Step 1: The back-end process
01/01/2001
Wafer dicing
Step by step
BY DIANNE SHI AND ILAN WEISSHAUS
 |
During the past 30 years, dicing systems and blades have been continuously improved to address process challenges and accommodate the requirements of different types of substrates. Recent equipment developments making the greatest impact on productivity include use of dual-spindle dicing systems that make two cuts simultaneously with minimal overtravel, automatic spindle torque monitoring and automatic coolant flow adjustment capabilities. Significant dicing blade developments include blades for wafers with very narrow streets and/or taller die sizes, wafers with copper (Cu) metallization, very thin wafers, and wafers with devices that require a polished surface finish after dicing. Many of today's demanding applications require processes in which both the equipment capabilities and blade characteristics are optimized to provide the highest possible yields at the lowest possible cost.
The Dicing Mechanism
The silicon wafer dicing process is the first step in "back-end" assembly. This process divides silicon wafers into single chips for subsequent die bonding, wire bonding and test operations.
A rotating abrasive disc (blade) performs the dicing. A spindle at high speed, 30,000 to 60,000 rpm (linear speeds of 83 to 175m/sec) rotates the blade. The blade is made of abrasive diamonds, which are embedded in an electroplated nickel matrix binder.
During the separation of dice, the blade crushes the substrate material (wafer) and removes the created debris simultaneously. Material removal occurs along dedicated dicing lines (streets) between the active areas of the dice. Coolant (typically deionized water) is directed into the cut to improve cut quality and extend blade life by assisting in the removal of debris. The width (kerf) of each street is proportional to the thickness of the blade.
Key Process Parameters
The goal in silicon wafer dicing applications is to maximize throughput and yield while minimizing cost of ownership. The challenge, however, is that increasing throughput often reduces yield and vice versa. The rate at which the wafer substrate is fed into the cutting blade determines throughput. As the feed rate increases, cut quality becomes more difficult to maintain within acceptable process windows. The feed rate also affects blade life.
The narrower street widths commonly encountered during the dicing of many wafers requires the capability to place each and every cut within several microns of the street center. This dictates use of equipment with high indexing axis accuracy, high optical magnification and advanced alignment algorithms.
A common recommendation when dicing wafers with narrow streets is to select the thinnest blade possible. Very thin blades (20 µm or less), however, are significantly weaker and more susceptible to premature breakage and wear. As a result, their life expectancy and process stability is inferior to thicker blades. The recommended blade thickness for 50 to 76 µm streets should be 20 to 30 µm.
Chipping
Top-side chipping (TSC), which occurs on the top of the wafer, becomes a yield concern when chips approach the active area of the die and is predominantly dependent on blade grit, coolant flow and feed rate (Figure 1a).
Back-side chipping (BSC) occurs on the bottom surface of the wafer when large, irregular micro-cracks propagate away from the bottom of the cut and join together (Figure 1b). BSC becomes a yield issue when these micro-cracks are long enough to cause removal of unacceptably large particles from the kerf.
Usually, the quality criteria for diced silicon wafers are as follows: if the size of the back side chipping is below 10 µm, it is disregarded. On the other hand, when the size is greater than 25 µm, it is considered potentially damaging. An average size of 50 µm may be accepted, however, depending on wafer thickness.
The tools and techniques available for controlling back-side chipping are blade optimization followed by process parameter optimization.
Blade Optimization
To address today's new dicing challenges, the synergy between the dicing system and the blade is essential. This is especially the case for high-end applications. The blade plays a major role in process optimization. To accommodate all the new dicing requirements resulting from rapid technology development, a tremendous variety of blades are available today. This makes selecting the right blade for the right process a more complicated task than ever before.
In addition to dimensions, three key parameters determine blade characteristics: the diamond (grit) size, the diamond concentration and the bond type. The bond is a matrix of various metals and/or binding materials in which the diamonds are dispersed. The combined effect of these components defines blade life and cut quality (TSC and BSC). Changing any of these parameters will directly affect blade properties and performance. Generally, blades with shorter life (i.e., faster wear) give better cut quality. Selecting the best blade for a given dicing process may require trade-offs between blade life and cut quality.
 Figure 2. Effects of feed rate, blade grit size and blade life during Si wafer dicing at 30,000 rpm. |
Other factors, such as the feed rate and the spindle speed, also can affect blade selection. The cutting parameters have a direct relationship to the material removal rate, which in turn affects blade performance and process efficiency. To optimize the blade for a process, the designed experiment (DOE) method can reduce the number of trials needed, and provide the combined effects of the blade properties and process parameters. In addition, the statistical analysis of DOE enables extrapolation of useful information to suggest directions for further process optimization with even higher throughput and/or lower cost of ownership.
Figure 2 shows the combined effects of feed rate, blade grit size and blade life (wear rate) during silicon (Si) wafer dicing at 30,000 rpm spindle speed. Figure 3 details the corresponding cut quality. As expected, within certain process windows, with increased grit size, blade life increases (or wear rate decreases), whereas the cut quality (TSC in this case) deteriorates. Finer grit size provides superior top-side cut quality at suitable feed rates, though blade life is significantly reduced. As mentioned, increasing throughput is a main concern for cost reduction in wafer dicing. Higher feed rates and/or spindle speeds are the parameters most frequently considered for this effect. However, depending on the three key blade parameters, the properties and thickness of the wafers, and the extent of metallization present in the streets, a "perfect blade" under one set of process conditions may not be suitable when the conditions are changed.
 Figure 3. Cut quality of Si wafer dicing at 30,000 rpm. |
As shown in Figures 2 and 3, with the increase in feed rate, both blade life and cut quality vary, regardless of the grit size. When optimizing a process, it is important to consider in parallel the various major process and blade parameters to maximize the throughput while keeping adequate blade life and acceptable cut qualities. A compromise may be required when selecting the blade type to maximize the reduction of cost of ownership.
Contrary to what may be common, slower feed rates do not always guarantee better cut qualities. Poor cut quality can be generated at feed rates that are too slow because of higher heat generation (Figure 3). The minimum feed rate to achieve acceptable cut qualities should be reached for a given dicing application. This is a general conclusion that can be applied for various types of wafer dicing. Higher BSC was observed in dicing Cu wafers when very slow feed rates were used.1 The same effects were observed when dicing at very low rpm.
The relative importance of the three key blade elements (diamond size, diamond concentration and bond hardness) depends on both blade grit size and process parameters. To select the blade best-suited for a specific application, an understanding of these relationships is essential.
 Figure 4. DOE evaluation of decrease in blade wear. |
Figure 4 is a DOE evaluation of the percent decrease in blade wear because of either increase in blade diamond concentration or bond hardness. Generally speaking, the effect of bond hardness on blade life is more pronounced with a finer diamond size. As grit size increases, the impact of bond hardness becomes less significant. The effect of diamond concentration, however, appears to be more important than bond hardness for all grit sizes. Depending on the extent of change in grit size, its effect on blade life can be most significant among all three blade parameters.
As a general rule, blades with finer diamonds are more sensitive to blade and/or process parameter changes. Softer bonds and/or lower diamond concentrations are often necessary when BSC needs to be improved. Changes in either bond hardness or diamond concentration can decrease blade life (Figure 5).
To select a blade, it is important to also understand the effect of the apparent hardness (often referred to as the matrix hardness) of the blade. This is an abstract measure of the hardness of a blade, and it reflects the way the blade "feels" when dicing a wafer. The matrix hardness is determined by the combined effect of the diamond grit size, diamond concentration and bond hardness. Usually, finer grit size, higher diamond concentration and harder bond will result in increased matrix hardness.
It is generally advised, together with other considerations, that harder materials require softer (matrix) blades to dice and vice versa. For example, gallium arsenide (GaAs) wafers normally require finer diamond grit size (harder blades), whereas lithium tantalate (LiTaO3) wafers are well-suited for coarser diamond size and lower diamond concentrations (softer blades). Deeper understanding of these types of advanced wafer dicing applications will arrive as the use of non-Si materials progresses.
Blade Load Monitoring
In dicing, or any other grinding process, the quality of the ground surface is related to the force applied by the blade. A relatively recent development is the use of this knowledge to control the process via dicing systems capable of force measurement, statistical data analysis and detection of process deviation.
To achieve the maximum feed rate without exceeding acceptable cut quality parameters, new generation dicing systems can automatically monitor the load, or torque, applied on the blade. For each set of process parameters, there is a limit torque value where dicing quality deteriorates and BSC appears. Correlation of cut quality with the blade substrate interaction force and measurements of its variations enable the determination of process deviations and damage formation. The process parameters can be adjusted in real-time so the torque limit is not exceeded and maximum feed rate is obtained.
A crucial part of the dicing procedure is the dressing of the cutting blade. In non-monitored dicing systems, the dressing procedure is established by a set of trial-and-error cycles. In blade load monitored systems, the end point of the dressing is detected by measured force data, which establishes optimal dressing procedures. This method has two advantages: no margin time is needed to ensure optimal blade performance and there is no yield loss because of poor quality that results from dicing with a partially dressed blade.
Coolant Flow Stabilization
System operation at steady torque requires stabilization of the feed rate, spindle speed and coolant flow. The coolant applies drag force on the blade, which contributes to the torque. The latest generation of dicing systems keeps the coolant torque effect steady by controlling the coolant flow to maintain a steady flow rate and drag force.
When the dicing machines have steady coolant flow, and all other parameters are controlled, a steady torque is maintained. Any deviations from steady torque, if recorded, are because of uncontrolled factors. These include changes in coolant flow because of nozzle clogging, changes in nozzle adjustment, blade-to-blade variation, blade condition and operator errors.
Summary
The dicing process is becoming more and more demanding. Cutting streets are narrower and may be filled with test pads, and the blade may need to cut through various coating layers made of different materials. Achieving maximum dicing process yield and productivity under these conditions requires careful blade selection and advanced process control capabilities.
Reference
- D. Shi, "Dicing of Cu Wafers," Advanced Technology Seminar I & II, SEMICON Singapore 2000, May 2000, pp.57-62.
DIANNE SHI is director of R&D dicing and grinding wheels, and Ilan Weisshaus is manager of dicing process development at Kulicke & Soffa Industries Inc. For more information, contact Dianne Shi, Kulicke & Soffa, 3025 Stender Way, Santa Clara, CA 95054-3216; 408-496-1092; Fax: 408-496-1091: E-mail:[email protected]