Structured Custom Design of MEMS Devices

A top down design is essential to minimize time-to-market of MEMS devices. A structured custom design approach for MEMS design based on the concept of using parameterized behavioral models as a means to improve and speed up the design process is presented in this paper. This design enables the user to quickly explore a larger design space on the interaction of the behavioral models with their surrounding system. Once the simulation is completed to the required device performance specifications, a device layout can be output for PDE verification on critical areas or to generate the masks for fabrication. INTRODUCTION The advent of Micro Electro Mechanical Systems (MEMS) technology has given rise to a need for a unique design environment. The design needs encompass multi-disciplines and multi-physical modeling capabilities. For example, a design engineer designing a MEMS based piezo-electric or capacitative pressure sensor would need to model solid deformation of a membrane and electromechanical or piezo-electric behavior. The design exploration would need to optimize the location of the sensing elements with respect to device geometry, which yields maximum sensitivity for the range of pressures required of the device. Another example is MEMS based mirror actuated electrostatically. In this case the modeling involves electro-mechanical behavior, gaussian optics and solid deformations. Finally, the design has to be consistent with the manufacturability. Manufacturability of a MEMS device typically involves mask creation, process descriptions involving selective deposition and photo-chemical machining / etching (lithography) by layers, typical of the integrated circuit (IC) manufacturing. The design iterations involve optimization of behavioral parameters for given performance constraints. Since the adoption of the MEMS technology about twenty years ago – the design process has undergone various generations of evolution as an approach to device design and manufacture. The Computer Aided Design (CAD) tools used for MEMS design have evolved in response to market drivers for this technology, which have emphasized a reduced time to market. The authors would prefer to categorize them as “Generation Zero”, “Generation One”, “Generation Two” and “Generation Three”, with the last one as a novel approach to this complicated issue. The categorization is based on the complexity and sophistication of the design process for MEMS applications as it evolved from its inception to its current state today. “Generation Zero” was the initial approach in the early days of MEMS design. The investments in CAD tools were minimal and restricted to mask design. This required multiple manufacturing iterations to tweak a device to its required performance – resulting in wasted time and resources. “Generation One” approach typically utilized point FEM (Finite Element Method) or CFD (Computational Fluid Dynamics) tools specifically for the purpose of studying device behavior in a single physical domain (electrical, mechanical, fluidic, chemical, optical, etc.). The tools were not specifically developed for MEMS applications and are used for exploring “macro-scale” physical phenomena. Hence, the tools provided very little information for exploring design space. Also the tools are very specialized requiring tremendous expertise in analysis techniques in order to get meaningful answers. In the past five years a host of CAD tools have been available in the industry specifically geared towards MEMS design (e.g., Coventorware, Memscap, Intellisense). The authors would like to classify these tools as the “Generation Two” CAD tools. These tools provide an integrated design flow starting from mask layouts in association with process descriptions that yield 3-D models of device. The CAD tools enable seamless integration of the manufacturing analyses with the device modeling and simulation tools to arrive at required performance constraints. Hence, the “Generation Two” CAD tools provide integrated design solutions that are linked and MEMS aware. In these tools it is relatively easy to move from 2-D mask layouts to 3-D solid models to FEM (or some multi-physics) analyses. With these design tools it is possible to explore the design space, albeit very tediously – which is very time consuming and often not very efficient. However, for these design environments solver expertise is still required of the designer in terms of detailed understanding of the numerical strategies and concepts underlying the solver tools. The benefit of this approach however is that the•FEM analysis (or other numerical solvers) can be used to extract behavioral models of the device which can be used in the design of MEMS enabled sub-systems. The extracted behavioral models – or what is typically called “macro-models” can be exported to a circuit simulator environment (e.g., Spice or Saber). In these circuit simulator environments the macro-models of the various subcomponents can be integrated together to study system level behavior of the whole device (see Figure 1). This type of design approach is categorized as “bottoms-up” approach or what is typically referred to as “full custom design”. This paper will explain a paradigm shift for MEMS design that the authors would categorize as the “Generation Three” or “Structured Custom Design”. This design process essentially relies on the “Generation Two” tools but provide MEMS device designer with tools to quickly explore large parameter space. This enables MEMS sub-system designer to explore design tradeoffs very early in the design iteration. In a way this obviates specialized knowledge set required of a designer and brings MEMS device design “to the masses”. As a result parameterized design becomes a possibility. Hence, FEM expertise (or specific solver expertise) is no longer required just to get started. Also, simulation runtimes are typically 10-100X faster than FEM analysis. Finally, design is complemented by FEM analysis. After design space is explored, FEM used for detailed analysis. In this paper, “full custom design” approach will be compared to the “structured custom design”. Two specific MEMS application examples will be used to demonstrate the concepts. STRUCTURED CUSTOM DESIGN Structured custom design is the process of designing a MEMS device (or MEMS sub-system) using a library of behavioral model building blocks. These building blocks are combined to produce the desired device behavior. In the structured custom design approach, the system parameters are specified in the integrated high-level system to obtain the required MEMS subsystem performance. This initial set of MEMS specifications is used to select a design and fabrication approach. Choices made at this point could include digital vs. analog control, surface vs. bulk micromachining, actuation method (thermal, electrostatic, magnetic, fluidic, etc.), and range of motion required. An initial system model of the MEMS component is then built from parametric primitives in a schematic. The system simulations are performed using a netlist of the component descriptions. A netlist typically consists of network description of the through variables (like current or flow) and cross variables (like potential or pressure drop) of device subsystem models. Once the system simulations are finalized through rapid iteration of device parameters (which could be geometric parameters, or physical parameters) a device layout can be automatically generated from this high level description in the context of the chosen process. For simple devices – the layout information is sufficient for taking it to fabrication steps. The more likely option would be to use the layout and process information to generate detailed 3-D PDE (Partial Differential Equation) solver analyses to verify the desired behavior and make any corrections necessary for the physical interactions not included in the parametric models, see Figure 2. Each library element in the parametric library represents two different information sets of the same substructure: a behavioral information (as mentioned before) and layout information representing the geometric parameters for the subsystem. Each set of information is driven and characterized by the same symbol parameters in the schematic level. The schematic is converted to layout by importing the corresponding netlist into the layout editor and 3-D model builder. The layout extractor parses the netlist information for interpretable model components. When a component is found – all geometry related properties are extracted and passed onto the appropriate layout generator. The layout extractor parses the netlist file for interpretable model components. When a component is found, all geometry related properties are extracted from the netlist and passed over to the appropriate layout generator. The layout generator uses these parameters to visualize specific information in one or more layers. All generators of the layout extractor have an additional user generator. These user generators are customizable extensions for the default generators, which can be used to add user specific layout information and/or additional layers, e.g. specific anchor geometries. The layout view of a model serves two different purposes. The first and maybe most important is to make photo masks so that the device can be manufactured. Second is to build a solid (or 3d model) that can be meshed and analyzed in a FEM driven verification step. The purpose of this design approach is to quickly traverse the parameter space based on the interaction of the device behavioral models with their surrounding system environment. For example, cross sensitivities for multi degree of freedom systems can be quickly evaluated. Similarly, the effect of manufacturing tolerances on device performance (like, resonant frequencies, pressure drops, etc.) can be investigated very quickly. An FEM or physical solver approach to estimating all of the criteria that define the technological process can potentially take days or weeks but can be investigated very rapidly using the structured custom design approach. Table 1 shows the comparison between full custom design and structured custom design. Table 2 shows the feature to benefits listing for structured custom design. The limitations of this methodology are that designs must be made from a pre-defined library of components. However, this can also be a benefit since the design is performed with well-understood components. The simulation environment is limited to current network simulators available in the industry (e.g., Saber, Cadence, Matlab, etc.) compared to the numerous PDE solvers conveniently available in the industry. Also, designers are not commonly familiar with network simulator design options. APPLICATION EXAMPLES Two examples are shown for comparing the full custom design approach to the structured custom design. The first example shows the electro-mechanical design of a MEMS mirror for Optical MEMS applications or MOEMS (Gunar et al., 2001). The second application example compares the two design strategies for micro-fluidic design of a micro-pump. TABLE 1 Full Custom Design Structured Custom Design •Designer has idea •Develops geometric model that embodies the idea •FEM expertise required to understand device performance •FEM simulation time/trouble limits exploration •Difficult to explore alternatives •Model hand off has no variables, subsystem engineer can’t explore variations of the device to see system impact •Designer has idea •Develops schematic model that embodies the idea •Explores design alternatives easily by modifying schematic and re-simulating •Easy to explore alternatives •FEM used to perform detailed analysis once design is solidified •Model hand off has variables, subsystem engineer can explore variations of the device to see system impact TABLE 2 Feature Benefits Quickly change design Allow engineer to test many different design ideas quickly Quick simulation runtimes Allow for more complete simulation of device performance Transient simulation Look at transient behavior of device, difficult with FEM Simulate device with surrounding subsystem components Understand behavior of MEMS device and sub-system Simulate without FEM Quicker results, easer to get results, FEM expertise not required Quickly change design Allow engineer to test many different design ideas quickly Quick simulation runtimes Allow for more complete simulation of device performance Transient simulation Look at transient behavior of device, difficult with FEM APPLICATION EXAMPLE 1: DESIGN OF A MEMS OPTICAL MIRROR: A simple mirror supported with two torsion beams is simulated in this example. The mirror is electro-statically actuated by two electrode located below the mirror (see Figure 3). The pull-in angle and the vertical displacement of the mirror was simulated for verification purposes. The paprametric library elements used in this schematic are 6 degree of freedom electro-mechanical beam element, mass element, electrode element and rigid plate element. The voltage was applied to a single electrode while the other electrode was grounded. Figures 4 and 5 show the resulting resulting tilting angle and the vertical motion of the plate center. The solid model (Figure 3a) was built by extracting the device layout into Coventor’s MEMSDesigner. After meshing the solid model with 682 and 1176 26-node bricks, respectively the structure was simulated with Coventor’s coupled FEM/BEM solver CoSolve. The accuracy of the coupled FEM/BEM simulations converges towards the results of the network model by increasing the number of elements. However, the most dramatic advantage of the presented simulation technique reveals by comparing the simulation times of the two different approaches, (see Table 3). TABLE 3 Simulation approach Simulation time for the DC transfer simulation (no contact) FEM/BEM using 682 elements 2.5 hours FEM/BEM using 1176 elements 8 hours SABER system model < 2 seconds!!! Table 3: Simulation time comparison (750 MHZ PC with 1G RAM) APPLICATION EXAMPLE 2: DESIGN OF AN INK-JET An electro-statically membrane actuated inkjet was designed using the parametric model library (see Figure 6). The elements used are rigid flow line, meniscus model, translation-to-flow element (flow piston), electro-static-membrane (“transcraptran”) and electrical source for voltage pulse. A voltage pulse actuated the membrane and the resulting droplet ejection was simulated. Figure 7 shows the droplet ejection behavior and oscillation of the liquid meniscus in response to the voltage pulse on the membrane. The graph shows that the membrane deflection is gradual with a slingshot release to create droplet. Since the membrane is extremely stiff it results in a ringing action of the membrane after the force (through the voltage pulse) is released. Figure 8 shows the corresponding layout, process description and 3-D model for the inkjet device. Figure 9 shows the results of the 3-D VOF (volume-offluid) analyses performed on the ink-jet device. The VOF analyses performed in this study typically requires simulation times of the order of hours whereas the system model simulations for the ink-jet device is solved in a few minutes on a workstation. CONCLUSION A new way to comprehensively design MEMS devices have been introduced in this paper along with two application examples. The new structured approach starts with system requirements and uses the library elements for initial designs that can meet these specifications. Optical, mechanical, electronic and fluidic aspects of the systems can be simultaneously simulated, enabling true system level design of applications incorporating MEMS. Using this methodology the design space can be explored quickly. This methodology also enables links to various PDE solvers to complete detailed analysis. Hence, second order and coupling effects can be verified in subsequent bottom-up verification or design-centering step. Masks for Fabrication Device Layout 3-D PDE Model for Verification Optics Library ElectroMechanical Library Fluidics Library System Model Electro Mechanical MEMS Requirements