Fully Integrated Pitch-Matched AlScN PMUT-on-CMOS Array for High- Resolution Ultrasound Images

This article presents a fully integrated pitch-matched piezoelectric micromachined ultrasonic transducers (PMUTs)-on-CMOS array with high potential in catheter-based ultrasound imaging systems and capabilities to obtain resolutions under <inline-formula> <tex-math notation="LaTeX">$100 \mu \text{m}$ </tex-math></inline-formula>. The system-on-chip (SoC) consists of <inline-formula> <tex-math notation="LaTeX">${7} \times {7}$ </tex-math></inline-formula> AlScN PMUTs connected in a 1-D configuration where the six external rows are used to generate the acoustic pressure through three high-voltage (HV)-pulser CMOS circuits, and the central row is used to sense the incoming ultrasound wave which will be amplified by a low-noise amplifier (LNA) CMOS amplifier. The experimental verification in a liquid environment gave, as a first result, a peak frequency of 7.7 MHz with a normalized pressure (ST) of 1.98 <inline-formula> <tex-math notation="LaTeX">$\text{k}{\text {Pa}}_{\text {pp}}$ </tex-math></inline-formula>*mm*<inline-formula> <tex-math notation="LaTeX">$\text{V}^{-1}$ </tex-math></inline-formula> and receiving sensitivity (SR) of 3.3 V/MPa, respectively, competitive sensitivities in comparison to the state-of-the-art. In the second part, ultrasonic imaging for different wires with a minimum diameter of <inline-formula> <tex-math notation="LaTeX">$25 \mu \text{m}$ </tex-math></inline-formula> was demonstrated as expected from numerical simulations. The system’s performance for ultrasound images was evaluated considering the product of the area and the resolution at 1 mm, giving competitive values compared with other reported ultrasound systems for catheter-based medical ultrasound imaging and the only one providing monolithic integration with the CMOS front-end circuitry.

The fast development of microelectro-mechanical systems (MEMS) technology has aroused significant interest for this catheter-based ultrasound imaging systems.While ultrasound-based medical diagnosis has proven invaluable in detecting and evaluating various diseases, the advancement of MEMS technology holds particular promise in enhancing their capabilities, specifically providing robust fabrication processes for phased-array solid-state catheter-based probes [2], [3].
Commercial IVUS catheters have greatly advanced, offering detailed evaluations of blood vessels and plaque morphology [2], [4], [5], [6].However, ongoing efforts within the scientific community have focused on improving catheter-based ultrasound devices, recognizing the importance of high-resolution ultrasound imaging.Traditional approaches, such as those using piezoelectric materials such as PZT in thickness mode resonance operation, have made significant contributions in ultrasound imaging [7], [8], [9], [10], [11], [12].In 2018, Janjic et al. [7] presented a 2-D PZT-based forward-looking IVUS (FL-IVUS) to perform volumetric image.This system with a maximum diameter of 1.5 mm was able to achieve a good lateral resolution of 560 µm at a 6.5-mm penetration depth; however, its integration in small catheters continues to be a challenge.
A high-definition IVUS (HD-IVUS) system has also been studied to evaluate the influence of blood on the HD-IVUS image quality [10].This system based on lithium niobate single crystal (LiNbO 3 ) in its thickness mode has a size of 0.6 × 0.8 mm 2 and a resonance frequency in water of 100.2 MHz, it achieves a lateral resolution of 324 µm by placing a 10-µm tungsten wire at 2.3 mm.As a final example of an IVUS system based on conventional piezoelectric transducers, in 2023 a novel dual-element was reported to improve the distortion suffering during the rotation of the catheter [11].In this case, dual PZT elements with similar performance and frequency (∼ 40 MHz) demonstrated their capability to resolve 10-µm wire imaging at approximately 246-µm axial distance with a lateral resolution of 167.3 and 184.6 µm for each piezoelectric element.
All these systems based on the conventional bulk ultrasound transducers face challenges such as: 1) complex fabrication processes for high-frequency devices in phased arrays, as well as combination with the front-end electronics with an application-specific integrated circuit (ASIC) and 2) need of matching layers to compensate for the mismatch between the acoustic impedances [2], [13].Besides, note that PZT-based systems are lead-based, which in real medical scenarios could be dangerous to human health and should be replaced in the near future with more sustainable materials.
With respect to MEMS-based solutions, two different types of micromachined ultrasound transducers (MUTs) have been reported for catheter-based ultrasound imaging probes [3]: capacitive MUTs (CMUTs) [14], [15], [16], [17], [18], [19], [20] and piezoelectric micromachined ultrasonic transducers (PMUTs) [21], [22], [23].Both are MEMS devices based on flexural membranes capable to produce and sense the ultrasound acoustic field by electrostatic actuation (CMUTs) or using a piezoelectrical layer as part of the membrane's layers (PMUTs).Both are microfabricated using robust microtechnology processes and offer distinct advantages.CMUTs, despite their requirement for high polarization voltages, demonstrate promise due to their small dimensions, compatibility with CMOS processes, and potential for high-density array fabrication.PMUTs, on the other hand, offer further advantages by eliminating the need for dc polarization voltages, simplifying the fabrication process (same element to transmit and receive) while maintaining the integration capabilities with CMOS circuitry [3], [24].Some examples from the literature using CMUTs or PMUTs for phased-array systems are described here.In 2013 a 1-D CMUT array was designed to be implemented in an IVUS on guidewire, being determined to provide details about vessel dimensions, plaque composition, and so on.[14].This system based on a 300 × 1000 µm CMUT array provides in water 277-µm lateral resolution at 2.4 mm with a frequency of 35.6 MHz.In 2020 [18], a highly integrated guidewire ultrasound imaging system-on-a-chip for vascular imaging was presented by the same group.It is based on a 1-D array of 12-element CMUTs working at 40 MHz in water and it is combined in a dedicated packaging with a complete front-end CMOS chip for transmission, reception, and signal processing of the ultrasound signals.The 12 elements consist of 40 square membranes with a 25-µm pitch (40 × 25 µm = 1 mm).The CMUTs array occupies an area smaller than 1 × 0.3 = 0.3 mm 2 .Using a three 100-µm wires' phantom sample, an ultrasound image is demonstrated obtaining a 560-µm lateral resolution at 8 mm in the axial direction with a dc voltage bias of 44 V.Although the final idea is to achieve a system-on-chip (SoC), here the ultrasound transducer and ASIC are in separate chips, which increases the total area and affects the signal-to-noise ratio.On the other hand, in 2018, Pekař et al. [17] presented a 4-mm 2 1-D CMUT array of 32 circular CMUTs with 60-µm diameter, 65-µm pitch, wired to a front-end-specific CMOS chip inside a catheter for an intracardiac applications.The CMUTs can be operated at different frequencies depending on the imaging mode desired.In the resolution mode, where an image of a tissue-mimicking phantom was used, the CMUTs were operated at 20.8 MHz using a very large dc voltage bias (−160 V), providing a lateral resolution of 0.035 radians at a penetration depth of 16 mm.
Dealing with PMUTs, one of the first works for catheterbased systems, was presented in 2014 [23], where PZT was used as the piezoelectrical layer together with the silicon layer and the electrodes to form the flexural membrane.Two PMUT arrays with different-sized rectangular apertures were fabricated by bulk micromachining in silicon-on-insulator substrates.The PMUTs are rectangular shape with a dimension of 110 × 80 µm and operate at 5 MHz in water.The array containing 1024 PMUT membranes in 64 × 16 PMUTs (arranged in 64 × 4 = 256 elements) occupies an area 11.2 × 1.9 mm 2 and presents a theoretical lateral resolution of 1.1 mm at 4-cm penetration depth.Experimental data using this array demonstrated a lateral resolution of 1 mm at 3-cm penetration depth in an ultrasound image of a phantom tissue.In this work, the PMUTs were actuated and sensed by nonintegrated electronics.In 2019, Lee et al. [22] present a 6 × 6 pitch-matched 2-D PMUT array bonded to a specific integrated CMOS circuit with the electronics for ultrasound transmission and reception of the 36 channels.Each channel occupies 0.0625 mm 2 and is formed by four PMUTs with a pitch of 250 µm.The overall 36-channel 2-D array equals 2.25 mm 2 .The PMUTs operate at 6 MHz in water.Using a phantom sample with three 500-µm diameter wires, they demonstrate ultrasound imaging using the PMUT and CMOS ASIC at different penetration depths, but the lateral resolution obtained is not quantified.An estimation of the lateral resolution can be done from the provided ultrasound image, giving approximately around 20 • (equivalent to 8.7 mm) at a 25-mm penetration depth.Although this system presents a dedicated CMOS ASIC, it requires wafer bonding between the PMUT chip and the CMOS ASIC, which increases the complexity on the fabrication processing and can affect the signal-to-noise ratio.To our knowledge, there has not been yet any work presenting a PMUT system integrated with CMOS intended for catheter-based applications.
In this article, we present a novel pitch-matched PMUTon-CMOS array tailored for low-dimensional catheter-based ultrasound systems.This integrated system has a small active area (430 × 430 µm) and a total area 0.46 mm 2 (including pads) conducive to easy integration inside small catheters with an area of 1 mm 2 .In addition, the monolithic integration reduces the interconnection wires, improves the signal-to-noise ratio, and provides independence of the cable length between the PMUT receivers and the acquisition system.The significance of this work lies in its potential to achieve high lateral resolutions and detect targets smaller than half the wavelength, thus being the first step toward advancing the capabilities for catheter-based ultrasound imaging.The article structure contains two main sections and the conclusions.In Section II, the ultrasound system will be described considering performance and fabrication process.In Section III, experimental results including the characterization as an actuator and as a sensor as well as an ultrasound image demonstration will be presented and discussed.

II. ULTRASOUND SYSTEM: DESIGN AND FABRICATION A. PMUT-on-CMOS Array Design
Ultrasound systems with narrow acoustic beams are essential to obtain ultrasound images with high resolutions.For catheter-based applications, probes with an outer diameter lower than 1.5 mm are required [7].This means that the transducer area must be small enough to ensure that all the systems (transducers, pads, bonding connections, and so on) fit within the size of the catheter.The aperture of the transducer is directly related to the resolution of the system.For a PMUT array with focusing, lateral resolution or beamwidth at −6 dB, BW −6 dB , is defined by (1) where λ is the wavelength, and F# is the f number defined as the ratio between focal length (F) and aperture of the array (L) [25].Based on this expression, for a specific transducer (frequency and dimensions fixed) and acoustic medium, the only method to obtain narrow beams and, consequently improving the lateral resolution, will be decreasing the penetration depth or focal length.Furthermore, if a resolution of λ/2 is desired, F# should be equal to 0.5, which means that the target must be placed at L/2.This condition is challenging for small apertures.In a PMUT array, focusing is achieved electronically by simply applying the corresponding delays (electronic focusing) to different PMUT rows, allowing different focus with the same device and, increasing the transmission sensitivity when the focusing distance decreases [26].In this sense and in comparison, with the use of fixed physical lens, different penetration depths, lateral resolutions, and acoustic pressures can be obtained in the same arrayed device with a simple modification of the applied waveform.Based on this, and considering our previous experience in high-frequency PMUT arrays [27], a PMUT array was chosen, with a total area of less than 1 mm 2 , where the active area (L × L) is 430 × 430 µm, and the dimensions considering all the pads are approximately 635 × 726 µm.With these dimensions, the theoretical lateral resolution at the natural focus (NF) of the PMUT array will be close to 100 µm, considering (1): BW −6 dB = λ • NF/L, and that the NF is given by the Rayleigh distance, NF = R 0 = Area/(4λ) = L 2 /(4λ).Moreover, sub-100-µm lateral resolution could be obtained using beamforming to focus at smaller penetration depths belonging to the near-field of the array

B. PMUTs-on-CMOS Array Characteristics
The ultrasound system is based on a 1-D PMUT array fabricated monolithically on top of a 130-nm high-voltage (HV) CMOS analog front-end circuitry using the MEMS-on-CMOS process developed by Silterra [27], [33].The array consists of 49 PMUTs arranged in seven rows connected through the top electrodes while the bottom electrode is common to all the systems.During the transmission of the ultrasound beam, the driving signals are generated by three HV CMOS transmitters based on a level-shifter topology generating 32-V pulses [34].Each of the three TXs is applied at the top electrode of two rows configured symmetrically from the central row allowing focusing on the center along the axial direction.On the other hand, the received signal is amplified by a CMOS LNA which is directly connected to the central row making the PMUT-on-CMOS received signals independent of the length of cables to the acquisition system (in our case the oscilloscope).Here, low-voltage switches are not required which reduces the noise level (improving the signal-to-noise ratio and therefore the image quality) [35].Besides, extra circuits are not necessary to produce the control signals, making the design and manufacturing process less complex and reducing the power consumption and area (if the electronic is on-chip).Finally, by avoiding switches the area is reduced facilitating the dimensional constraint to implement a pitch-matched system.The LNA presents a voltage-voltage gain of around 26 dB, a power consumption of 0.3 mW, and a bandwidth of 22 MHz [27].The LNA amplifier is followed by a 50-matched buffer, giving an overall reception area of 6 × 10 −4 mm 2 .Fig. 1 shows a diagram of the system where rows sharing the transmitter circuit are in the same color.This array arrangement with symmetry from the center is used to facilitate the beam focusing at different penetration depths.More complex configurations with TX/RX capabilities in each column will allow beam steering and beamforming for image reception enhancing the ultrasound image quality.These aspects are not included in this article.
Once the general system was presented, the single element in the array will be introduced.Each individual PMUT consists of a clamped multilayer membrane driven at its first outof-plane flexural mode following the technological process described in [27] and [36].The shape and size of the membrane are defined by the cavity, which in our case is a square of 40 × 40 µm.The multilayer membrane over the cavity is composed by a 0.6-µm piezoelectrical layer [AlN doped with 9.5 % of Sc (Sc 9.5% Al 90.5% N)], placed between two Al electrodes with thicknesses of 0.15 and 0.2 µm for the bottom and top electrodes, respectively.An etching step, through four external holes on the piezoelectric layer, allows the releasing of the membrane, defining the squared cavity with 600-nm height.Finally, a 1.5-µm thick Si 3 N 4 layer is deposited by plasma-enhanced chemical vapor deposition (PECVD) process over all the PMUT structure.This layer acts as a passive  Unlike conventional ultrasound transducers based on thickness mode, PMUTs work in the flexural mode where d 31 piezoelectric coefficient creates a mechanical deformation via bending [39].During the transmission (inverse piezoelectric effect), an ac signal between the top and bottom electrodes at the resonance frequency causes a deflection of the membrane and generates the acoustic wave.On the sensing mode (direct piezoelectric effect), the incoming ultrasound wave causes a vibration of the membrane which can be detected by measuring the electric output between both the electrodes.From the theoretical point of view, the resonance frequency for a PMUT device is described by (2a) where the value is determined by its physical characteristics: λ 2 i j depends on the vibration mode, the shape, and the boundary conditions (λ 2 i j = 35.99 for the first mode corresponding to a square clamped PMUT), a is the PMUT side, µ is the mass per unit area [see (2b)], and D is the flexural rigidity [see (2c)] [38], [39].From these equations, t n and ρ n define the thickness and mass density of the nth layer, respectively; E 11,n is the plate modulus and h n defines the location of the top of each layer relative to the bottom of the laminate According to these analytical equations, the resulting multilayer single PMUT will be resonating at its first out-of-plane flexural movement with an expected frequency of 19.4 MHz in air.This frequency is closed to the obtained with the real PMUTs layout using FEM simulations in COMSOL, 20.6 MHz.Compared with our previous system [36], by reducing the PMUT size by half, the resonance frequency increases (improving the axial and lateral resolutions) and is more challenging to design a pitch-matched PMUT-on-CMOS system.In addition, the use of 9.5% scandium-doped AlN ensures better efficiency of the piezoelectric material regarding their capability to transform electrical energy in mechanics and vice versa, described by the electromechanical coupling factor K 2 t [40], [41].This figure-of-merit is directly related to e 2 31, f /ϵ 0 *ϵ r (where e 31, f is the effective piezoelectric coefficient, ϵ r is the relative permittivity, and ϵ 0 is the vacuum permittivity), giving 34.56 GPa for AlScN and 13.48 GPa for AlN, which is a 2.56× improvement due to Sc concentration of 9.5% [42].
The PMUT resonant frequency will be changed if the PMUT is operated in a fluid environment.In this case, the resonance frequency is affected by the medium properties which add an extra mass causing a drop in frequency; see (3) [45], [46].This parameter is known as added virtual mass (β) and can be computed for a squared clamped device as shown in (4), which takes into account the fluid viscosity, η [47].In the case of using Fluorinert (FC-70: c = 685 m/s, ρ = 1940 kg/m 3 ) as fluid medium, with a nonneglecting viscosity (η = 24 cP), this added virtual mass factor is β = 4.5, and in consequence, the resonance frequency in the liquid is expected at around 8.3 MHz Fig. 3 illustrates the final layout and an optical image of the presented PMUTs-on-CMOS array where the pitch was set to 65 µm giving a fill factor of around 42%.The positions of the TXs and RX blocks shown in Fig. 1 have been highlighted [Fig.3(a)] where all the CMOS circuitry is under the PMUT array, creating a complete and compact pitch-matched ultrasound system.Fig. 3(b) is an optical image of the PMUTs-on-CMOS array, highlighting the vias from the PMUT electrodes to the CMOS circuitry (which is placed underneath).The inset in Fig. 3(b) is a zoom on four PMUTs, to clearly show the four holes outside the cavity used for the releasing of the membrane and which are covered by the passive Si3N4 layer to guarantee watertightness.

A. Acoustic Characterization in a Liquid Environment
The acoustic tests of the PMUTs-on-CMOS ultrasound system were done with the array immersed in Fluorinert (FC-70: c = 685 m/s, ρ = 1940 kg/m 3 ).The system was bonded to a PCB using wedge-wedge wire bonding, the inputs of the transmitter circuits were connected to a signal generator (81150A, Keysight, USA), and the voltage signal at the LNA + buffer output was acquired by an oscilloscope (DSO-X 3054A, Keysight, USA).A 200-µm-diameter needle hydrophone from ONDA (HNC-0200, ONDA, USA) was used to measure the generated acoustic pressure.
1) Frequency Response: The frequency response of the system was obtained considering two scenarios.The first one is based on a pulse-echo configuration where the liquid-air interface was used as a reflecting surface.In this case, the transmitter rows were driven considering four unipolar pulses with 32 V of amplitude, and the frequency was modified from 5 to 20 MHz with a step of 250 kHz; see the setup in Fig. 4(a).The temporal responses at each frequency were acquired at three different times of flight corresponding to acoustic paths (APs) of 4, 6, and 8 mm (i.e., 2, 3, and 4 mm of FC-70 thickness, respectively) reaching a maximum amplitude of 70 mV when the AP is 4 mm and 15 mV when the distance traveled is double.Based on the obtained results, the peak frequency ( f 0 ) appears at 7.7 MHz independently of the AP (or independently of the final liquid thickness).This frequency is close to the expected one computed in Section II (8.3 MHz) derived from the theoretical analysis (3).To corroborate these results, a second experiment was carried out by exciting the transmitter rows with one pulse of 65-ns width and computing the fast Fourier transform (FFT) from the echo acquired at 2 mm by the HNC-0200 hydrophone [see the setup in Fig. 4  the solid black line corresponds to the FFT, and red squares correspond to the pulse-echo point at an AP of 4 mm, showing a good correspondence between them.Based on the FFT, the bandwidth at −6 dB is 3.9 MHz, which corresponds to a fractional bandwidth of around 50% (computed as bandwidth (@-6 dB)/ f 0 *100).
2) Transmitting Sensitivity: The position in the plane and the axial distance was adjusted using a manual micro-positioner system to obtain the maximum pressure.All the transmitting rows were driven with four unipolar pulses at 7.7 MHz with 32-V amplitude.Electrical crosstalk is not expected in this case (see [26] for further details).Fig. 5(a) shows a schematic representation of the experimental setup.
The first acoustic measure was done after 3.6 µs corresponding to a distance between the hydrophone and the array surface of 2.5 mm.Fig. 5(b) inset shows the time response at this distance giving a peak-to-peak pressure of 29 kPa pp .From this position, the hydrophone was lifted every 50 µm and the output pressure amplitudes were measured at each point; see red circles in Fig. 5(b).To obtain the pressure dependence with the distance (A = P 0 * R 0 ), the measured peak-to-peak values were fit according to the following expression [29], [43]: where P 0 is the surface pressure, R 0 is the Rayleigh distance (R 0 = Transducer Surface/4λ), z is the axial distance, and α is the damping viscosity coefficient.The damping coefficient defined by ( 6) depends on the resonance frequency ( f 0 ), the longitudinal or acoustic viscosity (η), the density (ρ), and the sound velocity (c) of the acoustic medium.Higher attenuation values are reached when the frequency is higher than 5 MHz [44], and it is important to estimate this parameter to obtain a better adjustment.Replacing all the terms consid- Considering the damping coefficient and ( 5), the red dashed line in Fig. 5(b) shows the fit curve giving a pressure dependence of 80.41 kPa pp • mm.Normalizing with the applied voltage and considering a factor of 1.27 due to the effective amplitude of a square signal in comparison to a sinusoidal one, Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.the ST obtained is 1.98 kPa pp • mm • V −1 .Taking into account this value and the active area (0.185 mm 2 ), the acoustic pressure from 1-mm 2 PMUT array area at 1.5 mm from its surface when 1 V is applied gives 7.1 kPa pp • mm −2 • V −1 .3) Receiving Sensitivity: The receiving sensitivity (SR) was obtained by testing the system in a pulse-echo configuration.Here a metallic surface was used as a reflective surface and the position along the axial direction was modified giving an AP difference between every point of 100 µm; see Fig. 6(a).The transmitting rows were driven with four unipolar pulses at 7.7 MHz with 32-V amplitude, and the maximum value was acquired at each point.where P(z), obtained in the previous section needs to be divided by 2 because this value refers to peak-to-peak pressure.
Considering that, we obtained a sensitivity of 3.3 V/MPa.4) Comparison With the State-of-the-Art: Table I shows a comparison of the presented PMUT-on-CMOS array with other PMUT-based systems reported in the state-of-the-art.The normalization with the distance and the applied voltage (ST) was used to compare the generated pressure at the same distance by different ultrasound transducers no matter what size they are.However, this normalization does not include the area, which is determining for catheter applications, and because of this, the parameter ST was divided by the effective area of the transducer thus providing the acoustic output pressure of the PMUT area of 1 mm 2 to 1.5 mm of its surface when 1 V is applied (kPa/mm 2 /V).From this result (ST normalized), it can be seen how the PZT-based PMUTs achieve an improvement transmitting performance of 2.8× [48] as expected.Similar benefits have been presented recently comparing PZT PMUTs with AlScN (Sc at 15%) [49].
To estimate the best performance as an actuator and as a sensor, the product ST*SR was defined as figure-of-merit, considering in all the cases the ST at 1.5 mm.Based on these results, the proposed AlScN PMUTs-on-CMOS array achieves an improvement of almost a factor of 2 in comparison to the PZT array (the best reported result in the table) used for dynamic monitoring of the arterial walls [48].Compared with our previous PMUTs-on-CMOS arrays [27], [36], it can be seen how this system with small dimensions can generate 3.6× more pressure, doubling the frequency and with a slight increase in the SR than 7 × 7 AlN array (the size of a single PMUT is 80 µm) [36].This is thanks to the change in the piezoelectric material from AlN to doped AlScN.
Likewise, with respect to the 7 × 7 AlScN array in [27] where the differences are given in the thicknesses of the layers (1-µm Si 3 N 4 , 0.4-µm bottom electrode, and 0.35-µm top electrode) [27], the normalized ST here (7.1 kPa pp * mm −2 * V −1 ) is 2.38× higher using one row less to transmit, as well as the sensed amplitude at 1 mm (133.5 mV) over the array surface increases in a factor of 5.8×.

B. Ultrasound Imaging
The ultrasound imaging demonstration and the capabilities of the PMUTs-on-CMOS array in terms of resolution were performed by carrying out a pulse-echo experiment using as a target a 25-µm Al wire.This wire was placed at three axial positions on top of the system (2.5, 3.5, and 4.8 mm, respectively) and was manually displaced 2 mm along the active aperture with a step of 100 µm; Fig. 7 shows the experimental setup.
For comparison, simulated ultrasound images were obtained using Field II considering the same experimental conditions.The resulting ultrasound images are shown in Fig. 8, where (a)-(c) corresponds to the simulation results and (d) and (e) shows the experimental ones.ultrasound images were created by taking the temporal response (amplitude) at each lateral point (in steps of 100 µm).This temporal response was processed using a Hilbert transform to obtain the envelope, which was normalized regarding its maximum.Finally, all the results were plotted in a 2-D image where x-axis corresponds to the lateral distance and y-axis corresponds to the time.The measured results match with the image predicted by the simulations, demonstrating the PMUTs-on-CMOS ultrasound images' capabilities and the possibility of detecting targets with dimensions below 100 µm.Performing a cross section in a lateral direction at its maximum value, the obtained −6 dB beamwidth is 516, 649, and 926 µm for 2.5, 3.5, and 4.8 mm, respectively.To validate these values, (1) was used to compute the lateral resolution at the same positions, and we obtain at 2.5 mm a BW −6 dB = 5.8λ = 517 µm; at 3.5 mm The small beamwidth is obtained at the small distance, which implies that higher resolutions and then better quality in the image can be obtained at distances close to the array.
In the same context, a second ultrasound image experiment was carried out using four different wires as targets.The wires' diameter goes from 25 to 150 µm being made from different materials.Table II lists the properties of each one, and Fig. 9 bottom shows their optical images.The phantom was performed by fixing them in a plastic support side by side, and their axial position was slightly modified to obtain different times of flight and thus different start positions in the ultrasound image; see a schematic setup in Fig. 9 top.Such as in the previous experiment, the phantom was immersed in Fluorinert and a manual sweep with steps of 100 µm was carried out to displace the sample 9.3 mm.Fig. 10 illustrates the final ultrasound image where all the wires are clearly identified as well as the liquid-air interface.Based on the time of flight, the thickness of the liquid is around 7.4 mm, and the interfaces were placed at 2.9, 3.9, 3.7, and 3.4 mm for A, B, C, and D, respectively.As in the previous experiment, the same distortion problem appears, for instance, the 150-µm wire gives a BW −6 dB = 6.7λ = 600 µm being four times greater than the real diameter.Besides, focusing on the incoming echo from interface B, the amplitude is very small in comparison to A and C (other copper interfaces, but bigger diameter) and even with D with a smaller diameter size.We attribute this effect to a change in the acoustic impedance of the B wire, which can be due to some oxidation of the copper wire.The peak-to-peak envelopes give a maximum value of around 4 mV pp for the A interface and 2 mV pp for the liquid-air interface.
1) Resolution Improvement: The previous ultrasound images demonstrated the capability to detect targets placed in a few millimeters' distance with dimensions in the micrometer range, arriving even at a size below 100 µm.Despite the good correspondence between the simulated and experimental results (see Fig. 8), it is true that the real size of the phantom (25 µm) is distorted, at least 20.6 times (516/25 µm where 516 µm corresponds to a cross section along the maximum value of the experimental image at 2.5 mm).Based on (1), it can be seen how if the target is placed at distances greater than the transducer aperture (F > 5L), the resolution increases in the same factor with respect to the wavelength (BW −6 dB > 5λ), being possible to reach a λ resolution if the transducer aperture and the focal distance have the same length.Considering the experimental data, the sound velocity in the acoustic medium (Fluorinert: c = 685 m/s), and the resonance frequency of the  Schematic illustration of the experimental setup for the pulse-echo ultrasonic image (top).Optical images of each used wires where the letters and colors correspond to their position in the plane (bottom).system (7.7 MHz), the wavelength gives 89 µm which is a small value and could be interesting to achieve a resolution   Due to our experimental setup, these distances are not reliable (bonding is not straight enough).As a consequence, a theoretical and simulated analysis was performed that was experimentally validated in the far-field regions.
Fig. 11 shows the theoretical (blue solid line), simulated (dashed black line), and experimental (red circles) beamwidths as a function of the axial position, where focusing techniques are required in the near-field.
The experimental points were obtained by displacing the HNC-0200 hydrophone along the active aperture from −1 mm to 1 mm with steps of 100 µm and acquiring the acoustic pressure at each point; see the inset graphs.From the inset images, as expected, at large axial distances, the beamwidth increases, and the maximum peak-to-peak pressure decreases.The minimum distance at which this hydrophone can be placed is 2 mm to not overestimate the beamwidth, and not underestimate the pressure [51], [52].These results illustrate a good correlation with the measured beamwidth in simulation, and calculated analytically, allowing extrapolation of this behavior in the near-field, for instance, to obtain resolutions lower than 100 µm, the axial distance must be lower than 400 µm.
To demonstrate the resolution improvement in terms of ultrasound images, some simulations were performed in Field II.Fig. 12 illustrates the simulated normalized acoustic field from 100 µm to 5 mm along the axial direction and from −1 mm to 1 mm laterally without focusing (a), and focusing at 400 µm applying the corresponding delays (c).As it can be seen, when focusing techniques are used, the beamwidth decreases because the acoustic energy is concentrated in a narrow beam, improving the capability to detect objects with small dimensions.To validate it, the same 25-µm phantom was placed at 685 and 400 µm (electronic focusing was applied).The results are shown in Fig. 12(b) and (d); note that there is a clear improvement in lateral resolution when focusing techniques are used, and the phantom is placed at distances below 400 µm.
Table III presents a comparison from some state-of-theart ultrasound systems designed for catheter applications.This comparison includes: systems based on thickness mode piezoelectrical transducers [7], [10], [11] and systems based on flexural membranes using MEMS-fabricated ultrasound transducers, either with CMUTs [17], [18] or with PMUTs [ [22], [23], this work].To provide a comparison between them, we normalize the lateral resolution which is provided by different axial positions for each of the systems, and based on (1), we recalculated them at the same axial distance (1 mm).In addition, a figure-of-merit (FoM = Resolution • Area) is defined to clearly obtain which system achieves the best resolution with the smallest size (i.e., FoM smaller), making the system more suitable for catheter applications.In Table III, we have also included a column explaining the capabilities to be integrated in CMOS technology.According to Table III, the smaller FoMs are obtained with systems based in MEMS fabrication processes: based on CMUTs [18] and the system presented here which is based on PMUTs.In addition, both the systems can be integrated with CMOS, although only our approach presents a monolithically integrated system.Finally, our system has the lowest operation frequency, which prevents for signal attenuation at large distance, improving the signal-to-noise ratio and eventually the quality of the image.These results demonstrate the benefits of this tiny AlScN PMUT-on-CMOS array.

IV. CONCLUSION
In this article, the high resolution and IVUS imaging potential of an AlScN PMUTs-on-CMOS array with an area lower than 1 mm 2 is demonstrated.The system has been monolithically integrated with the CMOS analog front-end, achieving pitch-matched system with high performance in terms of (1.98 kPa pp −1 ) and sensed pressures (3.3 V/MPa) at 7.7 MHz Fluorinert.Ultrasonic images of a target with dimensions less than λ/2 (wire diameter 25 µm) were successfully demonstrated, being a great achievement for the practical implementation of this system in small IVUS catheters.Likewise, a second experiment with targets with different diameters was carried out and checked the capabilities of the PMUTs-on-CMOS array in front of different sizes and materials.The defined figure of merit (R'•A) allows us to compare the capabilities of our system with the state-ofthe-art to provide the best resolution with a very small area and monolithic CMOS integration.The use of PMUT arrays like this, as opposed to ultrasound systems with lens, offers a powerful tool for focusing at different points along the axial direction, achieving in this case, sub-100-µm resolutions for focusing distances less than 400 µm.

Fig. 2 .
Fig. 2. (a) 3-D structure of a single PMUT-on-CMOS Silterra technology.(b) Cross section of one row of seven PMUTs-on-CMOS.In both the schematics layers are not to scale.

Fig. 3 .
Fig. 3. (a) Layout and (b) optical image of the fabricated system.Inset: zoomed-in view to show the details of four PMUTs in the array.
(a)].Fig. 4(b) inset shows the results where Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

Fig. 4 .
Fig. 4. (a) Experimental setup for the frequency characterization in Fluorinert (FC-70) in a pulse-echo configuration.The acquired signal by the oscilloscope is recorded for each of the actuation frequencies to determine the resonance frequency of the PMUTs-on-CMOS system.(b) Frequency response using a pulse-echo configuration at an AP of 4 mm (red squares), 6 mm (blue squares), and 8 mm (green squares).Inset: FFT corresponding to the hydrophone time response to a 65-ns pulse (solid black line) and the normalized pulse-echo response when the AP is 4 mm (red squares).

Fig. 5 .
Fig. 5. Acoustic characterization as an actuator.(a) Schematic experimental setup.(b) Measured pressure at different heights from the array surface (red circles) and the fit curve (red dashed line), pink shadow corresponds to the experimental error.Inset: Temporal response at 2.5 mm.

Fig. 6 .
Fig. 6.Acoustic characterization as a sensor.(a) Schematic experimental setup.(b) Measured amplitude at different APs (blue square) and the fit curve (dashed blue line).Inset: Upper envelope of the received pulse-echo time response corresponding to two different APs, AP = 2.5 mm in red and AP = 4.5 mm in olive.

Fig. 6 (
b) (blue square) shows the measured amplitudes and the inset graph depicts the upper envelope in the first and last point.The results were fit taking into account the dependence of the amplitude with the AP and the losses due to the viscosity (α = 45 m −1 ), V(AP) = B/AP * e −αAP , where the coefficient B gives 133.5 mV * mm; see dashed blue line in Fig. 6(b).The SR can be computed using V(AP)/(P(z)/2),

Fig. 7 .
Fig. 7. Schematic and photograph of the phantom used to create the first ultrasound image.

Fig. 9 .
Fig. 9. Schematic illustration of the experimental setup for the pulse-echo ultrasonic image (top).Optical images of each used wires where the letters and colors correspond to their position in the plane (bottom).

Fig. 10 .
Fig. 10.Pulse-echo ultrasound image of four wires fixed in a plastic support and immersed in Fluorinert.Cross section along the time response placing the cut in the maximum of each interface (left).Red line: A target/wire; purple line: B wire; blue line: C wire; and green line: D wire.in this range for catheter-based ultrasound image applications.To obtain it, the target should be placed at 430 µm (F = L).

Fig
Fig. Theoretical (blue line), simulated (dashed line), and experimental (red circles) beamwidths versus axial distance.The inset images correspond to the experimental distribution at 3.5-mm and 4.5-mm axial distances.

Fig. 12 .
Fig. 12. Field II simulation results.Normalized pressure map (a) without focusing and (c) focusing at 400 µm.Pulse-echo ultrasonic image of a 25-µm Al wire placed at the corresponding focal points (maximum pressure points), at (b) 685 µm and (d) 400 µm.

TABLE I COMPARISON
OF SYSTEMS BASED ON PMUTS AS ACTUATORS AND SENSORS

TABLE II PROPERTIES
AND DIMENSIONS OF THE WIRES USED IN ULTRASOUND IMAGING EXPERIMENT a BW −6 dB = 8.1λ = 724 µm, and at 4.8 mm a BW −6 dB = 11.2λ= 993 µm.