SUGAR – A Computer Aided Design Tool for MicroElectroMechanical Systems
Web Page: bsac.berkeley.edu/cadtools/sugar/sugar/
Participating UC Berkeley Faculty:
James Demmel (CS/Math – contact – demmel@cs.berkeley.edu
),
Alice Agogino (ME – aagogino@me.berkeley.edu
)
Sanjay Govindjee (CEE – sanjay@ce.berkeley.edu
)
Kris Pister (EECS – pister@eecs.berkeley.edu
)
Participating UC Davis Faculty:
Zhaojun Bai (CS - bai@cs.ucdavis.edu
)
General Description of CAD Tools for MEMS on High Speed Networks. MEMS form the technological core of the Sensor Web application, as well as a 5 to 10 billion dollar and rapidly growing commercial industry. The continued rapid development of new devices and their applications depends on having adequate CAD tools for the design, simulation, measurement, and evaluation of MEMS devices. Current practice (back of the envelope calculations, or very detailed finite element models of small parts of large systems) is far from adequate for current and future needs. There is no robust and widely available system as there is in the integrated circuit world with SPICE..
SUGAR is our system level solution to this problem. Our goal is to close the design loop by enabling design, simulation, fabrication, comparison of measurement with simulation or other data, diagnostics, and then re-design. We are also developing design synthesis tools built on SUGAR to design MEMS devices with optimized configurations, designed to meet one or more performance objectives. All our software will be freely available.
Measurements will be done by a variety of devices at Berkeley and eventually other sites, all connected by the Internet, and supporting a variety of outside users. The measurement devices are capable of producing nanometer resolution real-time 3D images of operating
MEMS devices, as well as simpler measurements. The model is that a user would be able to use all the facilities (simulation, measurement, and data repositories) remotely at high speed. Speed is important because of the enormous measurement files produced and the ability to control and observe the devices being measured in real-time.
Improved Simulation Algorithms. (1) We will incorporate new and improved physical device models (like nonlinear beams and plates) for simulation, as well as for new physical phenomena (like contact). (2) We will improve efficiency via better numerical methods. First, we will accommodate multiphysics simulations more effectively, which cause conventional numerical methods to have scaling problems and for transient analysis to take overly small time steps. Second, we will better exploit sparsity and parallelism, in particular making use of the Millennium cluster on which the web-service version of SUGAR is currently running. (3) We will incorporate reduced order models to accelerate large simulations (reduced order models are automated ways of replacing a system with many degrees of freedom with a much smaller one while retaining overall system behavior). (4) We will incorporate sensitivity analysis to understand performance impacts of variations in system parameters, like stiffnesses, geometries, and so on. (5)
We will build models for design synthesis, exploiting our fast simulation capability. For example we will use genetic algorithms to search for optimal physical configurations that -have a given stiffness while minimizing size or another property. (6) We will build a library of test cases (devices with well developed simulations and measurement data) to explore the limits of the software and measure progress.
Coupling Measurement to Simulation. (1) We will build interface software to access the data collected by the 3 available measurement devices: Prof. Muller’s stroboscopic microscopic interferometer system (SMIS), the MIT Micro-Computer-Vision System, and the Polytec Laser Doppler Vibrometer. We will support access both live from the device and from data repositories. (2) We will build software to process the data from these systems to permit extraction of 3D information from 2D images, and to align and calibrate measured data with CAD models (necessary to compare measured and simulated data). We will start with easy problems (frequency response at a given point) and move on to full 3D transient response. (3) We will develop software and algorithms to compare and visualize the differences between simulated and measured data. We will coordinate with other investigators, since all this software will be based on shared interfaces. (4) We will develop algorithms to extract physical parameters (geometry, stiffnesses, etc.) from measured data, and feed these back into simulations to attain higher accuracy. (5) We will use the measured data to perform sensitivity analysis, such as by fabricating and measuring an array of test structures with varying parameters. (6) We will develop algorithms to support design synthesis, by building and measuring a spectrum of device designs and interpolating between them or extrapolate beyond them.
Design Synthesis for MEMS.
(1) We will identify and define common MEMS building blocks and assemblies of building blocks for use in automated design synthesis tools built on SUGAR. (2) We will design a framework for encoding MEMS layout, fabrication parameters and design constraints into a genotype for use by Genetic synthesis algorithms. (3) We will develop an indexed case base library of pre-existing MEMS components and sub-components that can be used as initial designs by a synthesis algorithm. (4) We investigate the issue of improving synthesis performance by including manufacturability (how suitable is a design to be manufactured with the expected performance) and suitability for simulation as objectives used in ranking the suitability of the design.
Making SUGAR widely available (1) We will make all our software freely available. (2)
We will make SUGAR available as a web service on the Berkeley Millennium, a multihundred processor cluster where jobs will migrate to least loaded machines. (3) In addition to simulating and measuring devices from our own local users, we will identify and support outside users to make sure that the system supports their needs. (4) Since not all users will have high-speed access, we will assess the usability of the system depending on users needs and their connectivity. (5) We will support situations where the user, measurement tool, data repository and simulator can all be in different places with different bandwidth networking connecting them. We will start with simple user scenarios where everything is co-located, and then target remote users.
Desired Support: Funding for grad student participants.