Niobate-on-Niobate Resonators with Aluminum Electrodes

The LN-on-LN LVR devices that we have developed feature an anchor-supported suspended piezoelectric thin film. A 0.9 $\mathrm{\mu m}$ thick stress free X-cut lithium niobate thin film is created on top of 1 $\mathrm{\mu m}$ oxide buffer layer. Oxide layer also serves as sacrificial layer to release the piezoelectric film. The suspended structure is designed with a length of L = 70 $\mathrm{\mu m}$ and a width of W = 44 $\mathrm{\mu m}$. This is integrated with meticulously positioned aluminum interdigital transducers (IDTs), aligned in a set of specific in-plane orientations to stimulate various acoustic modes. For symmetric configuration, IDTs consist of M = 3, 5, 7, 9, 11 electrodes, while for anti-symmetric configuration, IDTs contain M = 10 electrodes. The symmetric and anti-symmetic configuration will affect the coupling of different mode families respectively. Pitch $\mathrm{W_{p}}$ is chosen to be 4 $\mathrm{\mu m}$ and metallization ratio is set to be 50%. Compared with gold, which is notorious for its high internal friction loss, aluminum offers a lower intrinsic loss and a superior acoustic impedance match to lithium niobate. As a result, it has seen extensive use in LN-on-Si devices[5361520, 8439516]. With novel fabrication technology, aluminum IDTs are introduced to our LN-on-LN platform.

We carried out 3D finite element analysis using COMSOL on lithium niobate resonators with varying orientations. For SAW wave devices, the frequency response has a sinc dependce on IDT pitch[7005450]. In our case, where lamb waves are launched within a laterally vibrating resonator, the frequency response is complicated by the Fabry-Perot (FP) cavity formed by the vertical sidewall. The mechanical boundary conditions of FP cavity enforces more constraints, where the $\mathrm{N^{th}}$ order lamb wave with a wavelength $\mathrm{\lambda}$ satisfying $\mathrm{N\times\lambda/2=W}$ will emerge. However, only the waves with best match to IDT configuration have good coupling and high Q. Through simulations, it has been deduced that escalating the number of IDT fingers (denoted as M) facilitates enhanced coupling of the resonator. Upon reaching a state where the product of M and the IDT pitch ($\mathrm{W_{p}}$) is equal to the overall device width (W), an optimal coupling scenario is realized. Consequently, the wave with an order defined by (N = M - 1) exhibits the highest degree of coupling efficiency in the system. Notably, with varied orientation of IDTs, two types of lamb wave modes, SH0 and S0 , are detected in the simulation. When IDTs are aligned +9 degree relative to the +z axis of X-cut LN thin film, where “+” sign denotes counterclockwise rotation, the coupling of SH0 modes is optimized. The resultant acoustic waves travel at a +9 degree angle to the +y axis with SH0 modes of different orders spanning 300 MHz to 500 MHz. As illustrated in Fig. 3, SH0 modes predominantly feature $\mathrm{S_{23}}$ stress field component, signifying that $\mathrm{d_{15}}$ and $\mathrm{d_{26}}$ piezo coefficients are playing substantial roles in the excitation of SH0 modes. In comparison, when IDTs are aligned at a -30 degree to the +z axis (a “-” sign signifying clockwise rotation), different orders of S0 modes materialize between 700 MHz and 800 MHz with maximized coupling. In the case of S0 mode, the $\mathrm{S_{22}}$ component dominates stress field, attributed to $\mathrm{d_{22}}$ piezo coefficient.

The preparation of LN-on-LN sample entails precise alignment and bonding of lithium niobate samples with 1 $\mathrm{\mu m}$ oxide buffer layer to alleviate stress. The upper niobate layer is subsequently thinned to 0.9 $\mathrm{\mu m}$ using mechanical polishing. Following the sample preparation, we deposit 100 nm thick aluminum electrodes through evaporation and lift-off procedure. The ensuing step involves patterning the LN structure. Conventionally, lithium niobate is defined via fluorine-based reactive ion etching (RIE). This tends to result in sloppy sidewall, rough etching surface and significant redeposition of $\mathrm{LiF_{2}}$[6381492, benchabane2009highly]. To circumvent these issues, we utilize an ion mill procedure[7005450, wang2016multi] under argon plasma. The argon ions are directed onto our sample at a steep angle at 14° to etch through the niobate layer at a rate of 56 nm/min. A subsequent 70° shallow angle ion mill is performed to clean up the sample and remove any redepositted materials left by ion mill process. After completing the ion milling process, the sample is immersed in acetone, and a 5-minute sonication is carried out to remove the photoresist.

The pivotal step in our process flow is the protection of aluminum IDTs while releasing the niobate structure. We opt for vapor HF technology over traditional buffered oxide etch (BOE) technology, largely due to the incompatibility between aluminum IDTs and BOE solution. The latter can easily penetrate the mask through the pinholes in the photoresist layers, leading to erosion of the aluminum IDTs. Conversely, vapor HF does not have the capacity to etch aluminum in the absence of water, making it a suitable candidate for processes involving aluminum IDTs. Nevertheless, during our experiments, we observed that moisture could accumulate on the sample surface, exacerbating the etching effect of vapor HF when aluminum IDTs were exposed. Therefore, it is imperative to employ effective masking strategies to mitigate moisture accumulation. Prior to spinning the photoresist mask layer, the surface is treated with Surpass 4000 resist, which activates the surface and enhances adhesion. The presence of Surpass 4000 adhesion is crucial. It maintains the integrity of photoresist layer under vapor HF ambience. Without adhesion promotion, vapor HF tends to yield highly undesirable bubbling or leakages inside the photoresist layer, where HF and moisture accumulate and cause damages to alumiunum IDTs. Surpass 4000 can also ensure better mask coverage on the sidewall. Once the surface is treated, we then pattern a 7 $\mathrm{\mu m}$ photoresist layer on treated surface, followed by a hard bake at 110 °C. This is performed to reflow the photoresist, removing moisture and cure pinholes within the layer.

Once the patterning of photoresist is completed, the sample is transferred to an electrostatic holding chuck within a commercial HF-vapor etcher. In this study, the release windows are intentionally designed to have an asymmetric configuration. Previous attempts employing symmetric release windows resulted in partial release of the structures, leaving unetched pillars at the central region. This is attributed to the oxide from the peripheral areas consuming substantial amounts of HF, consequently impacting the dynamics of the HF flow. At the symmetric point, i.e., the center, the HF flow is diminished, leading to a reduced etch rate. To address this challenge, the mask windows are crafted with an asymmetrical layout - one side features a circular shape while the opposite side is designed as an elongated stripe. This asymmetry creates an imbalance in HF pressure, thereby directing the HF flow across the center, ensuring that the etch rate remains consistent throughout the structure. The vapor HF etch is carried out at a temperature 15 °C above room temperature (in our case, T = 315K) for 23 minutes, a condition intentionally set to achieve an optimal etch rate and minimal moisture concentration in the photoresist. After the vapor HF etch, the sample is carefully immersed in acetone and IPA to dissolve the photoresist layer. Finally, a critical point drying is performed. The IPA solution is replaced with liquid carbon dioxide, allowing the sample to dry while preserving the suspension of the lithium niobate structure.

The one-port scattering parameters of resonators are assessed using a network analyzer (Agilent PNA-L N5230A). Parasitic pad feed-through capacitance is canceled through de-embedding process[pozar2011microwave]. The scanning range is set at 700 MHz, with 20,000 sampling points. A resolution bandwidth of 10 KHz is employed, and the input power is maintained at -12 dBm. All measurements are conducted in atmosphere. Measured S parameters are transformed into admittance ($\mathrm{Y_{11}}$). The Q is computed using the method decribed in [4803696]. And effective coupling $\mathrm{k_{eff}^{2}}$ is calculated as: $\mathrm{k_{eff}^{2}=2\times(f_{p}-f_{s})/f_{s}}$, where $\mathrm{f_{p}}$ represents parallel resonance and $\mathrm{f_{s}}$ represents series resonance.

When IDTs are aligned from +0° to +40°, sheer horizontal modes (SH0) are detected. Different orders of SH0 modes emerge at varied frequencies between 350 MHz and 600 MHz. With 11-finger IDTs equipped, the $\mathrm{10^{th}}$ order SH0 modes typically exhibit higher effective coupling $\mathrm{k_{eff}^{2}}$ and Q. The edges are set at the zero points of displacement field, so the FP boundary conditions for best coupling SH0 modes are fulfilled. As shown in Fig. 8, the maximum product of $\mathrm{k_{eff}^{2}\times Q}$ is observed when the IDTs are aligned at +9°, where $\mathrm{10^{th}}$ order SH0 mode is found at 418 MHz. Q is 2117 and $\mathrm{k_{e}ff^{2}}$ reaches 13.9%. Thus the figure of merit (FOM) $\mathrm{k_{eff}^{2}\times Q}$ is computed to be 294, one of the highest on LN-on-LN platform. The response of $\mathrm{10^{th}}$ order SH0 mode is analyzed using Butterworth-Van Dyke (BVD) model illustrated in Fig. 8 (a). The extracted parameters $\mathrm{C_{0}}$, $\mathrm{C_{x}}$, $\mathrm{L_{x}}$ and $\mathrm{R_{x}}$ are listed in the Table I. The measurement results of $\mathrm{10^{th}}$ order SH0 agree well with finite element simulation.

The substantial influence of the frequency response of SH0 modes on IDT orientation is confirmed experimentally. As the IDT fingers move away from +9° to +0°, +20°, and +30°, we observe a decrease in coupling and an increase in resonant frequency, as shown in Fig. 9. These findings align with our simulation results and can be attributed to the anisotropy present in the LN piezoelectric thin film. As the IDTs’ alignment deviates significantly from +9°, the SH0 strength becomes less pronounced. However, maximum coupling of S0 mode is achieved when IDTs are aligned from -30° to -40°, as demonstrated in Fig. 10. At a -30° IDT orientation, the most prominent S0 mode emerges at 719 MHz with an effective coupling of 10.1% ($\mathrm{k_{eff}^{2}}$) and Q = 891. At a -40° IDT orientation, the most noticeable S0 mode appears at 713 MHz with 8.0% $\mathrm{k_{eff}^{2}}$ coupling and Q = 1344. We analyzed the response of S0 modes using the BVD model and compared the results with COMSOL finite element simulations. The extracted parameters are provided in Table II. It is worth noting that S0 modes are highly sensitive to the resonators’ boundary conditions[7005450]. Maximum coupling and Q can only be obtained when the edge of resonator is positioned at the stress field nodes, which is different from the boundary conditioned required by SH0 modes. Fabrication errors could affect the performance of the S0 mode resonator. In our case, although the resonant frequency aligns with the simulation results, the coupling coefficient decreases because the aforementioned FP boundary conditions are not fulfilled for S0 modes. Even slight mismatch between nodes of S0 mode and IDT finger configuration could impact the coupling.

Despite aluminum’s low inherent loss and superior acoustic impedance compatibility with LN, it offers lower conductivity compared to unreactive metals such as gold. To delve deeper into the impact of electrode loading on Q, we fabricated identical S0/SH0 mode resonator structures, maintaining the same dimensions but varying the number of IDT fingers. Our measurements suggest that S0 modes are more vulnerable to electrode loading. As shown in Fig. 11, when the IDT fingers are aligned at -30° relative to the z-axis, reducing the number of IDT fingers from 11 to 3 leads to a decrease in energy coupled to the acoustic regime from RF regime, which is represented by the effective coupling $\mathrm{k_{eff}^{2}}$ dropping from 8.6% to 0.6%, while Q elevates from 801 to over 2400. This implies a significant energy loss in S0 modes due to electrode loading. In contrast, for the SH0 mode resonator, electrode loading has a lesser impact. With a +9° IDT orientation and an increase in the number of IDTs from 3 to 11, we observe an increase in $\mathrm{k_{eff}^{2}}$ while Q remains consistently high, exceeding 1500.

We examined the temperature coefficient (TCF) of LN resonators of various orientations in their fundamental mode by incrementally adjusting the temperature from 300K to 370K in steps of 5K. Each measurement was conducted after allowing a 5-minute stabilization period at the desired temperature. The results, depicted in Fig. 12, reveal that both SH0 and S0 modes display a highly linear and stable TCF. The TCF for SH0 modes, recorded at -101 ppm/K, is slightly higher than that for S0 modes, which stands at -65.7 ppm/K. These TCF values align with those observed in uncompensated LN SAW devices[4803491]. Compensation for temperature can be achieved by incorporating an additional $\mathrm{SiO_{2}}$ layer, as explored in previous studies[6782639, 7863565, 7329342].

We conducted an empirical investigation into the nonlinearity behavior of LN resonators by measuring the frequency response across various power levels. An S0 mode resonator with 11 IDT orientations -30° from the +z axis, and dimensions of 70 $\mathrm{\mu m\times}$ 44 $\mathrm{\mu m}$, was subjected to a frequency sweep from 680MHz to 790MHz at input power levels ranging from -12 dBm to 10 dBm, incremented by 1 dBm. The device operated within its linear region, as evidenced by a Lorentzian-shaped frequency response near resonance when the input power was below -1 dBm. As the input power increased beyond this level, a progressively larger resonant frequency shift and distortion of frequency response became evident, signifying the transition of the resonator into its non-linear regime. A bifurcation point was identified at 6 dBm, marking the threshold of instability.

We investigated the power handling capacity of our LN-on-LN resonator devices by conducting a frequency sweep at varying temperature ranges from 300K to 370K and power levels spanning from -12 dBm to 6 dBm. Resonant frequencies of our LN-on-LN devices under these varied conditions are displayed in Fig. 13. It is discernible that for power levels below 0 dBm, the shift in resonant frequencies remains minimal. However, upon exceeding a power level of 0 dBm, substantial frequency shifts are observable, even though the temperature response maintains its linearity. Further study of the TCF at various power levels, depicted in Fig. 14, shows that the TCF variation remains under 0.1 ppm/K for power levels less than 0 dBm, suggesting a stable temperature response. Contrastingly, SAW LN-on-Si devices exhibited significant TCF alterations even at lower power levels [9385101]. We attribute the stability of our measured TCF in relation to input power to the minimization of residue stress in the LN-on-LN thin film resulting from thermal cycling induced by power in the matched substrate.