An analysis of potential 450mm CMP tool scaling questions
Since the invention of the integrated circuit, microchips have been fabricated on progressively larger silicon wafers to take advantage of the economies that can be realized from being able to process more die simultaneously. In the early 1970s, a 50mm wafer could hold about fifty 0.5 x 0.5cm dice. More than 36 times as many die of the same size can now be processed on a 300mm wafer. Even allowing for the greater cost of the larger wafer and the tools needed to process it, the growth in wafer area has historically been a significant factor in reducing the cost per die. This article discusses the considerations that arise with respect to chemical mechanical planarization (CMP) for 450mm wafer manufacturing.
Leonard Borucki, Ara Philipossian, Araca Inc., Mesa, AZ USA; Michael Goldstein, Intel Corp., Santa Clara, CA USA
A change to 450mm would involve a significant investment in tools and fabrication facilities, so questions naturally arise about whether tools that are currently used for 300mm could, in some cases, simply be scaled up and whether there are any problems created by doing so.
For example, in CMP processes:
• Would rotary tool scaling to 450mm lead to dangerous heating of the pad or wafer?
• Would slurry costs become prohibitive due to the increase in pad area?
• Could the polishing pad still be conditioned in the same way?
We investigated many of these questions for an example CMP process run on a hypothetical 450mm rotary polisher. For comparison, identical simulations were also run for similarly configured 200mm and 300mm tools.
Araca, Inc., developed the specialized programs that were employed. The software implements many of the key physical models needed for tool-scale CMP studies. For example, we use a three-dimensional (3D) load and moment balance model to simulate rough surface contact between the wafer and retaining ring and the pad. This makes it possible to estimate the slurry film thickness distribution under the retaining ring and the wafer. Since slurry has a high capacity to absorb heat, this is important for getting the correct wafer and ring temperatures.
The software also has a model for the injection of slurry at multiple points onto plain (un-grooved) pads that have rough surface textures, and for the subsequent flow over the pad, under the retaining ring and wafer, and ultimately off of the platen. A 3D thermal model simulates frictional heat generation by the wafer and retaining ring, and the transport and transfer of heat throughout the tool by the slurry and by the rotation of the polishing head and the platen.
|Figure 1. Rotary CMP tool schematic layout.|
A schematic layout of the hypothetical rotary polisher is shown in Fig. 1. Both the platen and head rotate counterclockwise. Slurry is applied to the pad within the wafer track using a bar applicator with nozzles spaced 1" apart. As the wafer size is scaled up, the number of nozzles is correspondingly increased. The total flow is divided equally between the nozzles.
The slurry viscosity and thermal properties are typical for slurries in current use. For the pad, the mechanical, thermal, and surface textural properties correspond to a commonly used commercial hard pad with a soft sub-pad. The polishing head poses a special thermal analysis problem. Because the head design for 450mm is not known and, therefore, cannot be simulated in detail, we assume a simple head for all platform sizes consisting of a ½"-thick aluminum plate. A thin polymer wafer backing film is interposed between the wafer and the head. The retaining ring is assumed to be 1" wide regardless of wafer size, mechanically independent of the wafer, and to have the thermal properties of PEEK (polyetheretherketone).
To compare the platforms, we use the same polishing time (1 min.), the same wafer and ring pressures (2psi for the wafer, 4psi for the ring) and a constant pad/wafer relative sliding speed of 1m/sec. Constant pressure and sliding speed are a starting point for obtaining the same material removal. For simplicity, we assume that the wafer and platen co-rotate so that the relative sliding speed is constant everywhere on the wafer.
On a rotary tool, the distance between the center of the wafer and the center of the platen must increase with the wafer diameter; therefore, a constant relative sliding speed implies that the platen rotation rate must decrease as the wafer size increases. Thus, at 1m/sec, the platen rotation rate is ~69rpm for 200mm, 47rpm for 300mm, and only 33rpm for 450mm.
What should the slurry flow rate be for a 450mm process? One idea would be to scale the slurry flow rate with the pad area. The proposed pad diameter for 450mm polishing is 1094mm, compared with 762mm for 300mm wafers and 508mm for 200mm wafers (some variations exist). Based on area, the slurry flow rate for a 450mm process should be about 2.1 times that of the same 300mm process and 4.6 times greater than a 200mm process. Scaled up in this way, a 200mm process consuming slurry at 150ml/min would require almost 700ml/min on a 450mm tool—a large increase.
There is a more subtle way, however, to analyze the question. Reducing the platen rotation rate has several effects, one of which is to reduce the slurry's outward centripetal acceleration. At a fixed distance from the platen center and at a fixed flow rate, reduced acceleration increases the slurry film thickness. For all tool sizes, however, the thickness of the gap between the wafer or ring and the pad is determined by the pad properties and applied pressures, and is, therefore, approximately constant. This suggests that one should try to scale the slurry flow rate so that the average slurry thickness on the pad outside of the ring and wafer also remains constant as the tool is scaled up.
A rigorous analysis is feasible when the slurry flow outside of the pressure ring is modeled using the thin film equation —a simplification of the incompressible Navier-Stokes equations that is suitable for modeling transport over a pad surface without grooving. A mathematical scaling analysis of the thin film equation suggests that the slurry flow rate should be scaled like the ratio of the pad areas times the ratio of the rotation rates. This scaling procedure reduces the flow rate for 450mm to 700 x 33/69 = 334ml/min in the example, or about half of what area scaling indicates.
Figure 2. Slurry film top view (left) and mean film thickness comparison (right)..
Figure 2 shows a top view of the steady state slurry film thickness on a 450mm tool. The bow wave is visible at the upstream side of the ring, and it is clear that the slurry thickness under the ring is smaller than the thickness under the wafer because of the applied pressure difference. The slurry thickness in the bow wave never exceeds 1mm. The graph in the same figure shows that the mean slurry thickness on the pad reaches steady state within a few platen rotations, and that the slurry flow rate scaling law has succeeded in producing nearly the same mean film thickness (~20μm) for all tool sizes.
|Figure 3. Wafer and ring radia ltemperatures (left) and a 3D-view (right) .|
Because the slurry used in a polishing process may be thermally activated, the temperature distribution of the wafer, or wafer body temperature, is of interest. The calculated steady state temperature rise above ambient for the surface of the pad, the ring, and wafer for a 450mm tool is shown in Fig. 3 (pg. 13). The largest temperature increase—17°C—occurs on the retaining ring near the pad center where cooling by fresh slurry is minimal. The ring temperature is also affected by the coefficient of friction, which is larger for PEEK than for the wafer.
Figure 3 also shows a scatter graph of the temperature increase as a function of radius on the wafer and ring for the three tool sizes. The wafer is slightly center hot, a prediction that has been verified experimentally . Surprisingly, the wafer temperature distribution for 450mm is predicted to be a smooth extension of the distribution for 300mm. At the edges of the wafer, there is a suggestion of wafer edge heating due to heat transfer from the ring; but again, it is the same for 450 and 300mm.
Another important quantity is the temperature rise of contacting pad summits, or the flash temperature increment . Most of the heat in CMP is generated at the contact points between pad summits and the wafer or ring. Mechanically mediated material removal of chemical reaction products formed on the wafer also occurs at the contacts. Since the real contact area is often less than 0.1% of the wafer area , contacting pad summits are probably hotter than the wafer, and their temperature may, therefore, dominate the chemistry.
|Figure 4. Ratio of the flash temperature increment for 450mm to that for 300mm.|
Figure 4 shows the ratio of the flash temperature increase predicted for a 450mm tool to the prediction for a 300mm tool. Since summits spend more time in contact with the ring and wafer on a 450mm tool due to the lower platen rotation rate, the flash temperature increment for 450mm is expected to be 10-20% higher than for 300mm, with a 20% increase over the trailing one-third of the wafer. Theoretically, this could increase the chemical reaction rate by a potentially significant percentage, depending on the activation energy of the process, and significantly affect removal rate uniformity in the case of chemically limited CMP processes.
For most polishing processes, the pad surface must be continually renewed or conditioned to prevent the buildup of polishing debris and to counteract plastic deformation and abrasive wear, which can lead to a decrease in material removal rate. This is usually done with a diamond conditioning tool. If the conditioning procedure is directly transferred as the tool size increases, however, the rate at which the conditioner refreshes the pad surface will decrease in proportion to the pad area. Thus, the vertical pad cut rate for 450mm should be about 20% of the rate for 200mm if the conditioner load and sweep frequency are kept the same.
One solution—increasing the load on the conditioner by the area ratio—will decrease conditioner life by the same ratio. Furthermore, with a slower platen rotation rate and a fixed duration process, the conditioner will make fewer passes over each point on the pad, suggesting that the sweep frequency should also be increased to compensate. Deeper study of this issue is required to determine whether there is a viable solution.
Simulations suggest that if the polishing pressure, relative sliding speed, and mean slurry thickness are held constant when scaling to 450mm, then slurry consumption will increase moderately and the wafer and pad temperature will be essentially unchanged. Because it is necessary to decrease the platen rotation rate to keep the speed constant, it is predicted that there will be an increase in the flash temperature at contacting summits, which may affect uniformity in chemically sensitive processes. The most difficult problem, however, is how to condition a much larger pad within the fixed time allotted to each polishing step.
1. L. Kondic, SIAM Review, Vol. 45, No. 1, pp. 95-115 (2003).
2. Y. Sampurno et al., J. Electrochemical Soc., Vol. 152, No. 7, pp. G537-G541 (2005).
3. Y. Li (ed.), "Microelectronic Applications of Chemical Mechanical Planarization," John Wiley & Sons, 2008, Ch. 6.
4. L. Borucki, Proc. 2008 CMP-MIC.
Leonard Borucki received his PhD in mathematics from Rensselaer Polytechnic Institute, and is CTO at Araca, Inc., 3831 East Ivy Street, Mesa, AZ 85205 USA; 480-748-5105; firstname.lastname@example.org
Ara Philipossian received his PhD in chemical engineering from Tufts U., and is president and CEO of Araca Inc.
Michael Goldstein received his PhD in chemistry from Ben-Gurion U., and is a principal engineer at Intel Corp., Santa Clara, CA, USA.