Modeling and Simillation for MEMS 7.5 Modeling and Simulating MEMS, i.e., Systems with Micro-(or Nano-) Scale Feature Sizes, Mixed Digital (Discrete) and Analog (Continuous) Input, Output, and Signals, Two- and Three-Dimensional Phenomena, and Inclusion and Interaction of Multiple Domains and Technologies In preceding sections we briefly described the current state -of-the-art in modeling and simulation in both the digital and analog domains. While the digital tools are much more developed, in both the digital and analog domains there exist standard, well-characterized technologies, standard widely available tools, and stable educational and prototyping programs. In the much more complex realm of MEMS, this is not the case. Let us compare MEMS, point by point, with digital and analog circuits. Is there a small set of basic elements? The answer to this question is emphatically no. Various attempts have been made by researchers to develop a comprehensive basic set of building blocks, beginning with Petersen's identification of the fundamental component set consisting of beams, membranes, holes, grooves, and joints . Most of these efforts focus on adding mechanical and electromechanical elements. In the SUGAR system, for example, the basic elements are the beam and the electrostatic gap. In the Carnegie Mellon tool MEMSYN , which is supported by the NODAS simulator, basic elements include beams and gaps, as well as plate masses, anchors, and electrostatic comb drives (vertical and horizontal). For the MUMPS process there is the Consol-idated Micromechanical Element Library (CaMEL), which contains both a nonparameterized cell database and a library of parameterized elements [which can be accessed through a component "generator," but not directly by the user). CaMEL supports the creation of a limited set of com- ponents, including motors and resonators, in a fixed surface-micromachined technology. But the bottom line for MEMS is that no set of basic building blocks has yet been identified which can support all the designs, in many different energy domains and in a variety of technologies, which researchers are interested in building. Moreover, there is no consensus as to how to effectively limit design options so that such a fundamental set could be identified. In addition, the continuous nature of most MEMS behavior presents the same kinds of difficulties that are faced with analog elements. Development of higher level component libraries, however, is a fairly active field, with, for example, ANSYS, CFD, MEMCAD, Carnegie Mellon, and MemsPro all providing libraries of previously designed and tested components for systems developers to use. Most of these compo-nents are in the electromechanical domain. As mentioned above, a few VHDL-AMS models are also available, but these will not be of practical value until more robust and complete VHDL-AMS simulators are developed and more experimental results can be obtained to validate these models. Is there a small set of well-understood technologies? Again the answer must be no. Almost all digital and analog circuits are essentially two-dimensional, but, in the case of MEMS, many designs can be developed either in the "2.5-dimensional" technology known as micromachining or in the true three-dimensional technology known as bulk micromachining. Thus, before doing any modeling or simulation, the MEMS developer must first choose not only among very different fabrication techniques but also among actual processes. Both the Carnegie Mellon and Cronos tools, for example, are based on processes that are being developed in parallel with the tools. MOSIS does provide central access to technology in which all but the final steps of surface micromachining can be done, but no other centrally maintained processing is available to the community of MEMS researchers in general. For surface micromachining, the fact that the final processing steps are performed in individual research labs is problematic for producing repeatable experimental results. For bulk micromac hining examples, fabrication in small research labs rather than in a production environment is more the norm than the exception, so standardization for bulk processes is difficult to achieve. In addition, because much MEMS work is relatively low-volume, most processes are not well enough characterized for low-level modeling to be very effective. In such circumstances it is very difficult to have reliable process characterizations on which to build robust models. Is there a well-developed educational infrastructure and prototyping facilities? Again we must answer no. Introductory MEMS courses, especially, are much more likely to emphasize fabrication techniques than modeling and simulation. In  a set of teaching modules for a MEMS course emphasizing integrated design and simulation is described. However, this course requires the use of devices previously fabricated for validating design and simulation results, rather than expecting students to complete the entire design-simulate-test-fabricate sequence in one quarter or semester. In addition, well-established institutional practices make it difficult to provide the necessary support for multidisciplinary education which MEMS requires. Are encapsulation and abstraction widely employed? In the 1980s many researchers believed that multiple levels of abstraction were not useful for MEMS devices. Currently, however, the concept of intermediate-level "macromodels" has gained much support [57,70], and increasing emphasis is being placed on developing macromodels for MEMS components that will be a part of larger systems. In addition, there are several systems in development that are based on sets of more primitive components. But this method of development is not the norm, in large part because of the rich set of possibilities inherent in MEMS in general. In Figure 7.2(b) we have given a partial classification of MEMS corresponding to the classification for digital devices in Figure 7.2(a). At this point it is not clear what the optimum number of levels of abstraction for MEMS would be. In Figure 7.8 we have attempted to classify.some of the tools from Section 7.4 in terms of their ability to support various levels (since these are simulators, they all support the "behavioral" view. MEMCAD, which allows fabrication process simulation, also supports the "physical" view). Note that VHDL-AMS is the only tool, besides the general-purpose Mathematica and Matlab, that supports a high-level view of MEMS. Are there well-developed models, mature tools, and integrated development systems which are widely available? While such systems do not currently exist, it is predicted that some examples should become available within the next ten years . 7.6 A "Recipe" for Successful MEMS Simulation A useful set of guidelines for analog simulation can be found in . From this we can construct a set of guidelines for MEMS simulation. 1. Be sure you have access to the necessary domain-specific knowledge for all energy domains of interest before undertaking the project. 2. Never use a simulator unless you know the range of answers beforehand. 3. Never simulate more of the system than is necessary. 4. Always use the simplest model that will do the job. 5. Use the simulator exactly as you would do the experiment. 6. Use a specified procedure foi- exploring the design space. In most cases this means that you should change only one parameter at a time. 7. Understand the simulator you are using and all the options it makes available. 8. Use the correct multipliers for all quantities. 9. Use common sense. 10. Compare your results with experiments and make them available to the MEMS community. 11. Be sensitive to the possibility of microlevel phenomena, which may make your results invalid. The last point is particularly important. Many phenomena, which can be ignored at larger feature sizes, will need to be taken into account at the micro level. For example, at the micro scale, fluid flow can behave in dramatically different ways . Many other effects of scaling feature sizes down to the microlevel, including an analysis of why horizontal cantilever beam actuators are "better" than vertical cantilever beam actuators, are discussed in Chapter 9 of . Chapters 4 and 5 of  also provide important information for low-level modeling and simulation. 7.7 Conclusion: Continuing Progress in MEMS Modeling and Simulation In the past fifteen years, much progress has been made in providing MEMS designers with simulators and other tools which will give them the ability to make MEMS as useful and ubiquitous as was predicted in . While there is still much to be done, the future is bright for this flexible and powerful technology. One of the main challenges remaining for modeling and simulation is to complete the design and development of a high-level MEMS description language, along with supporting models and simulators, both to speed prototyping and to provide a common user-friendly language for designers. One candidate for such a language is VHDL-AMS. In , the strengths and weaknesses of VHDL-AMS as a tool for MEMS development are discussed. Strengths include the ability to handle both discrete and continuous behavior, smooth transitions between levels of abstraction, the ability to handle both conservative and nonconservative systems simultaneously, and the ability to import code from other languages. Major drawbacks include the inability to do symbolic computation, the limitation to ordinary differential equations, lack of support for frequency domain simulations, and inability to do automatic unit conversions. It remains to be seen whether VHDL-AMS will eventually be extended to make it more suitable to support the MEMS domain. But it is highly likely that VHDL-AMS or some similar language will eventually come to be widely used and appreciated in the MEMS community.