capaci- tors, which are inconvenient for circuit-level design or simulation. In order to simplify the analysis when adopting FBAR as a lumped component, the Butterworth-Van Dyke (BVD) model 107 was developed to mainly describe the electrical behavior around the fundamental resonant frequency, while the higher-order harmonics were neglected. In real-world applications, an FBAR resonator suffers from not only mechanical/acoustic loss but also dielectric loss and resistive loss from the Si substrate and metal electrodes, thus J. D. Larson III et al. proposed a modified BVD (MBVD) model 19 during the commercialization of FBAR duplexers, which becomes the most used equivalent circuit model for SAW and BAW resonators for RF filter design, as shown in Fig. 12.
Fig. 12 MBVD equivalent circuit model for FBAR. Reprinted with permission from ref. 19. ©2000, IEEE. |
Thanks to the excellent material properties such as high acoustic ve- locity, AlN-based FBAR resonators offer high Q (> 2000) at high fre- quency (> 2 GHz) 108, low temperature coefficient of frequency (TCF) 109 and process compatibility with complementary metal oxide semiconduc- tor (CMOS) active circuits 109. At the time, it was difficult for SAW filters to deliver sufficient performance at high frequencies above 1.5 GHz, es- pecially when considering the standard bands (Band 1, Band 2, Band 3, Band 40, Wi-Fi 2.4 GHz, etc.) are extremely crowded in the sub-3 GHz frequency range, the FBAR technology achieved prominent commercial success in the 4G era due to its high Q and therefore sharp roll-offat the transition edge from filter pass band to stop band 110, as is shown in Fig. 13. In order to cope with even higher frequencies (> 3 GHz and up to mil- limeter wave) and significantly wider bandwidth (> 500 MHz) required by 5G, AlN-based FBAR manifests some intrinsic deficiencies, for example, its fractional bandwidth (FBW) is limited by the electromechanical cou-pling coefficient (∼7% for sputter-deposited AlN FBAR resonators 111), a parameter primarily determined by material properties and hard to be improved by design or process.
Fig. 13 a, Measured Band 25 duplexer S-parameters. b, Wideband response of the FBAR duplexer S-parameters. Reprinted with permission from ref. 110. ©2019, IEEE. |
One way to achieve higher frequency operation is to thin down the piezoelectric film. It is reported that an airgap-type FBAR can work up to 24 GHz 112, accounting for the fact that the thickness of AlN is only 120 nm. The resonant frequency, Q value at series resonant frequency (Qs), Q value at parallel resonant frequency (Qp) and electromechanical cou- pling coefficient (k2) were 24.7 GHz, 285, 291 and 6.01%, respectively. The other way is to utilize higher-order overtone modes. An overmoded bulk acoustic resonator (OBAR) was demonstrated through the fabrica- tion of a Pt-AlN-Al sandwich structure (70 nm, 140 nm and 90 nm thick- ness, respectively) with the k2 of 1.7% and the Qs of 110 at 33 GHz 113. When the AlN layer thickness is down to the 100 nm range, the crys- tallinity of the deposited thin films tends to be more and more important for the piezoelectric performance 109. Researchers have been trying to real- ize single-crystal piezoelectric films: the so-called XBAW 25 and Two-Step Method 114 have been reported. In 2022, X. Yi et al. fabricated high-quality AlN films on Si substrate adopting a mix of metal-organic chemical vapor deposition (MOCVD) and physical vapor deposition (PVD, i.e., sputter- ing) methods, as shown in Fig. 14, which resulted in a small value of 0.68 °for the full width at half maximum (FWHM) of the X-ray rocking curve. In comparison, the FWHM for sputtered AlN films on Si substrate is gen- erally above 1.4 °. The XBAW will be explained in detail in the following chapters.
Fig. 14 Cross-section drawing for a, sample 1: two-step AlN deposition; b, sample 2: one-step PVD AlN film deposition; c, the fabricated FBAR resonator. Reprinted with permission from ref. 114. ©2022, IEEE. |
To increase the electromechanical coupling coefficient (k2) of AlN resonators, scandium (Sc) doping to form AlScN has been proven to be an effective invention 115⇓-117. It is demonstrated that d33 up to 5 times larger than pure AlN can be achieved due to enhanced piezoelectric re- sponse in Sc-doped AlN 118, resulting in a significant improvement in k2 (2.6 times higher than AlN for Sc0.35 Al0.65 N) 119. The highest k2 reported to date for AlScN FBAR resonators is 18.1% by J. Wang et al. in 2020 120. This is also the first demonstration of frequency tuning and intrinsic polar- ization switching of FBAR resonators, based on sputtered AlScN piezo- electric thin films with 30% doping concentration. Besides, researchers also resorted to other piezoelectric materials, out of which lithium niobate (LiNbO3 or LN) stands out as a promising alternative for BAW appli- cations. LN-based FBAR exhibits higher k2 than AlN, which is simply ascribed to the fact that LN exhibits larger and more piezoelectric coef- ficients available from different crystal orientations, for example, k2 can be 39.2% for X-cut LN 121, 29.4% for Y + 163 °-cut LN 122 and 17.4% for X-cut LiTaO3 123. State-of-the-art results of recent FBAR resonators based on different materials are summarized in Table 1.
Table 1. State of the art results for FBAR resonators. |
| Ref. | Piezo Film | Type | f (GHz) | Q | k2 (%) |
|---|---|---|---|---|---|
| 120 | 900 nm Al0.7Sc0.3N | FBAR | 2.93 | 210 | 8.1 |
| 112 | 120 nm AlN | FBAR | 24.7 | 285 | 6.01 |
| 113 | 140 nm AlN | OBAR | 33 | 110 | 1.7 |
| 124 | 1.2 um Al0.91Sc0.09N | FBAR | 2.2339 | 513 | 9.53 |
| 124 | 1.2 um Al0.85Sc0.15N | FBAR | 2.1522 | 348 | 12 |
| 122 | 600 nm X-cut LN | FBAR | 2.986 | 250 | 39.2 |
| 121 | 600 nm Y+163°-cut LN | FBAR | 2.50 | 350 | 29.4 |
| 123 | 740 nm X-cut LT | FBAR | 1.58 | 400 | 17.4 |
| 125 | 410 nm Z-cut LN | FBAR | 2.9 | 73 | 5.8 |
| 24 | 400 nm Z-cut LN | XBAR | 4.8 | 300 | 25 |
SMR
The essence of the air cavity underneath the FBAR sandwich structure is to make sure that there is a large acoustic impedance mis- match at the interface, so that the maximum acoustic energy is reflected and confined in the resonator body itself. An alternative way to serve the same purpose is to use a mechanical Bragg reflector 126, such a BAW de- vice is called the solidly mounted resonator (SMR). As early as 1965, W. E. Newell 127 proposed a method to solidly mount a piezoelectric resonator on a substrate with the adoption of cascaded quarter-wavelength acoustic transmission lines without degrading the quality factor too much. The au- thor demonstrated its feasibility at low frequencies and predicted that this technique would be extensively useful for thin film structures operating at high frequencies. In 1994, R. J. Weber 128 applied for a patent claiming an acoustic isolator disposed between the resonator and the substrate, so that the thin film resonator sees an equivalent impedance close to air. In 1995, K. M. Lakin et al. 94 published a paper that used the term SMR for the first time.
SMR can exhibit two kinds of configurations:λ /2 mode 129⇓⇓⇓-133 andλ /4 mode 134,135, as shown in Fig. 15. For the λ/2 mode configuration, the top layer of the Bragg reflector shows low acoustic impedance so that the bot- tom surface of the piezoelectric film is traction-free, and therefore the piezoelectric film corresponds to a half-wave plate (in the z direction) at the fundamental resonant frequency. As for the λ/4 mode configuration, the top layer of the Bragg reflector should exhibit high acoustic impedance so that the bottom surface of the piezoelectric film is almost clamped, and therefore the piezoelectric film corresponds to a quarter-wave plate at the fundamental resonant frequency. For the two different SMR configura-tions, more Bragg reflector layers will contribute to an increase in the Q value, which will ultimately saturate, while k2 is less affected by the num-ber of reflector layers. The Q and k2 of the λ/4 mode configuration are theoretically lower than those of the λ/2 mode configuration, while the λ/4 mode configuration is endowed with the advantage of thinner piezoelectric film thickness 134.
Fig. 15 Two different SMR configurations. a,λ /2 mode configuration. b,λ /4 mode configuration. Reprinted with permission from ref. 134. ©2000, Japan Society of Applied Physics. |
Infineon Technologies (R. Aigner et al.) started volume manufac- turing of SMR filters for GSM mobile handset applications in mid- 2002 17,136,137. The Bragg reflector includes three pairs of tungsten and oxide layers, which achieves an excellent ratio of acoustic impedance (∼7) and very high reflectivity. Since tungsten is metal and plays the func- tion of avoiding parasitic electromagnetic coupling between neighboring resonators, the multi-layer reflector has to be patterned by a novel process called “Multi-CMP”138. Only one lithography step is required to pattern all layers at once, and the structure of the Multi-CMP SMR device is shown in Fig. 16.
In 2017, Qorvo introduced an SMR filter that can handle 5 W of RF average input power, with peaks up to 40 W, which verifies the excellent high power handling capability of SMR for base station applications 141. The SMR filter is optimized for 2575 to 2635 MHz (a sub-band of Band 41), and it has a small size of 5 ×5 ×1 mm3, which is 90% smaller than that of the traditional ceramic filters used in base stations.
In 2022, A. Tag and M. Schaefer et al. from Qorvo reported a new generation of SMR-type BAW technology 142, as shown in Fig. 17 a, which supports the frequency range from 1 to 8 GHz, covering the 5G NR and Wi-Fi 6E frequency bands. They outlined five key points in the future development of BAW technology, including reduced die size of filters, support of higher frequencies, much wider fractional bandwidth (BW), higher power handling and improved insertion loss (IL) and rejection. Fig. 17 b-c show that this new generation of Sc-doped SMR can simul- taneously reduce the device size by 50%, while improving the TCF by 50% and Q value by 100%. The demonstrated n75/n76 filter shows a performance improvement of 1.2 dB at both band edges, as illustrated in Fig. 17 d.
Fig. 17 Qorvo’s new generation of SMR-type BAW technology. a, Cross-section drawing of an SMR. b, TCF improvement of over 50%. c, 100% improvement on Bode Q. d, 1.2 dB improvement at both band edges for n75/n76 filter. Reprinted with permission from ref. 142. ©2022, IEEE. |
CMR
As has been explained, the resonant frequency of FBAR or SMR resonator is determined by the total thickness of the metal-piezo- metal thin film layer stack. And from the manufacturing point of view, only one frequency of operation can be obtained on the same wafer or chip with a fixed layer stack. However, in 3G and 4G applications, a cellphone usually has to support more than 10 bands, which requires more than 10 filter chips (working at different frequencies) that are separately manufac- tured and packaged 143. To reduce the overall complexity and area of such filter banks, researchers started to look for technologies that could enable multiple frequencies of operation on the same chip, and the contour-mode resonator (CMR) technology 29 has received a lot of attention for this pur- pose. Although both FBAR/SMR and CMR technologies have the same layer stack: metal-piezoelectric thin film-metal, in the case of CMR, the transverse piezoelectric coefficient (d31) is utilized to excite a mechanical vibration in the lateral (in plane) direction, while the voltage is applied in the vertical (z) direction. In comparison, the longitudinal piezoelectric coefficient (d33) is primarily the mechanism responsible for FBAR/SMR operation: vibration and voltage both in the vertical (z) direction. The resonant frequency of CMR is determined by the lateral dimensions of the resonator body, and therefore multiple frequencies can be enabled on the same chip or wafer. After years of researches, a variety of piezo- electric materials have been tried to implement CMR devices, including AlN 144⇓-146, zinc oxide (ZnO) 147⇓-149, lead zirconate titanate (PZT) 150⇓-152, lithium niobate (LiNbO3 or LN) 153,154 and recently scandium-doped AlN (AlScN) 155⇓-157 as well.
Researches on CMR type of devices were started in 1941, when G. Builder et al. 158 studied principal contour-mode responses in Y-cut quartz plates. Later in 1979, J. Hermann et al. 159 proposed a computer proce- dure to determine the frequency and piezoelectric coupling of rectangular contour mode resonators of any cr ystallog raphic orientation. G. Piazza et al. 29,160 published a series of papers on CMR resonators based on re- leased AlN thin films from 2004 to 2006, which laid out the foundation for extensive researches on CMR-type devices implemented by more ad- vanced MEMS processes. The in-plane shape can be rectangular, circular, ring, etc., as shown in Fig. 18 29. Teams from all over the world including G. Piazza et al. 145, 161-163, V. Yantchev et al. 164, 165 and R. Olsson et al. 166,167 have made significant contributions to this field, which also inspired the later researches on Lamb wave devices based on suspended single-crystal LiNbO3 thin films, as will be discussed in the following chapters.
Fig. 18 Three different designs of contour-mode resonators (CMR). a, Rectangular plate. b, Circular ring. c, Square-shaped ring. Reprinted with permission from ref. 29. ©2006, IEEE. |
Although the electrode configuration of CMR is similar to interdigital transducers (IDT) in SAW, the CMR fabrication process offers the pos-sibility of placing electrodes on the bottom surface of the piezoelectric thin film. In earlier researches, CMR can be mainly categorized into two electrode configurations: thickness field excitation (TFE) 168,169 and lateral field excitation (LFE) 161,170,171, which are shown in Fig. 19 a and b. The TFE configuration allows to obtain higher electromechanical coupling co- efficient (k2) by utilizing electric field in both thickness and lateral direc- tions at the same time, while its fabrication process is more complicated: at least three lithographic masks are required and there is misalignment issue between top and bottom electrodes. For LFE devices, k2 is sacri- ficed to some extent in favor of a simpler fabrication process and there are interdigitated electrodes only on the top surface of the piezoelectric thin film, so that the applied electric field is primarily distributed laterally in the resonator body. Another possible CMR configuration 172 is shown in Fig. 19 c, which looks like LFE but with a floating bottom electrode. The expected k2 lies between LFE and TFE, and this configuration fea- tures better piezoelectric thin film deposition due to a single flat bottom electrode when compared with TFE.
$ f_{0}=\left\{\begin{array}{l} \frac{1}{2 L} \sqrt{\frac{E_{p}}{\rho}} \\ \frac{1}{2 W} \sqrt{\frac{E_{p}}{\rho}} \end{array}\right.$
where, Ep represents the equivalent Young’s modulus of the metal-piezo- metal layer stack, ρ is the equivalent density and L (W) denotes the length (width) of the rectangular plate depending on which of the length- extensional or width-extensional mode we look at. As can be seen, the resonant frequency of CMR can be easily set by lateral dimensions (i.e., layout design), which makes single-chip multiple frequencies available. This also comes with an intrinsic bottleneck for CMR: the electromechan- ical coupling coefficient (k2) is below 3%, which is much lower than that of FBAR/SMR (typically ∼7%) 174, simply because the piezoelectric co- efficient d31 is less than half of d33 for AlN.
XMR
As part of the efforts to enhance the electromechanical cou- pling coefficient (k2) of both CMR and FBAR/SMR for future wideband filter applications, researchers started to investigate the vibration mode shapes that could combine the effect of two or more piezoelectric coeffi- cients. Traditionally in CMR or FBAR/SMR design, attention has always been focused on the geometrical dimensions (layout design) of the res- onator when looking at it from a top view. C. Zuo et al. 30 proposed a new approach to design the mode shape in the cross-sectional view, where the d33 and d31 piezoelectric coefficients could be coherently combined to ex- cite a two-dimensional (2D) mode that achieves higher k2 than either CMR or FBAR/SMR, as shown in Fig. 20. This kind of XMR provides effective leverage from the design side to improve resonator performance without changing the piezoelectric material, which is still under development by several research teams over the world 175⇓-177.
Fig. 20 Mode shape difference between FBAR, CMR and XMR. Reprinted with permission from ref. 30. ©2012, IEEE. |
There are two electrode configurations under the XMR concept: the first includes the so-called combined overtone resonator (COR) 178, cross- sectional lame mode resonator (CLMR) 177 and laterally coupled alternat-ing thickness mode resonator (LCAT) 179 ; the second includes the two- dimensional-mode resonator (2DMR) 180, couple bulk acoustic resonator (CBAR) 181 and two dimensional resonant rods (2DRR) 182. The COR ex- ploits the multimodal excitation of two higher-order Lamb waves (2nd and 3rd order asymmetrical Lamb waves) in a suspended thin-film AlN plate to transduce a 2D vibration mode. The difference between CLMR and LCAT is the ratio of the AlN film thickness (tAlN) to the electrode pitch (pele). For an ideal CLMR, the ratio is supposed to be 1, but for LCAT, it is around 0.5. In 2022, X. Zhao et al. reported an AlScN based two-dimensional resonant rods resonator (2DRR), with a high Sc doping concentration of 24%, exhibiting a record high k2 of 23.9% at 5 GHz 183.
The main difference between the two configurations is the way how electrical signals are applied to the electrodes. For the first configuration, signals of opposite polarities are applied to the adjacent interdigitated electrodes, and signals with opposite polarities are also applied to the electrodes on the bottom side of the piezoelectric layer, facing the ones on top; while for the second configuration, signals of the same polarity are applied to the adjacent interdigitated electrodes, but signals of oppo- site polarity are applied to all of the electrodes on the bottom 174. In order to illustrate the variations, the electrode configurations of different XMRs are summarized in Fig. 21.
Fig. 21 Comparison between different electrode configurations of XMR. |
5G: LTCC AND IPD
The 5G new radio (NR) bands (n77, n78 and n79) are defined with signif- icantly larger bandwidth (10 ×) when compared with 3G/4G bands. Con- ventional acoustic filters (either SAW or BAW) are unable to cover such wide bandwidth, so electromagnetic inductor and capacitor (LC) based filters naturally become the solution in the beginning phase of 5G deploy- ment. Although LC filters cannot provide sharp roll-off, which is mainly ascribed to the fact that inductors and capacitors at this frequency (3 to 7 GHz) have limited Q (< 300) as compared with acoustic resonators (Q > 1000 @1 to 3 GHz), at least they can offer wide passband (> 500 MHz) with low insertion loss (< 2 dB) as required by the 5G standards.
LTCC and IPD are the two major LC filter technologies that are widely adopted in today’s 5G systems. Due to its g reat ther mo-mechanical and electrical properties, ceramics have been used for more than 60 years for harsh environment applications such as aerospace, military, radar, auto- motive and other applications. LTCC is such a ceramic technology that en- ables multiple layers of conductors and dielectrics stacked on each other, for which the firing temperature is relatively low (< 1000 °C) 184. The multiple layers of conductors and dielectrics are adopted to design embed- ded inductors and capacitors within the same substrate or device. Man- ufacturing an LTCC multi-layer circuit involves many steps of printing, laminating, firing, etc., which is a relatively old process as illustrated in Fig. 22 185.
Fig. 22 Process flow of LTCC. Reprinted with permission from ref. 185. ©2023, University of Arkansas. |
The history of LTCC dates back to the 1980s when Hughes and DuPont first developed the technology for military applications 186. In 1985, M. Sagawa came up with the idea of designing an LTCC filter 187. A multi-layer LTCC technology enables RF modules to be reduced in size dramatically by taking advantage of the three dimensional (3D) flexibil- ity 188, and in addition, the LTCC technology can also offer other features which are ideal for multi-chip module packages, such as low resistivity conductor, low dielectric loss at high frequencies and good thermal sta- bility 189. LTCC has achieved rapid growth in the wireless communication industry: cellular phones (0.5 to 3 GHz), Wi-Fi and Bluetooth (2.4 and 5 GHz), the Global Positioning System (GPS, 1.5 GHz), broadband access connection systems (5.8 to 40 GHz), etc. LTCC can not only be used as standalone filters, moreover, it can serve as a supporting substrate for a multi-component RF module in which integrated circuit (IC) chips and discrete surface mount components can be wire-bonded or soldered on top, as illustrated in Fig. 23.
Fig. 23 Schematic view showing an LTCC module with multiple embedded components. Reprinted with permission from ref. 190. ©2015, IEEE. |
There are mianly two different circuit configurations for designing LTCC filters. The popular approach in the early days was to fold dis- tributed elements, such as microstrip transmission lines, coupled lines and electromagnetic resonators into multi-layer 3D LTCC structures. The re- cently developed more advanced design method is to construct lumped elements, including metal-insulator-metal (MIM) capacitors and spiral in- ductors, in a very compact 3D-stacking form and use through holes (vias) to connect the elements together. The distributed design is more suitable for higher frequencies (> 30 GHz, i.e., millimeter waves), for which the electromagnetic wavelength is relatively small and therefore the compo- nent size can be in the millimeter range. However, for cellular bands (1G to 5G: 0.5 to 5 GHz), it is ubiquitous to design lumped inductors and ca- pacitors (LC) to realize filters, diplexers and couplers in a miniaturized package (e.g., 1.6 ×0.8 mm2), tens of which are commonly used in to- day’s 5G smartphones.
LTCC 190⇓⇓-193 filter has demonstrated its cost performance in 5G ap- plications, however, it is also faced with two intrinsic deficiencies going forward. Due to the relatively old process technology of printing and lam- inating, it tends to be more difficult to improve the uniformity in line width/spacing and film thickness, which results in lower manufacturing yield when further shrinking the device size. LTCC technology relies on 3D stacking of multiple layers for high performance filter design, which makes the device thickness relatively large (0.65 to 1 mm). Oftentimes, filter components need to be integrated into a system-in-package (SiP) RF module, and in order to make the module thin enough for going into a smartphone, the filter package is generally required to be thinner than 0.5 mm or even 0.35 mm. These two reasons drive researchers and the industry to look for next-generation LC filter technologies, and this is why the integrated passive device (IPD) technology was developed with semiconductor-level process control and small device height.
IPD technology dates back to 1967 194, at that time tantalum, tantalum pentoxide and nichrome-gold were used to fabricate resistors, capacitors and inductors on an integrated circuit, respectively. The quality factor (Q) was reported to be about 15 at 50 MHz for a 50 pF capacitor, while the in- ductor Q was measured between 15 and 20 at 10 MHz for 1 μH occupying 0.35 inches square. Until the late 1990s, researchers were able to achieve inductor Q (Qmax) larger than 20 at GHz frequencies by adopting exotic substrates, like high-resistivity silicon (HRS), sapphire and quartz 195,196, or even fabricating MEMS structures to suspend inductor coils above sub- strate 197. Fig. 24 shows a typical process flow of IPD.
Fig. 24 Process flow of IPD. Reprinted with permission from ref. 198. ©2015, Elsevier. B. V. |
In terms of capacitors for RF application, IPD shows a natural advan- tage over LTCC, simply because it is based on thin film deposition instead of multi-layer ceramic sheet lamination 199 and the capacitance density in IPD is much higher than that in LTCC. Especially for higher frequencies above 3 GHz as required by 5G, higher capacitance density means smaller capacitor area and therefore less parasitic resistance and inductance com- ing from interconnects, which makes IPD thin film based MIM capacitors intrinsically exhibit higher Q and resonant frequency. In consideration of the better uniformity, smaller package height and higher performance, IPD stands out as a very promising technology for wide-bandwidth, low-loss and small-size filtering solutions at frequencies above 3 GHz for 5G and future wireless communications 3.
Fig. 25 a, Simulated transmission response of a hybrid Band n77 filer showing wide passband of 900 MHz, low insertion loss and high rejection 3. b, Performance comparison between two Band n41 filters with FBAR alone and FBAR + IPD hybrid. (Red line: FBAR alone; green line: hybrid FBAR and IPD) 237. Reprinted with permission from refs. 3,237 ©2019, 2022, IEEE.
The wafer substrate material is a critical factor that determines the performance of IPD filter. Compared with silicon IPD 201,204, glass IPD 202 is an alternative solution that provides superior RF performance due to the higher resistivity and lower dielectric constant of glass. For GaAs IPD 203,205,206, it is endowed with the advantage of being able to be inte- grated with active circuits on the same chip. Typical LTCC and IPD filter examples are given in Fig. 25. Although they have been widely adopted in the first phase of 5G, it can be easily seen from the electrical responses that they cannot deliver sharp roll-offdue to the limited Q (< 300), which would become an issue for future co-existence scenarios between 5G, 5.5G, Wi-Fi 7, etc. Therefore, some emerging filter technologies are pro- posed, as shown in Fig. 26.
EMERGING TECHNOLOGIES
Solidly mounted thin-film SAW devices
In the past decade, further innovation on acoustic resonators/filters has been achieved by the devel- opment of single-crystal thin film transfer and heterogeneous integration techniques. As a powerful supplement to epitaxial growth techniques (e.g., molecular beam epitaxy (MBE), MOCVD, PVD and magnetron sputter- ing), heterogeneous integration can transfer and bond single-crystal piezo- electric materials (e.g., LiTaO3 and lithium niobate) to a variety of sub- strates for addressing the limitations in traditional SAW devices 208⇓⇓⇓-212.
The layer transfer technique provides the possibility of selecting proper materials (layer stack) underneath the piezoelectric single-crystal thin film for different purposes. The layer stack can be engineered as in SMR devices to confine acoustic energy into the top piezoelectric layer and achieve higher Q than normal SAW devices. Murata pro- posed a solidly mounted thin-film SAW device called the incredible-high- performance (IHP) SAW featuring 4 times enhancement in Q (∼4000) 28. As shown in Fig. 27, a submicron thin film of LiTaO3 is bonded on a multi- layered substrate to improve acoustic energy confinement, while SiO2 is used in the substrate to reduce the temperature coefficient of frequency (TCF). In addition, the IHP-SAW technology also enables modes with higher acoustic velocity 213, which pushes the operating frequency of SAW into the 5 GHz range.
Fig. 27 a, Simulated surface concentration ration of acoustic energy. b, Measured Q values in different SAW devices. (Red line: IHP-SAW resonator with 3 thin-film layers; green line: IHP-SAW resonator with 2 thin-film layers; blue line: conventional 42 °YX-LT SAW resonator). Reprinted with permission from ref. 28. ©2017, IEEE. |
Instead of adopting Si substrate in IHP-SAW, researchers also tried to utilize quartz in the hetero acoustic layer (HAL) configuration, which is mainly ascribed to the fact that quartz substrate exhibits high acoustic ve- locity and positive TCF 214. This quartz-based HAL-SAW (LT-on-quartz) also achieved 1.2 times bandwidth (BW: 5%) and 6.4 times in Bode-(∼3000), when compared with a reference normal SAW resonator. In ad- dition to submicron thin-film LT, 5 up to 20- μm thick LT films were in- vestigated to be bonded on silicon substrates (LT-on-Si), and especially the LT layer was apodized (roughened) in thickness (a concept borrowed from the apodization of FBAR resonators via non-regular polygons) so that un- wanted high- Q spurious modes got “smeared”out by random variations in height determined by the top to bottom LT thickness, as illustrated in Fig. 28 86.
In addition to LT, thin-film LiNbO3, which is featured with higher piezoelectric coefficients, was also studied in the solidly mounted struc- ture to improve the performace of SAW. As shown in Fig. 29 a, silicon carbide (SiC) was chosen as the substrate for confining acoustic energy within thin-film LN 215. SiC is the second best material (next to dia- mond) that has a mix of properties with respect to LN. It is endowed with exceptionally high thermal conductivity (370 W/(m ·K)) and has been known for harsh environment applications. The LN-on-SiC thin-film SAW demonstrated a k2 as high as 26.9%, a Q of 1228 and a Figure of Merit (FoM = k2 ×Q) of 330 at 2.28 GHz. Despite the high performance of LN-on-SiC, when considering from large-scale manufacturing and lower cost, Si has to be considered as the substrate in LN based thin-film SAW 216. This has been carefully studied 217 : a thin layer of 400-nm amorphous Si was inserted between the LN piezoelectric layer and the crystalline Si sub- strate, so as to minimize the parasitic surface conduction (PSC) effect and also spurious acoustic modes, as shown in Fig. 29 b. The LN-on-Si SAW achieved a high k2 of 22.8% and a Q of 1208 at 1.6 GHz, which is not too far away from the LN-on-SiC case.
Suspended thin-film LiNbO3 devices
The solidly mounted thin- film SAW devices have achieved promising commercial success (a typ- ical application is in the Wi-Fi 2.4 GHz band), but the operating fre- quency and electromechanical coupling (k2) still remain insufficient for the next-generation wireless communication systems. To address the limi- tations, novel suspended acoustic devices based on thin-film LiNbO3 have been proposed and demonstrated with record-breaking Figure of Merits (FoM = k2 ×Q) to support passband filtering of over 10% fractional bandwidth 218⇓⇓⇓⇓-223. In consideration of the acoustic energy confinement, the suspended thin-film platform is featured with the unique advantage that the surrounding air can provide the maximum acoustic impedance mis- match. These suspended LN resonators have been demonstrated with vari- ous acoustic modes over an unprecedented wide range of frequencies from MHz to 60 GHz 207,224.
Around GHz, fundamental symmetric Lamb wave (S0) mode and shear horizontal wave (SH0) mode were proposed on single-crystal X-cut LN thin films with a demonstrated k2 over 20% and high Q 225, thanks to mode isolation and energy trapping of the suspended structures. As shown in Fig. 30, similar to SAW, the operating frequencies of S0 and SH0 mode devices are primarily defined by their lateral dimensions, which can be accurately controlled by photolithography 218,219. However, their moderate acoustic velocities, sub-7000 m/s for S0 and sub-4500 m/s for SH0 make it challenging to cover the entire sub-7 GHz spectrum 226.
For sub-7 GHz 5G NR applications, a single-crystal thin-film LN res- onator based on the first-order asymmetric Lamb wave (A1) mode was proposed by Y. Yang et al. in 2017 220. Subsequently, this resonator was commercialized under the name of XBAR by Resonant Inc. 223. The op- erating frequency of the A1 mode was determined by the thickness of the LN thin films and their lateral dimensions, enabling monolithic multi-band solutions with moderately wide electrodes for substantial power- handling capabilities. An alternative approach is to excite the first or- der symmetric (S1) mode in X-cut LN, allowing the resonator to oper- ate at 6.5 GHz with a moderate k2 of 4.79% 23. By appropriately selecting the Euler angle, a surprisingly high Qp (quality factor at parallel reso- nance) over 100,000 has been demonstrated, representing the largest leap even seen for acoustic resonators. However, the high piezoelectric co- efficients and anisotropic properties of LN lead to significant spurious modes around the passband, which is currently impeding the commer- cialization of suspended thin-film LN devices. Several approaches have been proposed to suppress the spurious modes, and one method is based on partially etching the top surface of LN thin film and filling the re- cessed grooves with metal electrodes. As shown by the SEM images in Fig. 31, the electrical and mechanical loadings of the electrodes can be balanced by the recessed grooves filled with electrodes for dispersion matching 227.
To scale the piezoelectric acoustic devices to millimeter-wave (mmWave) frequencies, the asymmetric Lamb wave modes can be ex- tended to higher orders, for which the operating frequency is determined by the thickness of LN thin film, lateral dimensions and mode order 228. As shown in Fig. 32, the fabricated device was measured with a Q over 100 at 56 GHz (the 13th asymmetric mode) 207, but the k2 remains lim- ited as it is inversely proportional to the square of the order. With further advancement in ion-slicing or film transfer techniques, the operating fre- quency can be scaled up by directly reducing the thickness of LN thin films without using higher order modes 229,230.
Fig. 32 a, mmWave acoustic resonator operating around 56 GHz. b, Demonstrated mmWave acoustic devices based on asymmetric Lamb wave modes. Reprinted with permission from ref. 207. ©2020, IEEE. |
Single-crystal AlN
Traditional FBAR and SMR BAW resonators are constructed with the adoption of thin film piezoelectric AlN materials, deposited by PVD techniques (e.g., sputter deposition), resulting in poly- crystalline AlN thin film. As single-crystal LT or LN thin films (either solidly mounted or suspended) showed significant performance enhance- ment, researchers started to conduct investigations on single-crystal AlN deposition and doping 231.
Single-crystal piezoelectric films should potentially exhibit higher acoustic velocity and piezoelectric coefficients than polycrystalline films 25. Besides, it is believed that single-crystal AlN thin films should exhibit better thermal conductivity than poly-AlN, which results in better power handling capability. Therefore, the XBAW technology was devel- oped by Akoustis in 2016 25,232. Single-crystal AlN films were deposited through MOCVD and the achieved FWHM was 0.03 °, compared with 1.4 °of poly-AlN films. As shown in Fig. 33 a, the maximum input power of XBAW resonators has been significantly improved 25. Measured power handling capability at 5.2 GHz of single crystal MOCVD AlN exceeds polycrystalline PVD AlN by 2.3 ×when compared with dies mounted on laminate and by 1.95 ×when compared with dies tested on wafer. The XBAW technology combines AlScN material of enhanced electrome- chanical coupling and resonator structures with the optimized Q, enabling miniature high-performance RF filters and improved design trade space. As shown in Fig. 33 b-d, several high frequency commercial level filters have been reported based on the XBAW technology, including a wide- band filter covering the UNII bands 1 through 3 233, a very steep transition UNII-5 filter 234 and a Wi-Fi 6E diplexer 235.
Fig. 33 a, Measured power handling capability at 5.2 GHz of single crystal MOCVD AlN compared to PVD AlN 25. b, Cross-section drawing of XBAW resonator using Al0.72 Sc0.28 N piezo 236. c, Bode Q plot of XBAW resonator 234. d, Measured frequency response of an XBAW diplexer 235. Reprinted with permission from ref. 236 ©2022, IEEE. |
Hybrid
From the above sections, it can be seen that traditional fil- ter technologies, either LTCC/IPD or acoustic, are not able to meet the new requirements set forth by 5G, Wi-Fi 7 and beyond. To realize high frequency, wide bandwidth and sharp roll-offat the same time, a hybrid filter design approach based on the combination of IPD and acoustic tech- nologies was proposed by C. Zuo et al. in 2019 3. In general, the hybrid design adopted IPD LC components to form a wide passband, which can be almost as much bandwidth (> 2 GHz at the center frequency of 4 GHz) as needed, while acoustic resonators can be inserted into the circuit to cre- ate sharp notches so as to give the fast roll-offand high rejection at the neighboring band.
It should be noted that a hybrid filter is not simply putting an IPD chip and an acoustic resonator chip connected together. Instead, the co-design between LC components and acoustic resonators should be conducted at the circuit schematic level, so that the electromagnetic and acoustic components, based on two drastically different physical principles, would cooperate with each other from the impedance point of view, as shown in Fig. 34. Consequently, one coherent band-pass filter characteristic is formed and exhibited at the input and output ports of the whole circuit.
Fig. 34 a, The basic circuit schematic and b, layout drawing of the hybrid filter using IPD, acoustic and substrate technologies. Reprinted with permission from ref. 3. ©2019, IEEE. |
The Band n77 filter, as shown in Fig. 35 a, was used as an extreme ex- ample to show that a hybrid design could achieve 900 MHz wide passband (3.3 to 4.2 GHz) with a low insertion loss of less than 2.5 dB, while the close-in rejection was 36 dB at 4.4 GHz (Band n79), only 200 MHz away from the passband, indicating the co-existence of wideband and sharp roll-off3. Fig. 35 b shows another hybrid design which was also used to broaden the bandwidth of a Band n41 filter while the composing FBAR resonators had limited k2 of only 6.5% 237. It was claimed that hybrid fil-ters should generally exhibit better performance than acoustic-only filters in terms of wide bandwidth, insertion loss, rejection, linearity and power handling capability.
Fig. 35 a, Simulated transmission response of a hybrid Band n77 filer showing wide passband of 900 MHz, low insertion loss and high rejection 3. b, Performance comparison between two Band n41 filters with FBAR alone and FBAR + IPD hybrid. (Red line: FBAR alone; green line: hybrid FBAR and IPD) 237. Reprinted with permission from refs. 3,237 ©2019, 2022, IEEE. |
CONCLUSION
As a core component in RF front-end circuits, filters are ubiquitous in base stations, cellphones, automobiles, electronics and radar. In the his- tory of RF filter, many organizations all over the world have made sig- nificant contributions for the development of wireless communication: University of California Berkeley, University of Pennsylvania, Chiba Uni- versity, Carnegie Mellon University, University of Illinois at Urbana- Champaign, University of Texas at Austin, Tsinghua University, Tianjin University, University of Electronic Science and Technology of China, Shanghai Institute of Microsystem and Information Technology, Shang- hai Jiao Tong University, Zhejiang University, Wuhan University, National Center for Nanoscience and Technology, South China University of Tech- nology, Hong Kong University of Science and Technology, University of Science and Technology of China, and companies like Murata, TDK, Sky- works, Broadcom (Avago), Qualcomm (EPCOS), Qorvo, etc.
This review delves into the history of invention, fundamental working principle, typical application and future trend of various filter technolo- gies, organized according to the timeline of their introduction throughout the evolution of wireless communication generations (1G to 5G). The pro- liferation of smartphones and the expansion of video content services have accelerated the growth of mobile data traffic and the demand for high- speed data rates. The next phase of 5G evolution presents new challenges, such as higher frequency, wider bandwidth, smaller size and sharper roll- off(high rejection). Several emerging filter technologies have been pro- posed to address these challenges, including IHP-SAW, suspended Lamb wave devices based on single-crystal piezoelectric thin films and hybrid filters, among others. Once upon a time, it was believed that acoustic devices could only work up to 10 GHz, which has been proven to be wrong by the demonstration of acoustic resonators at mmWave frequen- cies. Innovations will not cease, as wireless communication will continue to evolve towards 6G, satellite internet and beyond.
MISCELLANEA
Acknowledgments This work was supported in part by the National Natural Science Foundation of China 62231023, in part by the USTC Center for Micro and Nanoscale Research and Fabrication, in part by the USTC Institute of Advanced Technology, in part by the CAS Key Laboratory of Wireless-Optical Communications and in part by the Hong Kong Research Grant Council under Grant 26202122.
Declaration of Competing Interest The authors declare no conflicts of in- terest.
