We believe that a software design system can now be envisioned for MEMS that will be able to take a set of desired performance requirements as input and accomplish the following tasks: choice of the optimal technology; selection and optimization of mechanical and electronic components from which the system will be composed; and determination of appropriate couplings among the components, including the degree of monolithic integration and packaging. The output will be a set of mask descriptions and an optimized process flow. Such a software system will require structured design methods, although these will, by the nature of the design problem, be largely distinct from the methods used in VLSI design.
The study of microelectromechanical systems (MEMS) is a rapidly growing area of research with a large potential to accomplish useful tasks in numerous applications, such as the following: microsensors, microactuators, micro-accelerometers, microphones, cellular phones, and microelectromechanical filters. MEMS elements that have appeared to-date include rotary motors, linear motors and resonators, springs, gears, grippers, diaphragms, and arrays of mirrors for display technology. All of these elements and systems have mechanical structures on a size scale of a few to a few hundred microns. MEMS is a quintessential interdisciplinary field of electronics, bringing together studies in mechanical engineering, electrical engineering, electronics, fluid mechanics, optics, chemistry, and chemical engineering, with application areas across the entire spectrum of national and commercial enterprises. Partly because of its infancy, and partly because it involves such a large number of disciplines, there is not yet a developed science of design for MEMS. Teams of interdisciplinary researchers are needed with a common interest in establishing the required science and engineering for MEMS design, which we see as consisting of design synthesis and process planning. An important goal is the establishment of a set of methodologies for the design of MEMS that starts from a specification of desired function and leads to an optimized fabrication of a MEMS system.
The system we envision will be able to choose the optimal technology, select and optimize mechanical and electrical components from which the system will be composed, and determine appropriate couplings among the components, including the degree of monolithic integration and packaging. The input to the design system will be a set of desired performance specifications, and the output will be a set of mask descriptions and a semiconductor-process flow. To attain the goal of a function-driven MEMS design system, there will of necessity be a need to develop parallel codes.
A library of parameterized MEMS elements and functional building blocks will need to be established, including the interesting case of compliant mechanisms.
One existing approach to the construction of elements of a library of MEMS devices is that taken by CAEMEMS (Computer-Aided Engineering of MEMS). CAEMEMS is a framework for the design of MEMS that provides a high-level, Motif-based graphical user interface for the specification of specific instances of parameterized MEMS devices, generation of inputs to a finite- element analysis (FEA) system, launching of FEA runs on a serial workstation, retrieval of FEA results, stripping of results and storage in the database internal to CAEMEMS, capabilities for single-run analysis, multiple-run analysis, and sensitivity analysis, as well as plotting of families of line graphs of use for the designer.
The library of MEMS elements and blocks may take various forms, ranging from simple, e.g., the geometric and materials specifications of a simple beam element, to complex, e.g., a tensor spline model of a parameterized electrostatic motor, including possible couplings to other functional blocks. It may develop that the library will consist mainly of software that can construct models of elements and blocks, rather than simply consisting of the models of the elements themselves.
Since the elements of the traditional mechanical repertoire consisting of rigid links and joints, fail to meet atypical requirements of micro-regime such as (i) eliminating the need for assembly, (ii) restricting the entire machine systems to just one or two layers in a plane, (iii) alleviating the adverse effects of friction, and (iv) accommodating unconventional actuation techniques including thermal, piezo-electric, etc. Fully compliant mechanisms, a certain kind of generalized flexible structures, readily meet all of these requirements.
Compliant mechanisms are single-piece flexible structures that deliver the desired motion by undergoing elastic deformation. This is in contrast to the rigid-body motions of conventional mechanisms. It will be important to develop mathematical formulations for optimal design of compliant micromechanisms. One of the computational procedures used in solving the synthesis problems is based on the homogenization method. This method, based on cellular microstructure, has the ability to generate any topology, shape, and size that are optimal for given problem specifications, which are applied forces, desired output displacements, and the amount of material to be distributed in a prescribed design domain. The output of this method is a density image in gray scale that indicates the optimal material distribution within the design domain. The homogenized image is directly transformed into manufacturable form. Structured methods are needed for the application of the homogenization method in the MEMS domain.
An important part of the proposed research involves the determination of process sequences for the fabrication of desired MEMS devices. This activity will be led by Prof. Carlos Mastrangelo. The design of fabrication processes for micromechanical and microelectronic devices requires a working knowledge of semiconductor thin film processing, along with a knowledge of materials and processes and a great deal of ingenuity. Some commercial VLSI processes require as many as 500-700 process steps, and a comparable degree of fabrication complexity occurs in the fabrication of three-dimensional micromechanical structures using planar processes.
In the micromechanics area, programs such as MEMCAD and IntelliFab give accurate representations of the finished micromechanical devices. These simulation tools take a description of the fabrication process flow and a mask set of a device as inputs and generate simulated profiles of the finished device as outputs. These design tools are undoubtedly useful aids to the designer that allow him to correct potentially expensive mistakes before fabrication begins. Nevertheless, these tools are aimed at design verification, and they require the input of, or assume, a known fabrication sequence, typically developed by experienced designers, hence the process flow is very much subject to their ingenuity and knowledge background. A more useful tool for the future a geometrical description of the device as input and provides, as output, a fabrication process flow. Such a program will speed up the development of microdevices by removing the knowledge background requirement from the human designer. The program will have the added advantage that it can be coupled to existing simulation tools to provide accurate information about the finished device in a true rapid concept-to-manufacturing design-synthesis fashion. Such programs must select the microdevice materials and fabrication sequence, as well as determine the feasibility of the decomposition of the given structure into layers. Furthermore, such programs will be useful in the design of robust fabrication processes, estimation of yields and manufacturing variations, as well as automated design centering. It is possible that such tools may create fabrication sequences that could not be conceived by the limited scope of human designers.
Of necessity, extensive use will be made of parallel computation, and many aspects of the parallelization of relevant software for MEMS design will be important.
The development of a structured design methodology for MEMS, taken in the broad sense presented in this position paper, will require and enable an important set of areas of scientific research; including the design of computational experiments for response-surface generation over extended, possibly highly non-linear domains; methods for efficient and effective man/machine interaction; and algorithms and communication strategies involving highly parallel computation. One particular area that we wish to bring out here is the need for a unified theory of transduction that is coupled to design, and we close with two paragraphs on this important emerging area.
A unified theory of transduction has begun to emerge from the research of Middlehoek, Ylilammi, van Duyn, Kirschner, and others. This theory identifies a set of signal domains (e.g., electrical, radiant, mechanical, thermal, chemical, and magnetic) each of which is characterized by its own set of variables, equilibrium constitutive equations, and non-equilibrium constitutive equation (e.g., the relation between flux of charge carriers and the gradient of electric potential is Ohm's law, which characterizes an energy dissipative process). There is unity in the theory because, under an assumption of local equilibrium, there exist relations among the variables across the signal domains, such as Gibbs relations, equations of state, and a balance equation for entropy.
The unified, irreversible thermodynamic theory of transduction can be generalized from a design perspective to enable the initial abstraction of a desired transduction function to its generic thermodynamic basis. From this abstraction various methods can be applied to enumerate and select a device or system realization of the transduction function. Implementation of the function, that is, the specification of the exact process steps and masks required, is the final step in this virtual prototyping procedure. We believe that software can now be developed to implement each step.