NSF Sponsored Workshop on Structured Design Methods for MEMS

Thoughts for the MEMS Community

Carlo H. Sequin
CS Division
U.C. Berkeley

Here are some additional thoughts to accompany the paper: "VLSI Design and Fabrication" by Carlo H. Sequin and Sara McMains. These thoughts are aimed particularly at the community of MEMS designers and fabricators who will come to the Caltech workshop.

The Problem Domain

MEMS occupy an interesting niche between VLSI chips and miniature mechanical systems. Their structures can be far more complex in the vertical dimension than the VLSI circuitry used in computer or memory chips. No good widely accepted language currently exists to describe the 3D geometry of the finished structures. The complete set of all mask geometries together with the travel sheet that describes the process sequence, at best give an implicit description of the resulting product. From this implicit description, it is in principle possible to obtain an explicit description of the geometry of the finished chip if suitable modeling programs are available for all processes and a suitable geometric modeling language is used.

A traditional way to describe mask geometry in the field of VLSI is through the use of some mask geometry language such as CIF. Some mask description formats have been provided with extensions that allow to describe in more details the thicknesses and properties of the various layers in different regions. Most of these extended formats are still 2.5D description formats, rather than real 3D geometry formats. They are thus inappropriate to describe complicated overhangs or features that occur on vertical faces.

One could use some established solid modeling language (e.g., ACIS) to describe the geometry of the chip at various states during the sequence of process steps needed for manufacturing. However, commercial solid modeling languages have the wrong strengths. They provide compact high-level ways to define a desired solid shape, often focusing on constructive solids geometry (CSG) operations, but they have virtually no support for tracking a complicated manifold such as the exposed surface of a silicon dioxide layer as a function of time during the simulation of an etching process. What we need for the use with advanced IC chips and with MEMS is a descriptive geometry language that captures the state of a complicated physical structure consisting of several different regions with different materials properties, which can change as a function of time. The immediate needs are for primarily a descriptive language, rather than for a design language, since we are still far from the point where it will be possible to define an ideal shape or a complete mechanism, and then have it manufactured in a top-down manner by defining automatically the needed mask geometries and fabrication steps. For the foreseeable future, MEMS will have to be designed from the mask-set point of view based on a detailed manufacturing process plan. There will not be a single generic MEMS process! Every new application, and every new conceptual structure may need a new process sequence for its implementation.

Relations between CIF, SIF, L-SIF, and UniGrafix

Berkeley UniGrafix, is a simple geometric description language that we have been using in computer graphics research and instruction since the early 1980's. It was created as a 3D extension of CIF, the Caltech Interchange Format, for situations where a 2.5D layered description is no longer appropriate. It is almost a strict superset of the 2D CIF language, but some of the primitives specifically aimed at the needs of IC layout, such as "path", have been left out of the 3D language and have instead been relegated to the domain of a separate set of generator tools; e.g., the "UGworm"-tool will generate a properly mitered prismatic tube along a piece-wise linear path through 3D-space. Berkeley UniGrafix would be quite appropriate for the description of even the most complicated, static 3D MEMS geometries. It does not have adequate facilities, however, for describing shapes that change as a function of time.

We are now in the process of defining SIF, an interchange format for Solid Free-Form shape manufacturing (see article referenced above). Thus the question naturally arises: Could this also be the language of choice for the MEMS community ? The answer is: Probably not ! The needs of the two communities are too different. The SFF world needs a language to capture designs of 3D shapes. The MEMS world needs to capture descriptions of complex structures, which, however, may be time-varying. The L-CIF language is closer to the descriptive end of the spectrum, but it is again a layered language and thus does not overcome the key shortcomings of the extended 2.5D mask description formats that are used today.

In my opinion, the ideal MEMS description language would follow a strict boundary representation paradigm and be efficient to represent many extended complicated surfaces with many detailed features, probably in the form of a shared vertex array and a triangular or quadrilateral mesh referencing these vertices. It must have the provisions to attach reasonably complex volume specifications to any of the regions bounded by a combination of parts of these surfaces, e.g., a gradated doping in the z-direction. As in all interchange formats, it is absolutely vital to have a cleanly defined semantic meaning for all expressible constructs; syntactic choices are more a matter of taste.

Other CAD Tools

MEMS may have moving parts, e.g., electro-motor rotors that turn or levers that flip up after processing is completed. Representation of such operations will require the power to formulate 3D rigid body transforms. To be further able to do at least a kinematic simulation of such MEMS operations may require even more sophistication; since these motions are typically constrained in some way, a complete constraint solver system may be required.

An important CAD operation that is performed on a description of the expected geometry of a processed device is "extraction". This is the process of finding a higher-level description from a large -- possibly unstructured -- mass of lower-level primitives. In the IC domain, extraction is often used as a verification step after a lengthy multi-stage layout compilation to verify that the sum total of all layout rectangles indeed constitutes a circuit with the desired properties.

For a MEMS with an interesting 3D geometry such extraction is much harder than for the typical digital circuit chip. There may not be a rich enough set of predefined higher-level primitives in which the function of the MEMS structure can be expressed succinctly. The richness of the possible geometrical structures that can be formed is far too rich.

The extraction, simulation, and analysis of MEMS function are a challenging research domain that will not likely yield to some standardization in the near future. Efforts should thus best be spent towards achieving a lower goal, i.e., that of creating a simple geometry language extension that allows to adequately describe resulting or desired MEMS geometries. An extension of 2D CIF to a true 3D language in the spirit of Berkeley UniGrafix may be all that is required.

Return to NSF MEMS Design Workshop Index