<MNA/ODEs> Electrical & Mechanical = Coupled G ce2 C 2 V W 0 ce1 0 Q A Q B V 0 mx ms C1 0 V 0 0 V M 0 Gmna X Amx X Bms f X , X , t MNA matrix D q 0 K q F I q I 0 q 0 MNA solution vector where 1 B Gmna Bms Excitation vector X AX Bf X , X , t Input coupling matrix 1 A Gmna Amx Multi-domain state vector System matrix 1st order ODE G=conductance ce=constitutional eqs C=capacitance V=voltage Q=charge M=mass D=damping K=stiffness F=force q=displacement Transient Analysis, (TA) SUGAR uses Matlab’s ODE solvers to integrate the 1st order system. Assuming the “mass” (G1) of the ode is constant, it’s stored into the solver. So we only need to return the RHS. Options may be chosen whereby stability and accuracy may be traded for speed. Mechanical example: D M G1 I 0 dF dD K G2 dX dX 0 I X X F ( X , X , t ) G2 X 0 X Stored into ODE solver ODE integration Gap Test Case TA / Gap-Closing Actuator A) B) Transient response of a gap-closing actuator. A) shows a plot of displacement as a function of time. The voltage ramps from 5V at t=5usec to 12V at t=500usec, and then releases. As the voltage increases linearly during this time interval, the space between the gap decreases at a nonlinear rate due to electrostatic forces; likewise, the period of oscillation decreases. The amplitude of oscillations decrease exponentially due to the viscous layer of air between the device and substrate. TA w/ Reduced Order Modeling Z. Bai ROM dim n ode dim N Reduced order modeling simulation (blue) superimposed onto our other transient solver (red) demonstrates its accuracy. The lower dimensionality of the reduced order system decreased simulation time by a factor 1/60. ROM via Krylov/Lanczos. Sensitivity Analysis, (SA) L Schenato, W Wu Fstatic K g X K g nom K X Monte-Carlo Solves the above equation by drawing many samples from probability distributions. It produces the most likely outcomes of performance variables. Ellipsoidal Calculus Finds the extreme bounds on performance parameters, i.e. the worst case scenario. Geometric Variation Performance possibilities via Monte Carlo Worst case scenario via Ellipsoidal Calculus L Schenato, W Wu [Knominal + K] q = Fstatic <Model Basics> Mechanical example 3D linear beam matrices Matrix coordinate transformation Model functions EOM Many microelectromechanical systems can be represented by lumped models and their performance described by parameterized ODEs such as M q Dq Kq F (t , q ). The 3D displacement and excitation vectors for a system of N nodes (6N degrees of freedom) are q1 q 2 q3 q q4 q N F1 F 2 F3 F F 4 F N F xi q xi F q yi yi F zi q zi Fi where q i M xi q xi M yi q yi q M zi zi The qi’s consist of displacement translations and rotations about global axes. The Fi’s are the corresponding forces and moments. Any electrical elements are appended onto these elements creating vectors of length Nmechanical + Nelectrical. Stiffness Matrix EA L EA L Ki EA L 12EIz L3 12EIy 6EIy L3 6EIy 12EIy 6EIy L2 L3 L2 4EIy 6EIy 2EIy L L2 L 6EIz L2 6EIz L2 4EIz L 6EIz L2 GJ L EA L 12EIz L3 12EIy 6EIy 12EIy 6EIy L3 L2 L3 L2 6EIy 2 L 6EIz L2 12EIz L3 GJ L L2 12EIz L3 6EIz L2 GJ L GJ L 2EIy 6EIy 2 L 2EIz L 6EIz L2 L 4EIy L 6EIz L2 2EIz L 6EIz L2 4EIz L G E 2 * (1 ) J 2 ( wh ) 3 7(w 2 h 2 ) wh 3 12 hw 3 Iz 12 A hw Iy K i 1 21 2 J = polar 2nd moment o L = length E = Young’s modulus G = Shear modulus Iy = moment about y-ax Iz = moment about z-ax h = layer thickness w = width = Poisson’s ratio Ki = stiffness matrix Mass Matrix Mi 1 3 13 35 13 35 11L 210 11 L 210 AL 1 6 9 70 9 70 13L 420 13 L 420 1 6 I y Iz 11L 210 11L 210 9 70 9 70 3A 13L 420 L2 105 L2 105 I y Iz 6A 13L 420 1 3 I y Iz 6A 13L 420 L2 140 13L 420 13 35 13 35 I y Iz 3A L2 140 11L 210 11L 210 wh3 Iy 13L 12 420 hw3 13L Iz 420 12 A hw Mi 1212 L2 140 2 L 140 AL = density 11L 210 L = length 11L w = width 210 h = layer thickness Iy = moment y-axis L2 105 Iz = moment z-axis L2 Mi = mass matrix 105 Couette Flow Damping Matrix The ith damping matrix is given by Di Lw = viscosity, = fluid layer thickness, Di = damping matrix Coordinate Transformations Mi, Di, and Ki element matrices must be rotated into global coordinates before assembled into system matrices. T directioncos cos direction i directioncos cos direction sin y cos z sin z cos y cos xX cos xX cos xX 1 1 cos x sin x sin z cos z directioncos cos yX cos yY cos yY cos zX cos zZ cos zZ sin x cos x cos y 1 sin y E.g. T qglobal Tqlocal , Fglobal T T Flocal , Kglobal T T KlocalT Instead of 9 direction cosine angles, SUGAR requires x, y, z. Local reference implies having the x-axis glued along the length of the beam, z-axis along thickness, y-axis along width, and using the right-hand coordinate system. Beams originate lying in the xy-plane (substrate), pointing along the positive x axis. Rotation / Positioning y node1 oy y node2 x x z Beam initiates along the X axis. The global XYZ and local xyz axes coincide. Think of the XY-plane is the substrate. z Step 1: +Rotation about local y. Local x & z are repositioned in the global XY-plane. x x ox y y oz z Step 2: +Rotation about the new local z. Local y & x are effected. z Step 3: +Rotation (twist) about local x. Local y & z are effected. 3 positive rotations (right hand rule) are shown here. Important: rotations in SUGAR are performed in the order of y-z-x! Model Functions function [output] = MF_beam(flag, R, params, x, t, nodes, varargin); switch(flag) case 'vars‘ output.dynamic = {1 {‘x’ ‘y’ ‘z’ ‘rx’ ‘ry’ ‘rz’}; 2 {‘x’ ‘y’ ‘z’ ‘rx’ ‘ry’ ‘rz’}}; E.g. case ‘M’ Residual stress Thermal expansion output = beam_matrices('M', params); Accelerating frames case ‘D’ Electrostatic force output = beam_matrices('D', params); Nonlinear effects case ‘K’ output = beam_matrices('K', params); case ‘F’ [F, dFdx] = compute_forces(flag, params, x, t); output = F; case ‘dFdx’ [F, dFdx] = compute_forces(flag, params, x, t); output = dFdx; ETC… Modelfunctions contain a model’s information. The output (e.g. stiffness, electrostatic force, stress) is determined by the flag. <Accelerating Frames> For non-inertial reference frames, SUGAR solves the following ODE Mq Dq Kq Felectrostatic (t , q ) Finertial (t ) where the 6Nx1 inertial force vector is modeled by M ( r ) 2 M r M r FInertial MR M = mass matrix R(t) = substrate position vector (t) = angular frequency vector r(t) = node position vector Finertial = inertial forces Understanding Inertial Forces Fcoriolis m Fcentrifugal Fcentrifugal m ( r ) Perpendicular to rotation axis Ftransverse 0’ r r m Fcoriolis 2 m r only if node is moving; perpendicular to velocity R m Ftransverse m r If angular acceleration Ftranslational m 0’ 0 Ftranslatio nal mR If substrate translates Fnet FElec FInertial Simple Gyro on a Rotating Substrate Z t1rad/sec E Ex(t) -X Y The gyro is first set in motion along the y-axis. The plots show the Ey and Ex displacements of node E as functions of time. Midway through the simulation the left anchor is set to spin about the z-axis at 1 rad/sec. Though this angular velocity has almost no affect on Ey, Ex is significantly affected. Ey(t) 40us <Residual Stress> For residual stress , one static equilibrium state for a MEMS device is Kq FStress FStress A A wh process | netlist > 0 tensile (shortening), < 0 compressive (elongation) For strain gradient , one static equilibrium for the MEMS system given by Kq M Stress M EI y > 0 concave upward, < 0 concave downward where A = cross sectional area I = moment of inertia about y-axis = stress. =strain Residual Stress Gauge 8.4um C Pan, W Hsu SUGAR simulation of residual stress. MEMS devices are often subject to residual stress effects which may affect device performance. Simulated deflection of this residual stress gauge is within 0.59% of measured data. y=8.35um Residual Stress Gauge C Pan, W Hsu ADXL-05 / BiCMOS Close-up view of the residual strain effects. Analog Devices <Thermal Expansion> For thermal expansion, the static representation for a MEMS devis is Kq FThermal FThermal A ET A wh Note: Average beam temperatures specified in the netlist. where E = Young’s modulus T = Tbeam – Tambient = coefficient of thermal expansion = stress B Allen Heatuator Simulation T=600C T=150C T=150C y=4.82u T=600C In the real device there’s a heat distribution along the hot and cold arms. Averaging the temperature along the beam produces the same linear thermal expansion. SUGAR is within 0.5% of the measured deflection done by B. Allen, Hilton Head 1998. <Electrostatic Gap> L FElec,i dxH i ( x)hp( x) 0 1 oV (d ( x)) p( x) 2 2 d ( x) 2 This integral finds the equivalent nodal force and moments caused by the distributed surface forces, where H(x) is the Hermitian shape function. For this case, p(x) is the electrostatic pressure. TA Pull-in (ramp) Smooth V(t) = 0V to 20V ramp. Pulled in before it got to 20V Abrupt V(t) = 11V to 15V, ramp. Nonlinear resonance, then pull-in. <Digital to Analog Converter> R Yeh, KSJ Pister Using a subnet description for device building blocks, the nestlist description of this 4-bit MEMS DAC was reduced to just 14 lines text. SEM (left) and SUGAR visualization (right). DAC Data (nonlinear beams) Output displacement vs digital input of a DAC. The nonlinear beam model matches measured data to within 5%. <Mirror Scanner> KSJ Pister Two-degree-of-freedom optical scanner prototype. Mode 1 = 739Hz, mode 2 = 745Hz. Projection Display Texas Instruments Electrostatic gap TI mirror for flat panel projection displays. Electrostatic gap electrodes (not visible) are underneath the plates. The mass is a little off due to the overlapping plates. Mode Analysis of a Torsional Mirror a) b) BSAC c) d) The mode shapes and frequencies of modes 1, 3, 4 and 6 are shown in Figures a) through d). Respectively they are 15.5kHz, 31.1kHz, 41.7kHz, and 123kHz. < Gyro Modes> Design by Inertial Sensors Group UCB Seshia, Howe Mode Analysis of Accelerometer LLNL Design provided by: Jonathan Simon, LLNL a) b) c) a) schematic design. b) first mode shape corresponding to 27.73 Hz from SUGAR. c) third mode corresponding to 133.02 Hz from SUGAR, mMatches Simon’s analysis 132.46 Hz within 0.5% Induced Current (Multimode resonator) R Brennen This demonstrates steady-state analysis applied to a multi-mode resonator. The left figure shows the Bode and phase plot of the current induced on the sensing comb as a function of the frequency of the voltage at the driving comb of the figure on the left. The measured modes are with in 5% of experimental frequency given by Brennen et. al. <Nonlinear model> y EI s = distance along the beam from node 1 to node 2 r(s) = radius of curvature at s y(s) = angle of the beam at s with respect to the horizontal M(s) = moment at s E = modulus of elasticity I = moment of inertia about the z-axis (out of plane) x and y are the in-plane horizontal and vertical axes with the origin at node 1. Fo = external force at node 2 L = beam length = projected beam shortening L x 1 ds dy F0 Y dx 2 L- Y0 Nonlinear stiffness model To obtain nonlinear stiffness, we first assume that the curves can be approximated by a third order polynomial of the form 2 F0 L q q q A B C D EI L L L 2 3 were q stands for , x, y, etc. Seeing that the solution has odd symmetry, we only need to keep constants B and D, which also eases iterative computations. Absorbing the material and geometric terms into B and D, respectively K1 and K2, we find that F0 K1,i q K2,i q 3 The coefficients of these polynomial curves are associated with the linear stiffnesses K1,i and the cubic nonlinearities K2,i. In order to maintain accuracy, it is applied in a continuous piecewise fashion by dividing the total physical range into, say, 4 intervals, were i(q)=[1,4]. (see plot) Nonlinear Model Results vs Theory L F0 L2 EI L y0 /2 Circles are theory. The red, blue, & green curves are nonlinear model <9-Node Plate Model> Excitation 1 9-Node Plate Model, Excitation 2 Conclusion SUGAR is . . . Simple Text netlist description Matlab environment Conclusion SUGAR is . . . Accurate Simulated results < 5% of actual device performance e.g. gap pull-in, Heatuator, residual stress, multimode resonator Conclusion SUGAR is . . . Extensible User-definable modelfunctions Easy to add more Matlab functionality Conclusion SUGAR is . . . FREE! www-bsac.eecs.berkeley.edu/~cfm cfm@bsac.eecs.berkeley.edu some near-term goals CIF in/out, Squeeze film damping, Nonlinear stiffness, Curved beams, Plates, Buckling, Fringing fields, Trapezoidal beams, Initially loaded beams, Capacitive sensing, Friction, Collisions, Charge transfer, Noise analysis, Genetic Algorithms for design optimization, AutoMacroLevel Algorithms for large dynamic systems, Joints, Strain limiting, Heat transfer, Plasticity, Piezoelectric, Piezoresistive, Resistance(temperature), Thermoelectrics, Multiple time scale handling, Fluidic systems, Flight, Performance analysis, Assembly, Magnetics, Rigid, Bimetalic, Shape-memory, charge leakage, repulsion, Millennium web service, etc.