A Hand-Held 190+190 Row–Column Addressed CMUT Probe for Volumetric Imaging

This paper presents the design, fabrication, and characterization of a 190+190 row-column addressed (RCA) capacitive micromachined ultrasonic transducer (CMUT) array integrated in a custom hand-held probe handle. The array has a designed 4.5 MHz center frequency in immersion and a pitch of <inline-formula> <tex-math notation="LaTeX">$95~\mu $ </tex-math></inline-formula> m which corresponds to <inline-formula> <tex-math notation="LaTeX">$\approx ~\lambda $ </tex-math></inline-formula>/4. The array has a <inline-formula> <tex-math notation="LaTeX">$2.14\times2.14$ </tex-math></inline-formula> cm<inline-formula> <tex-math notation="LaTeX">$^{2}$ </tex-math></inline-formula> footprint including an integrated apodization scheme to reduce ghost echoes when performing ultrasound imaging. The array was fabricated using a combination of fusion and anodic bonding, and a deposit, remove, etch, multistep (DREM) etch to reduce substrate coupling and improve electrode conductivity. The transducer array was wire-bonded to a rigid-flex printed circuit board (PCB), encapsulated in room temperature vulcanizing (RTV) silicone polymer, electromagnetic interference (EMI) shielded, and mounted in a 3D-milled PPSU probe handle. The probe was characterized using the SARUS experimental scanner and 3D volumetric imaging was demonstrated on scatter and wire phantoms. The imaging depth was derived from tissue mimicking phantom measurements (0.5 dB MHz<inline-formula> <tex-math notation="LaTeX">$^{-1} \text{cm}^{-1}$ </tex-math></inline-formula> attenuation) by estimating the SNR at varying depths. For a synthetic aperture imaging sequence with 96+96 emissions the imaging depth was 3.6 cm. The center frequency measured from the impulse response spectra in transmit and pulse-echo was 6.0 ± 0.9 MHz and 5.3 ± 0.4 MHz, and the corresponding relative bandwidths were 62.8 ± 4.5 % and 86.2 ± 10.4 %. The fabrication process showed clear improvement in relative receive sensitivity and transmit pressure uniformity compared to earlier silicon-on-insulator (SOI) based designs. However, at the same time it presented yield problems resulting in only around 55 % elements with a good response.

apodization scheme to reduce ghost echoes when performing ultrasound imaging. The array was fabricated using a combination of fusion and anodic bonding, and a deposit, remove, etch, multistep (DREM) etch to reduce substrate coupling and improve electrode conductivity. The transducer array was wire-bonded to a rigid-flex printed circuit board (PCB), encapsulated in room temperature vulcanizing (RTV) silicone polymer, electromagnetic interference (EMI) shielded, and mounted in a 3D-milled PPSU probe handle. The probe was characterized using the SARUS experimental scanner and 3D volumetric imaging was demonstrated on scatter and wire phantoms. The imaging depth was derived from tissue mimicking phantom measurements (0.5 dB MHz −1 cm −1 attenuation) by estimating the SNR at varying depths. For a synthetic aperture imaging sequence with 96+96 emissions the imaging depth was 3.6 cm. The center frequency measured from the impulse response spectra in transmit and pulse-echo was 6.0 ± 0.9 MHz and 5.3 ± 0.4 MHz, and the corresponding relative bandwidths were 62.8 ± 4.5 % and 86.2 ± 10.4 %. The fabrication process showed clear improvement in relative receive sensitivity and transmit pressure uniformity compared to earlier siliconon-insulator (SOI) based designs. However, at the same time it presented yield problems resulting in only around 55 % elements with a good response.

I. INTRODUCTION
I N the recent years, there has been a growing interest in developing 2D ultrasonic transducer arrays for performing 3D volumetric imaging with resolution comparable to conventional 2D imaging made with linear 1D arrays.
For good focusing and high resolution 3D ultrasound imaging a large probe with a significant number of elements, N , is needed. 2D arrays with elements in both the azimuthal and elevation direction are typically fabricated as fully populated matrix (FPM) arrays, where the number of elements and therefore also the number of individual connections each scale with N 2 . A linear probe might have 190 elements, which for a matrix array having the same resolution in both directions would result in 190 × 190 = 36,100 connections. This would require highly impractical bulky cables connecting the probe to the scanner in addition to electronics capable of handling the data rates, which would be several terabytes per second for such an array.
The focusing ability of ultrasound probes is proportional to the wavelength and the ratio of the imaging depth to the width of the probe. The width is equal to the pitch times the number of elements N . Maintaining the same resolution in both dimensions necessitates the same number of elements, hence N 2 connectors. Also maintaining a good resolution demands large probes to maintain the ratio between depth and width.
State-of-the-art FPM array probes based on lead zirconium titanate (PZT) crystal materials can already be acquired commercially from e.g., Philips (X6-1 xMATRIX array transducer with 9212 elements) and for research solutions through Verasonics (Matrix Array Transducer with 1024 elements (32 × 32)). To achieve a higher channel count an effort has been made to design electronics performing pre-beamforming that fit inside the probe handle [1], [2], [3], which has also been made possible in the Philips X6-1 xMATRIX probe. Integrated electronics and the use of PZT materials, however, still have problems concerning probe heating under continuous use [4], [5].

A. ROW-COLUMN ARRAYS
Recently, a new type of array scheme has been developed which can reduce the number of channels needed for imaging and keep the same array footprint without resorting to e.g., using sparse arrays [6], [7]. This is the so-called row-column addressed (RCA) array which was proposed by Morton and Lockwood in 2003 [8] and reiterated by Démoré et al. in 2009 [9]. An RCA transducer is composed of two 1D arrays placed orthogonal to each other, with sub-elements in overlapping segments. Imaging sequences are likewise different, in the sense that instead of actuating each single sub-element and selecting sub apertures in the matrix array, entire row or column elements encompassing potentially hundred of sound emitting sub-elements are excited simultaneously. Row or column elements are used for transmitting a line focus, while the orthogonal elements are grounded. For receiving, opposite elements are often used to focus in a perpendicular line while the previously transmitting elements are grounded. This creates a focus point in the volume where the two lines overlap, which can be used for volumetric imaging. Since only 2 × N elements are used this significantly reduces the number of connections needed. Morton and Lockwood [8] state that beamforming two perpendicular planes with an RCA array results in a loss of signal strength compared to an N × N FPM array, and that 4 × N elements are needed to get a comparable resolution. It has, however, been shown by Rasmussen and Jensen [10] that RCA arrays always have a higher resolution than FPM arrays for the same number of matrix elements (32 + 32 rows and columns compared to 8 × 8 matrix elements). More details about the imaging abilities of RCA arrays can be found in [11].
Row-column arrays have been fabricated using PZT or other piezoelectric materials, including a 1-3 ceramic RCA 64 × 64 array [12], an RCA 256 × 256 array [13], and a 7.5 MHz dual-layer transducer with 256 PZT elements used for transmit combined with 256 orthogonal P[VDF-TrFE] copolymer elements for receive [14]. An alternative version of the row-column scheme named top-orthogonal-to-bottom electrode (TOBE) using an electrostrictive ceramic PNM-PT was presented in [15]. Recently, a 62+62 PZT row-column array integrated in a hand-held probe was demonstrated [16]. Commercial row-column probes using PZT are available from Vermon and Verasonics (RC6gV Row-Column Array Transducer), using 128+128 rows and columns with a 6 MHz center frequency. A similar prototype probe with 128+128 rows and columns with a 12 MHz center frequency was introduced by Daxsonics.
PZT probes are often fabricated using the dice-and-fill method [17], where the kerf is limited by the width of the dicing blade to around 15 µm [18]. This imposes restrictions on the transducer design, especially at higher frequencies where the element width becomes comparable to the kerf for small pitch probes. Furthermore, high frequency probes require a small thickness of the PZT material, which makes such probes increasingly difficult to fabricate. Therefore, alternatives to fabricating probes using PZT are needed.

B. CAPACITIVE MICROMACHINED ULTRASONIC TRANSDUCERS
One alternative to the piezoelectric transducer is the capacitive micromachined ultrasonic transducer (CMUT) based on silicon microfabrication. It was first demonstrated using sacrificial release methods [19], and later with fusion bonding (also called direct bonding) [20]. This structure uses a thin vibrating plate or membrane for transmit and receive. CMUTs, when compared to conventional PZT based transducers, do not have any significant self-heating due to low internal loss and high thermal conductivity [21], [22] and also provides higher frequency bandwidth [23]. These transducers can be made with microfabrication and the single CMUT units, referred to as cells, can be defined through the use of simple lithographic processes. The devices can therefore be made smaller, and the pitch can more easily be controlled than what is usually possible with dicing for piezoelectric transducers. This allows a pitch of λ/2, which minimises grating lobes [24], to be more easily kept for higher center frequencies, and easier integration with integrated circuits on chip.
Since the first use for imaging [25] CMUTs have been successfully used for 1D linear array probes by several academic and industrial research groups [23], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35] and more recently such probes have become commercially available. The first 2D array using CMUTs based on the row-column addressing scheme had 32+32 elements fabricated using wafer fusion bonding, where the cavities and membranes were both made with silicon nitride. Since then several row-column CMUTs VOLUME 2, 2022 have been designed and fabricated with a variety of different methods, such as sacrificial release [36], [37] using a TOBE array architecture, adhesive wafer bonding [38], [39], fusion bonding [16], [40], [41], [42], [43] and anodic bonding [44], [45], [46]. Some of these arrays have been integrated in handheld probes, presented in [16] as a 62+62 RCA 2D CMUT probe together with a corresponding PZT probe for comparison, and in [43] where two 92+92 RCA 2D CMUT probes with and without integrated diverging lenses are presented. These probes showed the potential for using CMUTs for RCA arrays and demonstrated that such probes can attain similar performances as a PZT probe made using the same dimensions. However, it was found that capacitive substrate coupling [47] and the resistance of the electrodes [48] used for the CMUT devices is important for the performance of large CMUT arrays and needs to be optimized.

C. REQUIREMENTS FOR LARGE ARRAY DESIGNS
When designing large RCA arrays there are a number of important criteria to ensure optimal performance. The pressure emitted along each element, rows and columns, and across neighbouring elements should be uniform. If this is not the case, the pressure field of the array when performing imaging will not match simulated fields and can be detrimental to the image quality. Likewise, the receive sensitivity of both rows and columns should be equal to ensure even performance.
Considering the first requirement, the distribution of the electric potential along the CMUT elements becomes important. When a high frequency excitation (AC) signal is applied on a long element the electrode resistance can have significant influence on the transmit pressure uniformity. If the resistance is too high the system will behave as a low-pass filter and attenuate the applied signal along the element. This effect can be modelled as a delay line [48]. The value of the dimensionless product ωRC, being a combination of the angular excitation frequency, ω, the element electrode resistance, R, and the total capacitance of the element, C, can be used as a criterion to minimise adverse attenuation. To keep a uniform pressure distribution, the potential drop along the element is set to a maximum of 1 %. This corresponds to an ωRC value of This can also be expressed in terms of the sheet resistance, R , the element or electrode length, L, and the capacitance per area, C This allows one to predict the signal drop and adjust the design parameters without the need to directly measure the electrode resistance. It furthermore shows that the choice of R = ρ/t, which depends on the electrode resistivity, ρ, and thickness, t, becomes much more crucial for larger arrays as the effect scales with the area, L 2 .
The second requirement regarding the sensitivity depends on how well the substrate coupling or cross-talk between elements [47] can be suppressed. CMUT elements which are fabricated with a silicon substrate often utilise a siliconon-insulator (SOI) wafer during fabrication with an insulating layer to separate bottom electrodes from the handling substrate [36], [41], [49], [50]. This allows the elements to capacitively couple to each other through the substrate, giving rise to parasitic capacitance which will lower the receive sensitivity of the array [47]. A way to mitigate this could be to increase the signal path length between the elements by partially separating them with an etching process [51]. An alternative solution could be removing the electrical path through the substrate completely [38], [52], which can be realized by utilising an insulating handling substrate as an alternative to silicon.
For the first effect, the ωRC criterion sets requirements for the resistivity and thickness of the material used to fabricate the top and bottom electrodes and the fabrication techniques needed. Two main types of electrodes determines which techniques that can be used; metal electrodes, which will require processes with a low processing temperature or thermal budget, and silicon-based electrodes allowing for a higher thermal budget.
If silicon is used as bottom electrode material, a limit of around 10 21 cm −3 is set as the highest doping level, which can be achieved, corresponding to a resistivity of 10 −4 cm [53]. This can prove problematic for long arrays when considering ωRC. CMUTs fabricated with a silicon substrate using fusion/direct bonding process [54], [55] requires a high temperature post annealing step at > 1000 • C to fuse the dielectric insulation layer with a top plate. Such a device is illustrated in Fig. 1(a) fabricated on an SOI wafer. The temperature step makes this technique incompatible with metal bottom electrodes, and one has to rely on the low resistivity of doped silicon for the bottom electrodes and increase the electrode thickness to keep the resistance sufficiently low.
Bottom electrodes made of metal are typically fashioned e.g., in Al, Cr, or Au. In comparison to doped silicon, metal electrodes have a resistivity on the order of 10 −6 cm [56], which being two magnitudes lower than doped silicon, will reduce resistance problems and be sufficient for most applications, depending on the electrode shape. The low thermal budget limits the number of techniques which can be used for CMUTs to mainly three types. 1) Adhesive polymer bonding [57] using e.g., BCB [38], see Fig. 1(b). This is a versatile method for joining different types of substrates and top plates with a polymer spacer, which can be patterned as a standard UV-sensitive resist and used with both silicon and metal bottom electrodes. 2) Sacrificial release methods [58], which can be used with both metal and silicon electrodes, see Fig. 1  metal electrodes are deposited, and the substrate is bonded to a silicon top plate, see Fig. 1(d). This has a maximum processing temperature of 375 • C.
In contrast, CMUT top electrodes are typically made using a wafer-bonded plate of silicon, which is then covered with a thick metal layer of low resistivity. For typical electrode designs the electrode resistance will not affect the pressure field [51]. The metal deposition happens as one of the last process steps and is not affected by previous high temperature processes. It can therefore be used in combination with various types of CMUTs.
Ideal structures for low resistance signalling would be combining a glass wafer patterned with CMUT cells, using a metal bottom electrode [44], [46], [59]. Such devices using dielectrics for the membrane or etching for structuring cell cavities can, however, display stability issues and dielectric charging of the device [60], [61], [62]. However, a fabrication process based on an SOI substrate together with the local oxidation of silicon (LOCOS) technique [55] and fusion bonding have been shown to exhibit little to no charging [50]. The structure is illustrated in Fig. 1(a). The LOCOS process, when used for CMUT fabrication, allows for reduced parasitic capacitance, high dielectric strength, good uniformity and tightly controllable CMUT dimensions and vacuum gap height down to under 10 nm precision due to the predictability of the oxide thickness [63].

D. NEW CMUT DESIGN
In this paper, we present a hand-held RCA CMUT based probe for volumetric imaging with a potential use in medical applications. This device is based on a LOCOS process fabricated on a doped silicon wafer, in combination with fusion bonding to an SOI wafer and a physical separation of the bottom electrodes to reduce cross-talk between the elements [52]. This is realised using a deep reactive-ion etching (RIE) based process. For improved mechanical stability, while also providing element-to-element insulation and low capacitive substrate coupling, a borofloat glass wafer is anodically bonded to the backside of the elements after the separation process. The separated electrodes have a cross-section of almost 100 × 100 µm 2 which combined with a highly doped silicon substrate will mitigate delay-line effects. The fusion-anodic double bonded structure is illustrated in Fig. 1(e) and as a cross-sectioned 3D sketch in Fig. 2. This device should benefit from stable high-performing CMUTs while also exhibiting low element-to-element coupling and high transmit pressure uniformity.
The paper is organized as follows: Section II is divided into a part A and B. The first part introduces the general design parameters of the presented CMUT array. The second part describes in detail the cleanroom fabrication of the device. Section III describes the assembly of the transducer probe. Results from the thermal, electrical and acoustical characterizations are presented in Section IV and discussed in Section V. Ultrasound imaging and measurements of sensitivity and penetration depth performed with the probe is described in Section VI. Finally, a conclusion is drawn in Section VII.

A. DESIGN
The RCA array design is a symmetrical square with 190 row elements and 190 column elements used for beamforming, divided into rectangular segments of 95 odd and even elements of alternating pad numbering on each side of the array.  A 3D sketch of the corner of an array is shown in Fig. 2. This cut-off of the array shows four orthogonal top and bottom electrodes highlighted in orange and blue, respectively, as well as the underlying cells and element separation. Bordering the central region of the array are four apodization regions as shown in the boxed section in Fig. 3, which each are accessible from a single contact pad in the corners of the array not shown in Fig. 2. This raises the total channel count to 192+192. The apodization is incorporated into the design to round off the signal towards the edges of the array with a Hann window function. This reduces the relative signal amplitude from 1 to 0 at the edge of the apodization region and thereby suppresses side lobes and ghost echoes. The signal is decreased over nine cells with an increasing intercell distance. The RCA CMUT array is designed with an element pitch of 95 µm. The elements are 92.5 µm wide and contains a single row of circular cells which are 70 µm in diameter. If the plate thickness is chosen as 4.0 µm and the vacuum gap height is 196 nm this will give a center frequency of 9.15 MHz in air and ≈ 4.5 MHz in immersion using a pull-in voltage of 190 V. The center frequency in immersion was found using the following equation obtained by Lamb [64] and revisited by Amabili and Kwak [65]. Here, ω r /ω 0 is the ratio between the resonant frequency in water and vacuum, β is the added virtual mass incremental factor (AVMI) consisting of the medium to plate density ratio, ρ m /ρ p , the radius, a, and the thickness of the plate, h. is the non-dimensionalised added virtual mass incremental (NAVMI) factor which can take multiple values depending on the clamping conditions. The value stated by Lamb [64] of = 0.6689 is chosen. The center frequency in immersion is predicted to be around 49 % of the center frequency in air (9.15 MHz at 80 % of V pull-in ).
The center frequency, gap height, and plate thickness were determined using the finite element method (FEM) simulators OnScale and COMSOL.
The remaining design parameters are stated in Table 1. The previously discussed delay-line effect above in Section I, will theoretically for one element of the 190+190 RC array, using C = 23.6 pF and R = 556 , approximately equal ωRC ≈ 0.38 (4) at a frequency of 4.5 MHz. This corresponds to a drop in signal along the element of approximately 1.2 %. This is deemed an acceptable loss considering that the silicon substrate used for fabrication has a resistivity of 0.025 cm or less. For the top electrodes, the resistance is R = 15 and an ωRC value of less than 0.01 is found showing that metal electrodes are efficient in suppressing this effect. The four inch wafer design features eight RCA arrays each with a footprint measuring 2.14 × 2.14 cm 2 . Furthermore, 16 smaller 16+16 RCA arrays as well as linear arrays are included for electrical and acoustical testing.

B. FABRICATION
The fabrication of the CMUT array was based primarily on the LOCOS process [55] for structuring the cells and fusion bonding for encapsulation. As mentioned in Section II-A a physical trench separation was used to isolate the bottom elements of the array. The separation etching process used was developed at DTU Nanolab and is a modified 3-step Bosch process called deposit, remove, etch, multistep (DREM) [66]. The multiple steps have been fine-tuned to eliminate maskerosion during etching (achieving so-called ''infinite'' selectivity [66]) and preserve scallop and hole uniformity even for high aspect ratios.
The fabrication process illustrated in Fig. 4 started with a highly doped 525 µm thick silicon wafer having a resistivity of ρ < 0.025 cm corresponding to a donor doping level, N d , of 10 19 cm −3 . First a silicon dioxide layer of 375 nm, used for insulation, was grown in a dry thermal oxidation process at 1100 • C, then a low pressure chemical vapour deposition (LPCVD) silicon nitride of 55.6 nm and an LPCVD polycrystalline silicon (poly-Si) layer was deposited on top (using Tempress horizontal furnaces), see step 1). The poly-Si layer was then patterned with circles representing cells in a photolithography step with the diameter stated in Table 1 and etched using a poly-Si etching solution (HNO 3 :BHF:H 2 O (20:1:20)), step 2). The pattern was transferred into the nitride using hot phosphoric acid (H 3 PO 4 at 160 • C) and the poly-Si masking layer was stripped using RIE (SPTS Pegasus). This will leave the nitride masking pads on top of the oxide in place of the cavities, step 3).
At this point the oxide surface was patterned with resist and trenches were etched using RIE (advanced oxide etcher (AOE) STS MESC Multiplex ICP) into the oxide in the kerf between the elements to expose the underlying silicon substrate, step 4). Then, the DREM process was performed at a temperature of −19 • C using the SPTS Pegasus for a trench etch depth of around 100 µm, step 5). These can be seen in Fig. 5, where deep straight trenches have been etched in silicon.
The wafer was then cleaned in RCA cleaning solution, and a second thermal oxidation process was performed in a wet oxidizing environment at 1100 • C to grow the post oxide to a total thickness of 825 nm, step 6). This formed the cavities through the LOCOS process with a gap height of 196 nm, excluding the nitride pad thickness. The device wafer was RCA cleaned again together with a poly-silicon-on-insulator (PSOI) wafer [67], which has been custom made to match the desired plate thickness. The structured device wafer and the PSOI wafer were then fusion bonded together, illustrated in step 7), at 400 • C with a tool pressure of 4 bar (performed on a Süss SB6 wafer bonder) and then subsequently annealed at 1100 • C for 70 min to form permanent silane bonds.
The oxide layer on the backside of the bonded wafer stack was removed with a BHF solution, step 8). Most of the silicon substrate was removed using lapping (Logitech PM5 Lapping & Polishing System), leaving approximately 150 µm to 180 µm. To completely separate the bottom electrodes from the backside the remaining 50 µm to 80 µm was etched using RIE (advanced silicon etcher (ASE) STS MESC Multiplex ICP)), step 9), exposing the trenches and oxide from step 6), seen in Fig. 6.
The backside surface was then polished to remove the free-standing oxide and to reduce the roughness necessary for anodic bonding, step 10). This was performed using a Logitech CM62 Orbis CMP (Chemical Mechanical Polishing) machine. A 500 µm thick borosilicate glass wafer was anodically bonded to the 100 µm thick electrodes on the backside using a four step voltage ramp (200 V/400 V/600 V/800 V) at an elevated temperature of VOLUME 2, 2022 FIGURE 6. Scanning electron microscope image of a test structure illustrating the cross-section of the backside of the wafer after step 9), before polishing. The exposed DREM trench oxide walls separating the bottom electrodes are clearly visible. This was also previously presented in [52]. 375 • C, step 11). The top poly-Si layer, two buried oxide (BOX) layers and the bulk of the PSOI handle wafer were then removed in a combination of dry and wet etching using RIE, BHF, KOH at 80 • C, and BHF, which left only the poly-Si top plate, step 12). The dashed line illustrates a shift in the perspective between the cross-section of the bottom electrode to the left of the line and the top electrode to the right, respectively.
Holes for contacting the bottom electrode pads were made by etching through the poly-Si plate and post oxide using a resist mask with a RIE ASE process tool, step 13). During this process it was discovered that not all post oxide was etched for some of the contact pads due to the formation of a sulphur compound, see Fig. 7. Multiple cleaning steps with RCA, HCl, HNO 3 , and Piranha were tried without success. As a result, this prevented a complete access to the incomplete pads making wire bonds unreliable and lowering the electrode yield. The wafer surface was then coated in 400 nm aluminium (utilizing a Temescal FC-2000 e-beam evaporator), which was patterned using a PES Al etching solution [68] to form the top metal electrodes. Finally, the plate was etched through on the ASE to separate the top electrodes completely, step 14).
The last step was dicing the wafer using a (DISCO DAD-321) dicing saw.

III. ASSEMBLY
The diced array chip was mounted and glued to a rigidflexible four-armed printed circuit board (PCB), and wire bonding was performed to connect the individual element contact pads on the chip to the PCB. The top and bottom electrodes were designated as rows and columns, respectively, and each of the four sides of the chip, of either odd or even elements, were wire bonded to each PCB arm for a total of 384 channel connections, depicted in Fig. 8. The assembly of the transducer probe was done at the facilities of BK Medical (State College, PA, USA). A tall glob-top dam was glued to the edges of the rigid PCB to ensure that neither the wire bonds or the chip were damaged by external mechanical stresses during scanning or assembly. The array was then encapsulated by filling the dam with room temperature vulcanizing (RTV) silicone, RTV664, described in more detail in [16], [69]. The thickness of the RTV directly on top of the chip was 0.485 mm on average. On top of the shield, the final layer of RTV was 0.595 mm on average. The attenuation of the RTV was 2.3 dB/mm at 4.5 MHz, so the total one-way attenuation is 2.5 dB for this array. An aluminised polymer film (12.5 µm polypropylene with a sub-micron thick aluminium layer), used as electromagnetic interference (EMI) ground shielding, was applied before the silicone cures and the PCB was mounted in a 3D milled PPSU probe nose-piece. Another layer of silicone was applied to the array surface and levelled to the probe edge, thus completely sealing and insulating the array.
The four sections of the rigid-flexible PCB were folded and connected to four preamplifier boards. The boards are each equipped with six MAX14822 16-channel high voltage (HV)-protected transimpedance low noise active amplifier ICs with a bandwidth of 45 MHz. These amplifiers are capable of supplying high voltage AC in transmit (TX) and receive (RX) individually on each channel of the probe superimposed on a DC bias. Each board has support for 96 channels, and two boards, pairing either the odd or even element, are connected with board-to-board connectors. Each pair of amplifier boards were connected to a 192 odd or even channel coaxial scanner cable (BK Medical, Herlev, Denmark). The board pairs were then shielded in Kapton and copper tape. Two cooling hoses for inlet and outlet were then encapsulated along with the boards within the probe shell by two 3D milled shell pieces, effectively sealing the transducer probe handle. The assembled probe is seen in Fig. 9.

A. THERMAL
The probe handle is fitted with an inlet and outlet tube for air cooling of the amplifier boards during scanning. An experiment was performed to evaluate the cooling capability during idle mode and when emitting a suitable imaging sequence. Temperature measurements were performed in air, see Fig. 10, on the side of the assembled probe body and on the front sole of the nose piece using two FLIR C2 thermal imaging system cameras mounted on a frame. These measurements showed that the idling temperature of the probe when supplied with a DC voltage of 190 V, corresponding to 160 V at the chip level, gives an average temperature at ≈ 31 • C on the sole (blue) and ≈ 36.5 • C on the side (red) after 25 min. An imaging sequence with a synthetic aperture (SA) using 18 elements, 192 emissions, a pulse repetition frequency of 12 kHz, and 75 V peak-to-peak was then used  as excitation signal. The surface temperature of the sole rose ≈ 2 • C during the measurement, reaching a maximum average of 32.8 • C (yellow) after 5 min. The probe side had a stable surface temperature of ≈ 36.5 • C (purple) throughout the measurement which is caused by the underlying voltage regulator on the amplifier boards. The temperature rise and surface temperature of the transducer body and sole when idling and imaging were both found acceptable for experimental external use below the FDA and DS/EN limits for scanning [70]. The airflow inside the current probe design is, however, likely restricted by the preamplifier boards and their shielding and the internal temperature is probably higher than what is measured on the surface.

B. ELECTRICAL
The electrical response of the CMUT array has been characterized before it was mounted in the probe handle using impedance measurements. These were performed using an Agilent 4294A Precision Impedance Analyzer. During the measurements the bias voltage was supplied through a bias tee by a Keithley 2410 sourcemeter.
Measurements were performed on separate linear test arrays and single test elements located in the corners of the 192+192 arrays to provide an accurate pull-in voltage for the array. The DC voltage supplied by the sourcemeter was varied between 0 V to 200 V while a 50 mV AC voltage from the impedance analyser was superimposed on top. The voltage sweep could then be used to find the pull-in voltage. With a constant supplied DC bias the array showed a stable performance and exhibited no charging. This showcases the effectiveness of the LOCOS fabrication process. Fig. 11 shows measurements, performed on a test element, of the impedance magnitude and phase from 1 MHz to 25 MHz for an applied voltage of 150 V. A pull-in voltage of 186 V was measured which compares well with the designed pull-in voltage of 190 V. The measured center frequency was 9.25 MHz at a bias of ≈ 80 % of the pull-in voltage which is in good agreement with the COMSOL simulated value of 9.15 MHz in air. Using the same bias voltage the electromechanical coupling factor is calculated to be k 2 = 4.5 %.

C. ACOUSTICAL
Acoustical characterization of the assembled probe was performed on the experimental research scanner SARUS [71] using the approach in [72] and the transducer impulse responses were determined as described in [73].

1) IMPULSE RESPONSE
The one-way impulse and two-way pulse-echo responses were measured by submerging the probe in deionized (DI) water with either a hydrophone or a planar steel reflector placed 35 mm from the transducer surface, respectively. Emitting and receiving was done with one element at a time for all rows and columns, respectively. The DC bias was set to 160 V and the AC excitation voltage up to 150 V peak-to-peak (±75 V). The voltage levels were set to reflect the measured pull-in voltage mentioned in Section IV-B. The hydrophone used to measure the pressure in immersion was an Onda HGL-0400 hydrophone connected to an Onda AH-2010 amplifier. The elements have been grouped into three categories depending on their maximum impulse response values from the acoustic measurements in transmit and in receive, see Fig. 12. The overall element yield of the finished array in transmit has been measured to 109 (57.4 %) for the rows and 123 (64.7 %) for the columns. This was based on elements with an impulse response maximum value over 1.1 × 10 15 Pa V −1 s −2 , designated as functional elements. The semi-functional elements have a lower response between 2.9 × 10 14 Pa V −1 s −2 and 1.1 × 10 15 Pa V −1 s −2 (52 rows and 32 columns) and the defect or not connected (NC) elements have less than 2.9 × 10 14 Pa V −1 s −2 (29 rows and 35 columns). Only the functional elements have been used for further acoustic characterization. The element yield measured in receive (pulse-echo) was 65.8 % and 43.7 % for functional rows and columns, respectively, with a cut-off at 1.4 × 10 −4 V V −1 . For semi-functional elements it was 1.1 % and 12.6 % for rows and columns, respectively, with a cut-off at 6.5 × 10 −5 V V −1 , and for NC elements the yield was 33.2 % for rows and 43.7 % for columns. The yield problems seem to be related to the formation of the sulphur compound mentioned in Section II-B.
The average transmit impulse response of the functional rows and columns can be seen in Fig. 13. The solid blue and red lines represent the response measured in Pa V −1 s −2 of the rows and columns, respectively, and the dashed lines in the same colour represent the envelope of the signal. The maximum impulse response values are between 1.3 × 10 15 Pa V −1 s −2 to 2.5 × 10 15 Pa V −1 s −2 . It can be seen from the data that the signal transmitted from the rows is ≈ 55 % of the response of the columns. The received pulse-echo signal in Fig. 14 shows a longer pulse with a more pronounced ringing for the rows compared to the transmit measurements. The ratio between the row and column main peaks is around 2.4 with the rows now having a higher amplitude than the columns. The average transmit sensitivity of the rows and columns was found to be 1.3 ± 0.1 × 10 15 Pa V −1 s −2 (rows) and 2.4 ± 0.6 × 10 15 Pa V −1 s −2 (columns), with a total average of 1.9 ± 0.7 × 10 15 Pa V −1 s −2 . The sensitivity for the pulse-echo measurements was 3.8 ± 0.7 × 10 −4 V V −1 (rows) and 1.8 ± 0.2 × 10 −4 V V −1 (columns), with a total average of 3.0 ± 0.2 × 10 −4 V V −1 .

2) CENTER FREQUENCY AND UNIFORMITY
The center frequencies of the rows and columns were found by calculating the weighted mean of the frequencies in the Fourier transformed impulse responses of the transmit and pulse-echo signals plotted in Fig. 15 and Fig. 16 using the expression: with f s being the sample frequency of 70 MHz and N the number of frequency bins in the spectrum, S. In Fig. 17 the extracted center frequency f c is plotted for each element showing the uniformity across the array in transmit ( Fig. 17(a)) and pulse-echo ( Fig. 17(b)). The average center frequencies of the functional row-and column elements were 5.0 ± 0.1 MHz (rows) and 6.85 ± 0.30 MHz (columns) with the total average being 6.0 ± 0.9 MHz in transmit, and 5.3 ± 0.2 MHz (rows) and 5.5 ± 0.5 MHz (columns) with the total average of 5.3 ± 0.4 MHz in pulse-echo. The peak frequency was also extracted from the spectral data and gave an average of 5.0 ± 0.6 MHz and 4.5 ± 0.5 MHz in transmit and pulse-echo, respectively, also seen in Table 2.
These values correspond reasonably well with the predicted resonance frequency in immersion of 4.5 MHz in both transmit and pulse-echo. However, the center frequency of the columns in transmit and the rows in pulse-echo both lie between 1 MHz to 1.5 MHz higher when biased at 86 % of V pull-in .

3) BANDWIDTH
The frequency bandwidth (BW) of each element was found at the −3 dB and −6 dB points of the Fourier transformed impulse responses of the transmit and pulse-echo measurements, respectively. The mean BW for the functional rows and columns in transmit was 3.3 ± 0.2 MHz and 4.2 ± 0.4 MHz, respectively, with a total mean of 3.7 ± 0.5 MHz. In pulse-echo the mean BW was 4.2 ± 0.1 MHz and 5.3 ± 0.8 MHz for rows and columns, respectively, with a total average of 4.6 ± 0.7 MHz. The relative fractional bandwidth for the rows and columns in transmit was 65.2 ± 3.7 % and 60.7 ± 4.0 % with a total mean of 62.8 ± 4.5 %. In pulse-echo this was 79.5 ± 3.7 % and 96.4 ± 87.7 % for the rows and columns, respectively, with a total mean of 86.2 ± 10.4 %.

4) ωRC -PRESSURE UNIFORMITY
The effect of the electrode resistance on the transmit uniformity has been estimated by measuring the pressure field along the top and bottom electrodes. The pressure field was mapped in immersion by moving the Onda HGL-0400 hydrophone in the x-y plane in steps of 0.5 mm with a distance of 5 mm from the transducer surface. This creates a grid of 49×49 measurements with an area of 24.5 mm × 24.5 mm which completely captures the array. A 4-cycle sinusoidal 6.5 MHz pulse with an amplitude of V AC = ±75 V has been used to excite three evenly spaced row elements and three column elements. The pressure maps shown in Fig. 18 are each an average of three elements and the plots have been normalized to their maximum value and log compressed. In Fig. 18(a) for the top electrodes (rows), it is seen that the pressure distribution is even along the elements. The mean value of the first and last 1/4 part of the element has been calculated as an estimate of the uniformity as −2.4 dB and −2.9 dB. Fig. 18(b) depicts the bottom electrodes (columns) for which similar values have been calculated as −1.9 dB near the contact pads and −3.0 dB near the end of the elements. This corresponds to a drop to 88 ± 16 % of the initial amplitude, or 12 % attenuation. The ωRC value for the measurement frequency is 0.55, which corresponds to a drop to 97.6 % of the initial value or 2.4 % attenuation. The measured attenuation is larger than the predicted value which could be caused by the low yield of the orthogonal elements providing the grounding during the measurement.
This value can be compared to an attenuation of 74 % measured on a previously fabricated 92+92 RCA array with a center frequency of 4.5 MHz and a bottom electrode resistivity of < 0.1 cm [48].
The data depicted in Fig. 18, which is an average of three emitting row and column elements, exhibited large variations in the recorded pressure and this likely causes the large standard deviations of the attenuation.

V. IMAGING
The imaging capabilities of the probe was evaluated using the SARUS system by scanning a wire phantom, a cyst phantom, and a stereolithography 3D printed hydrogel phantom with isolated 205 × 205 × 80 µm 3 embedded cavities, which function as point scatter targets [74]. The scatterers were placed in a 6 × 4 × 4 grid with a spacing of 2.05 mm. 3D imaging was performed using a SA sequence with 96 row and 96 column emissions and three orthogonal planes are shown in Fig. 19. From this the resolution of the point spread function (PSF) could be determined in the axial direction as 0.82 λ (0.1880 mm) and in the lateral direction as 1.72 λ (0.3936 mm). The side-to-main lobe level was fairly high at −11.90 dB due to scattering from the phantom surrounding the point cavities.
The resolution is visualized in Fig. 20, which depicts three different planes from the volumetric scan of a wire phantom. The phantom is a matrix of wires immersed in water with little attenuation. The x-z (lateral) plane shows the wires as horizontal lines, the y − z (azimuthal) plane shows two columns of wire cross-sections as dots, and the x −y (transverse) plane depicts a single wire. The resolution in this phantom was in the axial direction 1.16 λ (0.2654 mm) and in the lateral direction 1.56 λ (0.3562 mm). The contrast was −16.90 dB due to the many missing elements.
The penetration depth was found at the point when the signal to noise ratio (SNR) attains a value of 0 dB, shown in Fig. 21, and was experimentally obtained by imaging a tissue mimicking cyst phantom with an attenuation of 0.5 dB/[MHz cm] using a SA sequence with 96+96 emissions at 6.5 MHz. This excitation frequency was chosen to use the full bandwidth of the probe to gain the best possible resolution. The transducer reaches a depth of 150 λ, where the wavelength λ in this phantom is 0.2375 mm. This corresponds to a penetration depth of 3.6 cm again due to the many missing elements.

VI. DISCUSSION
The amplitude of the impulse response of the row elements during pulse-echo measurements was found to be a factor of 2.4 times higher than the columns. This reduction in receive sensitivity for columns is likely due to the aforementioned parasitic capacitance in Section I-C. Since a glass wafer is used as the substrate instead of an SOI wafer during fabrication this effect was hypothesised to be lower as the coupling should be significantly reduced. However, the 100 µm thick and ≈ 2.1 cm long bottom electrodes separated by a 2.5 µm wide trench can potentially contribute to an increased parasitic capacitance, which ultimately will lower the sensitivity. During the transmit measurements, a reduction in the transmit sensitivity for the rows by a factor of 1.92 compared to the columns was observed. It is also observed that in transmit, the columns have a higher center frequency than the rows even though all CMUT cells have the dimensions, and the plate thickness is uniform across the array. These effects are currently not understood and is the subject for future work.
A previously fabricated 92+92 RCA CMUT probe [75], which is based on an SOI wafer substrate with a 20 µm device layer, exhibited a reduction of the bottom electrode receive sensitivity by a factor of around 3.3. The presented 190+190 fusion-anodic bonded probe shows a clear improvement in receive sensitivity over the previous SOI wafer based fusion bonded design. The presented probe has, however, shown problems with element yield and demonstrated only around 55 % working elements. This is believed to be due the formation of a sulphur compound in the processing chamber when etching an opening to the bottom electrodes. Ultimately, this prevented some wire bonds from making proper contact and lowered the overall yield and performance of the probe. Solving this problem requires further investigation and tuning of the cleaning processes, the chamber conditioning, and the gas composition used when performing the silicon oxide etching process on the samples.
Ultrasound imaging showed that the probe is capable of performing 3D volumetric imaging, but the low SNR limits the penetration depth to 150 λ corresponding to 3.6 cm. The  imaging performance of the probe is naturally also limited by the low number of working elements, which affects the main-to-side lobe levels. It is, however, demonstrated that full volumetric imaging is possible, and the point spread function is isotropic in all three directions with a volume rate comparable to normal 2D imaging with a linear array probe.
The probe has sub-λ/2 pitch and normally there would be no advantage in linear array imaging, but for second harmonic imaging grating lobes could be avoided at the double center frequency. Such measurements have, however, not yet been conducted.

VII. CONCLUSION
The developed hand-held 190+190 RCA CMUT ultrasound probe with integrated edge apodization and active cooling was fabricated using a standard fusion bonding process utilizing LOCOS in combination with an anodic bonding process and an etch separating the bottom electrodes. The array was wire bonded and mounted in a 3D milled PPSU probe handle. The element yield of the elements with an impulse response higher than 1.1 × 10 15 Pa V −1 s −2 and 1.4 × 10 −4 V V −1 in transmit and pulse-echo, respectively, was a total of 61.1 % and 54.7 %. This was in part due to the formation of a sulphur compound in the bottom electrode contact holes during fabrication. To improve the yield the microfabrication process can be further optimised. The formation of the sulphur compound can be avoided by further optimising the process conditions including temperature control, chamber pre-conditioning and gas flow. Characterization of the probe in transmit showed that the maximum value of the averaged impulse responses of the columns was a factor of 1.8 times higher than the rows in transmit. In pulse-echo the maximum value of the averaged impulse responses of the rows was a factor of 2.4 times higher than the columns due to the substrate coupling effect. Compared to a previously fabricated probe, where the averaged maximum impulse response of the rows were a factor of 3.3 times higher than the columns, the substrate coupling effect has been reduced. The weighted center frequencies were 6.0 ± 0.1 MHz in transmit and 5.3 ± 0.5 MHz in pulse-echo with relative fractional bandwidths of 62.8 ± 4.5 % and 86.2 ± 10.4 %, which is close to the designed center frequency of 4.5 MHz. The attenuation of the transmit pressure along the top electrodes was found to be insignificant. For the bottom electrodes the transmit pressure was attenuated by 12 %, which is a clear improvement when compared to previous results where an attenuation of 74 % was measured. This demonstrates that using a highly doped (< 0.025 cm) 100 µm thick substrate in the fabrication process can solve the previously mentioned problems concerning high electrode resistances and the formation of delay lines.
The imaging performance of the probe was determined using a 3D printed point scatter phantom, a wire phantom, and a cyst phantom showing for the first two a reasonable contrast of −11.90 dB to −16.90 dB, considering the many missing elements, and the resolution was 0.82 λ to 1.72 λ (axial) and 1.56 λ to 1.72 λ (lateral), demonstrating the 3D volumetric capabilities with a near isotropic point spread function. The SNR measured on the tissue mimicking cyst phantom was low, resulting in a penetration depth of around 3.6 cm due to many missing elements. In conclusion, the fabrication process using thick highly doped silicon bottom electrodes and an insulating glass substrate has improved the pressure uniformity when emitting with the bottom electrodes and reduced the substrate coupling effect.