Optimizing processing equipment using virtual prototyping
09/01/2003
In the past, issues related to fluid flow distribution, heat transfer, and chemical reactions in semiconductor-manufacturing equipment were addressed by trial-and-error methods whose cost and lead time made it very difficult to optimize equipment and design new processes. Recently, the combination of flow-modeling software and powerful desktop computing has provided a means to quickly and inexpensively evaluate the performance of a wide range of design alternatives while minimizing the expense of building prototypes.
Flow modeling
The model begins with a geometric representation of the equipment to be analyzed. A three-dimensional (3-D) solid model may be available from an existing CAD database or it can be built using tools provided with the flow-modeling software. The next step is mesh generation, in which the 3-D model is subdivided into many small "control volumes" or elements, similar to the mesh created for traditional finite element analysis.
It is within each control volume that the flow velocity, pressure, temperature, etc., are computed. Mesh creation tools are also provided with the flow-modeling software. Next, the problem definition is completed by inputting the material properties and operating conditions. Now the analysis problem is fully described and the solution can be calculated. Finally, the results are post-processed or visualized. Post-processing can include examining qualitative features such as large-scale flow patterns or temperature distributions, as well as obtaining specific quantitative values, such as the photoresist film thickness variation across the wafer.
 Figure 1. CFD analysis of a spin-coating station: a) 3-D solid model of spin coater; and b) flow path lines traced from the inlet reveal the complex air-flow patterns. |
null
Rapid thermal processing
In a model of an RTP chamber where the radiant energy source is a plasma arc tube, the radiation is collected and focused by a highly reflective and contoured surface. The radiation passes through a transparent quartz window and is absorbed by the underside of the wafer surface and the internal walls of the chamber. Uniform heating of the wafer by the radiant energy is critical to prevent thermal stresses and warping, and to obtain uniform deposition/etch rates. The shape of the reflector surface must be carefully designed to achieve this goal. The exact shape of the reflector is incorporated into the CFD model and the resulting radiant energy distribution is calculated. If the uniformity is found to be unacceptable, the shape of the reflector can be adjusted in the model until it provides satisfactory results.
Another influence on the radiant energy distribution is the quartz window. The orientation, thickness, location, and material properties of the window cause the radiation to be distorted as it passes through the window. This refraction
eflection is calculated by the CFD code using Snell's Law — the net influence will be seen on the wafer irradiation uniformity.
null
Photoresist application
The airflow distribution in a spin coater has a critical effect on film thickness uniformity because it strongly influences the drying rate. Additionally, a large number of resist particles are generated during the spin-off stage. If they are not immediately removed by the exhaust, they can be deposited on the walls of the chamber or on the wafer. There is no practical experimental method for accurately and completely determining flow patterns within a spin coater. Flow-modeling technology has advanced to the point where it can be used to solve this design problem.
When the wafer spins, it draws air from both the top opening and the bottom gap, and forces it outward radially (Fig. 1). This results in a positive pressure buildup underneath the baffle that contributes to a flow blockage. It also explains why there are no recirculation zones underneath the baffle, as was widely presumed. CFD results provide a much clearer understanding of how the flow pattern in the spin bowl is affected by complex interactions between the spinning wafer, cup geometry, exhaust suction, and the gap between the spin head and the surrounding baffle. The new design improved the photoresist uniformity and minimized the particulate contamination of the wafer surface.
The latest work in this area goes one step further by modeling the movement of the photoresist liquid over the surface of the wafer. When comparing the spreading rate of a coating droplet as predicted by flow modeling with experimental results, very good agreement is shown.
Etching
Because electrons and ions react on tremendously different time scales, modeling their interaction with the flow of the neutral species is especially challenging. Recent efforts have focused on coupling transient plasma chemistry models with steady-state fluid flow simulations that include a reduced-order description of the plasma chemistry.
Figure 2a shows the electron density and energy distributions in an inductively coupled plasma system. The plasma is centered on the axis of the reactor as indicated by the electron density, and the average electron temperature is >4eV. The electron temperature distribution is uniform inside the reactor due to low operating pressure. Figure 2b shows the overall chamber clean rate in a P5000 reactor using fluorocarbon discharges. The simulation results indicate that at high pressures, a more spatially confined plasma is generated for cleaning the electrode region, which has the heaviest buildup of deposits. At low pressures, the plasma is more spread out, allowing it to clean the more remote chamber surfaces. The simulation results predicted the trends correctly, and shed light on the change in plasma constituents with respect to process parameters.
Conclusion
Using flow modeling, it is possible to evaluate more design alternatives, which can lead to substantial improvements in equipment performance. At the same time, the lower cost and shorter lead times achieved using flow simulation result in faster time-to-market and reduced development costs.
Balaji Devulapalli, Christopher LaRivere, Eric Grald, Fluent Inc., Lebanon, New Hampshire
Balaji Devulapalli is a senior engineer at Fluent Inc., 10 Cavendish Court, Lebanon, NH 03766; ph 603/643-2600, fax 603/643-3967, e-mail [email protected].