Ultrasonic imaging applications are being driven to higher frequencies. Extremely high frequency ultrasonic imaging systems, important for enhanced diagnostic applications in ophthalmology and dermatology, await the development of arrays working above 20 MHz. These high frequency arrays need small spatial scales (<10 µm) which cannot be achieved using standard fabrication techniques. Innovative materials and novel fabrication methods are therefore required.
Prototype arrays developed in this frequency range comprise of a pair of 20 MHz PZT arrays , a 100 MHz array incorporating a sapphire lens and thin film ZnO , and a piezoelectric polymer array with built-in transmit and receive circuitry . Additionally, O’Donnel et al described the operation of a 20 MHz phased array imaging system for catheter use .
Linear arrays, which typically do not use beam-steering, are capable of tolerating a much bigger element pitch than phased arrays. It is therefore practical to focus first on the development of linear arrays with 1λ to 2λ pitch. Although integrating the array with the electronics offers advantages, a more flexible and conventional approach of coupling the array elements to a 50 ohm imaging system has been adopted. Broad bandwidth (minimum of 40%) is preferred, both to suppress grating lobes and to advance the axial resolution. Crosstalk levels of near –30 dB are acceptable for a linear array not incorporating Doppler. Mild elevational focusing is thus desired for enhanced resolution in the elevation direction.
PZT “strip” vibrators (length >> width or height) with free boundary conditions need a width to height ratio of less than approximately 0.6 for proper pulse performance. For 30 MHz operation each PZT strip must be able to measure approximately 30 µm wide by 50 µm in height. For a 2-2 composite comprised of interleaved polymer and ceramic strips, optimized operation requires that the spatial scale of all constituents be much less than a wavelength. For instance, the epoxy between each ceramic must be less than 10 µm to push spurious lateral resonances above the passband of a 30 MHz array.
Standard dice and fill technology cannot presently be employed to manufacture this structure. An alternative fabrication technique is to stack polymer and ceramic layers to form a block of composite material, then slice sections from this block. Variants of this technique have been proposed earlier for arrays operating above 10 MHz. The difficulty lies in controlling the ceramic and polymer dimensions. A solution to this problem is described next.
Fine grain PZT-5H equivalent material (TRS 600, TRS Ceramics, State College, PA) was lapped to a thickness of 33 µm using a precision lapping process. Numerous plates of this material were stacked and then bonded together using Epo-Tek 301-2 epoxy (Epoxy Technology, Billerica, MA). The plate to plate spacing was controlled by combining polystyrene spheres into the bonding epoxy. These spheres (#PS06N, Bangs Laboratories, Fishers, IN) possessed a nominal diameter of 6.20 µm, with a standard deviation of only 0.09 µm.
The spheres were integrated into the epoxy at a volume fraction of 5%. Plates of 5 mm x 5 mm ceramic and a consistent amount of loaded epoxy were stacked in an alternating fashion. The stack was constrained from lateral motion and light, uniform pressure was then applied during the room temperature overnight cure. Thin sections of this stack were diced from the block, followed by being lapped to a thickness of 62 µm, electroded with 4000 A° of Au over a thin Cr layer, and poled at 2000 V/mm and 50 °C.
Analysis of the performance was encouraging and the results have been listed in Table I. The small standard deviation in kerf indicates that this technique can provide a controlled polymer width to within 0.3 µm. The observed value for the lateral mode frequency corresponds closely to the d31 vibration of the ceramic plate, as expected for a 2-2 composite with a volume fraction exceeding 75% . A theoretical investigation of the composite performance demonstrated exceptional agreement with these experimental results. The theoretical predictions were based on a dynamic model of guided wave propagation in the composite plate . Table II below lists the theoretical results. Using this model a dispersion curve predicting resonant frequencies over a range of thicknesses was obtained.
Table I. Measured 2-2 composite properties.
|Standard deviation of kerf
|Standard deviation of ceramic
|Dielectric constant, e33 s/ eo
|Thickness velocity at Fp
|1st lateral mode frequency
|Coupling coefficient kt
Fp is the parallel resonance frequency.
Table II. Predicted 2-2 composite properties.
|Thickness velocity at Fp
|1st lateral mode frequency
|Coupling coefficient kt
Fp is the parallel resonance frequency.
This composite fabrication technique may be adapted to single element transducers despite being developed for high frequency arrays. A wide diversity of polymers, ceramics and particles can be used to achieve desired properties. Additionally, it is possible to produce nonuniform structures for suppression of lateral resonances resulting from spatial periodicity.
Although the piezoelectric composite is the heart of an array transducer, matching, backing and lens materials are needed in order to increase device bandwidth and sensitivity and provide focusing. A brief list of the tested materials and their properties is given in reference . Materials were prepared as part of an effort to approximately match the impedances determined from one-dimensional modeling.
A conductive epoxy backing material (E-Solder 3022, Von Roll Isola, Inc, New Haven, CT) was chosen with sufficient attenuation in order to eliminate backing echoes over a round trip path length of 4 mm (twice the backing height). The longitudinal acoustic impedance of this material was 5.5 Mrayls at 30 MHz. This same material was centrifuged to provide a higher acoustic impedance (5.9 Mrayls) and then used as the first matching layer in a dual matching layer design.
Finding a suitable lens material proved challenging. Requirements included a longitudinal impedance close to tissue and a longitudinal velocity significantly different from tissue. TPX (Mitsui Plastics, Inc., White Plains, NY) met these requirements with a longitudinal velocity of 2170 m/s and an impedance of 1.8 Mrayls. Since this velocity is higher than the ultrasound velocity in water, a concave lens design was needed.
This is in contrast to a silicone lens material which typically needs a convex design due to an intrinsically low longitudinal sound velocity. The convex geometry may be preferred clinically since it is effortlessly coupled to the human body, but the high attenuation in silicone materials results in extreme low device sensitivity. The concave TPX design was preferred acoustically due to the inherently low attenuation. In addition, any attenuation in the lens may serve to favorably apodize the beam .
Besides acoustic impedance matching, electrical impedance matching was also addressed. The electrical impedance of the elements in the 30 MHz array was more than 100 ohms. A transformer was desired in order to match these devices to a 50 ohm system. For this array exceptional results were achieved using the coaxial cable connecting each element to the system.
Using transmission line theory, an optimum length of low capacitance micro-coax from Precision Interconnect (Portland, OR) was selected in order to transform the impedance and tune out reactive components. In addition, the coax helped in increasing bandwidth and “fine tune” the center frequency. This technique is considered to be best suited for devices in the frequency range of 20 to 60 MHz where the length of the coax will differ from approximately 3 m to 1 m, respectively.
The impedance transformation achieved with coaxial cable was examined using well known transmission line equations. The first step in quantifying this transformation dealt with characterizing the impedance (Zo) and propagation constant (g) of the coaxial cable. This was accomplished by using an HP4194 impedance analyzer and the expression for the impedance transformation from a length x of coax :
where Zload symbolizes the load at the end of the coax and Zx is the transformed impedance measured by the impedance analyzer. The open circuit values of Zload and values of Zx for short circuit were measured at 30 MHz and equation (1) was used to solve for g and Zo. The distributed network representation of a transmission line then resulted in the following expressions:
The quantities r, l, g and c are the per unit length values of series resistance, series inductance, shunt capacitance and shunt conductance, while v is the frequency in radians per second. The propagation constant g was then examined in terms of the real part (a or attenuation) and imaginary part (b or phase constant). The phase constant was used for obtaining the velocity of propagation from the relationship:
The coax selected for this array was completely characterized using the technique described above and the results have been listed in Table III below. The characteristic impedance of the coax lies between the system (50 ohms) and the impedance values of an array element (150 to 200 ohms). It was thus possible to broaden the bandwidth of the element using this coax. The impedance of the coax should be ideally optimized for the array impedance.
Table III. Coax properties measured at 30 MHz.
|Characteristic impedance Zo
||85 – 5.2i ohms
|Propagation constant γ
||0.051 + 0.78i
|r (resistance per unit length)
|c (capacitance per unit length)
|g (conductance per unit length)
|l (inductance per unit length)
Once the coax was characterized the impact of varying the coax length was explored using equation (1). In this case, the array element impedance was used as Zload. The transformed impedance across the expected passband of the device was examined for different lengths of coax above and below a quarter wavelength (2.0 m) at the center frequency. A range of lengths were found where the phase angle within the passband reached zero degrees. The bandwidth was optimized by understanding the band edges as the points where the phase angle fell to +/- 45°, a reasonable assumption when the phase angle passes through zero degrees. The results of this tuning are displayed in Figure 1 below.
Figure 1. Impedance and phase with and without coax for lensed element #1. The solid line is impedance and the dotted line is phase.
Transducer Design and Fabrication
Three different array designs were fabricated and then tested. The most recent design integrated three ceramic “subelements” per element, a dual matching layer, an “air” kerf separating each element and a TPX lens. The fabrication of this device has been described below. Both one dimensional equivalent circuits and multi dimensional finite element models were employed for analyzing the design. The Redwood equivalent circuit in PSPICE was used to select backing and matching materials and provide general design guidelines. For a more accurate representation of the performance the time domain finite element code PZFLEX (Weidleinger Associates, Inc, Los Altos, CA) was used.
A 2-2 composite was fabricated as described in section 2 and then lapped to a thickness of 0.100 mm. The ceramic posts were 0.025 mm wide and the kerf was 0.008 mm wide. Next, the conductive epoxy backing, acoustic impedance of 5.5 Mrayls, was cast in place on the back of the composite. This backing was lapped flat and a frame was bonded to the edges of the whole assembly. This frame was fabricated from a machinable ceramic with a low dielectric constant. It was used for providing a surface for the conductive traces connecting each element on the face of the array. An additional 0.047 mm was lapped from the face, for a final composite thickness of 0.053 mm, after this frame was in place.
Interconnect reliability was a significant problem with the first prototypes. Sputtered Au/Cr electrcodes were used for wrapping the traces around the edges of the machinable ceramic. A dicing operation was then used in order to separate the wraparound traces. This method resulted in a loss of interconnect through small breaks in the metallization. An electroless deposition process proved to be more reliable, but adhesion was still problematic. Most recently, posts of brass have been included to the machinable ceramic for conductivity in only the z-direction. Preliminary results are encouraging.
A conductive epoxy (5.9 Mrayl) was employed as the first matching layer to decrease the possibility of losing electrical connection to one of the elements. A concern with this epoxy was the attenuation of 112 dB/mm at 30 MHz. This high value was perhaps because of the large silver flakes combined into the epoxy matrix. A 3 dB reduction in the sensitivity was expected since the desired matching layer thickness as determined from modeling was .014 mm. This attenuation, although undesirable, could be tolerated, and may even help in decreasing near field clutter in the image .
After casting the first matching layer in place and lapping to thickness the elements were seperated using a dicing saw and a 0.012 mm blade. The thickness and depth of the kerf was critical. The depth of the kerf affected crosstalk and ringdown . A larger kerf meant a reduced impedance, smaller element size and an increased grating lobe amplitude in the focused beam. If a kerf filler was used, such a wide kerf could also translate into lateral modes within the filler falling within the passband of the device.
As expected from the work of Dias , the depth of separation into the backing was discovered to be crucial. Finite element analysis was used for investigating this effect. Six different separation depths were modeled and the impulse response was inspected for ringdown and amplitude. A long “tail” on the waveform showed a resonance within the backing due to the periodicity of the dicing. A higher than expected amplitude was a result of crosstalk. A separation depth of at least 18 µm into the backing was required for reducing the amplitude and ringdown to near the levels for a completely separated backing. Further modeling was carried out using differential dicing, where kerf depth alternated between two values. Dicing 18 µm and 36 µm into the backing was found to offer the best result, a ringdown of slightly more than two cycles and a 64% bandwidth. This model did not yet integrate the TPX lens.
The element to element spacing was 0.099 mm, or almost two wavelengths at the design center frequency. This would place grating lobes in the array at 30°. The directivity pattern of the array was calculated in order to determine the impact of the grating lobes. The directivity of each element was established using the non-rigid baffle condition . Discrete frequencies throughout the passband were analyzed as well as focusing throughout the depth of field. The worst case scenarios showed the grating lobes at -10 dB relative to the main lobe for the one-way response.
This would be unacceptable in most imaging systems. However, broad bandwidth may help to ameliorate this problem. If a pulsed response is considered, the 1-way amplitude is reduced by the ratio of the number of cycles in the pulse to the number of active channels . For the array, this reduction would be at least -12 dB, taking into account a two cycle pulse and a simple 8 channel system. Additional channels would result in a larger reduction. Grating lobes should thus be at a manageable level assuming a two cycle impulse response is attained.
A second matching layer (3.1 Mrayls) was bonded to the face after separating the elements. A special bonding technique was produced so that adhesive would not flow into and fill the kerfs. This air kerf was used to reduce the element-to-element crosstalk within the array. The matching layer was prepared by lapping to 0.018 mm thick and carefully waxing to a flat base. Shims, several microns thicker than the matching layers, were then placed next to the layer. A high viscosity epoxy (Insulcast 501, American Safety, Roseland, NJ) was used as the adhesive. A uniform, thin film of the epoxy was scraped over the matching layer using the shims as a guide for a flat edged tool. The epoxy was slightly cured for 20 minutes in order to raise the viscosity and the array was pressed into place. Light pressure was applied and the assembly was cured. Analysis of the bond line (using scanning acoustic microscopy) and the kerf (using electron microscopy) revealed uniform adhesion across the face and minimum infiltration of the epoxy into the kerf.
Addition of the TPX lens to the array proved to be extremely challenging. TPX, or Polymethylpentene, is a rigid plastic with low values of acoustic impedance (1.8 Mrayls at 30 MHz) and attenuation (6 dB/mm at 30 MHz) . Chemically it is similar to Polypropylene and requires surface preparation prior to bonding. Two methods of preparation are Corona discharge and a Toluene based chemical adhesion promoter . Experiments demonstrated that adhesion promoter 459T (Lord Chemical, Erie, PA), when applied to a clean TPX surface, resulted in dramatically improved adhesion. Epo-Tek 301 (Epoxy Technology, Billerica, MA) was then used to bond the TPX to the face of the second matching layer. The bond line between the TPX and the second matching layer was less than 1 µm.
The lens was machined subsequent to bonding in order to provide proper alignment of the lens with the elevation aperture. A ball end mill was used along with an alignment fixture to mill the cylindrical curvature in the face. The depth of cut was controlled to within 0.005 mm. After machining, a polishing operation was used for providing a mirror finish to the lens.
The connection of the coax to each element was attained using a precision soldering station, low melting point Indium based solder, and a 50 x microscope. In the future, a flex circuit may be used for connecting all the individual coaxes, but for the four element prototypes this simple method proved to be adequate.
Three different prototype transducers, each comprising of four active elements, were fabricated and then tested. The first device incorporated a single matching layer and 0.066 mm element-to-element spacing. The second device comprised of a single matching layer and 0.099 mm element-to-element spacing. Both of these prototypes were tested using a flat plate target at a distance of 1 mm from the face of the device. The third prototype incorporated the dual matching layers and TPX lens described in the section above. The elements in this last device were tested with the help of a flat plate target placed at the focal point.
The simplest design incorporated a single 3.3 Mrayl impedance matching layer and an element pitch of 0.066 mm. This resulted in two ceramic subelements per element. A shallow isolation cut, 5 µm deep and 17 µm wide, was used for separating the elements. Most of the element-to-element separation was thus provided by the composite filler. This epoxy filler was extremely hard and possessed minimal attenuation (9 dB/mm). Crosstalk was expected to be an issue for this design, but if the test results proved promising the fabrication process could be significantly simplified. A photolithographic process could be used on a plate of composite material instead of dicing in order to provide element isolation.
PZFLEX was considered to be a powerful tool not only for predicting array performance but also for troubleshooting problems and determining design constraints. Issues such as the required velocity and attenuation in the kerf filler, the required depth of the isolation cut (between the major elements of the array), and the impact of bond lines were determined using this software. The finite element model for this first prototype predicted a maximum element-to-element crosstalk of -21 dB, while the experimental results showed -25 dB of crosstalk. Figure 2 shows a comparison of the modeled and actual pulse echo impulse responses. Of the four responses, the one with the lowest bandwidth is displayed. The amplitudes of the responses were all within 15% of the average and the pulse shapes were similar. These first results displayed the utility of finite element modeling and demonstrated that 30 MHz arrays could be both fabricated and tested.
Figure 2. Normalized Impulse Response, First Test Array.
A second test array integrating a single matching layer was also constructed. The element pitch was 2l, with three ceramic subelements per element. This resulted in a bigger element and a lower electrical impedance when compared to the 1.3l spaced array. This lower impedance was more efficiently transformed to 50 ohms using the coax and broader bandwidth resulted. Experimentally, the pulse length was less than 2 cycles and the bandwidth was 65%. Figure 3 displays the modeled and experimental impulse responses for this array. Although there is some disparity between the modeled and actual results, the agreement is still good. Slight deviations in actual materials values, such as a shift of 1 µm in the matching layer thickness, may account for the deviation.
Figure 3. Normalized Impulse Response, Second Test Array.
The third and last test array incorporated a TPX lens and two matching layers. The fabrication of the array was described in section 5. For this array, a finite element model of the final design was not attempted. In order to account for the lens in the elevation direction, a three-dimensional model, as opposed to the two-dimensional model employed for the other arrays, was needed. Modeling of the array without the lens was carried out to guarantee that the two matching layers provided a compact impulse response and that the array would focus properly. The finite element model predicted a bandwidth of 64% without the lens. The TPX lens used a radius of curvature of 2.38 mm for a predicted focal point of 7.7 mm. The focal point of the array was experimentally determined to be 7 mm. The impulse response for element 1, tested at this focal depth, is displayed in Figure 4.
A loss of connection to the elements on one side is a problem with this third array. As a result, crosstalk data was unavailable. A design change integrating a more reliable interconnect method is being implemented in order to correct this problem.
Arrays for applications above 20 MHz and methods for fabricating 2-2 composites have been developed. The composites possessed high coupling (kt > 0.65) and lateral mode frequencies near 60 MHz. The arrays incorporated matching, backing and elevational focusing. An air kerf separated the elements and coaxial cable was employed for electrical impedance matching.
Finite element and one-dimensional modeling was used to examine and then refine the design. Experimental results demonstrate that 70% bandwidth can be achieved. Agreement between experiment and model was exceptional.
Figure 4. Normalized Impulse Response, Third Test Array Experimental.
Several issues have been identified which need to be addressed before the construction of a fully sampled array. A number of improved interconnect methods are being tested. A first matching layer material with a slightly greater impedance and lower attenuation is being explored. Acceptable crosstalk levels will be confirmed experimentally and the directivity of individual elements will be measured in order to assure proper array performance. Shielding will be added to the array between the outer matching layer and lens. After completing these investigations, a 48 element array incorporating a flex circuit for interconnect will be fabricated.
 M. Lethiecq, et al.., “Miniature High Frequency Array Transducers Based on New Fine Grain Ceramics,” 1994 Ultrasonics Symposium, 1994, pp.1009-1013.
 A. Nguyen-Dinh et al., “High Frequency Piezo-Composite Transducer Array Designed for Ultrasound Scanning Applications,” 1996 IEEE Ultrasonics Symposium, 1996, pp. 943-947.
 Y. Ito et al., “A 100 MHz Ultrasonic Transducer Array Using ZnO Thin Films,” IEEE Transactions on UFFC, vol. 42, no. 2, pp. 316-324, March 1995.
 P. A. Payne. et al., “Integrated Ultrasound Tranducers,” 1994 Ultrasonics Symposium, 1994, pp.1523-1526.
 M. O’Donnel and L.J. Thomas, “Efficient Synthetic Aperture Imaging from a Circular Aperture with Possible Applications to Catheter Based Imaging,” IEEE Transactions on UFFC, vol. 39, no. 3, pp. 366-380, May 1992.
 Endosonics Corporation
 J. Stevenson et al., “Fabrication and Characterization of PZT/Thermoplastic Polymer Composites for High Frequency Phased Arrays,” J. Am. Ceramic Soc., 77, pp. 2481-2484, 1994.
 X. Geng, “Numerical Modeling and Experimental Study of Piezocomposite Transducers”, PhD thesis in Materials, Penn State University, December 1997.
 W. J. Hughes. Personal Communication
 A. Macovski, Medical Imaging Systems, N.J., Prentice Hall Inc., 1983.
 T. Ritter, K.K Shung, X. Geng, H. Wang, and T.R. Shrout, “30 MHz Medical Imaging Arrays Incorporating 2-2 Composites,” presented at The 1998 IEEE Ultrasonics Symposium, Sendei, Japan, October, 1998.
 A. P. Albrecht, “Transmission Lines,” Electronic Designer’s Handbook, L.J. Giacoletto ed., N. Y., McGraw-Hill, pp. 8.1-8.78, 1977.
 A. R Selfridge, “The Design and Fabrication of Ultrasonic Transducers and Transducer Arrays,” PhD thesis in Electrical Engineering, Stanford University, July, 1982.
 J.Fleming Dias, “An Experimental Investigation of the Cross Coupling Between Elements of an Acoustic Imaging Array Transducer,” Ultrasonic Imaging, vol. 4, pp. 44-55, 1982.
 Manny Mafilios, Matsui Plastics, Inc., personal communication
 R. McKeighan, “Design Guidelines for Medical Ultrasonic Arrays,” Medical Imaging 1999: Ultrasonic Transducer Engineering, vol. 3341, K. Kirk Shung editor, pp. 2-18, SPIE, Bellingham WA, 1998.
This information has been sourced, reviewed and adapted from materials provided by Epoxy Technology, Inc.
For more information on this source, please visit Epoxy Technology, Inc.