I. Introduction
In recent times, surface acoustic wave (SAW) filters have been extensively employed in miniaturized microwave integrated systems due to their low manufacturing cost, good performance, and design flexibility [1]. They are especially used in next-generation radio frequency (RF) mobile systems, where not only must a very large number of bandpass filters (BPFs) be integrated into a small die area but also the performance of these BPFs plays a critical role in overall system performance. As a result, the design optimization of BPFs has emerged as a key RF system design concern [2, 3].
By applying an electric signal to the surface of a piezoelectric material, a SAW interdigital transducer (IDT) efficiently confines the acoustic energy between reflectors, forming standing waves that result in resonance. However, most horizontal and vertical acoustic waves do not contribute to this resonance, and may leak out or cause undesirable spurious resonances [4]. Moreover, the viscoelastic attenuation of longitudinal waves can be another source of energy loss. Since a BPF is composed of numerous SAW IDTs, inaccurate estimation of the loss of a BPF may lead to catastrophic design failures, as shown in Fig. 1. Thus, the accurate characterization and modeling of SAW resonators is an integral part of BPF design. As a result, energy loss mechanisms within SAW IDTs have been thoroughly investigated to improve circuit performance [5, 6]. To mitigate acoustic leakage, new process technologies [7, 8] and modeling techniques [9–13] have been developed.
Once a process technology is established, modeling for design optimization is one of the most important tasks from the circuit designer’s perspective. The simplest technique is to represent energy loss through the use of lumped resistance, such as a modified Butterworth-Van Dyke (mBVD) model [9]. However, this model is too simple to be accurate. A SAW IDT can be modeled more accurately using a four-port transmission line (TL) model [10] or a coupling-of-modes (COM) model [11, 12]. The COM model is particularly effective in describing multiple modes and coupling effects along the propagation direction [13]. However, its primary limitation lies in its difficulty in directly incorporating loss mechanisms [11]. In contrast, the four-port TL model can more realistically and efficiently describe the physical behavior of a SAW IDT. Nonetheless, although the four-port TL model significantly improves the accuracy of a SAW resonator loss model, the brute force TL model, which uses constant resistance and conductance, imposes some inherent limitations with regard to accuracy. The energy loss within SAW IDTs is frequency-variant. This frequency dependency has a decisive effect on the skirt characteristics of a BPF [14, 15]. Furthermore, spurious resonances must also be carefully taken into account since they may have a significant effect on the neighboring bands of the BPF.
II. Physical Modeling of SAW IDT
1. Mathematical Formulation of SAW IDT
The physical behavior of surface acoustic wave propagation can be mathematically formulated in terms of stress T(z, t) and particle velocity v(z, t) [16–18] as follows:
where
Here, the acoustic phase velocity of the piezoelectric material in the presence of an electric field is indicated by vo, whereas va denotes the material’s acoustic phase velocity without additional stress due to the electric field [17]. Furthermore, c, ρ, and k2 refer to stiffness tensor, mass density, and the electromechanical coupling coefficient, respectively
2. Circuit Representation of SAW IDT
The physical behavior of a SAW device can be explained by the propagation and interaction of acoustic waves and electromagnetic waves. Theses physical phenomena can be well modeled using a four-port distributed TL circuit (i.e., two ports for electromagnetic waves and two ports for acoustic waves). As shown in Fig. 2, the unit cell for the four-port distributed circuit comprises three cascaded subnetworks [10, 18]. Furthermore, a set of parameters is necessary for the four-port TL circuit model, defined as
ℳ ≡ { f r , f a , C o , v o , v m , k 2 , Z c A , r i } . Notably, the resonant frequency (fr), anti-resonant frequency (fa), and finger capacitance (Co) can be directly determined from the measured S-parameters. In contrast, the acoustic wave phase velocity of the SAW IDT without loading effect (vo) and with loading effect (vm), the electromechanical coupling coefficient (k2), and the turns ratio (ri) have to be determined in an iterative manner. The network equations for the i-th cell can be expressed as follows:
where the system matrix [Ui ] can be readily determined by cascading the three subnetworks, as follows:
Assuming a lossless system, the components of the submatrix can be readily determined using the following equation [10]:
where x ∈ {left, center, right}, Sh ≡ sinh(θA), and Ch ≡ cosh(θA). Moreover, the factor ζ ∈ {0,1} is 1 for stiffened parameters; otherwise, it is 0. The acoustic propagation constant and acoustic characteristic impedance can be defined as follows:
Furthermore, the turns ratio for an ideal transformer model can be represented using the following expression:
In Eqs. (7) and (8), the bar indicates the stiffened parameters resulting from the additional stress caused by the IDT finger metals. Thus, the acoustic wave equation interacting with electromagnetic waves can be solved without directly performing numerical calculations of the differential equations, by exploiting a more versatile circuit theory.
III. Lossy Acoustic Wave TL Model
Traditionally, in a TL circuit model, the physical behavior associated with electromagnetic wave propagation can be expressed based on the characteristic impedance and propagation constant. They can be represented using the following per-unit-length circuit model parameters: series resistance (R) and inductance (L), parallel conductance (G) and capacitance (C). Therefore, the loss of an electromagnetic wave can be considered a real part of the propagation constant, as follows:
However, the lossy mechanisms of acoustic waves within SAW IDTs are completely different from those of electromagnetic waves. In SAW IDTs, the main source of energy loss is leaky waves. Longitudinal waves may also lose energy during propagation due to viscoelastic damping [16]. Therefore, it is necessary to test whether the TL model is suitable for evaluating the physical behavior related to the loss mechanism of acoustic wave propagation. To investigate acoustic wave energy loss, we tested the lossy circuit model presented in Fig. 2. This test, which involved a single-mode resonator, allowed us to readily determine the lossy model parameters that matched the measurement data.
As shown in Fig. 3, there are three representative loss model parameters: resistive loss (
R o i ), conduction loss (
G o i ), and attenuation (
α o i ). In the ensuing subsections, these parameters are investigated in more detail.
1. Generic Loss Model using Resistance
First, we investigated the energy loss associated with acoustic wave propagation by varying the series resistance (
R o i ) without considering the other model parameters. As evident from Fig. 4, the energy loss near the resonant frequency (fr) can be accurately represented only by the series resistance. Therefore, a suitable resistance
R o i ( p ) corresponding to a given grating periodic length (p) that matches the measurement data can be readily determined in an iterative manner, as follows:
Here, the arrow indicates “to set the
R o i to
R o i ( p ) .” The physical phenomena beyond the resonant frequency (fr), however, cannot be explained by the series resistance alone, as shown in Fig. 4.
2. Frequency-Variant Conductance Model
The parallel conductance (
G o i ) variation is strongly correlated with energy loss near the anti-resonant frequency (fa). As discussed in the previous subsection, the series resistance (
R o i ) has little effect at high frequencies above fr, whereas the parallel conductance (
G o i ) has only a negligible effect at low frequencies below fr. Therefore, acoustic wave loss from fr to fx can be more accurately explained by the frequency-variant conductance,
G o i ( f , p ) , determined as follows:
where G′(f, p) can be approximately expressed as:
where G(fa, p) is determined in an iterative manner:
As shown in Fig. 5, the lossy TL circuit model that features a combination of the series resistance and the frequency-variant parallel conductance can more accurately represent the propagation loss associated with acoustic waves in the broad frequency band (i.e., up to frequency higher than fx).
3. Frequency-Variant Damping Model
At high frequencies, spurious resonances and unexpected spikes arise due to interactions between the IDT fingers and various types of acoustic waves. Unfortunately, the lossy TL circuit model cannot explain all the loss mechanisms at very high frequencies. As shown in Fig. 6, spurious resonances of a SAW resonator may appear within the passband of the wideband BPF. Notably, the spurious resonance of individual IDTs may lead to significant design errors associated with insertion loss.
Therefore, to accurately reflect spurious resonances, a frequency-variant damping factor must be considered. To address this, going back to Eq. (7), the formulation for the acoustic wave propagation constant can be rewritten as:
The real part of the acoustic wave propagation constant (αA) is concerned with attenuation of the acoustic wave during propagation. The damping phenomenon of spurious resonances and unexpected spikes can be adequately controlled by representing the acoustic wave propagation constant as a frequency-variant model parameter, as follows:
Furthermore, the damping factor can be roughly modeled as a linear function at high frequencies (i.e., f > fa), expressed as follows:
where
Here, α(fa, p) can be determined in an iterative manner, as follows:
As shown in Fig. 7, the frequency-variant damping model effectively suppressed the unrealistic spurious resonances, thus enabling practical implementation.
In summary, the lossy TL circuit model associated with acoustic wave propagation can be accurately modeled using three frequency-variant circuit model parameters. As schematically summarized in Fig. 8, the series resistance can accurately model the loss mechanism at low frequencies, where loss is found to mainly arise from metal fingers. Meanwhile, energy loss due to leakage increases linearly up to a certain frequency (around fx), and then saturates. This leakage phenomenon can be accurately modeled by frequency-variant conductance. Finally, the damping phenomena of spurious resonances can be accurately modeled using the frequency-variant acoustic damping factor.
4. Cascaded Matrix for the TL Circuit Model
To account for all loss mechanisms, the lossless circuit matrix in Eq. (5) had to be modified. The lossless cascade matrix can be expressed as follows:
Upon incorporating series resistance and parallel conductance, the cascaded matrix for the unit cell was modified as follows:
where
χ ≡ [ R o i / ( 1 + ξ 43 R o i ) ] . Here, the superscript R&G refers to the lossy model using both series resistance and parallel conductance.
IV. Experimental Verification
To investigate the physical characteristics of SAW resonators, various test resonators and BPFs were designed. They were fabricated on a piezoelectric substrate composed of 42°YX-LiTaO3 (42°-rotated Y-cut X-propagation LiTaO3). Notably, we acquired the network parameters (S-parameters and capacitances) for the test resonators using wafer-level measurements. For accurate acquisition of the experimental data, parasitic effects (pads and access lines) between the measurement reference plane and the device under test (DUT) were de-embedded using the well-known Y-parameter de-embedding method [19]. The S-parameter data were acquired from 50 MHz to 10 GHz using a vector network analyzer (VNA: Anritsu MS46122A). The VNA was calibrated by employing the short-open-load-thru (SOLT) calibration method and an impedance standard substrate (FormFactor ISS 101–190). The capacitances were measured using an LCR meter (Agilent 4284A Precision LCR Meter). To verify the proposed loss model, we designed a BPF to compare the results obtained using the proposed model with the measured data. Notably, since parasitic capacitances due to layout interconnect lines may have an effect on system performance, they were included in the equivalent circuit. Fig. 9(a) and 9(b) depict the equivalent circuit of the test filter considered for evaluation and a photograph of the BPF chip, respectively. As shown in Fig. 9(c), the results of the calculation conducted using the proposed modeling technique exhibits excellent agreement with the experimental data.
V. Conclusion
SAW IDTs are widely used in RF circuits and system designs. However, they are limited by the inherent leakage and loss that occur during acoustic wave propagation. In this study, we present a modeling technique for addressing the energy loss occurring within a SAW IDT resonator during the propagation of acoustic waves. The physical behavior of surface acoustic waves was mathematically formulated using four-port TL circuits. The losses during the propagation of surface acoustic waves were modeled based on three model parameters: series resistance, frequency-variant parallel conductance, and frequency-variant damping factor. To verify the validity of the proposed model, a test band 20 Rx BPF was designed. The results obtained using the proposed model demonstrated excellent agreement with the measurement data.








