Introduction

Standard digital data interchange representations (e.g., Caltech Intermediate Form (CIF)) have had a beneficial effect on the development of structured design methods for VLSI. It enabled the development of design tools to address issues at all levels of abstraction, and provided access to technology that was otherwise unavailable to a wide community of electronics designers. The resulting innovations ranged from powerful new digital design tools, to computer architectures that spawned new corporations. The simplified design tools in turn made VLSI design accessible to all university students interested in computer science and electrical engineering. The standard interface to manufacturing produced an economy of scale that could not be achieved by individual researchers or educators.

Some of the lessons that were learned in VLSI might be beneficially applied to the MEMS community. An examination of a typical design and manufacturing flow in mechanical systems reveals certain similarities in data interfaces between major modules where the VLSI approach might be emulated. Figure 7 (Mechanical Design Process Flowchart.) illustrates a typical mechanical system design process. However, the mixture of mechanical and electronic components and functions will require new research to develop the multi-domain, multi-dimensional representations required.

Five specific areas where research work on standard representations will benefit the development of MEMS are:

1. 3-D solid modeling representation;
2. Interfaces between design tools (e.g., between 3-D modeling, analysis, and process design);
3. Representations for mask-layout, process specification, and metrology;
4. Interfaces between design and fabrication (e.g., between process design tools and fabrication);
5. Representation for function ({e.g., the MEMS equivalent of digital VLSI's Boolean logic).

Findings and Recommendations

Common data representations which are widely used are critical to rapid innovation in MEMS.

• Finding: In MEMS it is important to describe the design in such a way as to maintain a "clean separation" between the design and fabrication processes in order to achieve the benefits of a structured design methodology. This requires that the designers describe the desired object in a well-defined language that is easy to learn and that is suited to hierarchical descriptions, while the fabrication community takes responsibility for conversion of that description to masks used to build the desired object. Standard processes, whose capabilities (design rules) are known in the design and fabrication communities, can facilitate this separation.
• Recommendation: Research efforts should be focused on design methodologies which support the "clean separation" of design and fabrication. That is, well understood standard processes with proven design rules should be supported with design hierarchies and design tools which encourage the designer to describe the layers of the final object resulting from the fabrication process rather than the mask set.
• Finding: 3-D Solid Modeling. Existing 3-D solid modelers currently used for macro-mechanical system design may not be adequate and appropriate for MEMS.
• Recommendation: Identify and adopt a standard for data representation of 3-D solid models for interchange between design tools. [3-years]
• Finding: Mask-layout. Two clear areas for extension of currently available mask-layout software are: an improved capability to define non-polygonal geometry; and a mechanism for interaction between the designer and the pattern generation system to produce an optimal match.
• Recommendation: Develop and adopt extensions to CIF or GDSII to accommodate non-polygonal geometries in order to provide a more efficient interface with pattern generation equipment. [3-years]
• Finding: Fabrication Process Language. In automated (CNC) mechanical fabrication, a standard language (G-Codes) is nearly universally used. In semiconductor manufacturing, MAP is commonly used. An extensible, hierarchical fabrication process specification is needed for MEMS (including flexibility in defining the materials and layers of the final product) to provide designers and fabricators with a common language for communicating manufacturing instructions.
• Recommendations:
  • Develop and adopt a standard for the specification of simple sequences of unit processes sufficient for simulation and fabrication of surface machining. [3-years]
  • Develop and adopt a standard for the specification of simple sequences of unit processes sufficient for simulation and fabrication for detail within unit processes. [5-years]
• Finding: Metrology. It is critical to assure that MEMS fabrication processes have been properly carried out for a given design. This will aid in debugging a new development by determining whether a failure is due to mis-fabrication, or a design error. Test structures that are implemented for this purpose will be highly sensitive to process variations and are easily measured at the end of the process. In addition, a second use of these test structures provide model parameters for designers who may reuse the same process steps.
• Recommendations:
  • Identify and document existing process monitors of use to the MEMS community and make available over the WWW. [1-year]
  • Develop verify and make available test structures, procedures and programs to quickly determine process quality, such as endpoint completion in MEMS processes. Provide a metric of technical progress in order to have a benchmark to measure improvement over time. [3-years]
  • Work with the industry to agree upon standards for test structures and test methodologies for MEMS processes. Identify opportunities for in-process monitors to more quickly evaluate the progress of MEMS process steps. [5-years]
  • Develop sufficient understanding of the methodology and materials to predict reliability of MEMS components and to provide tools for improving materials technology for maximum lifetime. [10-years]
• Finding: Language for Function. Digital VLSI utilizes Boolean logic to describe function. The existence of a formal language for function has many advantages, among them are the ability to represent and analyze function mathematically. Macro-mechanical systems, at present, has no single language for function, but instead utilizes many different representations for the many different energy domains. Currently MEMS also lacks a unified representation of function.
• Recommendation: Develop a representation for MEMS function, perhaps based on Bond-Graphs. This representation provides a framework for systems involving several energy domains. It not only provides a common representation for elements belonging to different energy domains, it also permits couplings among different domains.

Mask Layout Geometry vs. Desired 3-D Shape

In VLSI, currently, the mask describes the final desired (2-D) geometry. The fabricators take into account the details of their fabrication process and pre-distort the mask so that once fabricated, the desired shape emerges. This establishes the "clean separation" between design and fabrication.

In MEMS, currently, the required pre-distortion is geometrically complex (due to the 3-D nature of most MEMS) and generally unknown. Therefore MEMS mask layout geometry is used as the mask directly. The required pre-distortion (compensation, etc.) must be determined by the MEMS designer, generally over many prototype cycles and with much experimentation. Furthermore, this experimentation does not usually, at present, result in a generalized understanding that can be applied to different MEMS devices.

At issue is whether the MEMS designer can be (or should be) insulated from the details of the fabrication processes. If this is desirable and/or possible, then the geometry that is communicated to the fabricator should be the desired (3-D) geometry. Conversely, if the relationship between mask layout geometry and final shape is too complex, MEMS designer must have available accurate fabrication simulations so that the required pre-distortion of the mask can be determined with the smallest number of fabricated prototypes possible.

• Finding: It appears possible to maintain a clean separation between design and fabrication for some classes of MEMS devices. Establishing this clean separation, where possible, appears to be the most direct route to achieving the benefits of a structured design methodology for MEMS.
• Recommendation: To broaden these classes and encourage the practice outside these classes, research efforts should be focused on design methodologies which support the clean separation of design and fabrication. That is, well understood standard processes with proven design rules should be supported with design hierarchies and design tools which encourage the designer to describe the attributes of the final object resulting from the fabrication process rather than the mask set.

Closure

While the challenges in developing structured design methods for MEMS that preserve the "clean separation" are significant, the benefits of such methods will greatly enhance MEMS developments in those areas where the design and fabrication communities can make use of such methods.

Summary

Much about the historical development of digital VLSI suggests that a common interchange language (CIF) greatly facilitated the rapid advances. While MEMS developers commonly use CIF to describe mask-layout geometry, CIF lacks many elements needed for a complete data interchange language. The development of a common interchange language for MEMS is likely to play a pivotal role in accelerating the advances in MEMS.