Challenges in 300-mm wafer assembly
07/01/1997
Challenges in 300-mm wafer assembly
Cary H. Baskin, Kulicke & Soffa Industries, Willow Grove, Pennsylvania
While the economics of front-end wafer fabrication continue to drive the development of 300-mm wafers, unexplored back-end assembly issues raise many questions and may pose roadblocks. It is time for users and designers of assembly machines to consider how machines will have to be modified to process these extremely large disks of silicon.
Wafer mounting, dicing, and die-attach machines will require new wafer handling systems to accommodate the increased diameter. The process issues resulting from the larger wafer diameter need to be explored to determine which processes are readily transferable from 200-mm machines and which will require new designs, inventions, or innovations.
Material handling
Assembly lot size for 300-mm wafers. The production lot size for assembly may not be the same as the 12- or 24- wafer lot size standard being adopted for the front end. The maximum lot size for today`s dicing and die-attach machines is 25 wafers, and a standard for 300-mm assembly has not yet been defined. Lot size and wafer handling must also be determined for both probe and back-grinding machines, though they are not generally considered part of assembly.
Lot size selection could create ergonomic and process issues that would necessitate equipment modification. A wafer cassette sized to support mounted 300-mm wafers is larger and more cumbersome than a front-end cassette carrier. A conventional film frame adds 79 mm (2.9 in.) to the wafer diameter (Fig. 1).
An assembly production lot of 25 300-mm wafers translates to a cassette weighing an estimated 13.6 kg (30 lbs), which must be lifted and carried by an operator. By comparison (Fig. 1), a cassette of twenty-five 200-mm wafers weighs only half as much [1].
 |
Figure 1. Size and weight of 300-mm film frame.
SEMI S8-95 safety guidelines [2], which refer to NIOSH tables [3], specify 8 kg (18 lbs) as the maximum lift weight for a female operator 45 years or older. To comply with this specification, a manually loaded cassette would have to contain fewer than 12 300-mm wafers.
For greater than 12-wafer lots, automated material handling systems will be required to transport, lift and load these very bulky cassettes. If automated material handling systems are to be adopted for assembly, then new designs and standards will be required for the electrical and mechanical interfaces between the wafer mounter and dicer, and between the dicer and die-attach machines.
Given the large number of dice available from a 300-mm wafer, lot size may be reduced to some quantity less than a full cassette of wafers. This will help minimize the lot cycle time and lower the amount of work-in-process.Assumptions:
Some semiconductor manufacturers have indicated that a 300-mm assembly lot size could be a single wafer. However, single wafer lots negate the utilization benefit of being able to process several wafers for each manual cassette loaded. Dicing and die-attach machines can be designed to process a single wafer or a cassette of multiple wafers. Configuring a machine to do both is possible, but the additional required mechanisms would increase cost.
Preparing 300-mm wafers for assembly. The current film-mounting process of pressing a film to a wafer introduces thermal and mechanical forces to the backside of the wafer. Force is required to achieve good adhesion and to prevent any air from being trapped between the wafer and the film. Excessive wafer warpage may degrade the mounting quality.
If larger wafers are more fragile than today`s 200-mm wafers, more breakage may occur. Process studies will have to be conducted to determine the maximum mounting force and minimum required wafer thickness/strength to prevent fracture during mounting. Currently, many manufacturers manually mount the wafer to the film. With the increased value and possible fragility of 300-mm wafers, the increased process control of automated mounting may be required.
 |
Dicing process issues
Increased dicing time. A larger-diameter wafer will be exposed to the dicing process for a greater amount of time (Table 1). This increased dicing time raises some concerns:
Surface wetness. The larger a wafer gets, the more difficult it is to keep the entire surface wet during dicing. A surfactant additive may be needed to maintain wetness over the entire surface area of 300-mm wafers.
Blade behavior. Longer cuts across 300-mm wafers will increase the amount of time a blade is exposed to the mechanical and thermal stresses of dicing. Process control procedures may have to be modified to reduce the risk of broken blades.
Wafer corrosion. With the possibility of increased exposure time to deionized (DI) water at high resistivity, some wafers may corrode. Controlling DI water resistivity may become more critical.
 |
Figure 2. Geometric model of maximum sag-angle after dicing before die-edge contact.
Removal of slurry. Pressurized water is sprayed onto the wafer surface to lift and push-off the silicon slurry created during dicing. In the transition from 150- to 200-mm machines, the spray bar was lengthened and more holes were added to increase the total volume of fluid flowing over the wafer. A similar design change for 300-mm wafers is expected.
Wafer cleaning/drying. Conventional cleaning and drying involves spinning a wafer up to 3000 rpm on a table of size proportional to the wafer. Since table balance is more affected by the rotational velocity than the increase in diameter, current table designs should be adequate for larger diameters as long as the rpm rate does not have to increase significantly to clean 300-mm wafers. The greater volume of a larger cleaning chamber (or "well") is more difficult to control, and thus more of a challenge in developing a drying process.
Chuck design. A poor interface between the vacuum chuck and the film could result in backside chipping during dicing. The 200-mm vacuum chuck flatness specification has evolved over the years to be around 5 ?m. The 300-mm specifications may be even tighter, to compensate for potential wafer warp and nonuniformity of the backside film. In addition to chuck flatness, the turntable and x-slide planarity will need to be better than 15 ?m (0.6 mil) to maintain linearity over a longer cut.
If the chuck flatness specification is not achievable, or if a wafer is excessively warped, a thicker film may be needed to compensate for surface irregularities. A thicker film allows for a deeper cut, thus ensuring that dice are completely cut through.
Depth of cut and film thickness play a role in backside chipping, but the mechanism is not yet understood. A given combination of wafer, blade, film type, and depth of cut will not always yield the same results. Deeper dicing has been seen both to reduce and increase backside chipping.
A large-diameter sawn wafer must remain reasonably horizontal when it is released from the vacuum chuck. Nonuniform release can result in chipping if the edges of the cut dice come into contact.
Film frame sag experiments
300-mm wafers weigh more than twice as much as 200-mm wafers, and more than four times as much as 150-mm wafers [1]. This increased weight, along with a larger film area, causes additional film frame sag. Excessive sagging after dicing can cause chipping if the edges of the sawn dice come into contact.
The maximum sag before die contact was calculated as a function of wafer diameter, die size, and wafer thickness. Empirical results were obtained from 200-mm wafers mounted on 200- and 300-mm film frames. Precut sag was measured on both frames; diced sag was measured on 200-mm frames.
 |
Figure 3. Model of deflection margin before top edge of dice would make contact.
Geometric model. The sag angle, at which the top edges of two adjacent dice (located in the exact center of the wafer) would make contact, was determined by a geometric model. The sag angle, A, formed by a triangle with sides a, b, and c (Fig. 2), was calculated as:
 |
For a 1.4-mil kerf width, the resultant sag angle was calculated for die thicknesses of 381, 765, and 762 ?m (15, 25, and 30 mils), and projected for die lengths of 2540 and 6350 ?m (100 and 250 mils). Based on these results, the maximum allowable film sag was calculated (Table 2).
The geometric model supports two conclusions:
1. Die size has minimal effect on the maximum sag.
2. Wafer weight (increasing with diameter) and film tensile strength are the main factors in finding the maximum sag.
 |
Figure 4. Uncut sag vs. sag margin on 3-mil, medium-tack film.
Wafers were mounted on 200- and 300-mm film frames to measure the uncut wafer film-sag. 150- and 200-mm wafers were then diced on 200-mm film frames to measure the amount of sag introduced by dicing. The difference between the measured sag and the maximum calculated allowable sag before top die edge contact (at the center of the wafer), is the calculated margin of allowable sag (Fig. 3).
Uncut wafer sag. After mounting medium tack 3-mil-thick film to a 200-mm film frame using standard techniques, the natural preload sag (prior to adding the mass of the wafer) was measured at a nominal 51 ?m (2 mils).
Next, standard techniques were used to mount 150- and 200-mm blank polished wafers to 200-mm film frames, and 150-, 200-, and 300-mm wafers to 300-mm film frames. Each mounted wafer and film frame was then suspended in air to allow the film to sag from the weight of the uncut wafer (Fig. 4).
A 200-mm wafer on a 200-mm film frame had a measured precut sag of 147 ?m (5.8 mils) with repeatability of 56 ?m (2.2 mils), consuming about 8% of the 1778 ?m (70 mils) of margin available for sag. The 300-mm wafer on a 300-mm frame consumed nearly 55% of the available sag margin.
Diced wafer sag. To determine the amount of sag introduced by dicing, 200-mm wafers were mounted and diced on 200-mm film frames. The results (Fig. 5) were then compared to the maximum sag margin specified in Table 2.
 |
Figure 5. Diced wafer sag vs. die size on 3-mil, medium-tack film.
Post-dicing sag, on a 200-mm film frame, was measured at 1143 and 1219 ?m (45 and 48 mils) for index steps of 100 and 200 mils, respectively. Thus, dicing introduced about 1016 ?m (40 mils) of sag and left about 635 ?m (25 mils) of allowable sag margin before die top-edge contact. Film tension variation will alter the exact extent of sag.
The measured precut 300-mm wafer sag of 1473 ?m (58 ?2 mils) on a 300-mm frame, consumed a little more than half of the calculated margin of allowable sag before die top edge contact (at the center of the wafer). Dicing wafers on 300-mm film frames may require a film thicker than 3-mil and/or a stronger film material to minimize sagging.
The initial 300-mm film sag before wafer mounting was not measured (as was done with the 200-mm film), so it is not clear how manual film mounting may have contributed to the results. Future experiments should include measurement of the film tension at the center of the film frame before wafer mounting.
Die-attach process issues
Increasing the wafer diameter should not introduce any process issues at die-attach. However, a design change will be needed to stretch a 300-mm wafer to the desired tension.
Film stretch. A film holding a diced 200-mm wafer needs to stretch about 6.4 mm (0.25-in.) to insure that the edges of all dice are completely isolated from nearest neighbors during die eject. Separation is typically 1 mil additive to the kerf width. A 300-mm film will need to be stretched about twice as much, an estimated 13 mm, to maintain the same die separation.
Approximately 29 kg (65 lbs) of force is required to stretch a 200-mm wafer mounted on a 200-mm film frame by 6.4 mm. The force required to stretch the film for a 300-mm wafer may be as much as 45 kg (100 lbs), which will require larger actuators and mechanics. The actual separation force required will depend on the film material, film thickness, depth of cut, and street width.
Current die-attach machines release the film frame from the clamp after die-attach, and then transfer the film to a heat station. Heat shrinks the film back to its approximate original tension, to enable insertion of the film back into a cassette. Due to the increased stretch of a 300-mm film, the film may need to be heat shrunk before removing the frame from the clamp.
Plunge-up dynamics. The amount of tension after stretch is approximately the same for 200- and 300-mm films, so the plunge-up force at die eject is expected to be similar. Since a vacuum system will continue to hold the film in place before the die eject process begins, a small amount of additional sag should be acceptable.
UPH optimization for dicing and die-attach
A machine`s productivity must be balanced with the amount of floor space it consumes. A 300-mm assembly machine will have an inherently larger footprint due to dicing-blade or die-attach bond-head stroke length, increased transfer-mechanism travel, greater wafer cassette volume, etc.
300-mm systems must maintain the same wafer/hr (at dicing) and units/hr (at die-attach) per m2 as 200-mm machines. 300-mm productivity should be inherently greater than equivalent 200-mm machines because there are relatively fewer handling motions to process the same number of dice.
Dicing productivity. Dicing time depends on the number of index steps, the linear speed of the blade as it travels through the wafer, the average length of each cut, and the number of wafer transfers and alignment steps. A machine dicing 300-mm wafers will have 2.5 x fewer wafer-alignment and load/unload steps per dice output. Therefore, a production lot of eleven 300-mm wafers will have 14 fewer alignment and load cycles than a lot of twenty-five 200-mm wafers, saving substantial time (Table 3).
The following expression can be used to estimate the units-per-hour (UPH):
UPH = QW (tC + tA + tL) [2]
where
UPH = the number of wafers diced/hr
QW = the quantity of wafers in the lot
tC = the cut time/wafer
tA = the alignment time/wafer
tL = the time to load/unload a wafer
Cleaning is considered a parallel process and does not impact UPH. Based on the above equation, the average UPH can be calculated as a function of lot size, wafer diameter, and die size. A lot of eleven 300-mm wafers is 8-15% more productive than the equivalent lot of twenty-five 200-mm wafers (Table 3). Thus a 300-mm dicing machine can have about a 10-15% larger footprint while maintaining constant UPH/m2.
Die-attach productivity. The time it takes to die attach a sawn wafer depends on the number of wafer transfers and alignments, the number of good dice to be picked (wafer yield), and the speed of the transfer bond-head. A machine die bonding 300-mm wafers will have 14 fewer wafer load/unload and alignment steps/wafers processed.
As in the dicing productivity calculation above, die-attach load and alignment are required only once per 2.5 wafers when compared to 200-mm wafers for the equivalent number of die. 300-mm die-attach will be only 1-9% more productive, and UPH/m2 for die-attach machines will degrade more rapidly than dicing machines with increasing footprint.
Business considerations
According to data published by VLSI Research Inc. [4], there will be 0.3 million 300-mm wafer-starts in the year 2000 (0.1% of the 205 million wafers forecast), growing to 1.8 million wafers in the year 2001. The amount of dicing and die-attach capacity required to support this level of demand is minimal.
Potential equipment revenues through the year 2005, based on anticipated 300-mm dicing and die-attach machine prices, may not provide sufficient return on investment (ROI) to support the development of these new machines. ROI concern is based on two factors:
The number of fabs that will run significant 300-mm wafer production volumes over this time period is uncertain.
The lack of lot-size standards may necessitate two different machine designs: manual stand-alone, and integrated material handling.
 |
Table 3. Dicing UPH vs. lot size, wafer diameter and index size
Cost-of-ownership model inputs for new machines must include productivity/m2, manpower requirements based on lot size and conversion frequency, and depreciation cost for the more expensive machines. Cost-benefit analysis is also needed for any ergonomically mandated automated material handling systems.
Conclusion
Some of the identified assembly issues that will require further investigation can be addressed by equipment manufacturers during machine development. However, some of the issues need to be faced by IC manufacturers, and resolved before 300-mm wafers can be assembled with predictable yield.
The main concerns are wafer flatness during dicing and the ability of the film to support sawn wafers adequately. Manual wafer mounting may not be acceptable for 300-mm wafers. Standards for assembly lot size, and handling requirements for film frames and cassettes are also needed. Failure to reach industry consensus will force each equipment manufacturer to make inefficiently independent design decisions.
In the front-end, IC manufacturers and equipment suppliers closely cooperate to establish design guidelines and industry standards. A similar paradigm of mutual cooperation and trust for 300-mm wafer assembly would provide the same industry benefit. Since the development cycle for a new machine takes about two years, machine requirements and process questions must be quickly resolved to stay on development roadmaps.
Acknowledgments
The author wishes to acknowledge the technical assistance and support of the following individuals: Robert H. Smith, senior applications engineer; Bob Larson, mechanical engineer; Gerry Michaud, engineering manager; John Huck, product manager, Kulicke & Soffa Industries; and Rami Langer, deputy managing director, Kulicke & Soffa Ltd; and M. Matsuda, general manager Hachioji Engineering Dept., Tokyo Seimitsu Co. Ltd.
This article is an update of "Process and Machine Considerations for Assembling 300-mm Wafers," which was first published at SEMICON/Japan 1996 by Semiconductor Equipment and Materials International Inc., 805 East Middlefield Road, MountainView, CA, 94043.
References
1. Weight was based on data for 300-mm wafer characteristics from SEMI/STEP: Standard for the 300-mm Era, 1995.
2. SEMI S8-95 Safety Guidelines for Ergonomics/Human Factors Engineering of Semiconductor Manufacturing Equipment.
3. NIOSH Work Practice Guide for Manual Lifting, US Department of Health and Human Services, Cincinnati, OH 45226, 1981.
4. VLSI Research Inc., Forecast of Worldwide Silicon Demand by Wafer Size, June 10, 1996.
CARY BASKIN received his bachelor`s degree in electrical engineering from Pratt Institute, his master`s degree in bioengineering from Polytechnic Institute of New York, and his MBA from Pace University. He is VP of Kulicke & Soffa`s Value-Added Products & Services Group, and has served as VP of the company`s Die Bonder business. Prior to joining K&S, Baskin spent more than 20 years in a variety of engineering and marketing positions for companies involved
in medical instrumentation and automatic identification technologies. Kulicke & Soffa Industries Inc., 2101 Blair Mill Rd., Willow Grove, PA, 19090, ph 215/784-6000, fax 215/659-7588.