High-uniformity miniaturized PMUT array with broadband and high-sensitivity for wearable ultrasound imaging

Abstract

Among the diverse ultrasound transducers deployed in wearable continuous imaging applications, piezoelectric micromachined ultrasonic transducer (PMUT), developed via microelectromechanical system (MEMS) technology, exhibit significant potential for wearable use, owing to their device miniaturization and low power consumption. However, PMUT arrays often suffer from poor imaging performance and no efficient method exists to evaluate large-scale arrays’ performance other than post-fabrication measurement. This study proposes a mathematical model of mutual radiation impedance to predict the vibration amplitude of each cell in a PMUT array, thereby guiding optimal cell arrangement for simultaneous enhancement of sensitivity, bandwidth, and acoustic uniformity. The elevation aperture of linear PMUT arrays is reduced based on the beam-formed acoustic output, improving the integration and wearability of rigid silicon-based transducers. An ultra-narrow 64-element linear AlN PMUT array with a 7:1 aspect ratio (1 cm × 0.15 cm, aperture area 0.15 cm²) was fabricated for experimental validation. The device demonstrates a pulse-echo fractional bandwidth exceeding 60% at a center frequency of 7 MHz, along with an image contrast of ~50 dB. Its lateral and axial resolutions are measured to be ~0.26 mm and ~0.25 mm, respectively. Continuous and real-time clear phased-array imaging of multiple superficial organs in volunteers, including carotid artery, thyroid gland, and dorsalis pedis artery, demonstrates the qualified ultrasound performance of our designed PMUT array. Our work establishes a systematic framework for developing large-scale PMUT arrays enabling the realization of wearable ultrasound patch with compact form factor, high fidelity imaging quality, superior spatial resolution and low power consumption.

Introduction

Wearable ultrasound imaging offers unprecedented capabilities for real-time assessment of the functions of many vital organs¹. In an ultrasound system, the transducer’s performance (including frequency, sensitivity and bandwidth) and the array configuration (element count and pitch) determines image resolution, signal-to-noise ratio, and penetration depth, ultimately affecting the image quality². Also the wearable ultrasound transducer must be miniaturized to ensure patient comfort during long-term wear³. Superior ultrasound imaging generally necessitates the transducer arrays with elevated center frequency, enhanced sensitivity, broader fractional bandwidth, reduced inter-channel spacing (pitch), and improved channel performance uniformity^4,5. Current wearable ultrasound patches, which rely on traditional bulk piezoelectric transducers, face challenges including limited element density, large pitch, and a compromise between wear-comfort and performance⁶.

Micromachined ultrasound transducers (MUTs) are microelectromechanical system (MEMS) devices and they can be categorized into capacitive and piezoelectric types (CMUTs and PMUTs) based on their operating principles⁷. CMUTs operate by modulating the capacitance between a micromachined membrane and a bottom electrode. PMUTs, in contrast, leverage the piezoelectric effect to convert mechanical vibrations into electrical signals. MUTs exhibit key features including miniaturization, high structural uniformity, scalability, batch-fabrication capability, and seamless integration with multiple systems. Additionally, they eliminate the need for a backing layer to absorb backward waves and avoid stringent impedance matching requirements for efficient energy transmission⁴. These features enable the creation of ultra-thin patches, essential for wearable imaging⁸. In contrast, PMUTs operate without high-voltage DC drive circuits, simplifying system integration and making them ideal for wearable applications where compact size and low power are priorities⁹.

Compared to piezoelectric bulky ultrasound probe, PMUT often suffers from poor imaging performance and no efficient method exists to evaluate large-scale arrays’ performance other than post-fabrication measurement. Reports on PMUTs for human imaging remain limited so far and they are mainly in low frequency ranges (generally smaller than 3 MHz). Savoia et al. introduced a PZT-PMUT from STMicroelectronics with a 2.5 MHz center frequency and 81% -6 dB fractional bandwidth (2 MHz absolute bandwidth) for carotid artery scanning¹⁰. Zhang et al. developed a PZT-PMUT 2D array for carotid artery 3D imaging, with a 25.5% -6 dB bandwidth at 2.9 MHz resonance¹¹. Pandit et al. created a ScAlN-PMUT array with 117% bandwidth at 2.4 MHz resonance for cardiac imaging¹². However, as Gami et al. showed by comparing B-mode images of the common carotid artery, a PZT-PMUT array (from TDK Americas R&D Center, Milan, Italy, with ~28% bandwidth at 4.8 MHz center frequency) produced noisier images with more reverberation artifacts and lower signal-to-noise ratio than a clinical L7-4 linear transducer, due to its low acoustic output and narrow bandwidth^13,14. Zhao et al. developed a high-frequency PZT-PMUT array (~11 MHz center frequency, 68% bandwidth) for carotid artery imaging, utilizing a rectangular diaphragm and mode merging¹⁵. van Neer et al. demonstrated flexible PVDF-TrFE ultrasound arrays with an 8.2 MPa center frequency and 78% -6 dB bandwidth¹⁶. While both methods achieve wide bandwidth, they exhibit reduced sensitivity.

Superficial vessels lie within 3 cm of the skin surface. Continuous monitoring of them helps prevent and early diagnose cardiovascular and cerebrovascular diseases. Clinically approved their imaging necessitates ultrasound operation at high frequencies of 5 ~ 12 MHz with a pulse-echo fractional bandwidth exceeding 60% to achieve axial resolution better than 0.3 mm for better diagnosis^4,17. Developing PMUT arrays at these high frequencies with large bandwidth and high acoustic output is essential. However, existing literature primarily focus on monitoring vessel diameter with single-element probes rather than imaging with transducer arrays. The main reasons can be summarized into three aspects.

The first challenge is ensuring uniform performance across array elements. In phased array ultrasound, amplitude and phase errors degrade beam-forming, reducing focusing gain, widening the main lobe, and raising side lobe levels. This leads to poorer resolution, lower contrast, and more clutter¹⁸. Both simulation and experiments show that small phase errors or channel response differences can cause visible artifacts in B-mode images¹⁹. For PMUTs, one channel usually comprises many parallel-connected elements. Element-to-element variation is more significant due to stress differences and film-related manufacturing factors compared to conventional bulky ones²⁰. This inherent variability makes it very hard to ensure highly uniform acoustic responses in large-scale PMUT arrays with hundreds or thousands of elements. Thus, performance uniformity is a key challenge for PMUT arrays in high-resolution medical imaging.

The second challenge is the sensitivity-bandwidth trade-off in PMUTs. High-resolution deep imaging demands both high sensitivity (for detection depth and contrast) and wide bandwidth (for axial resolution), but improving them together is difficult. Many PMUT designs aim to boost sensitivity or bandwidth, but most trade off one for the other. For example, Wang et al. used rectangular PMUTs to couple vibration modes and widen bandwidth. But higher-order modes’ out-of-phase vibrations reduce acoustic output, causing significant sensitivity loss²¹. Similarly, Nistorica et al. used a multi-size PMUT array to widen frequency response, but the lower fill factor inherently reduces sensitivity²². Hajati et al. developed a curved PMUT structure that boosts acoustic output but reduces bandwidth (~15%)²³. Xu et al. added resonant cavities to suppress crosstalk and improve sensitivity, but simulations showed this also narrows bandwidth²⁴. Simultaneously achieving high sensitivity and wide bandwidth remains a major challenge for PMUTs.

A third challenge for PMUTs is achieving a truly wearable form factor without sacrificing imaging quality. Most PMUTs are silicon-based and inherently rigid. Many wearable ultrasound systems use flexible arrays: transducer elements on a flexible substrate that conforms to the skin and accommodates motion²⁵. However, skin stretching and bending can shift the array elements’ relative positions. This leads to motion artifacts and geometric distortion in images, reducing imaging accuracy^4,26. In contrast, conventional large rigid probes offer stable imaging but suffer from poor long-term skin contact and wearability. Thus achieving both geometric stability with superior imaging quality for rigid PMUT arrays and meeting the comfort requirements of wearable imaging poses a significant challenge.

Beyond performance constraints, PMUT arrays intended for imaging also face limitations in their design methodology. Currently, the design process relies heavily on finite element modeling (FEM) which is prohibitively time-consuming, creating a major bottleneck for overall design efficiency. This work introduces a mathematical approach, leveraging the equivalent circuit model (ECM), to rapidly compute the vibration amplitude and phase of each element in large-scale PMUT arrays. These parameters enable the assessment of uniformity, calculation of ultrasound output, and analysis of frequency response characteristics. We also developed a reduced equivalent circuit model (reduced ECM, RECM) to quantify the effects of inter-cell acoustic coupling on array performance. To meet the conformability and comfort requirements critical for wearable imaging, we seek to minimize the array size while maintaining its acoustic performance. Following these design guidelines, we developed an ultra-narrow 64-element linear array with a 7:1 aspect ratio (1 cm × 0.15 cm, aperture area 0.15 cm²). The device demonstrates a pulse-echo fractional bandwidth exceeding 60% at a center frequency of 7 MHz, ultrasonic penetration depth over 3 cm along with an image contrast of approximately 50 dB. Continuous and real-time phased-array imaging of multiple superficial organs in volunteers, including the carotid artery, thyroid gland, and dorsalis pedis artery, were carried out to demonstrate the imaging behavior our designed PMUT array. This work presents a wearable ultrasound transducer array featuring compact size, high imaging quality, and enhanced wear comfort. Its successful implementation could enable next-generation ultrathin monitoring patches.

Results and discussion

Design of highly uniform PMUT array

The vibrating membrane of an PMUT cell in the array comprises a multilayer stack of Mo (100 nm), AlN (500 nm), Mo (100 nm), and SiO₂ (1200 nm) from bottom to top (Figure S1). The cavity diameter is 40 µm and the resonant frequency in water is ~7 MHz (Note S1 and Table S2).

An equivalent circuit model of the array (ECMA) containing more than one thousand PMUT cells were created to investigates the uniformity of cells’ vibration within the array (Figure S2). It allows efficient computation of the instantaneous vibration of all cells in the array. The following balance Eq. (1) describes the behavior of all individual cells within the array.

$$\left[\begin{array}{l}{F}_{1}\\ {F}_{2}\\ \vdots \\ {F}_{N}\end{array}\right]=\left[\begin{array}{cccc}{Z}_{11} & {Z}_{12} & \cdots & {Z}_{1N}\\ {Z}_{21} & {Z}_{22} & \cdots & {Z}_{2N}\\ \vdots & \vdots & \ddots & \vdots \\ {Z}_{N1} & {Z}_{N2} & \cdots & {Z}_{NN}\end{array}\right]\left[\begin{array}{l}{v}_{1}\\ {v}_{2}\\ \vdots \\ {v}_{N}\end{array}\right]$$

(1)

where F is the effective piezoelectric driving force, Z is the impedance, v is the volume velocity, and N is the number of PMUT cells in the array. Equation (1) can be rewritten as Eq. (2). Detailed mathematical transitions are provided in Note S2 of the supplementary information.

$${v}_{N\times 1}={({Z}_{{m}_{N\times N}}+{A}_{eff}^{2}({Z}_{sel{f}_{N\times N}}+{Z}_{mult{i}_{N\times N}}))}^{-1}({\eta }_{N\times 1}\odot {V}_{i{n}_{N\times 1}})$$

(2)

where ${v}_{N\times 1}$ is the velocity vector of all PMUT cells, ${\eta }_{N\times 1}$ is the electromechanical coupling vector, ${V}_{in\,N\times 1}$ is the input voltage vector, and $\odot$ denotes the Hadamard product. ${Z}_{mN\times N}$ is the mechanical impedance matrix, ${Z}_{selfN\times N}$ is the self-radiation impedance matrix, and ${Z}_{multiN\times N}$ is the mutual radiation impedance matrix. The full derivation and the expressions of these parameters are provided in Note S2. We obtain the vibration velocity of each cell From Eq. (2) and then the displacement Disp at frequency f can be calculated via Eq. (3).

$$Dis{p}_{N\times 1}=-{\text{j}}\frac{{v}_{N\times 1}}{2\pi f}$$

(3)

As shown in Fig. 1, we computed the array performance for different cell spacings and configurations, and extracted the displacement (Fig. 1a) and phase (Fig. 1b) profiles along the central line of the array. An increase in the inter-cell spacing from 50 µm to 150 µm leads to a marked deterioration in the uniformity of both vibration amplitude and phase across the array. Quantitative analysis confirms this trend. After excluding the edge cells to avoid boundary effects, the standard deviation of the displacement measures 0.0097 nm, 0.041 nm, and 0.14 nm for the respective spacings of 50 µm, 100 µm, and 150 µm. The phase show standard deviations of 0.68°, 2.3°, and 5.2°, respectively. Thus a smaller inter-cell spacing and a correspondingly higher fill factor are conducive to achieving a more uniform response across the array.

Fig. 1: Uniformity of cell vibration in a PMUT array and the role of acoustic coupling.

a Displacement amplitude of cells in arrays with different cell spacings. Each point means a cell. b Phase distribution of cells in arrays with different cell spacings. c Variation of mutual radiation impedance and acoustic coupling strength with cell spacings and the total coupled cells on one side

During operation, each cell in a PMUT array not only emits its own acoustic wave but also interacts with waves generated by neighboring elements. This phenomenon, termed acoustic coupling, can significantly influence the instantaneous vibrational state of the entire array. The strength of the acoustic coupling is governed by three key layout parameters: the inter-cell spacing, the overall array dimensions, and the specific location of each cell. To quantitatively analyze how these layout parameters govern the final uniformity, we developed a reduced equivalent circuit model (RECM) (Figure S5) derived from the full equivalent circuit model of arrays (ECMA). This approach was necessitated by the computationally prohibitive demands of modeling intricate acoustic coupling for every cell in a large-scale array. Figure 1 and Figure S6 also indicates that in large arrays with small spacing (e.g., array of 40 × 40 and 20 × 20 elements with spacing of 50 µm), the vast majority of interior cells exhibit highly consistent amplitude and phase responses. A reasonable simplification in RECM can be adopted that all interior cells share identical vibration velocities as shown in Eq. (4).

$${v}_{i}={v}_{j}$$

(4)

When the array has only two PMUT cells ${P}_{i}$ and ${P}_{j}$, their mutual radiation impedance ${Z}_{{ij}}$ can be written as Eq. (5)²⁷.

$$\begin{array}{l}{Z}_{ij}=\frac{9\rho c}{5{A}_{eff}}A(k{a}_{eff})({R}_{ij}+j{X}_{ij})\\ =\frac{9\rho c}{5{A}_{eff}}A(k{a}_{eff})\frac{\sin (kd)+j\,\cos (kd)}{kd}\end{array}$$

(5)

where $d$ is the center-to-center spacing between two cells, $A$ is a polynomial function of $k{a}_{{\rm{eff}}}$ (Figure S18), and ${R}_{{ij}}$ and ${X}_{{ij}}$ are the real and imaginary parts of the mutual radiation impedance, respectively. For an interior cell $i$ in a large array, the total mutual radiation impedance ${Z}_{{multi}}$ contributed by all other cells can be expressed as Eq. (6). Then a nonlinear multi-cell coupling relation is converted into a simpler linear one. These simplified calculations may not produce very exact values about the mutual impedance, but they can provide a reasonable explanation for the variation trend of ultrasonic behaviors.

$${Z}_{multi}=\mathop{\sum }\limits_{j=1,j\ne i}^{N}{Z}_{ij}\frac{{v}_{j}}{{v}_{i}}=\mathop{\sum }\limits_{j=1,j\ne i}^{N}{Z}_{ij}$$

(6)

Figure 1c shows how the total mutual radiation impedance ${Z}_{{\rm{multi}}}$ of a single interior cell $i$ changes with the cell spacing $d$ in an array with $2N+1$ cells. The real part of ${Z}_{{\rm{multi}}}$ represents an effective damping that directly decreases the element′s vibration amplitude, while the imaginary part reflects an effective inertance that elevates the element′s resonance frequency when it decreases. These curves exhibit differences when $N$ is small. As the cell number increases, both the real and imaginary parts tend to a limit, indicating that crosstalk contributions from more distant cells diminish significantly. And the inter-element crosstalk becomes negligible when N exceeds 5, which directly contributes to more uniform vibration across large-scale arrays.

For cell spacing $d$, reducing it increases the number of cells within a fixed array area. Although intensified mutual damping suppresses individual cell displacement, it significantly enhances vibration uniformity in the interior region (Fig. 1a). This improvement should facilitate precise beam-forming and increased ultrasound output. We then proceed to analyze the acoustic field distribution and frequency response to ascertain the array’s performance.

Design of a high-sensitivity and wide-bandwidth PMUT array

Once the vibration behavior of each cell in a PMUT array is determined by ECMA model, the resulting radiated acoustic field can be subsequently calculated as the following Eq. (7).

$$P(r,\theta ,\psi )=\mathop{\sum }\limits_{i=1}^{N}{p}_{i}=\mathop{\sum }\limits_{i=1}^{N}j\omega \frac{\rho {v}_{i}{{a}_{eff}}^{2}}{2{r}_{i}}\left(\frac{48{J}_{3}(k{a}_{eff}\,\sin \,{\theta }_{i})}{{(k{a}_{eff}\,\sin \,{\theta }_{i})}^{3}}\right){e}^{j(\omega t-k{r}_{i})}$$

(7)

where $r$, $\theta$, and $\phi$ are the radial coordinate, polar angle, and azimuthal angle in spherical coordinates, and ${J}_{3}$ is the third-order Bessel function. Figure 2a presents the calculated acoustic fields at the resonance frequency for arrays with increasing spacing but fixed array area (2.2 mm × 2.2 mm). A perfect match between the calculated acoustic field and that simulated via the finite element method (Figure S4) validates the credibility of our proposed ECMA model. As shown, the focal position and acoustic field remains nearly constant since it is determined by the aperture size. However, the corresponding on-axis pressure at the focus as well as the entire field markedly decreases as the cell spacing $d$ increases.

Fig. 2: Acoustic performance of PMUT arrays with varying spacing but fixed array area (2.2 mm × 2.2 mm).

a Simulated acoustic fields at the resonance frequency for arrays with cell spacing of 50 µm, 100 µm, and 150 µm respectively. b Ultrasound pressure responses versus frequencies at 1 cm depth for arrays with cell spacing varying from 40 µm to 150 µm. c Theoretical and experimental normalized ultrasound sensitivity, and (d) the normalized −6 dB bandwidth as a function of cell spacing at 1 cm depth of the array

Figure 2b plots the ultrasound pressure responses versus frequencies at 1 cm depth for arrays with cell spacing varying from 40 µm to 150 µm. Obviously, the output pressure increases as the cell spacing decreases, exhibiting an inverse relationship. This can enhance the signal-to-noise ratio (SNR) of the ultrasound pulse-echo signals, resulting in images with improved contrast. The center frequency increases with cell density in the array and the response curve shows a widening profile, indicating an increase in the ultrasound transmission bandwidth.

Figure 2c, d summarize the dependence of the array sensitivity and −6 dB bandwidth on cell spacing. As confirmed by theoretical calculations and experimental measurements, both the output pressure and bandwidth reach their respective maxima at the smallest studied spacing of d = 50 µm. This result indicates that although smaller spacing reduces the displacement of individual cell, the increased fill factor and effective vibrating area, together with improved uniformity, lead to a higher overall acoustic output. Also stronger acoustic coupling in denser arrays increases the effective damping for cell vibration, consequently broadening their operational bandwidth.

We further characterized other variants of PMUT array: (i) maintaining a fixed cell number (20 x 20) with various overall area by varying the cell spacing from 50 µm to 150 µm (Figures S6–8) and (ii) having a fixed array area (2.2 mm × 2.2 mm) but varying cell numbers via adjusting the spacings from 50 µm to 150 µm (Figures S9–11) with a step of 20 µm. Consistently a decrease in cell spacing results in more uniform vibrational responses across the array (Figs. S6 and S9). From the normalized transmitting ultrasound pressure and bandwidth (Figures S8 and S11), reducing cell spacing in the array improves both the ultrasonic sensitivity and the bandwidth. Under a fixed cell count, the optimal sensitivity does not necessarily coincide with the smallest spacing. But the largest bandwidth consistently occurs at the minimum studied spacing (Figure S8). Also smaller cell spacing elevates the working frequency. To further validate the model, we fabricated an additional PMUT array (1.6 mm × 1.6 mm area) with cell spacing modulated from 90 to 105 µm which operates at ~3 MHz. The experimental data again matches our computational forecasts quite well (Figure S12).

Therefore large-scale PMUT arrays with reduced cell spacing are strongly favored for high-resolution ultrasound imaging. First, the densely arranged cells harnesses the acoustic coupling to promote superior vibration uniformity. Also this array configuration elevates both the acoustic sensitivity and operational bandwidth, thereby concurrently improving the imaging depth and spatial resolution.

Design of an ultra-narrow and wearable phased-array PMUT probe

Wearable ultrasound patch’s geometry critically governs the long-term wearability and signal stability. Wide-format patches frequently require substantial adhesive force to ensure skin contact (Figure S13). Furthermore initial edge lift-off not only degrades image quality but also accelerates complete delamination. Comparatively, narrowing the patch to conform to skin curvature will ensure the consistent comfort and reliable acoustic contact (Fig. 3a).

Fig. 3: Development of a PMUT array for wearable application.

a Schematic of the ultra-narrow PMUT array with high wearability. b Simulated acoustic field distributions of PMUT arrays with different aspect ratios (AR). c Pressure profiles along the x, y, and z lines through the focus point (20 mm from the device surface). d Optical image of the fabricated miniature and wearable PMUT array with the AR = 7:1

Informed by these critical insights, we designed a large-scale narrow silicon-based PMUT array as the core component to enable intrinsic conformity to the skin’s natural contours for wearable applications. This design leverages the substrate’s inherent stiffness to guarantee precise inter-element positioning and geometric integrity during operation. This synergistic approach successfully resolves the traditional trade-off between imaging stability and long-term wearing comfort.

To quantify the difference between ultra-narrow and wide arrays in terms of acoustic performance, we employ our developed ECMA to calculate the three-dimensional acoustic fields of a 64-linear PMUT arrays with different aspect ratios (AR) and compared their main-lobe width as well as sensitivity. We keep the array length $a=1$ cm and reduce the width $b$ to obtain three layouts with AR = 1:1 (square), 3:1, and 7:1 (ultra-narrow), as shown in Fig. 3b. The working frequency of the PMUT elements is 7 MHz. They were in water medium and excited by a single pulse.

The acoustic field (main lobe) and pressure profiles along the X, Y, and Z axes through the focus positions (20 mm above the device) are plotted (Fig. 3c). The 64 linear channels are arranged along the x-axis, functioning as a phased array to generate and focus the acoustic field. This configuration produces an exceptionally narrow beamwidth (the profile width at the −6 dB acoustic pressure) in this dimension that remains nearly independent of the aspect ratio. And the acoustic pressure for the AR = 7:1 design is ~5 dB lower than the other two configurations. Along y-axis, the measured beamwidth demonstrates clear dimensional dependence, being 10.6 mm, 3.1 mm, and 3.4 mm for the AR = 1:1, 3:1, and 7:1 array configurations respectively. A narrower main lobe beamwidth enhances lateral resolution. The beamwidth data demonstrates that designs AR ratios of 3:1 and 7:1 yield comparable resolution. In terms of Z-axis ultrasound output (linked to sensitivity), the ultra-narrow AR = 7:1 array shows ~5 dB and ~3 dB reductions compared to AR = 3:1 and AR = 1:1, respectively. It may impair imaging in deeper tissues but is still acceptable for superficial ones (depth < 3 cm from the skin).

The power consumption of a PMUT array is approximately proportional to its radiating area. Under the same drive voltage, the AR = 1:1 and 3:1 arrays consume 7× and 2.3× the power of the AR = 7:1 array, respectively. Consequently, our developed ultra-narrow AR = 7:1 PMUT array achieves a marginal trade-off in acoustic output while greatly reducing the power consumption, enhancing its wearing comfort, and ‌preserving the imaging resolution. These render it highly suitable for wearable ultrasound applications.

Figure 3d shows a digital photo of our fabricated 64-channel linear PMUT array. Figure S20 provides an SEM cross-section of a single cell. Each channel has 30 × 3 cells in parallel, totaling 5,760 cells. The cell diameter is 40 µm, with 50 µm spacing between cells. The channel pitch is 167 µm (< 80% of the 7 MHz ultrasound wavelength 214 µm). The active area measures 1 cm×0.15 cm. A 5 µm trench between channels minimizes mechanical crosstalk (see Figure S17).

Acoustic and imaging characterization of the PMUT array

Figure 4a illustrates the experimental setup for measuring the ultrasound transmit performance of a single channel in the PMUT array. A 2 mm needle hydrophone (Precision Acoustics, UK) and a DPR500 pulser-receiver (Imaginant, USA) were employed. The PMUT array was driven by a 50 ns negative pulse, while the hydrophone was positioned 10 mm in front of the center of the linear element to record the time-domain pressure waveform. The recorded curve exhibits a peak-voltage of ~44 mVpp. FFT analysis reveals a resonance frequency of ~7 MHz and a −6 dB fractional bandwidth of 90%. Based on the hydrophone sensitivity, the estimated focused pressure under phased-array excitation is ~400 kPa. This value is approximately 40% of the acoustic pressure (>1 MPa) required for high-contrast imaging with conventional bulky probes. Figure 4b illustrates the pulse-echo measurement setup. Under identical drive conditions, one linear element generated a 250 mVpp echo signal with a -6 dB bandwidth of 63%. This value exceeds the 60% of minimum requirement for medical imaging, thereby facilitating high-resolution imaging of superficial tissues.

Fig. 4: Ultrasound imaging characterization of the PMUT array.

a Experimental setup and results for ultrasound transmission performance of one channel. b Experimental setup and results for pulse-echo performance of one channel. c Schematic and imaging results for three human hairs, including the measured imaging resolution in both x- and z-axes

Ultrasound imaging performance was assessed using a Verasonics ultrasound platform (Verasonics, USA), as shown in Fig. 4c. Three ~60 µm human hairs were placed side by side at a depth of 20 mm as microscale targets, with the distances between adjacent hairs being approximately 3 mm and 1.5 mm, respectively. The Verasonics console utilized phased-array imaging across all 64 linear PMUT channels and produced B-mode images. As shown three hairs are clearly resolved, with their cross-sections appearing nearly circular and closely matching the true geometry. The measured full width at half maximum (FWHM) indicates a lateral resolution of ~0.26 mm and an axial resolution of ~0.25 mm for this PMUT array. Non-beamforming mode was also conducted to compare the image contrast and resolution with beam-formed ones (Figure S15). In contrast, non-beamformed images exhibit a signal amplitude reduction of ~15 dB in both the x- and z-axes. The x-axis resolution of the non-beamformed image (0.43 mm) is approximately half that of the beamformed image. Therefore, we employ beamforming mode for human imaging validation with our developed PMUT array.

Multi-organ imaging and central blood pressure monitoring by the PMUT array

The carotid artery, thyroid gland, and dorsalis pedis artery are clinically significant for cardiovascular disease, endocrine disorders, and diabetic foot complications. Convenient imaging of these sites facilitates early detection of subtle lesions and enables timely intervention thereby slowing disease progression and improving patient prognosis.

Carotid artery imaging

Carotid ultrasound is a key tool for evaluating cardiovascular and cerebrovascular risks. It helps visualize atherosclerotic plaques, artery narrowing, and blood flow dynamics. If real-time continuous imaging of the carotid artery becomes possible, it could enable early detection of subtle changes, like blood pressure variations, artery wall thickening, and unstable plaques, leading to more precise and timely treatments.

Carotid system includes the common carotid artery (CCA), carotid sinus (CS), internal carotid artery (ICA), and external carotid artery (ECA). Figure 5a shows carotid imaging results with our developed PMUT patch on a healthy volunteer. In the long-axis CCA view, the anterior and posterior vessel walls and the intima–media complex are clearly visible, and the CCA diameter is ~6 mm. In the short-axis view, the CCA appears as a near-circular lumen with a similar diameter (~6 mm), consistent with the long-axis measurement and indicating self-consistent dimension estimation. The jugular vein (JV) is seen superior to the CCA. When the probe is moved cranially along the neck, the CS appears at the bifurcation as two partly overlapping circular structures. Further upward, the CCA fully bifurcates, and the ICA and ECA appear as two separate circular or near-circular cross-sections.

Fig. 5: Representative ultrasound imaging of superficial tissues with PMUT array.

a Carotid artery. b Thyroid gland. c Dorsalis pedis artery

Movies S1–S4 showcase PMUT array’s capacity for real time continuous monitoring of the CCA long axis, CCA short axis, CS, ICA, and ECA. The rhythmic motion of the vessel wall is observed to synchronize with each heartbeat, while vessel boundaries and surrounding tissues are clearly visible.

Thyroid imaging

Figure 5b shows the ultrasound images of thyroid obtained with our developed PMUT array. The transverse view of the left thyroid lobe reveals a homogeneous, slightly hyperechoic structure with a wing-like shape, while the strong anechoic circular structure on the right side of the thyroid corresponds to the CCA. In the longitudinal plane, the thyroid manifests as a flattened, ovoid structure with uniform contour symmetry, and homogeneous echotexture. They are concordant with normative thyroid morphology in healthy individuals. Movies S5 and S6 are videos about the imaged thyroid for observing its more comprehensive information.

Dorsalis pedis artery imaging

Systemic diseases like rheumatoid arthritis and diabetes can decrease blood flow in the dorsalis pedis artery (DPA). In diabetics, a weak or absent DPA pulse often indicates foot ischemia and microcirculation issues, raising amputation risks. DPA ultrasound imaging aids in assessing disease severity and enabling early intervention. As depicted in Fig. 5c, the DPA represents the distal continuation of the anterior tibial artery. This vessel courses through the superficial subcutaneous layer of the foot dorsum, typically positioned at a depth of < 10 mm from the skin surface. Using the PMUT array, DPA can be clearly imaged: vessel’s anterior and posterior walls are sharply defined, and the measured diameter is about 2.2 mm. A bright high-echo region beneath the artery corresponds to the underlying bones and their acoustic shadow.

Central blood pressure monitoring

Throughout each cardiac cycle, the carotid lumen diameter undergoes periodic variations in response to instantaneous blood pressure fluctuations. When systole, carotid pressure increases as blood is ejected into the systemic circulation, leading to lumen expansion. During diastole, ventricular relaxation and refilling result in arterial pressure decline and lumen recoil. As demonstrated in Fig. 6a, real-time carotid imaging with the PMUT array enables frame-by-frame lumen diameter extraction, facilitating the generation of diameter-time curves and reconstruction of carotid blood pressure waveform.

Fig. 6: Representative central blood pressure monitoring through ultrasound imaging with PMUT array.

a Schematic of reconstructing central blood pressure from real-time carotid images. b Carotid diameter–time curve over two cardiac cycles. c Central blood pressure–time curve derived from the diameter measurements

A continuous ultrasound video about carotid artery was recorded and images were extracted (Fig. S22). Stable video as well as the clear definition of the lumen and arterial wall boundaries show the potential to continuously predict central blood pressure variation. We analyzed two consecutive cardiac cycles and measured the carotid lumen diameter frame-by-frame (Fig. 6b). At end-diastole, the carotid lumen diameter reaches its minimum value of ~5.2 mm, subsequently expanding to ~6.1 mm during systole. The measured values show a 7.7% deviation from those acquired using a clinical ultrasound system (EPIQ 5, Philips, Netherlands), with an example image provided in Fig. S24a. This confirms the reliable imaging accuracy of our PMUT array. Leveraging the established diameter-pressure relationship and blood pressure measurements from a Yuwell wrist monitor (YE8800AR), we derived the arterial stiffness coefficient β (Note S5). By integrating β with the real-time vessel diameter obtained from ultrasound imaging, the central blood pressure waveform is reconstructed (Fig. 6c). More than 5 volunteers were recorded and only one representative curve from our studied volunteers is presented here.

Comparison between our proposed PMUT array and references

To evaluate the imaging capability of our developed PMUT array, we compare its ultrasound images with those generated by a gold-standard clinical 160-channel linear transducer (L12-3ERGO) paired with a clinical ultrasound system (EPIQ 5), as well as reference PMUTs fabricated with PZT or AlN thin films, using a research-grade imaging system (Fig. S24). The key parameters of the proposed PMUT array, including center frequency, bandwidth, array size, imaging resolution, and power consumption, are thoroughly compared with those of reference PMUTs and piezoelectric bulk ultrasound transducers, as detailed in Tables S4, S5 and S6. These comparative analyses offers a detailed assessment of the application feasibility of our proposed transducers in wearable ultrasound imaging.

The transducer’s center frequency of L12-3ERGO linear probe is ~7 MHz. As presented in Figure S24, the clinical diagnostic system generates ultrasound images of the carotid artery and dorsalis pedis artery with excellent contrast resolution and high spatial resolution. Detailed structures surrounding the arteries are also clearly visualized. Comparatively, the images of arteries obtained using PMUT arrays connected to a research-grade imaging system is less graceful (Figure S24c, d, g, h). One reason is that the commercial EPIQ 5 ultrasound image processing system offers superior performance compared to the research-grade platform (e.g., Verasonics) which has minimal image processing capabilities. Another reason is the lower ultrasound output of the PMUT array, resulting in a smaller signal-to-noise ratio (SNR) and reduced image contrast compared to the commercial L12-3ERGO probe.

In comparison to images from PMUT array, those obtained from our developed PMUT array have higher contrast and resolution, probably resulting from a greater number of array elements, increased operating frequencies, enhanced ultrasound power, and expanded bandwidth. Table S4 quantitatively compares the center frequency, ultrasound pulse-echo bandwidth, and SNR between our PMUT array, a typical stretchable piezoelectric bulk ultrasound transducer array, and several representative PMUT arrays. The reference values were derived from the figures in the literatures. Without well-designed backing and acoustic impedance matching layers, the ultrasound pulse-echo bandwidth of bulk piezoelectric transducers is significantly constrained. It can be less than 25% at a center frequency of 2 MHz. PMUTs typically exhibit narrow bandwidths (<20% at about 6 ~ 7 MHz) and poor SNR (~20 dB below bulk transducer), compromising their suitability for high-quality imaging. However featuring an extremely high channel cell density (i.e., fill factor), our proposed PMUT array features a much greater bandwidth (>60% at ~7 MHz) and higher SNR (~10 dB high vs. reference PMUTs), resulting in superior ultrasound imaging quality.

Table S5 lists the transducer size and resolution of this study, along with data from representative reports for comparison. This study employs the smallest elevation size (0.15 cm) yet one of the largest aspect ratios (7:1). Its center frequency (7 MHz) ranks among the highest. Among reported piezoelectric bulk ultrasound transducers integrated into wearable patches, the most advanced device measures 2 cm² in size, with resolutions of 0.38 mm (lateral) and 0.23 mm (axial), and an imaging depth exceeding 6 cm. In contrast, the selected PMUT arrays operating at similar frequencies show suboptimal resolution. Our proposed PMUT array, however, achieves the smallest form factor while demonstrating enhanced lateral resolution (0.26 mm) and comparable axial resolution (0.25 mm) relative to piezoelectric bulk transducers. Moreover it attains a penetration depth exceeding 3 cm, enabling high-resolution wearable imaging of superficial tissues.

Beyond imaging quality, the wearable ultrasound patch must maintain thermal stability during operation. Moreover, it should feature low power consumption to ensure prolonged wear use. As illustrated in Figure S21, the temperature-induced variations of the ultrasound signals of the PMUT array remain nearly constant between 10 °C and 40 °C. During continuous operation, the surface temperature of our developed PMUT array keeps at ~31.4 ± 0.5 °C (Table S6), falling within the temperature range required for signal stability and human skin comfort. When driven by the Verasonics system, the array power consumption is ~15.6 mW (Table S6), calculated based on the drive voltage, device impedance and duty cycle. This power consumption is lower by several times to an order of magnitude than that of piezoelectric bulk ultrasound transducers. Also each imaging session involved a half-day (~5 h) of operating the PMUT probe using the Verasonics system. These experiments were conducted over several half-days, and no measurable degradation in ultrasound signals was observed, preliminarily confirming the reliability of PMUT array for long-term monitoring. Compared to PZT bulk transducers and PZT-PMUTs, our developed AlN-PMUTs are lead-free which offers superior biocompatibility for wearable patch applications. These findings pave the way for the development of high-performance wearable ultrasound patch for continuous monitoring systems.

Conclusion

This study presents an effective design of a very large-scale PMUT array with equivalent circuit model, along with the experimental evaluation of its acoustic performance and human imaging outcomes. The equivalent circuit model facilitates the rapid calculation of each cell’s vibration amplitude and phase in a very large-scale array. Accordingly, a smaller inter-cell spacing promotes a more uniform response across the array and thus enhances the ultrasound beams by phased array technique. Reduced spacing enhances acoustic coupling, thereby broadening the operational bandwidth, though it concurrently reduces individual cell displacement. Nevertheless, uniform cell vibration combined with a larger effective vibration area enhances the overall ultrasound output. Consequently large-scale PMUT arrays with small cell spacing facilitate better imaging depth and resolution. Though narrowing the elevation size slightly lowers PMUT array’s acoustic output, it maintains resolution comparable to larger arrays while significantly enhancing wearing comfort and reducing power consumption. We developed an ultra-narrow 64-channel linear AlN-PMUT array with a area size of 1 cm × 0.15 cm and a center frequency of 7 MHz. This ultra-narrow PMUT array features exceptional uniformity, broadband response, and high ultrasound output. It achieves a pulse-echo fractional bandwidth exceeding 60%, a penetration depth in human tissue of over 3 cm, and lateral/axial resolutions of 0.26 mm and 0.25 mm, respectively. Continuous in vivo imaging of the carotid artery, thyroid gland, and dorsalis pedis artery demonstrates its capability for high-resolution wearable imaging of superficial tissues. During continuous imaging, the surface temperature of our developed PMUT array exhibits minimal variation, while its performance remains nearly constant. The power consumption is approximately 15.6 mW, which is several times to an order of magnitude lower than that of piezoelectric bulk ultrasound transducers. This study lays the foundation for developing high-performance wearable ultrasound patches for continuous monitoring systems.

Materials and methods

Design of PMUT cell

A PMUT cell in the array consists of a bottom-to-top multilayer stack of Mo (100 nm), AlN (500 nm), Mo (100 nm), and SiO₂ (1200 nm) (Figure S1). Each cell has a 40 µm diameter, with the top electrode covering 70% of the cavity to maximize vibration efficiency. Its resonant frequency is ~7 MHz. Note S2 describes more details. The PMUT array is fabricated by sacrificial layer process (Figure S19).

Encapsulation of the PMUT array for wearable application

Encapsulation layer employs a bilayer silicone structure (Figure S16): a 15 µm-thick PDMS film (SYLGARD 184, mixed at a 10:1 ratio) serves as the primary seal, while an Ecoflex coating (Ecoflex™ 00-30, mixed at a 1:1 ratio) provides secondary encapsulation and direct skin contact. Both the PDMS layer and the PMUT surface underwent oxygen plasma treatment to enhance bonding. Ecoflex 00-30 precursors were mixed, degassed, and cast over the device. The coated device was cured at room temperature to form a compliant outer layer.

The silicon PMUT array was mounted on a rigid ~0.5 mm thick PCB which was subsequently integrated onto a flexible PCB. A rigid PCB is placed beneath the PMUT array to prevent silicon breakage, while a flexible PCB is employed to enhance the wearability of the ultrasound probe. The encapsulated ultra-narrow PMUT array was employed to image all locations mentioned in this study. Ultrasound gel was used for enhanced ultrasound transmission between the transducer and human skin when imaging.

Analysis of the PMUT array using equivalent circuit and finite element models

All performance simulations by ECMA were conducted using MATLAB (MathWorks, USA), including the calculations of cells’ vibrations, their acoustic field as well as ultrasound output of large PMUT array. Analyses of acoustic coupling between cells were studied by the simplified model (RECM).

Finite element modeling (FEM) was also performed for comparison to validate the accuracy of the ECM calculated results. A linear channel comprising 20 cells connected in parallel with the inter-cell spacing being 90μm was employed for comparison. Figure S4 displays the ultrasound field obtained from ECM and FEM. The Z-axis acoustic pressure profiles extracted from the FEM and ECM simulations were plotted versus distance from the device surface. The acoustic field achieved by these two analytical methods are nearly the same and the Z-axis acoustic pressure variations from these two analytical methods exhibit nearly identical trends, with the values at each position being highly consistent. This validation confirms the ECM’s ability to produce accurate results, thereby enabling optimized array design. Moreover, the FEM calculation time for the 20-cell array is ~30 min, whereas the ECM requires only ~3 s. This demonstrates the ECM’s significant computational efficiency. And the computational efficiency advantage becomes much more pronounced as the number of cells in the array increases. Field II was employed to optimize the beam-forming parameters for imaging.

The ultrasound imaging platform and specific experiments

A linear PMUT array was coupled to a Verasonics Vantage 256 system via a custom high-shielding coaxial cable bundle for ultrasound excitation and reflected signal reception. The applied signal was a 40 V, single-cycle sinusoidal wave. Signal processing was implemented in MATLAB, encompassing focused excitation control, echo acquisition and storage, and subsequent image reconstruction. Ultrasound images were acquired from over 5 young healthy subjects, including both males and females. Each volunteer was imaged more than two times: once by doctor Zhenzhen Wang (an ultrasound expert) with our PMUT array. Another time was operated by ourselves. As known, ultrasound imaging is strongly dependent on operation experience. Doctor Zhenzhen Wang is an ultrasound expert, she consistently achieves optimal image quality, whereas our imaging of the arteries, although successful, yields comparatively lower quality. The volunteer was seated at rest during measurement. The study was conducted under a protocol approved by the Ethics Committee of Tianjin University (approval number: TJUE2025-H-S-033). In the manuscript, only results from a 26-year-old healthy male volunteer were presented, and statistical analysis was not performed due to varied ultrasound image quality caused by operator dependency.

Processing of the ultrasound videos and images

The ultrasound images acquired via the PMUT array underwent post-processing to remove the random speckle noise and enhance clarity as well as readability. Figure S23 shows examples of the raw unprocessed ultrasound images about the carotid artery, thyroid gland, and dorsalis pedis artery. Linear interpolation was applied to enhance spatial sampling density and smoothness, thereby minimizing artifacts from discrete sampling. A differential denoising strategy of 10 frames was then implemented to suppress random noise. In specific, multiple frames were continuously acquired at the same position. We compared the gray levels across different frames and selected the minimum value as the final output to form one pixel. This approach effectively suppresses random background noise while preserving true echo information. Moderate brightness and contrast enhancement were applied to the reconstructed images, followed by mild smoothing to reduce graininess and enhance visual continuity, thereby accentuating important organs and tissue structures.