Transferred P3F: LiNbO₃ & AlScN Devices

• Transferred P3F is a domain-engineered piezoelectric film that alternates crystal orientations to boost electromechanical coupling while minimizing charge cancellation.
• The approach involves bonding and transferring thin LiNbO₃ or AlScN layers with reversed polarities, enabling efficient mode matching and higher-order operation.
• Key results include high Q factors at mmWave and lower frequencies, compact device footprints, and improved thermal and mechanical stability through optimized fabrication techniques.

Searching arXiv for papers on transferred periodically poled piezoelectric film (P3F), especially LiNbO₃/AlScN resonators, filters, and PMUTs. Transferred periodically poled piezoelectric film (P3F) denotes a transfer-enabled piezoelectric stack in which the effective piezoelectric tensor changes sign from layer to layer, typically by bonding or transferring thin films of alternating crystal orientation, and in some implementations by electrically inverting ferroelectric domains before or after transfer. In thin-film lithium niobate (LiNbO$_3$), P3F has been realized as bi-layer, tri-layer, and four-layer stacks on amorphous silicon (a-Si)/sapphire, Si, or cavity substrates, with the explicit goal of enabling efficient higher-order Lamb or thickness-extensional operation while retaining high quality factor $Q$, high electromechanical coupling $k^2$, and compact footprint (Kramer et al., 2023, Cho et al., 2024, Barrera et al., 27 Jun 2025). Related transferred P3F embodiments have also been reported in X-cut LiNbO$_3$ PMUTs, 36° Y-cut LiNbO$_3$ power resonators, and 4-layer AlScN bulk acoustic wave devices, indicating that P3F is best understood as a domain-engineered material platform rather than a single device topology (Yao et al., 13 Aug 2025, Izhar et al., 2024, Chulukhadze et al., 8 Dec 2025).

1. Structural concept and stack archetypes

The canonical LiNbO$_3$ P3F structure in the millimeter-wave literature is a bilayer 128° Y-cut stack on a-Si on sapphire. In the 50.74 GHz resonator, the two LiNbO$_3$ layers have measured thicknesses of approximately $105$ nm and $80$ nm, each layer acts as a single ferroelectric domain of opposite polarity, and the effective domain period along thickness is $\Lambda = t_1 + t_2 \approx 185$ nm (Kramer et al., 2023). The two layers are bonded with a 180° rotation about the axis in the intersection of the LiNbO$Q$0 128°-cut plane and the YZ-plane, which inverts the spontaneous polarization between adjacent layers (Kramer et al., 2023).

Subsequent LiNbO$Q$1 implementations generalized this bilayer concept to tri-layer and four-layer stacks. The 19.3 GHz tri-layer filter uses three individually bonded 128° Y-cut films, each approximately $Q$2 nm thick, in a $Q$3 sequence, for a total LiNbO$Q$4 thickness of $Q$5 nm on a $Q$6 a-Si sacrificial layer over sapphire (Barrera et al., 27 Jun 2025). The 50 GHz filter uses a four-layer 128° Y-cut P3F stack with layer thicknesses measured by STEM as $Q$7 nm, $Q$8 nm, $Q$9 nm, and $k^2$0 nm, for an overall thickness of $k^2$1 nm, with the through-thickness periodicity given as $k^2$2 (Barrera et al., 27 Jun 2025).

Transferred P3F is not restricted to sub-200 nm LiNbO$k^2$3. In power conversion, two 36° Y-cut LiNbO$k^2$4 wafers of opposite in-plane polarity, each $k^2$5 thick, are bonded into a $k^2$6 stack (Yao et al., 13 Aug 2025). In PMUTs, two X-cut LiNbO$k^2$7 wafers are thinned to approximately $k^2$8 each and directly bonded after a 180° in-plane rotation of the top wafer, forming a bimorph P3F LN active layer without intermediate electrodes (Chulukhadze et al., 8 Dec 2025). A related PMUT prototype uses two $k^2$9-thick X-cut LN plates separated by a $_3$0 nm SiO$_3$1 interlayer (Yao et al., 30 Aug 2025).

A recurring ambiguity in the literature is the meaning of the “period.” In bilayer and multilayer mm-wave LiNbO$_3$2 P3F, the periodicity is through-thickness and set by the stacked layer sequence (Kramer et al., 2023, Barrera et al., 27 Jun 2025). In the A$_3$3 resonator, by contrast, the chosen poling period is $_3$4, matching the acoustic wavelength $_3$5 (Barrera et al., 2023). This suggests that P3F nomenclature spans both through-thickness alternation and wavelength-matched ferroelectric patterning, provided the transfer or bonding process preserves complementary orientation.

2. Transfer, bonding, and poling implementations

The fabrication literature reports multiple P3F realization routes. In the bilayer 128° Y-cut LiNbO$_3$6 platform, NGK Insulators Ltd. forms and poled each LiNbO$_3$7 layer, deposits them onto an a-Si/sapphire substrate, and performs a direct (oxide-free) bond or adhesive-assisted bond; post-bond anneal in the typical range $_3$8–$_3$9 improves bond strength, and CMP or ion milling thins the LiNbO$_3$0 to the target $_3$1–$_3$2 nm thickness per layer (Kramer et al., 2023). Device definition then proceeds by photolithography and ion milling for mesas and release windows, e-beam lithography and $_3$3 nm Al evaporation for IDTs, and XeF$_3$4 isotropic etch of sacrificial a-Si to suspend the film (Kramer et al., 2023).

The 23.8 GHz filter extends this transferred P3F flow by local thickness trimming. Starting from a bi-layer 128° Y-cut LiNbO$_3$5 film of total thickness approximately $_3$6 nm on $_3$7 a-Si/sapphire, ion-beam-assisted Ar cluster trimming locally thins the LiNbO$_3$8 to $_3$9 nm top/$_3$0 nm bottom for series resonators and $_3$1 nm/$_3$2 nm for shunt resonators, after which $_3$3 nm Al is evaporated for IDTs and an additional $_3$4 nm Al is added over routing lines and probe pads (Cho et al., 2024). The structures are then released by XeF$_3$5 etching of Si in the release windows (Cho et al., 2024).

Other LiNbO$_3$6 embodiments use explicit sequential transfer. The 50 GHz four-layer device is fabricated by a smart-cut or ion-implantation/wafer-bonding process in which a bulk 128° Y-cut LiNbO$_3$7 wafer is implanted with H$_3$8 or He$_3$9, directly bonded to polished a-Si/sapphire, split by annealing, and then planarized by CMP and low-damage ion-beam smoothing; each exfoliated layer is rotated 180° before bonding the next layer, and no electric-field domain inversion is used (Barrera et al., 27 Jun 2025). The 19.3 GHz tri-layer device uses repeated thermo-compression bonding, donor-handle removal by grinding/polishing, sacrificial a-Si deposition, and final dry removal of all sacrificial a-Si before IDT patterning and local trimming (Barrera et al., 27 Jun 2025).

A distinct route is electrical periodic poling before transfer. In the A$_3$0 resonator, a bulk X-cut LiNbO$_3$1 wafer is periodically poled by electric field pulses through lithographically patterned electrodes; with coercive field $_3$2, a $_3$3-thick wafer uses approximately $_3$4–$_3$5 kV per pulse, after which the poled wafer is bonded to a host silicon carrier and thinned to a final total thickness of $_3$6 (Barrera et al., 2023).

A useful corrective point follows from these reports: “periodically poled” does not always denote on-chip high-voltage domain inversion. In several transferred LiNbO$_3$7 and PMUT demonstrations, periodic poling is realized by bilayer or multilayer stacking with alternating orientation rather than by post-transfer electrical switching (Kramer et al., 2023, Chulukhadze et al., 8 Dec 2025).

3. Electromechanical basis and mode selectivity

The electromechanical description of P3F devices follows the standard coupled acoustic-piezoelectric relations,

$_3$8

with the critical modification that $_3$9 alternates in sign from layer to layer (Kramer et al., 2023). In a uniform film, $105$0, whereas in a periodically poled bilayer a more general form is

$105$1

which makes explicit that coupling is controlled by the overlap of strain, electric field, and the sign-alternating piezoelectric tensor (Kramer et al., 2023).

Across reported devices, the central P3F mechanism is cancellation avoidance. In the bilayer 128° Y-cut LiNbO$105$2 platform, periodic inversion cancels the coupling reduction that normally accompanies alternating displacement in higher-order symmetric modes, allowing the device to maintain the high intrinsic $105$3 of the fundamental A$105$4 mode even when operating at S$105$5 or S$105$6 (Kramer et al., 2023). In the 36° Y-cut power-conversion resonator, inversion of $105$7 in alternating layers restores constructive piezoelectric coupling for even-order modes while canceling odd modes (Yao et al., 13 Aug 2025). In the A$105$8 resonator, opposite polarizations mitigate the charge cancellation arising from opposite stress of A$105$9 in the top and bottom piezoelectric layers (Barrera et al., 2023). This suggests that P3F acts as a through-thickness mode-matching scheme for electromechanical overlap.

Parameter extraction in these devices is usually based on resonance splitting or MBVD fitting. Recurrent definitions are

$80$0

with variants such as $80$1 in single-branch MBVD models (Cho et al., 2024, Yao et al., 13 Aug 2025). The multi-layer P3F stack also increases capacitance density: for $80$2 identical poled sublayers,

$80$3

and the 23.8 GHz study reports a measured $80$4 of $80$5 fF for the P3F shunt resonator versus approximately $80$6 fF in a single $80$7 nm film (Cho et al., 2024). The same work states that the multi-layer P3F stack promises smaller footprints and better nonlinearity than single-layer counterparts, thanks to the higher capacitance density and lower thermal resistance (Cho et al., 2024).

4. High-frequency LiNbO$80$8 P3F resonators and filters

Transferred LiNbO$80$9 P3F has been developed most aggressively for FR3, mmWave, and 50 GHz-class filtering. The bilayer resonator reported at $\Lambda = t_1 + t_2 \approx 185$0 GHz achieves $\Lambda = t_1 + t_2 \approx 185$1, $\Lambda = t_1 + t_2 \approx 185$2, and $\Lambda = t_1 + t_2 \approx 185$3, while the same platform shows $\Lambda = t_1 + t_2 \approx 185$4, $\Lambda = t_1 + t_2 \approx 185$5, and $\Lambda = t_1 + t_2 \approx 185$6 for the $\Lambda = t_1 + t_2 \approx 185$7 GHz S$\Lambda = t_1 + t_2 \approx 185$8 tone (Kramer et al., 2023). The same report attributes the $\Lambda = t_1 + t_2 \approx 185$9 value partly to the suspended structure obtained by XeF$Q$00 release, which removes substrate leakage (Kramer et al., 2023).

The first piezoelectric acoustic filter in P3F LiNbO$Q$01 at $Q$02 GHz is a third-order ladder filter implemented with electrically coupled resonators in $Q$03 nm bi-layer P3F 128 rotated Y-cut LiNbO$Q$04, operating in the second-order symmetric Lamb mode (Cho et al., 2024). It reports insertion loss of $Q$05 dB, $Q$06-dB fractional bandwidth of $Q$07, and compact footprint of $Q$08 (Cho et al., 2024).

The tri-layer and lattice-filter reports then broaden the topological space. The first tri-layer P3F LN filter operating at $Q$09 GHz reports low IL of $Q$10 dB, $Q$11-dB FBW of $Q$12, and $Q$13 dB close in rejection, with local top-layer trimming used to activate adjacent modes and generate built-in transmission zeros (Barrera et al., 27 Jun 2025). Lattice XBAR filters in P3F TFLN report $Q$14-dB FBWs of $Q$15 and $Q$16 and low ILs of $Q$17 dB and $Q$18 dB at approximately $Q$19 GHz for the direct and layout-balanced lattice filters, respectively, under conjugate matching, with footprints smaller than $Q$20 (Anusorn et al., 16 Feb 2026).

At the upper end of the reported frequency range, a third-order ladder filter based on twelfth-order symmetric S$Q$21 mode XBARs in a 4-layer P3F 128 Y-cut LiNbO$Q$22 stack reports center frequency $Q$23 GHz, insertion loss $Q$24 dB, $Q$25-dB FBW $Q$26, and footprint $Q$27 (Barrera et al., 27 Jun 2025). The same work states that these results represent the highest frequency acoustic filters reported to date (Barrera et al., 27 Jun 2025).

Implementation	Stack and mode	Reported performance
Bilayer LiNbO$Q$28 resonator	128° Y-cut P3F, S$Q$29 and S$Q$30	$Q$31 GHz: $Q$32, $Q$33; $Q$34 GHz: $Q$35, $Q$36
Third-order ladder filter	$Q$37 nm bi-layer P3F LiNbO$Q$38, S$Q$39	$Q$40 GHz, IL $Q$41 dB, FBW $Q$42, $Q$43
Tri-layer FR3 filter	$Q$44 nm $Q$45 LiNbO$Q$46	$Q$47 GHz, IL $Q$48 dB, FBW $Q$49, close in rejection $Q$50 dB
Lattice XBAR filters	Bi-layer P3F TFLN at $Q$51 GHz	IL $Q$52 dB, FBW $Q$53, $Q$54
Four-layer 50 GHz filter	4-layer P3F LiNbO$Q$55, S$Q$56	$Q$57 GHz, IL $Q$58 dB, FBW $Q$59, $Q$60

Taken together, these reports indicate that transferred LiNbO$Q$61 P3F supports multiple filter architectures—ladder, lattice, and multi-overtone XBAR—while preserving practical footprints and reported operation from FR3 through 50 GHz.

5. Extension beyond mmWave: A$Q$62, power resonators, AlScN BAW, and PMUTs

Transferred P3F is not inherently a mmWave-only platform. At low frequency, the A$Q$63 resonator on X-cut LiNbO$Q$64 reports $Q$65 MHz, $Q$66, $Q$67, and $Q$68, with measured phase velocity approximately $Q$69 and acoustic wavelength $Q$70 (Barrera et al., 2023). The same study compares this to single-layer A$Q$71 on LiNbO$Q$72, where $Q$73, $Q$74–$Q$75, and $Q$76 (Barrera et al., 2023).

For piezoelectric power conversion, the first P3F thickness-extensional LN resonator operates at $Q$77 MHz with $Q$78, $Q$79, $Q$80, and $Q$81 (Yao et al., 13 Aug 2025). Its high-power testing shows that up to approximately $Q$82 dBm, $Q$83 stays within $Q$84 of the small-signal value; from $Q$85 dBm to $Q$86 dBm, $Q$87 deviates up to $Q$88, with permanent shift indicating bonding damage; and full material failure occurs at approximately $Q$89 dBm $Q$90 (Yao et al., 13 Aug 2025).

The P3F idea also extends to nitride BAW technology. A 4-layer AlScN P3F stack, formed by sequential PVD co-sputtering of AlScN and in-situ Al electrodes followed by one-step electric poling of the third nitride layer and Akoustis XBAW film transfer, achieves $Q$91 and $Q$92 at $Q$93 GHz (Izhar et al., 2024). Filters synthesized from these resonators report IL $Q$94 dB and $Q$95 dB, bandwidths $Q$96 MHz and $Q$97 MHz, and in-band IIP$Q$98 values of $Q$99 dBm and $k^2$00 dBm for 3-element and 6-element filters, respectively (Izhar et al., 2024).

In ultrasonic transduction, a transferred P3F X-cut LN PMUT prototype demonstrates an out-of-plane mode near $k^2$01 MHz with electromechanical coupling $k^2$02, and laser Doppler vibrometry shows peak center displacement of $k^2$03 pm/V (Yao et al., 30 Aug 2025). A thicker bimorph PMUT with a mechanically robust $k^2$04 thick P3F LN active layer reports a $k^2$05 kHz flexural mode device with $k^2$06, extracted $k^2$07, and high transmit efficiency of $k^2$08 nm/V, together with stable device operation up to $k^2$09 and survival up to $k^2$10 (Chulukhadze et al., 8 Dec 2025).

6. Residual stress, process limitations, and optimization directions

As transferred P3F films become thinner and more multilayered, mechanical stability becomes a central issue. In 128° Y-cut transferred thin-film lithium niobate, cantilever curvature measurements show normalized gradient stress $k^2$11 ranging from $k^2$12 to $k^2$13 at $k^2$14 nm, $k^2$15 to $k^2$16 at $k^2$17 nm, and $k^2$18 to $k^2$19 at $k^2$20 nm, with stress-free orientations shifting from approximately $k^2$21 at $k^2$22 nm to approximately $k^2$23 for $k^2$24–$k^2$25 nm films (Kim et al., 22 Apr 2026). A $k^2$26 nm bilayer P3F reduces the equivalent normalized $k^2$27 to $k^2$28 to $k^2$29, resulting in significantly reduced deformation (Kim et al., 22 Apr 2026). This establishes bilayer P3F as not only an electromechanical strategy but also a stress-compensation strategy.

The dominant reported non-idealities are thickness non-uniformity, bonding misalignment, and process-induced damage. In the bilayer mm-wave resonator, TEM shows $k^2$30–$k^2$31 nm thickness variation, linked to spurious modes (Kramer et al., 2023). In the tri-layer FR3 filter and four-layer 50 GHz filter, maintaining uniform film thickness below $k^2$32 nm is explicitly identified as a fabrication challenge, and the 50 GHz report notes that thickness non-uniformity introduces spurious adjacent modes such as A$k^2$33 and A$k^2$34 (Barrera et al., 27 Jun 2025, Barrera et al., 27 Jun 2025). The four-layer 50 GHz device also reports bonding misalignment of $k^2$35 (Barrera et al., 27 Jun 2025). In the lattice-filter work, domain-wall roughness, thickness non-uniformity, ion-milling damage, thick bus-line lift-off, and EM resonances in the interconnect are all identified as limits on $k^2$36, skirt shape, and out-of-band rejection (Anusorn et al., 16 Feb 2026).

The published optimization agenda is correspondingly concrete: tighter control of film thickness to further raise $k^2$37 and suppress spurs; optimizing IDT apodization and pitch to maximize overlap with the desired mode while rejecting spurious modes; refining poling electrode design and waveform; using low-damage thinning methods such as CMP over ion-mill; exploring direct oxide bonding rather than a-Si adhesive to reduce interfacial losses; adding more alternating layers to access higher-order modes; and integrating temperature compensation layers or monolithic electronics (Kramer et al., 2023, Anusorn et al., 16 Feb 2026, Barrera et al., 27 Jun 2025). A plausible implication is that transferred P3F has matured from a mode-enabling trick into a broader co-design framework linking crystal orientation, stack thickness, bonding interface, residual stress, and circuit topology.