Process simulation is the conversion of two dimensional mask layout geometry and process information to a three dimensional representation of an output shape for the purposes of visualization and FEM and other CAD tools. MEMS research currently involves a large number of different processes and structures, and there is a need for accurate and flexible process simulators. State of the art MEMS design involves several design cycle iterations; the incorporation of process simulators in rapid design for MEMS would reduce the number and cost of design iterations. Even more than VLSI, the growth of MEMS will be dependent on accurate process simulations.
While VLSI process simulation tools are available, their generality and applicability to MEMS process simulation has not been fully explored. Existing VLSI tools have not yet been fully integrated in MEMS CAD systems. Existing tools can be divided into three broad groups: inexpensive basic tools, high end commercial systems, and university research/experimental systems. A study of existing process simulators should be undertaken to evaluate their applicability to MEMS processes. These should be extended to incorporate second- and third-order effects (mask undercutting, etc.).
Both VLSI and traditional macro-mechanical design have an established knowledge base of acceptable engineering approximations for different engineering domains, i.e., rules of thumb. How detailed spatially and temporally, must a MEMS process simulation be? MEMS design tools need to be developed that allow the designer to easily specify the level of approximation of simulation for their designs. Note that different structures within a design and different energy domains (e.g., thermal vs. mechanical) may have different levels of detail. Additionally, design techniques need to be developed in the MEMS community that guide the user in selecting the appropriate level of detail. These techniques may be application dependent. Basic design tools are needed in the immediate to short-term future, with greater flexibility to follow on a longer time scale.
The complexity and cost of process simulations can vary widely. What distinguishes them is the amount of separation between designer and fabrication process. Novice designers may have little need or desire to know the details of the fabrication process, while experts may need extensive knowledge. The type of simulator a user requires depends on whether the designer is concerned with standardized systems or custom systems. Thus there may be two broad categories in the need for process simulators: an inexpensive basic simulator which operates on PC's, and a high-end complete simulation package which is computationally intensive.
Figure 8 (Simulation of Fabrication.). summarizes one potential approach. The expert designer who is doing custom or experimental design must be aware of the details of the fabrication. Thus he would need an expert process simulator. The process definition file (PDF) would then include a large number of process parameters. A novice designer would prefer a cleaner separation between design and fabrication, and would use a simpler designer process simulator, which would shield him from most process details. The associated PDF would be much simpler, perhaps generated by an expert designer to give a reasonable, simpler approximation to the process. In both cases there is a trade-off between process detail and clean separation. Different users will have different needs and it is important that the two tracks be maintained. The results of the process simulators are evaluated using 3-dimensional visualization tools and/or FEA tools. The designer may have to run the process simulator, examine the results, and modify his design a number of times in order to achieve an acceptable solution. If no acceptable solution is found by standardized systems, a custom or expert process may be needed.
The process simulator constructs the 3-dimensional representation of a shape using a number of inputs. These inputs include the mask data and the PDF (process definition file) which contains a sequence of process steps that represent the fabrication process. The PDF must reference both layer information contained in the mask file and simulation code modules in the process simulator. Note that the PDF's for expert/custom process designers will have a much higher degree of detail and information than will PDF's for standardized process designers. The high level of detail in the expert/custom simulators will be used to design the process itself, and may contain proprietary information. Standardized PDF's will be an approximation designed to capture the essential result of the process, while separating the designer from the fabrication. Standardized PDF's may be created by the fabricator from a detailed knowledge of the process, or by the designer based on test fabrication and measurement. PDF's should be extensible since new processes will continue to be developed and a process which is custom today may become standardized next year. Specific investigations need to be carried out on the format of the PDF as well as methods to ensure a smooth interface between simulator modules and PDF's.
Compared to VLSI, MEMS has a greater need for 3-dimensional partial etch time snapshots and animation. This greater need is due in part to 3-dimensional etching (lateral/underetch and vertical etching). While there are many different MEMS fabrication techniques (surface, bulk, LIGA, deposition) their visualizations needs should be attainable by one set of tools. Some MEMS shapes change non-linearly in time, that is the interesting shape changes can occur in short periods of time: e.g., compensating structures, intersecting shapes. Time evolution tools need to be developed and distributed to the MEMS community which will help answer the questions: When is the shape defined? What time portion is most critical to the definition of the shape?
No process is ideal and variation or noise will always exist. For example the etching process may depend critically on such factors as temperature, pressure, concentration and timing, all of which may have slight variations. Who has the burden of keeping track of process noise? The answer depends on the application: novice designers with standard processes will want to be shielded, while expert designers will want the full capability to simulate process variation. In VLSI, the design rules are conservative, so the user has little noise burden. MEMS may be closer to analog electronics where some burden remains with designers. What types of statistical tools are needed to model process variations? How is robust design to be incorporated into the process simulation, i.e., how do we develop designs that are insensitive to process variations? For example in micro-electronics, an analog differential pair is designed so that the pair is relatively insensitive to process variations. How do we design "self-correcting" MEMS devices or "mechanical differential pairs", that is mechanical devices that are insensitive to process variations. VLSI designers are trained to design robustly, MEMS needs the same design experience and techniques and such approaches should be explored.
In order to evaluate the suitability of process simulators, it is necessary to measure the difference between the actual fabricated shape and the predicted shape. This will give a measure or metric of the fidelity of the simulator. What are the performance criteria, by which process simulators are judged? Standardized benchmarks or test suites should be developed that provide a means of evaluating simulator performance. Process simulators would then include a summary of benchmark results, these would be a record of past performance, not a guarantee or warranty of future performance. This is so because the performance criteria for simulators and the suitability of simulators will be application dependent. Thus the benchmarks should cover a wide range of application so that users may select the benchmark which best reflects their application. A set of benchmarks for MEMS process simulators should be developed.
There is a continuum of MEMS designers, which will require a range of process simulation tools. The two extremes of process simulations types are the library-based designs utilizing standardized foundry processes and the custom processes that involve the concurrent development of both the design and the process. We believe that a clean separation between design and fabrication can be attained in the first case, while clean separation is neither desirable nor attainable in the second case. The standardized processes lead to time and cost effective designs by taking advantage of economies of scale. The custom processes develop new processes, and custom processes of today are the standardized processes of tomorrow.
We recommend that the development of process simulation tools to support both tracks of designers be pursued. We feel that these tools will be an enabling technology in the realization of the full potential of MEMS.