THE invention of the printing press in the 15th century is

Transcription

1 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST Acoustic Picoliter Droplets for Emerging Applications in Semiconductor Industry and Biotechnology Utkan Demirci Abstract This paper presents the theory of operation, fabrication, and experimental results obtained with a new acoustically actuated two-dimensional (2-D) micromachined microdroplet ejector array. Direct droplet based deposition of chemicals used in IC manufacturing such as photoresist and other spin-on materials, low- and high- dielectrics by ejector arrays is demonstrated to reduce waste contributing to environmentally benign fabrication and lower production cost. These ejectors are chemically compatible with the materials used in IC manufacturing and do not harm fluids that are heat or pressure sensitive. A focused acoustic beam overcomes the surface tension and releases droplets in air in every actuation cycle. The ejectors were operated most efficiently at 34.7 MHz and generated 28 m diameter droplets in drop-on-demand and continuous modes of operation as predicted by the finite element analysis (FEA). Photoresist, water, isopropanol, ethyl alcohol, and acetone were ejected from a D micromachined ejector array. Single photoresist droplets were printed onto a silicon wafer by drop-on-demand and continuous modes of operation. Parallel photoresist lines were drawn and a 4-in wafer was coated by Shipley 3612 photoresist by using acoustically actuated 2-D micromachined microdroplet ejector arrays. [1513] Index Terms Acoustic radiation pressure, deposition, droplet ejection, finite element analysis (FEA), inkjet, microfluidic channels. I. INTRODUCTION THE invention of the printing press in the 15th century is a major force in the Renaissance and Reform movements [1]. It burst the chains that fettered knowledge and revolutionized society advancing civilization toward the freedom of mind from which humanity benefits today [2]. The capability to controllably eject ink droplets onto a white sheet of paper is a descendant technology of this very old and beneficial concept of the printing press. Droplet generation techniques have become even more valuable and significant as numerous new applications are posed by today s technology. This concept will be elaborated in the following paragraphs focusing on the semiconductor and biotechnology applications. Biotechnology applies enabling technologies from other disciplines to biology facilitating significant discoveries. The direct impact of these achievements on the human-beings makes this field a leading research area. Almost all of the experiments in Manuscript received January 26, 2005; revised November 8, This work was supported by NSF/SRC Engineering Research Center for Environmentally Benign Semiconductor Manufacturing. Subject Editor C. H. Mastrangelo. The author was with the E. L. Ginzton Laboratory, Stanford University, Stanford, CA USA. He is currently with Harvard-MIT Health Sciences and Technology, Harvard Medical School, Boston, MA USA ( utkan@stanfordalumni.org). Digital Object Identifier /JMEMS biotechnology are conducted in fluidic environments. Thus, capability to generate droplets of various fluids has emerged as an attractive technology enabling various biological applications. For instance, bioarrays for drug testing could easily utilize controlled droplet generation techniques; two-dimensional (2-D) arrays of cells laid on a surface can be tested with small amounts of drugs delivered by microdroplet ejectors [3]. Writing DNA arrays by drop-on-demand ejection is another example [3]. The semiconductor industry is one of today s largest existing markets, for example wafer fabrication equipment market was $18 billion market, flat display coating was $22 billion market in 2003 [4]. The business model of semiconductor companies relies on Moore s Law, which can be summarized as scaling, which enables packing more and more devices per unit area, decreasing cost and increasing functionality and speed. Since success in scaling depends on success in lithography, droplet generation applications that could improve or revolutionize lithography are of major interest. The deposition of photoresist thin film is the first step in lithography, followed by masking and developing steps. In 2003, photoresist consumption worldwide was $735 million [4]. Most of this photoresist is deposited by tracker systems that utilize spin coating technology [5], [6]. However, this technology wastes up to 95% of the disposed resist. When the cost of hazardous waste disposal is included, the cost associated with photoresist consumption increases to approximately $1 billion. A droplet ejection methodology that prints photoresist onto wafer surface drop-by-drop has the potential to minimize photoresist waste [7]. This ejection technology relies on surface tension of the fluid to provide thin film quality and uniform thickness of the deposited photoresist without spinning [5] [7]. Reducing operational cost associated with lithography would be a significant improvement, and this work will specifically target this application. Although there are various droplet generation techniques currently available, the semiconductor and biotech applications of microdroplet generation require certain design parameters for successful delivery of fluids. One of the most important factors is avoiding damage to pressure and heat sensitive fluids. This requirement eliminates inkjet technology, where the temperature and pressure in the ejection fluid can rise. Human and many other living cells do not survive at high temperatures or pressures. Second, the directionality of ejected droplets is a primary concern. The ejection directionality of traditional inkjet devices is limited by nozzle geometry. This limit may pose a problem on ink printing applications, since the human eye is sensitive up to a resolution limit [8]. The current applications that are being targeted require knowing exactly where the droplet is going to land /$ IEEE

2 958 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST 2006 Fig. 2. Schematic of the physical operation of an ejector array. Leaky surface acoustic waves form an acoustic focus and droplets are ejected from an open pool of a microfluidic channel etched into a silicon spacer. Fig. 1. Geometry of a 2-D ejector array. Signal and ground pads are used to actuate the interdigitated transducers. on a surface. Another significant parameter is the reliability of ejection for drop-on-demand actuation. An unreliable ejection method would cause non-uniform film thickness and non-uniform film quality, if it were used for photoresist deposition onto a wafer [9]. There has been previous work on droplet ejection by utilizing an acoustic focus [10]. These approaches use an acoustic lens or a fresnel lens approach to form a focal point, from which droplets are released. Huang and Kim demonstrated a focused ejector that utilizes the direct coupling of acoustic waves to a fluid in order to form a focus [11]. This method could suffer from fabrication complexity such as alignment of multiple layers and nonuniformity of piezoelectric film deposition, which may cause variation from one array element to another, satellite droplets, and problems with ejection stability. Acoustically Focused 2-D Micromachined Microdroplet Ejector Arrays that are demonstrated in this paper do not utilize an acoustic lens approach. The formation of the focal point is by surface acoustic waves which leak into the fluid medium and interfere to form an acoustic focus. They are easily and repeatably fabricated. The substrate uniformity and fabrication ease ensures identical and stable operation of ejector array. This type of acoustic wave is previously employed by Farnell et al. in order to build a planer acoustic microscope [12]. We demonstrate controlled droplet generation by utilizing leaky surface acoustic waves. This paper demonstrates a new micromachined acoustically actuated 2-D microdroplet ejector array and initial coating of a 4-in silicon wafer by drop-on-demand photoresist ejection. Since ejection takes place from an open-pool, nozzleless reservoir, droplet directionality does not depend on the nozzle geometry, and is easily controllable. Furthermore, the device releases acoustic waves at every cycle of actuation, which assures reliability of ejection. These novel ejector arrays fulfill all design parameters for sensitive fluid ejection applications. II. THEORY A. Theory of Device Operation The basic building block of an acoustically actuated microdroplet 2-D ejector array is an interdigital ring transducer as shown in Fig. 1. The rings are periodically spaced on a piezoelectric substrate. This unit cell can be repeated in a 2-D geometry as shown in Fig. 1. The interdigital ring transducers launch surface acoustic waves as explained in [13] and [14]. If another medium is placed on the piezoelectric substrate, such as a fluidic environment as shown in Fig. 2, the launched surface acoustic waves will leak into this medium. The angle of leakage follows Snell s Law, based on the ratios of two media indices (or velocities in acoustics) [13]. As waves travel through fluid, they reach a point where they interfere constructively forming a focus. If this focal point is right at the surface of the fluid and there is enough force exerted by acoustic radiation at that point to overcome the surface tension forces of the fluid, then a droplet will be ejected as shown in Fig. 2. Chu et al. gives the Langevin radiation pressure by the mean energy density of an acoustic beam at the surface of air liquid boundary [15], [16]. The time averaged energy density can be computed [14] [16]. Hunter et al. gives the average intensity of an incident wave as in (1) and the Langevin radiation pressure as in (2). where is the pressure amplitude of an incident acoustic wave, is the bulk density of the liquid, and is the velocity. The radiation pressure acts for a time, which generates an initial momentum per unit area. We assume that the initial momentum is given to a cylinder of fluid and is uniformly distributed in that cylinder such that the full width at half maximum (FWHM) of the acoustic focus is at the midpoint of the focal fluid cylinder. This momentum results in formation of a fluid droplet. B. Explanation of the Finite Element Analysis (FEA) The FEA of the device is performed using ANSYS 5.7 (ANSYS Inc., PA). The initial objective of this analysis is to determine the location, width and height of the focal point, and the best frequency of operation for the device, where the acoustic pressure at that focal point is maximized per unit voltage input. The secondary objective of the FEA is to investigate effect of the number of interdigitated finger pairs on device performance, in order to finalize the device design and determine how many metal finger pairs are needed for successful droplet ejection or for a certain focal point geometry. The third objective is to determine the evolution of acoustic wave as it travels from the device surface to the focal point, facilitating design of better (1) (2)

3 DEMIRCI: ACOUSTIC PICOLITER DROPLETS FOR EMERGING APPLICATIONS 959 Fig. 3. Schematic for FEA of ejection. The figure is axisymmetric around the y axis. One of the four interdigitated finger pairs is marked on the figure. The pressure distribution from point A to B determines the theoretical focal point and droplet size generated by the device can be estimated by the lateral pressure distribution at the focal point. devices and microfluidic channel holding silicon spacers that do not interfere with device performance. A harmonic analysis of the 2-D axisymmetric structures shown in Fig. 3 was performed. The mesh had at least ten nodes per wavelength in the structure. The device was simulated by loading one side of a piezoelectric substrate with an infinitely long fluid space, eliminating undesired reflections (FLUID 129 for absorbing boundary conditions). The structure/fluid interaction was taken into account by solid/fluid interface elements (FLUID 29). The characteristic piezoelectric material values of the fabricated device and distilled water were used for simulations [14]. C. Real and Imaginary Impedances of the Device The resonance frequency is inherent to the device geometry and the substrate and depends on the periodicity of the rings. Once the resonance frequency is known by simulations under a certain fluid loading, location of the focal point by simulations under that fluid loading can be directly determined. We simulated the impedance of a single ring device with various fluid loadings. The simulations can be run with various fluids for the same device geometry. When the fluid loading is changed, the velocity of acoustic wave and the location of the focal point change. It is important to run simulations with various fluid loadings to be able to predict the device behavior. We searched for frequencies where the real part of impedance is maximized under fluid loading in order to determine the optimum frequency of operation, i.e., resonance. Another way to determine the resonance frequency is to sweep frequency in a range of 1 50 MHz and obtain plots for pressure distribution over the line from point A to point B (as shown in Fig. 3) for each frequency point. The highest pressure levels would be obtained on the line when device is operated at its resonance. Also, the same curve would show where the focal point is located. Moreover, comparison of simulated and experimental real parts of the impedance indicates presence of an 8 series resistance due to interconnect metal strips guiding the signal and Fig. 4. (a) Real part of the impedance, theory, and experiment in air. (b) Imaginary part of the impedance, theory, and experiment in air. Fig. 5. Pressure on the axisymmetry line from A to B, as shown in Fig. 3, in the fluid medium for 4- and 8-pair ejector devices at 34.7 MHz. ground pads to the ring ejector. The comparison of simulated and experimental imaginary parts of the impedance indicates the presence of a parasitic capacitance on the device. This parasitic capacitance should be minimized to achieve maximum power delivery efficiency to interdigitated rings of the ejector. The simulation and experimental results for real and imaginary

4 960 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST 2006 Fig. 6. Theoretical lateral pressure distributions from the center of to the device to 1.5 mm radius outside the device are shown at various heights away from the piezoelectric substrate showing the evolution of the focal point for a 4-finger pair device and leakage pathway of the acoustic waves into the fluid medium. parts of the impedance of a 4 finger pair device in air are shown in Fig. 4(a) and (b). There is agreement between the simulation results of the FEA and the experimental results. D. Focal Point and Operation Frequency of the Device FEA provides the pressure distribution at any point in the fluid. The focal point location is defined as the highest pressure point in the fluid space. As a sinusoidal signal of 1 V amplitude was applied on metal signal lines at varying frequencies, the pressure distribution in the fluid medium from the piezoelectric substrate surface into the fluid on a vertical path from A to B of Fig. 3 was monitored as a function of frequency. The simulations predicted that devices generated the highest acoustic pressure at the focal point at 34.7 MHz. The number of finger pairs per ejector element is another crucial design parameter. The pressure distribution in the fluid on a vertical path from A to B as shown in Fig. 3 was monitored for devices with 4- and 8-finger pairs (Fig. 5). The pressure is maximal at the focal point, which is 575 away from the surface of the piezoelectric substrate for a device with 4-finger pairs. The device can achieve efficient droplet ejection when the fluid surface height is located between 550 and 585, as demonstrated by Fig. 5. This point is located above the center of metal rings. The focal point of an 8-finger pair device can also be obtained from Fig. 5, and is 1500 above the piezoelectric substrate surface. The lateral pressure distribution at the focal point is significant, since the focal point geometry determines the droplet size. The lateral pressure distribution at the optimal operating frequency of 34.7 MHz at 575 away from the piezoelectric substrate surface for a 4-finger pair ejector device is shown in Fig. 6 for distilled water. At the focal point of 575 pressure reaches a maximum of 0.75 MPa as shown in Fig. 6. The simulation predicts the pressure amplitude to be half of its peak value of 0.75 MPa at 15 away from the focal point. This implies an effective droplet diameter of 30 for distilled water. Moreover, the FEA predicts a droplet diameter of 26 for isopropanol loading of the device. E. Theory of Leaky Waves The evolution of acoustic waves as they travel in the fluid to form a focal point is significant, since it is a key to design better devices and microfluidic channel spacers that do not interfere with device performance. The lateral pressure distribution from the center of the device to 1.5 mm distance at various heights of 10, 50, 250, and 575 above the substrate is shown in Fig. 6. This figure demonstrates the evolution of the focal point from the piezoelectric substrate to the focal point as acoustic waves leak into fluid. The outer ring of a 4-finger pair device has a radius of 755. The acoustic pressure increases both inside and outside the rings due to interference, but a focus is formed only inside, due to the circular geometry. Outside the ring, waves travel away from the device. F. Simulation of Coverage The coverage of a wafer has to satisfy important parameters of throughput and uniformity to compete with the current existing ejection technologies. The time required to cover a wafer surface with photoresist is calculated by a simple program assuming that the droplet size on wafer is four times the droplet diameter in air; uniform coverage is achieved by surface tension forces spreading the droplets uniformly when the edges of the ejected droplets barely touch; the wafer is moved beneath the ejector, while a single ejector is ejecting droplets. Various algorithms for a wafer moving under an ejector array or an ejector array moving over a wafer can be developed for achieving uniform coverage, since there is reliable control over the droplet size and directionality in an open pool ejection, such as in this device. The emphasis is on how fast an ejector array can generate necessary number of droplets to cover a wafer. In the calculations, one drop is ejected per spot; however, by ejecting multiple drops to each location, the same system can be used to obtain various film thicknesses. On the other hand, conventional spin coating depends on various photoresist formulations with varying viscosities to achieve various film thicknesses. As the 2-D ejector array size increases, time to generate the necessary number of droplets for coating decreases. The predicted results

5 DEMIRCI: ACOUSTIC PICOLITER DROPLETS FOR EMERGING APPLICATIONS 961 TABLE I PREDICTED COVERAGE RESULTS FOR AN N 2 NEJECTOR ARRAY COATING A 4-in WAFER for the time required, fluid consumed, and number of droplets required for a 4 wafer being covered by an ejector is shown in Table I considering three droplet sizes of 15, 24, and 30. This model can be improved by accounting for photoresist solvent drying. These numbers are directly proportional with area and will be approximately nine times larger for 12 wafers. III. FABRICATION A. Device Design The basic building block of an acoustically actuated picoliter 2 D ejector array is an interdigital ring transducer as shown in Fig. 1 and Fig. 7(a). The rings are periodically spaced on a piezoelectric substrate. This unit cell can be repeated in a 2 D geometry as shown in Fig. 7(a). As shown in Figs. 2 and 7(b) the microfluidic channels have the function of stabilizing the fluid surface for a stabilized fluid focal point. The channels prevent the fluid from leaking by surface tension forces such that we could eject in all angles. Moreover, the microfluidic channel openings vary from 50 to 300 and are wider than the focal point diameter. This still enables open pool ejection and ensures directionality and size uniformity of droplets independent of nozzle geometry. B. Device Fabrication The device is fabricated using integrated circuit (IC) manufacturing techniques. First, a piezoelectric substrate is covered with photoresist and resist is patterned so that circular active areas are inscribed on a substrate. In this step, the substrate is covered with photoresist everywhere except at locations where the gold circular lines will be deposited. Second, 3000 A of thin gold film is deposited on to this patterned substrate by evaporation. A standard lift-off process etches the photoresist in acetone and leaves behind the circular gold rings on a piezoelectric substrate surface. A single ring ejector and a 4 4 ejector array are shown in Fig. 7(a). The top and bottom spacers are fabricated by using potassium hydroxide (KOH) etching of silicon substrate through the crystal planes to achieve the desired opening on the other side. A thick and wide top spacer is bonded to a thick and 1.5 mm-wide bottom spacer to form a fluidic channel that stabilizes the fluid surface at a height of 550 from the interdigitated ring array surface as shown in Fig. 7(b). Fig. 7. (a) Fabricated ejector arrays, and a unit cell of the array is shown. The yellow lines are the deposited 17-m-wide interdigitated gold lines on the surface of the piezoelectric substrate and are separated by 17 m. (b) 1.5-mmwide microfluidic channels are etched to a 350-m-thick bottom silicon spacer, and m wide microfluidic channels are etched to a 200-m-thick top silicon spacer. The two silicon spacer pieces which have the microfluidic channels defined are directly bonded together. This total spacer height is designed to match the fluid surface height to the focal point location of the device, and enables droplet ejection from a focal point. The bonding of two microfluidic spacers is achieved by bonding the two by an initial silicon to silicon surface touch, which achieves an initial bond between the two pieces. This bond is strengthened by placing ultraviolet light curable epoxy around the two silicon microfluidic spacers. IV. EXPERIMENTAL METHODS AND RESULTS A. Capacitance Measurements The impedance of a single ejector device in air were presented above in Section II. The ejector device is then loaded with various solvents in order to understand its characteristics. The capacitance of the device changes with fluid loading, since the

6 962 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST 2006 TABLE II DIELECTRIC CONSTANTS OF VARIOUS FLUIDS AND CAPACITANCES MEASURED BY A ARRAY Fig. 9. Schematic of the setup for pitch-catch measurements. The vertical and lateral distance can be controlled down to 0.1 m by a micrometer stage. Fig. 8. Ratio of (area/distance) in meters of the same device under various fluid loadings. capacitance of finger pairs can be considered as a parallel combination of the piezoelectric plate and the loading medium capacitances. The dielectric constants of loading fluids and capacitance measured by a network analyzer (HP 8752A, CA) for a 1 3 array are shown in Table II. In order to understand the variation of capacitance with fluid medium, a simple capacitance calculation approach (area/distance) is utilized. The ratio of (area/distance) is observed to be constant over the same device under various fluid loadings as shown in Fig. 8. The mean and variance of (area/distance) are calculated to be m,, respectively. The rapid change in impedance and high sensitivity to dielectric constant and conductivity changes in fluid medium make these devices attractive for biosensor applications such as viral, bacterial, or cellular detection and quantification. Once the impedance and the optimum operating frequency of the device are determined, it is possible to design a matching network to maximize the power delivered to the ejector. The device impedance changes with various fluid loadings as can be seen from Table II, which requires the matching network to be tuned for a specific fluid ejection. In our experiments, it was observed that the ejector ejected various solvents and photoresist without a matching network. However, a simple power reflection calculation, which takes into consideration the impedance of the ejector as measured before and a 50- input to the ejector indicates that on average 70% of initial power delivered is being reflected from the ejector array. Therefore, a matching network should be implemented for efficient use of power. B. Pitch-Catch Measurements The focal point measurements were obtained by placing a single 4-finger ejector array element in water facing another, forming the angle of a square. The top ejector transmitting, the bottom receiving the acoustic signals through the fluid volume, in which due form the pitch catch measurements were performed. The normal through the center of the receiving bottom ejector should intersect with the center of the transmitting top ejector and ejectors have to be facing parallel to each other in order to receive maximal signal amplitude as shown in Fig. 9. The bottom ejector is placed on a controllable x, y, z, m stage and distance is incremented, as the acoustic signal received is recorded. Movement in the z direction gives the focal point location as demonstrated in Fig. 10(a). The bottom ejector is brought to the focal point, where the pressure is maximized and then the ejector is moved in x and y directions to obtain the lateral pressure distribution at the focal point as shown in Fig. 10(b). The simulated pressure amplitude is half of its peak value at 15 away from the focal point. This implies an effective droplet diameter of 30 for distilled water. The simulation of the focal point location overlaps with the experimental results. Moreover, the insertion loss of a single ejector element was measured to be 11 db. C. Imaging and Wafer Coating Stroboscopic imaging techniques were used to view ejected droplets. Light emitting diodes that shine light on ejected droplets were turned on and off by a periodic square pulse waveform, which was synchronized with the drive signal to ejectors. Using this stroboscopic imaging technique, a jet of droplets could be viewed by an LCD camera (Sony, SSC-CD33V, Japan) with a microscopic lens (5, NA 0.13, Olympus, Japan) on the monitor screen as shown in Fig. 11. The input signal to a ring ejector array is a 10- -long 34.7-MHz tone burst. The burst repetition rate determines the drop ejection rate. The ring array ejected various fluids such as photoresist, water, and solvents (acetone, isopropanol, methanol, and ethanol), at a drop ejection rate of 1 khz to 0.1 MHz, such as Fig. 12 shows 28 in diameter isopropanol droplets ejected upward into the air at 1 khz which are generated at different periods of the driving signal. The size of a pixel on screen is determined by imaging a known sized object by the

7 DEMIRCI: ACOUSTIC PICOLITER DROPLETS FOR EMERGING APPLICATIONS 963 Fig m in diameter isopropanol droplets are ejected upward from an open pool of fluid. The imaging system had a large viewing area and each droplet is generated at a different period of the driving signal. Fig. 10. (a) Theoretical and experimental results for pressure on the axisymmetry line from A to B in the fluid medium for various frequencies for distilled water. The maximum pressure is achieved around 34.7 MHz. (b) Theoretical and experimental results for lateral pressure distribution at the focal point of 575 m at 34.7 MHz for a four-finger pair device. Fig. 11. The silicon spacer consisting of the microfluidic channels are aligned on top of the ejector array and the stroboscopic imaging setup is demonstrated. imaging setup. The number of pixels that a droplet covers on screen reveals the droplet size. Moreover, current microscopes can automatically calculate size of an object on a screen by using imbedded calibration data for a known objective size. In order to achieve ejection not only upward but in all directions (sideways, downwards), and control focus location without being effected by evaporation or the device tilt, silicon microfluidic channel spacers were designed so that the Fig. 13. Photoresist solvent droplets are ejected downward from a 100-m-wide spacer opening. The first generated droplet has traveled in air when the second droplet generated by the second driving signal is following the same path. The ejector generated droplets on drop-on-demand without satellite droplets. fluid could continuously fill the microfluidic channel ejection openings through the fluidic channels, keeping the fluid level constant. Downward ejection of photoresist solvent droplets through a microfluidic channel opening of 100 is shown in Fig. 13. The ejection of a droplet through a neck of acoustically raised fluid focus followed by a second droplet is also seen in Fig. 13. The ejector generated droplets on drop-on-demand without satellite droplets. Various viscosity and surface tension fluids such as water (1 Centistoke at 20 C), isopropanol, ethyl alcohol, ethylene glycol (18 Centistokes at 20 C), and acetone are ejected from D micromachined ejector arrays [17]. Shipley 3612 photoresist was also ejected through microfluidic channel openings onto the surface of a silicon wafer at varying rates from 1 to 10 KHz. The setup for coating a wafer with photoresist was similar to the imaging setup that was demonstrated in Fig. 11. A wafer on top of a controllable micrometer stage was placed under an ejector, 1 cm away and parallel to the ejector surface. The wafer was moved fast enough under the ejector so that single photoresist droplets could land on the wafer surface as shown in Fig. 14(a). The profile through the center of a single droplet is shown in Fig. 15. This profile is observed to be the same for all droplets. By changing ejection frequency or decreasing wafer movement

8 964 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST 2006 Fig. 16. Selective photoresist coverage of a 4-in wafer with three thickness values by varying the overlap of printed individual photoresist lines. Three regions are shown with thicknesses of 3.6, 2.9, and 2.4 m from left to right. Fig. 14. (a) Ejected single photoresist droplets in air cover 165 m in diameter area on the silicon wafer surface. (b) Single photoresist droplets ejected every 50 m overlap and form a 200-m-wide line by surface tension forces. Fig. 15. Single photoresist froplet and a line profile on wafer surface. speed the location where droplets land can be exactly determined. Moreover, a line can be formed by overlapping single droplets and achieving a uniform thickness with the help of surface tension forces as shown in Fig. 14(b). Lateral profile of the printed photoresist line is shown in Fig. 15. The wafer was moved at a speed of at an ejection rate of 1 KHz, which corresponds to a single droplet ejected onto the wafer surface every 50 during the photoresist line printing. This also indicates that the wafer coverage can be performed rapidly. We demonstrated that the photoresist lines can be repeated many times side by side and a 4-in diameter silicon wafer can be coated with photoresist as shown in Fig. 16. The wafer is moved at a speed of as the photoresist ejection is performed at 1 KHz. This corresponds to one droplet per 20 separation for a single line. Photoresist lines were written in parallel to perform full coverage. The separation between two lines was set to 80, 100, and 120 at three regions on the wafer by the automatic controlled x-y stage. This resulted into photoresist thicknesses of 2.4, 2.9, and 3.6 on wafer. The separation between droplets and lines can be modified, which eventually determines the overall photoresist thickness. The photoresist thin film thickness uniformity was measured to vary peak-to-peak at the most uniform coated areas of the wafer and at the areas with the worst uniformity in all three regions. There was not any spinning involved in the process and coverage experiments were done in a dry laboratory environment. The surface roughness was measured to vary from 14 to 600 A over the wafer. These measurements are taken by a Dektak profilometer (Veeco Inc., Woodbury, NY). The thickness of the photoresist film can be decreased by decreasing the number of droplets per location, or by decreasing the overlap between two photoresist lines drawn side by side. We coated surfaces with photoresist drop-by-drop and results were repeatable given the same ejection rate and wafer movement speeds. V. DISCUSSIONS Finite element theory indicates that the number of rings affects the location and width of focal point. A device with a larger number of rings will be able to maintain wider vertical areas of high pressure as can be seen in Fig. 5. In a sense, the number of rings lowers the ejection sensitivity to the height of the fluid surface. An 8-finger pair device has a wider range of fluid heights that it can eject at. Since controlling fluid height is not a serious challenge and can be easily handled, increasing the number of finger pairs does not offer a significant advantage. Moreover, pressure amplitude at the focal point of an 8-finger pair ejector is comparable to the 4-finger pair ejector. Furthermore, 8-finger pair ejector does not provide a different lateral pressure distribution at focus from 4-finger pair ejector. The droplet sizes generated by 4- and 8-finger pair devices are also comparable, since droplet size is determined by the acoustic wavelength and the interference geometry at the focal point. As the number of finger pairs in the design does not play a significant role for current applications, 4-finger pairs are chosen as the design, which according to the FEA results, can generate a focal point and eject droplets. The droplet diameter depends on the wavelength in fluid and focal point geometry as described above. The theoretical focal point diameter was found to be 30 and 26 for distilled water and isopropanol, respectively. The speed of sound in water is 1480 m/s, and 1100 m/s in isopropanol. Therefore, the acoustic wavelength in isopropanol is smaller resulting into smaller droplets. Moreover, the theoretical 26 focal point diameter agrees with the experimental value of 28 in diameter for ejected isopropanol droplets. The capability to eject in all directions, control of the focus location without being affected by evaporation, and nozzle independent open-pool ejection were desired objectives. In order to achieve these goals, the microfluidic channel spacers

9 DEMIRCI: ACOUSTIC PICOLITER DROPLETS FOR EMERGING APPLICATIONS 965 were designed to continuously fill the microfluidic channel openings through the microfluidic channels by capillary action and maintain a constant fluid level. The height of the fluid level and the focal point which coincide with the microfluidic channel opening was stabilized and rendered independent of the device tilt by the surface tension forces. Moreover, this open-pool ejection ensures nozzle independence. The nozzle dependent ejection methods such as inkjet may suffer from clogging. The droplet size and directionality in inkjet depends on the nozzle, where as in open pool ejection, the ejection pool can be as wide as 300 m and clogging is not an issue. We have not observed any clogging for our ejectors. Any residue in the microfluidic channels can be removed by some photoresist solvent or by acetone flow through the channels, or if a cleaning step is intended before loading a new type of a fluid. The micromachined ejector arrays were able to eject photoresist. However, since the experiments were carried out in a dry laboratory environment at room temperature ( 20 C), evaporation of the photoresist solvent and a rapid increase in viscosity of the photoresist took place. This evaporation degraded thickness uniformity of the deposition. It should be possible to achieve better thickness uniformity in a solvent saturated environment. Moreover, presence of diameter dust particles in the experimental environment also degraded the thickness and film uniformity. This would not be a problem when experiments are performed in a clean room. Furthermore, post deposition spinning was not performed in order to avoid any waste of photoresist. The presented FEA results were performed with a signal input at 34.7 MHz. The simulation for the focal point location overlaps with the experimental result. However, a wider band of ejection is observed with the experimental results than simulation results. This could be explained by the fact that the parallelism between the transmitting and the receiving ejectors is hard to control. An angle variation from horizontal could cause variation on the received signal. Moreover, the closer the transmitter and receiver get to each other the harder it gets to separate the acoustic signal from the electromagnetic signal feedthrough. This causes an interference of electromagnetic waves and acoustic waves and the signal amplitude varies, which results in a change in measured bandwidth. Outside the metal rings, waves travel away from the device as shown in Fig. 6. These waves also leak into the fluid at a critical angle, however without focusing and with attenuation. In the presence of an attenuating environment these outside traveling waves are quickly attenuated. They do not reach the next array element or cause any undesired crosstalk effect that could impede the focal point formation and deter simultaneous droplet generation by all array elements. Moreover, the capability to determine the pathway for leaky acoustic waves facilitates the design of the microfluidic channel spacers. The acoustic waves traveling into the fluid medium should not be intercepted by the presence of these microfluidic spacers. The reliable control over the droplet size and directionality in an open pool ejection, such as in this device enables various deposition algorithms for a wafer moving under an ejector array or an ejector array moving over a wafer. These algorithms can be developed for achieving uniform coverage, but these are beyond the scope of this paper. The main emphasis is on how fast the ejector array can generate necessary number of droplets to cover a wafer. As the number of ejectors of a 2-D ejector array increases, the time it takes to generate necessary number of droplets decreases. The capability to eject from many ejector arrays simultaneously could enable linear coverage algorithms as utilized for printing of single photoresist droplets demonstrated in Section IV. In theory, hundreds of these ejectors could be in a linear array or 2-D array format, printing a single line wide enough to cover a whole wafer by a single pass over the wafer. The wafer movement speed and the droplet ejection rate can be increased. There is potential to coat a wafer in less than 5 s given that there are not other problems due to wafer speed. This would remarkably increase the throughput of coating. The possible increase in the throughput is clear, when we consider that there is no spinning or edge bead removal processes necessary after the coating step, which take about a minute to complete in today s tracker systems by using spin coating method. The surface tension and viscosity of the ejection fluid are important parameters, since they affect the droplet size and thickness of the deposited resist. The droplet size is affected since the acoustic wavelength changes with fluid as explained earlier in the theory. A fast evaporating fluid would have changing properties, which makes a solvent saturated environment very important. The droplet size eventually influences thickness of the photoresist film. In theory, a device could place as many droplets as possible to the same location for thicker resist coating. We observed that a single droplet can spread on a surface such that the thickness can be as small as 300 nm with Shipley This indicates that a wafer could be coated at nanometer scale heights given that the droplets could be placed such that they just barely touch each other. This could be possible with the presented droplet ejector, since the ejection directionality is uniform. We observed thicknesses as low as 300 nm for single droplets on wafer surface and as high as 8 by increasing the overlap of the droplets or lines. The temperature increase in the fluid reservoirs is not desired, since it could increase evaporation rate and affect photoresist uniformity, fluid properties, and droplet sizes. We did not experience any significant fluid temperature increase or temperature related problems in our experiments during ejection at 1 KHz. Heat generated on the device surface would be more readily absorbed by the substrate and surrounding silicon and metal before it influences the fluid reservoir. We also observed droplet sizes to be constant in size over 10 min of ejection. Our goal was to demonstrate coating of a wafer with photoresist by using micromachined ejector arrays, which did not require elaborate temperature control initially. We made simple assumptions and calculations for a worst case situation considering the interconnect resistor of 8 ohms in order to get a quantitative feel of temperature generated per droplet in a fluid reservoir. A more detailed model considering the acoustic waves could be developed. Here, the power dissipated on the resistor can be calculated by. The maximum voltage on the whole ejector device was 17 V during a droplet generation cycle, which lasts 10. The temperature raise in the reservoir can be approximately calculated by, where is the rise in temperature, m is the mass of fluid in kg and c is the specific heat given as 1000 calories/kg for water. The mass of the fluid in the fluid reservoir can be calculated from the microfluidic channel dimensions. If we assume that all the heat is uniformly given to the reservoir, it

10 966 JOURNAL OF MICROELECTROMECHANICAL SYSTEMS, VOL. 15, NO. 4, AUGUST 2006 is not necessary to concentrate on the distribution of heat waves traveling in the fluid reservoir. These simple calculations indicate that the temperature of fluid will increase per ejected droplet in the worst case situation, if all the heat generated by the 8 ohm resistor is given to the reservoir at once. It is not a correct approach to multiply this energy per droplet temperature value with 1000 in order to obtain a total temperature increase per second for an ejector operating at 1 KHz. If the ejector is operating at 1 KHz, this means that the ejector will generate one droplet every The length of the droplet generation signal is 10. So, the fluid reservoir is left to cool for 990 after one droplet is ejected. This could explain why temperature is not building up experimentally when 1000 droplets per second are ejected. The increase, which is very small, is already being dissipated during the relaxation time of 990. At 10 KHz, the relaxation time is 90 which is still 9 times the device actuation time. The temperature could be an issue, if we had a much larger ejector array or operated at continuous mode of ejection at higher frequencies for a long time. We initially considered ways such as an electronically controlled cooling pad at the back of the ejector chip or having a bottom layer fluid that flows close to the device surface that would take any possible heat away. These would complicate the device design and was not necessary for the ejector array we used. If this device was to become a commercial product, an electronically controlled cooling pad could be placed at the back of the device to keep the temperature constant in the fluid reservoir. These cooling devices are available for temperature control. Hence, there are many simple ways to control the temperature. These ejectors have several major advantages: the direction of ejection is not dependent on nozzle geometry; the acoustic ejector is simple, repeatable and performs reliably; and droplet ejectors do not harm the heat and pressure sensitive fluids. Moreover, these ejectors can be packed uniformly as arrays and easily addressed individually. Finally, these devices could eject a broad range of fluids, polymers, cell and protein suspensions, which allows them to target various applications, in semiconductor and biotechnology fields, which predict a promising future for acoustically actuated 2-D micromachined microdroplet ejector arrays. VI. CONCLUSION The theory of operation, fabrication and experimental results obtained with a novel acoustically actuated 2-D micromachined picoliter droplet ejector array is demonstrated. Photoresist, water, isopropanol, ethyl alcohol, and acetone are ejected from D micromachined ejector arrays. The ejector operation at 34.7 MHz and generation of 28- diameter droplets in drop-on-demand and continuous modes of operation at 1 10 KHz are demonstrated. Single printed photoresist droplets and lines drawn onto a silicon wafer are demonstrated and characterized. A 4- in wafer was fully coated with photoresist. Acoustic picoliter droplets have great potential to impact the bioengineering field. Current research focus is on interesting biological applications in tissue engineering field for regenerative medicine, such as cell-by-cell 3-D vascularized tissue printing and drug testing, cell sorting for cancer and HIV, and improving the device performance by analytical and FEA methods. ACKNOWLEDGMENT The author would like to thank Prof. K. Saraswat, Prof. F. Shadman, Prof. G. Kovacs, Prof. E. Haegstrom, Dr. G. Percin, Prof. R. Reis, and Prof. M. Toner for their invaluable support. REFERENCES [1] T. L. De Vinne, The Invention of Printing. New York: F. Hart, [2] C. H. Builder, Is it a transition or a revolution?, Futures: J. Forecasting, Planning and Policy, vol. 25, no. 2, pp , Mar [3] A. V. Lemmo, D. J. Rose, and T. C. Tisone, Inkjet dispensing technology: applications in drug discovery, Curr. Opin. Biotechnol., vol. 9, no. 6, pp , Dec [4] Freeman, D. Johnson, R. Stromsburg, M. Rinnen, and Klaus, Midyear 2003 Semicond. Manufact. Market: Wafer Fab. Equipment. New York: Gartner, Aug [5] G. Percin, Micromachined piezoelectrically actuated flextensional transducers for high resolution printing and imaging Ph.D. dissertation, Stanford Univ., Stanford, 2002, ch. 1, ch.6. [6] U. Demirci, G. Yaralioglu, E. Hæggström, G. Percin, S. Ergun, and B. T. Khuri-Yakub, Acoustically actuated flextensional Si N and single crystal silicon 2D micromachined ejector arrays, IEEE Trans. Semicond. Manuf., vol. 17, no. 4, pp , Nov [7] U. Demirci, G. Yaralioglu, E. Hæggström, and B. T. Khuri-Yakub, Photoresist deposition by single reservoir 2D micromachined ejector arrays, IEEE Trans. Semicond. Manuf., vol. 18, no. 4, Nov [8] B. Wandell, Foundations of Vision., MA: Sinauer, [9] J. D. Plummer, M. D. Deal, and P. B. Griffin, Silicon VLSI Technology Fundamentals, Practice and Models. Englewood Cliffs, NJ: Prentice-Hall, [10] B. Hadimioglu, S. Elrod, and R. Sprague, Acoustic ink printing: an application of ultrasonics for photographic quality printing at high speed, in Ultrasonics Symposium, 2001, pp [11] D. Huang and E. S. Kim, Micromachined acoustic-wave liquid ejector, Jour. of Microelec. Syst., vol. 10, no. 3, pp , [12] G. W. Farnell and C. K. Jen, Planar acoustic microscope lens using rayleigh to compressional conversion, Electron. Lett., vol. 16, pp , [13] B. A. Auld, Acoustic Fields and Waves in Solids. Malabar, FL: Krieger. [14] G. S. Kino, Acoustic Waves: Devices, Imaging, and Analog Signal Processing. Englewood Cliffs, NJ: Prentice-Hall, ch. 1, p. 17, and Appendix B, p [15] B. Chu and R. E. Apfel, J. Acoust. Soc. Amer., vol. 72, p. 1673, [16] J. L. Hunter, Acoustics. Englewood Cliffs, NJ: Prentice-Hall, [17] CRC Materials Science and Engineering. Boca Raton, FL: CRC. Utkan Demirci received the B.S. degree in electrical engineering as a James B. Angell Scholar (Summa Cum Laude) from University of Michigan, Ann Arbor, in 1999 and the M.S. degree in electrical engineering, in 2001, the M.S. degree in management science and engineering, in 2005, and the Ph.D. degree in electrical engineering, in 2005, all from Stanford University. He is an Assistant Professor at Harvard-MIT Health Sciences and Technology and Harvard Medical School, Boston, MA. He spent two years at Massachusetts General Hospital, Harvard Medical School, as a Research Fellow. His craft stands on 2 pillars of truth and relief, and follows philanthropy. His research interests involve biological applications of microelectromechanical systems (MEMS) and acoustics, especially microfluidics for low cost CD4 counts for HIV in resource-limited-settings for global health, acoustic picoliter droplets for cell-by-cell 3D tissue generation and semiconductor applications, and capacitive micromachined ultrasonic arrays (CMUTS) for medical imaging applications. Dr. Demirci is a Member of Phi Kappa Phi National Honor Society. He is one of the few recipients of the prestigious Full Presidential Fellowship given by the Turkish Ministry of Education. He is a corecipient of the 2002 Outstanding Paper Award of the IEEE Ultrasonics, Ferroelectrics and Frequency Control Society. He is the winner of Stanford University Entrepreneur s Challenge Competition in 2004 and Global Start-up Competition in Singapore in 2004.