774 Proceedings of the 26 th Chinese Control Conference July 26-31, 2007, Zhangjiajie, Hunan, China On MEMS Design Automation * Zhao Xin, Sun Guangyi 1 , Ren Liang, Lu Guizhang Inst. of Robotics and Automatic Information System, Tianjin 300071, P. R. China 1. E-mail: sungy@robot.nankai.edu.cn Abstract: General approach to the design of MEMS process flow and mask layout relies on purely experience and prior knowledge of the similar devices. It is a quite challenging and hard task, since a variety of professional knowledge and itera- tive attempts are required. This paper puts forward a novel approach based on feedback-based expert system (FBES) to the auto-design of process flow and layout according to the abstract geometry description of MEMS device. FBES involves feed- back verification facility on the basis of conventional expert system, and has shown significant superiority on applicability, flexibility, and expansibility in comparison with the latter, especially for MEMS process flow auto-design. This paper de- scribes the representation of 3D geometry of device, framework of FBES, rules of MEMS process matching, and an example of layout autogeneration and process matching system. Key Words: MEMS design automation, feedback-based expert system (FBES), layout autogeneration, process matching 1 INTRODUCTI ON 1 Micro-Electro-Mechanical Systems (MEMS) are fabri- cated through various micromachining techniques, many of which build on IC processing technologies to produce micron-scale mechanical and electro-mechanical de- vices. Technologies for fabricating a variety of MEMS devices have developed rapidly [1] , so design tools that allow engineers to quickly design and optimize these micromachines are critical to the future and the growth of the MEMS market. Moreover, engineer must ensure the desired device can be actually fabricated through available micromachining techniques after system-level and physical-level simulations, so the design of process flow and layout is especially important [2] . General approach to the design of process flow and cor- responding layout relies on purely MEMS experts’ ex- perience and prior knowledge of the similar devices. It is a quite hard and challenging task, since the desired structures are geometrically complicated, electrome- chanically coupled, and inherently three dimensional. The fabrication techniques for MEMS involve a large number of processes such as etching, deposition, sacrifi- cial etching, or wafer bonding. Additionally, multi-domain physics and chemic for MEMS and complex material behavior need to be considered to correctly predict de- vice performance. Thus, a design tool which can assist engineer by de- signing process flow and layout is critically needed. This tool should include autogeneration of process flow and layout as well as autotest of fabrication conditions and material behaviors. In response to these, this paper firstly puts forward the concept of auto-design of proc- ess flow and layout, and then describes the implementa- tion of MEMS layout autogeneration and process matching system. The objective is providing a powerful CAD tool for MEMS designer and exploring the auto- mation of MEMS design. In this project, there are three key techniques: represen- * This work is supported by National Nature Science Foundation under Grant 60674068 and Chinese National 863 Plan under Grant 2006AA04Z304. tation of desired geometry of device, setup of repository, and implementation of inference engine in FBES. Rep- resentation of geometry consists of high level (abstract) description and low level solid model. Both of them are layer-based structure. Abstract description is used as the system input, and the later process inference is per- formed on this level. Solid model is only used for the visualization of input geometry. Repository is the basis of decision-making in FBES. Knowledge in repository is represented in rule-based format "if <condition> then <action>." Inference engine is the decision-making part in FBES. It takes charge of designing layout and process flow based on abstract geometry description and rules in repository. In our previous work [3,4] , we had successfully imple- mented Virtual MEMS Fabrication Process (Virtual Process) prototype system. It was a transfer process from process flow and 2D layout to 3D geometry of MEMS device. The system introduced in this paper can be intuitively regarded as the converse of Virtual Proc- ess, and representation of geometry used here also comes from Virtual Process. In the following section, we introduce the logical archi- tecture of layout autogeneration and process matching system. In Section 3, three key techniques are detailed. Section 4 presents a prototype system and an example of process flow and layout auto-design for a device fabri- cated through surface micromachining techniques, which verifies the feasibility and efficiency of current system. 2 LOGICAL ARCHITECTURE The system discussed in this paper performs inference based on insufficient information. The input to system is abstract geometric description for device, and the out- puts are corresponding process flow and layouts used to build actual device. It is obviously difficult to perform such conversion using traditional mathematical deduc- tion due to its inherent characteristics:  Difficult to get algebra solution.  Need to perform inference based on inaccurate and