Analyzing suitable methods for applying flux with die bonding
07/01/2002
By Stefan Behler, Dominik Hartman, ESEC, Cham, Switzerland
Overview
A thorough analysis is performed of the many different flux printing systems with an eye to which work best when integrated with automated die bonding. Such an equipment combination is desirable for C4 and solder bump flip-chip assembly operations where system footprint and integrated layout are very important considerations for productivity, cost, and throughput.
Flip-chip assembly using "C4" high-melt or eutectic PbSn solder bumps requires a flux — applied either to bumps or substrate pads — to remove oxides during reflow. The flux also holds die in place during transport. Flux is applied either to bumps (dip fluxing) or to pads during die bonding (flux printing). The latter method can increase throughput. While flux is applied to one pad, another die can be picked and placed without stopping for dip fluxing (Fig. 1).
 Figure 1. Comparison of a) dip fluxing and b) flux printing work flows. |
There are several system requirements when integrating flux printing with die bonding:
Speed. Foremost, an integrated flux printer must keep up with die bonding. A typical process cycle that integrates flux printing with die bonding starts with moving the substrate to a new pad (150-250 msec) and alignment imaging (50-100 msec). Then, the print head is moved into position and prints (100-200 msec). While the substrate is indexed to the next pad, the printhead moves out of camera vision (100-200 msec) and the process repeats (possibly including a printhead-load cycle).
Assume that substrate transport takes 200 msec, image capture 70 msec, and the move to print position 150 msec. To reach a productivity of 5000 units/hr (i.e., 720 msec cycle time), the print cycle cannot exceed 300 msec.
Dosing. Flux can be applied either as dots or as a thin film. Dots minimize the amount of flux, but require high print accuracy that may need optical alignment. Thin-film dispensing may be better suited for high throughput, but leaves more flux residues after reflow.
Volume control. Time, pressure and position make the control of dispensed volume a necessity. This also enables fast changeover from one product to another and compensates for flux viscosity changes. Some systems require a tool change (i.e., stencil, nozzle, etc,) for adjusting dispense volume, which can lead to a decline in productivity.
Single pad printing. To build a die bonder as compact as possible, the flux printer must be close to the bonding stage. Often they share the same substrate positioning system. Printing and bonding might even be performed on the same substrate. Since die attach is done one-die-after-another, a flux printer needs to be able to print single pads. This limits the size of the printer and the appropriate printing method.
Size. Advanced die bonders have been optimized to fit as much functionality as possible into a limited space. A flux printer must not affect the work range of neighboring modules. For example, the lowest part of the print head (~5mm above the substrate) should not use more space than a single substrate pad so neighboring pads or substrate transport devices are not touched.
Chip shift vs. viscosity. Because moving substrates on a die bonder "pulls" a few Gs, which might cause a lateral die shift, a flux printer must handle fluxes that provide enough tackiness measured in grams of vertical force/mm2 (e.g., Indium's FC-NC-HT-D high viscosity flux with h >20,000 cps). If liquid fluxes are used, enough time must be provided after printing for solvent evaporation. Liquid fluxes with a 85-98% alcohol content are best suited for printing thin flux films. Adequate evaporation time could be a few seconds to minutes, depending on film thickness and solvent chemistry. The substrate layout (e.g., matrix arrangement) determines the time between flux print and die bond.
Solvent evaporation. Active evaporation by moderate heating can be applied to reduce drying time. Bottom-side substrate heating is the preferred method because it avoids bubble formation from solvent entrapment by a dry, high viscous flux surface (i.e., a diffusion barrier to solvents). For stable production, control of the solvent evaporation rate is essential. Also, for safety and environmental reasons, appropriate measures have to be taken to remove ignitable solvent vapor from the equipment. We must also consider the susceptibility of open flux reservoirs, such as a dip fluxing module, to solvent evaporation, which changes the viscosity of the flux over time.
Inspection. Solder fluxes are often transparent and cannot be measured by standard optics so inspection of a fluxed pad can be awkward. Ultraviolet fluorescence is a new technique to measure position and in some cases volume of applied flux [1]. However, optical inspection after printing can reduce system throughput.
 |
Flux application methods
The flux application method (see table) must also be carefully considered when integrating flux printing with die bonding:
 Figure 2. Working principle of a) ink jetting and b) ultrasonic spraying for flux application. |
Ink jetting separates a fluid into small volumes and forces it through a nozzle. When it leaves the nozzle, it forms a droplet (nozzle diameter controls dot size) with a certain kinetic energy (generated by a piezoceramic transducer, Fig. 2a) that contacts the surface of the substrate. Droplets <100μm in dia. have more surface energy than kinetic energy so they do not splash when hitting a surface. Dosing heads are commercially available [2]; these have typical diameters of 10-12mm and weigh about 100 grams. Jetting can be combined with nozzle heaters to reduce fluid viscosity, thus improving "jetablity."
Flux dispensing can be performed on-the-fly at distances <1mm with droplet speeds of 2m/sec. However ink jet systems can not operate at lateral accelerations typically >12 m/sec2. If the acceleration is too high, air gets sucked into the nozzle and no ink is dispensed.
Consider a 100 dot layout where an ink jet dosing head has to cover eight 5mm long tracks on a chip (i.e., two parallel perimeter arrays of 14 and 11 per side of 150μm pads at 300μm pitch). Each track starts and ends at a chip corner (i.e., zero speed). At an acceleration of 12 m/sec2 reaching a max. speed of 0.2 m/sec, one track takes 50 msec. Printing 100 dots on 8 tracks takes 400 msec. With 0.3mm pad spacing, the minimum time between two shots is 1.5 msec. Thus, the time resolution of the hardware board that triggers the dosing head needs to be at least 0.25 msec to provide 50μm placement accuracy.
Lateral shift due to dosing head motion also has to be considered (shift can be 0.05mm with a 0.5mm writing distance, 2m/sec droplet speed and 0.2m/sec dosing head speed). Very sophisticated and fast hardware is required to provide adequate synchronization between moving and jetting, when maximum print speed and accuracy has to be achieved.
 Figure 3. Jetting with coaxial air for flux application. |
Jetting with coaxial air uses a mechanical piston to generate the kinetic pulse. The piston also acts as a needle valve that cuts the fluid into small portions (Fig. 3). Low viscosity fluids (<5000 cps) result in thin, wide dots (i.e., printed dot diameter 2.5-5mm, <10μm thick [3, 4]. High viscosity fluids give ~1mm dia. and 200μm thick dots, but this method is not suitable for flip-chip fluxing.) If these thin dots are placed next to each other, they flow together and form a uniform film. A coaxial stream of air can be emitted along with the droplet. If the air stream is triggered right after the droplet leaves the nozzle, the droplet is broken up ("diluted") into a jet of smaller droplets. Another method triggers the air after the droplet contacts the substrate so that the droplet is pressed flat on the substrate to help better wet a surface and to form a uniform film. Both techniques are used for applying flux in flip-chip assembly. Stand-alone systems are available that offer a productivity of up to 4000 units/hr.
Ultrasonic spraying atomizes a fluid into <100μm droplets. Typically, this is done by pressurizing the fluid and forcing it through a nozzle producing a 10 m/sec spray velocity. The spray can be precisely controlled and produces droplets with relatively low 0.1 m/sec velocity. A nozzle vibrates longitudinally at its resonant frequency (20-100 kHz) from a piezo-ceramic actuator (Fig. 2b). The nozzle includes a specially shaped "atomizing surface" at its free end that is wetted by the fluid from the nozzle feed tube. The film picks up the vibration and forms standing waves on the surface. When the vibration amplitude exceeds a certain threshold, the waves collapse and tiny drops of fluid are ejected to the surface. The amount of spray is controlled by a fluid delivery system.
Ultrasonic spraying has several advantages compared to pressure spraying, including less overspray because kinetic energy is lower, very low flow rate, and controlled drop size. In addition, the spray can be focused by a flow of air around the nozzle. However, the application of ultrasonic spraying is limited to low-viscosity fluids (5-50 cps).
Needle dispensing is one of the most widely applied printing techniques in semiconductor assembly. It is used for delivering single dots, lines, or patterns of adhesive or solder paste. A fluid reservoir is connected to a small-diameter needle capillary that is manipulated via motorized positioning to deliver a dot pattern. When a pressure pulse is applied to the fluid, a small portion of it is forced through the capillary and forms a ball at its end. The capillary is moved down towards, but not touching, the substrate pad to transfer the fluid as a dot. The smallest achievable dot size is slightly larger than the capillary diameter (Fig. 4).
Needle dispensing dot writing speed is limited by the reaction time of the dispensing system (viscosity, capillary diameter, and air volume above fluid [5]). The time for moving the capillary and syringe from one dot position to the next is another limiting factor. Commercial systems for needle dispensing of flux and solder paste are available, some capable of 80 msec/dot dispensing speeds.
The productivity of needle dispensing can be increased by using multinozzle dispensers ("showerheads"). For 150μm capillaries, pitches down to 500μm can be achieved, assuming a realistic 75μm capillary wall thickness and a 200μm gap between nozzle tubes. For these small capillary diameters, dot volume will not be as uniform as that from a single nozzle, since variations in flow resistance cannot be avoided (i.e., it varies with fourth power of diameter). Also, the showerhead and the substrate must be held parallel to within ±25 μm. The pitch limitation and nonuniformity of the dot volume mean showerheads are not attractive for flip chip flux dispensing.
Tampon printing allows printing on non-flat surfaces. A highly elastic silicone rubber stamp ("tampon") picks up ink from etched cavities in a metal stencil (Fig. 5). Then the stamp is moved to the printing location and transfers the ink (along with a few ppm of tampon material residues). Stencil holes typically are <50μm deep and down to 20μm in diameter. A blade spreads ink over the stencil. Printed dots are <10μm thick. Areas up to 100 x 100mm can be printed with this method; if the surface to be printed is flat, the print is an exact copy of the stencil.
 Figure 5. Working principle of tampon printing. |
Ink transfer in tampon printing is supported by solvents; ink tackiness depends upon its solvent content. After the ink has been delivered to the stencil cavities, solvents start evaporating from the exposed surface, which makes it stick better to the tampon than to cavity walls. When the tampon is lifted, the new exposed surface of the ink dries up and gets tacky enough to stick to the part that needs printing. Unfortunately, evaporation takes time and results in print cycles >3 sec. If evaporation times are set too short, stencil cavities clog and make printing results unrepeatable. Depending on the ink and the desired print quality, the tampon needs to be cleaned with sticky tape every 50 cycles. The lifetime of a tampon is <100,000 cycles.
 Figure 6. Principle of flux printing by pin transfer. |
Pin transfer transfers a paste via a plate with a series of pins (Fig. 6), where the pins correspond to pad positions on the substrate. The simplest pin transfer method involves dip fluxing the flip chip bumps on the way to placement, since no special tooling is required [6]. Solder-ball mounters for BGA assembly use pin transfer for fluxing more than 10,000 pads on a strip at once. Pin transfer has also been used in epoxy stamping.
 Figure 7. a) Silicone rubber stamp with 300μm high pins at a pitch of 250μm and b) corresponding flux dots printed on glass. Dot size 200μm, center height 30μm. |
We have performed flux printing by pin transfer on glass substrates. This method used conical shaped cast and cured silicone rubber 300μm high pin stamp (Fig. 7a); they are elastic and adapt well to any surface. The pin stamp is dipped into a 60μm thin film of 18,000 cps flux paste. The flux wets the pins and stays attached to them with a predictable amount when the stamp is lifted from the film. The flux stamp is then pressed lightly onto a glass substrate to transfer the flux. The film thickness and the amount of pin compression during printing can adjust dot size and height (Fig. 7b).
Our conclusion from this work was that the flux transfer is very precise and reliable. Print speed was tested by mounting the pin stamp on a pick-and-place system that moved at 200m/sec2. The stamp was dipped into the flux film on one position and touched down on a glass substrate at a second position. With a dip-print time of <100 msec, we achieved perfect print results. To test reliability, we compressed the stamp 150μm in a repeating dry cycle more than 1,000,000 times without damage. In mass production the film thickness needs to be controlled and "tall" pins must be used to avoid filling the gaps between the pins with flux.
Stencil printing of solder paste is well established in semiconductor assembly. Recently, it has also been applied to flip chip wafer bumping [7]. Fine pitch stencils with 50μm apertures at 100μm pitch are available at foil thicknesses down to 25μm. The printed dot volume is well defined by aperture volume. Apertures can have a conical form that supports the separation of the substrate from the stencil at about 2mm/sec.
In this method, a blade spreads the paste uniformly over the stencil and presses the stencil against the substrate. High-viscosity pastes are printed at blade speeds in the range 2-200mm/sec. Stencils, however, do not adjust well to substrate surface variations (e.g., photoresist or thickness variation). In many cases only the use of an under-stencil cleaner guarantees stable production. The cleaning cycle has an impact on the productivity of the system.
Perhaps most significant, a stencil printer requires a certain amount of parking and working space for the blade. Therefore, it is not realistic to integrate a miniature stencil printer on a die bonder.
The size problem could be avoided by using a 2-5mm thick stencil with a downset at the pad position. However, with such a setup, dot volume is no longer controlled by stencil thickness, but rather depends on the nonreproducible breaking of the flux column within a stencil cavity when the stencil is separated from the substrate. Furthermore, fine pitches and small apertures — required for flip-chip fluxing — cannot be realized with thick stencils.
Conclusion
Each of the described flux printing systems has its specific advantages and disadvantages. From a print speed point-of-view, clearly the dot mode systems cannot compete with film spraying-jetting techniques. Ink jetting and pin transfer are capable of <500 msec print times for a productivity of up to 5000 units/hr. For film dispensing, one has to keep in mind that an additional process step, the controlled evaporation of solvents, has to be added. On the other hand, dot dispensing requires optical pad alignment.
In principle, all methods except a stencil printer could be integrated on a die bonder with certain efforts. According to this analysis and our experience as equipment manufacturers, a realistic integration is only feasible for film spraying and coaxial air jetting techniques and for dot dispensing by pin transfer.
References
1. S. Kalisz, High Density Interconnect, p. 22, Nov. 2000.
2. D.J. Hayes, et al., Proceedings of 1999 International Conference on High Density Packaging and MCMs, pp. 242-247.
3. W.E. Donges, K.J. Fox, Proc. Nepcon West 1998, p. 181 ff.
4. W.E. Donges, K.J. Fox, High Density Interconnect, p. 40, October 2000.
5. M. Krieger, S. Behler, Proceeding of Semicon Singapore 1998, pp. 105-111.
6. W. Prinz von Hessen, "Flip Chip - Integrated In A Standard SMT Process," Universal Instruments, Binghampton NY.
7. J. Kloeser, et al., Proc of 1999 Intl Symposium on Microelectronics, pp. 1-7.
Stefan Behler received his MS in experimental physics from the University of Göttingen and his PhD in physics from the University of Basel. He is a senior process engineer at ESEC, Hinterbergstr. 32, CH-6330 Cham, Switzerland; ph 41/41-749-5111, fax 41/41-741-64-84, email [email protected].
Dominik Hartmann received his BS in mechanical engineering from the Zentralschweizer Technikum in Luzern, Switzerland, and his MS from McGill University in Montreal, Canada. He is senior robotics engineer at ESEC.