CARNEGIE MELLONDepartment of Electrical ~nd Computer Engineering__

Development of MEMSTesting Methodology

Abhijeet Kolpekwar

1998

Development of MEMS Testing Methodology

Abhijeet Kolpekwar

Department of Electrical Engineering & Computer Engineering

Carnegie Mellon University

Pittsburgh, PA 15213-3890

May 1998

Submitted in partial fulfillment of the requirements for the degree of Master of Science in

Electrical and Computer Engineering.

ABSTRACT

Microelectromechanical systems (MEMS) are miniature electromechanical sensor and actuator

systems developed from the mature batch-fabricated processes of VLSI technologies. Projected

growth in the MEMS market requires significant advances in CAD and manufacturing for MEMS.

These advances must be accompanied with testing methodologies that ensure both high quality and

reliability. Effectiveness of any testing methodology depends on the accuracy of the fault models

applied. So, the first step towards development of comprehensive testing methodology involves

generating accurate fault models for MEMS. Creating accurate fault models requires a complete

knowledge of all the possible failure mechanisms in MEMS. A systematic study of the impact of

these failure mechanisms on the final functionality of MEMS devices provides important guidelines

for MEMS fault model generation.

A CAD tool called CARAMEL (Contamination And Reliability Analysis of MicroElectromechanical

Layout) has been developed that can be used for fault model generation for MEMS. CARAMEL

is based on the IC contamination analysis tool CODEF and is capable of analyzing the impact

of contamination particles on the geometrical and material properties of microelectromechanical

systems. Output generated by CARAMEL indicates that a wide range of defective structures are

possible due to the presence of particulate contaminations. Moreover, electromechanical simula-

tions of CARAMEL’s mesh representations of defective layout has revealed that a wide variety of

faulty behaviors are associated with these defects.

Several hundred contamination simulations were performed using CAI~AMEL on the surface

micromachined comb-drive resonator. The results obtained were classified under three broad fault

classes: catastrophic, parametric, and harmless. The defect statistics generated by CARAMEL

indicates the comb drive as the most defect-prone region of the microresonator. Most of the

comb drive defects resulted in the catastrophic failures. Contaminations introduced during PolyO

deposition caused the maximum number of faults fotmd in the microresonator. Such defect-to-fault

mappings provide an essential tool for the MEMS fault model generation process. Simulation results

presented in this report confirm that CARAMEL can be used as an effective tool for development

of realistic fault models for MEMS.

ACKNOWLEDGMENTS

Several people at Carngie Mellon University have contributed towards this work. First and

foremost among these is my advisor, Shawn Blanton. I am very much grateful to him for his

invaluable support and guidance throughout this project. It would be difficult to find a more

understanding, patient and friendlier advisor. I would like to thank Dr. Gary Fedder for giving

useful advice and reviewing this report. I am also thankful to Christopher Kellen for his valuable

contribution. My thanks are due to Dr. Tamal Mukherjee and Sitaraman Iyer for providing useful

tips at every stage of the project.

This work would not have been possible without the help of all my friends at CMU: Kumar

Dwarkanath, Alok Jain, Srihari Cadambi, Piyush~ Singh, Vishnu Patankar, Anand Dixit, Anan-

dasivam Gopal, to name a few. A big thanks to all of you for an exciting and memorable stay in

Pittsburgh. My thanks are also due to Ashit Talukder and Srihari Cadambi who are always there

when I need their help.

I am deeply indebted to my parents and my fiancee Sonali for offering constant support and

encouragement.

Finally, I would like to thank my sponsors. This research is sponsored by the National Science

Foundation under grant MIP-9702678 and the Defense Research Projects Agency under Rome

Laboratory, Air Force Materiel Command, USAF, under grant F30602-97-2-0323,

III

Contents

1 Introduction 1

2 Related Work 11

2.1 Open Problems in MEMS Testing ............................ 11

2.2 Fault Modeling and Simulation of MEMS ........................ 13

2.3 Test Generation for MEMS ................................ 15

2.4 BIST and Fault Tolerance for MEMS ........................... 17

2.5 Summary .......................................... 21

3 CARAMEL 23

3.1 Micromachined Microresonator .............................. 23

3.2 Resonatcr Application: Accelerometer .......................... 26

3.3 CARAMEL Implementation ................................ 28

3.3.1 Process Simulation ................................. 28

3.3.2 Structure Extraction ................................ 31

3.4 Simulation Results ..................................... 35

4 Conclusions and Future Work 43

V

Lis¢ of Figures

1.1 (a) A SEM and (b) top view of a surface-micromachined comb-drive microresonator.

1.2 Effect of stiction on surface-micromachined microresonator ............... 7

1.3 MEMS contamination analysis performed using CARAMEL .............. 8

2.1 (a) Cross section of piezoresistive accelerometer die and (b) motion of doubly

ported mass under acceleration. (Cap is not shown for clarity.) ............ 18

2.2 (a) Schematic of accelerometer with thermal self test and (b) deflection of seismic

mass due to thermal expansion of the actuator beam .................. 20

3.1 (a) Top view of the microresonator and pertinent dimensional parameters of the (b)

comb drive, (c) shuttle, and (d) folded-flexure ...................... 25

3.2 Accelerometer basics:(a) Sensing operation of a typical accelerometer. (b) Signal

sensing and processing circuitry for transducing a mass deflection into an output

voltage ............................................ 29

3.3 Abbreviated flow for MCNC’s MultiUser MEMS Process service. (a) Cross-sectional

view. (b) Top view (layout) ................................. 30

vi

3.4 Illustration of element splitting for mechanical mesh construction: (a) A set of regions

before split and (b) the same regions after x-y and z splits ............... 33

3.5 Three-dimensional representations of defective resonators and their corresponding

mechanical mesh for particle contaminations located (a) between comb fingers, (b)

on a beam, and (c) under the shuttle ........................... 36

3.6 Three-dimensional representations of defective resonators and their corresponding

mechanical mesh for particle contaminations located (a) outside of the resonator’s

active area and (b) on the shuttle ............................. 37

vii

List of Tables

3.1 Resonator parameters (and typical values) that define the modeled operation of the

microresonator 26

3.2 Five representative cases of defect analysis using CARAMEL .............. 37

3.3 Coarse fault categorization based on deviation in resonant frequency fx ........ 40

Defect categorization based on the MUMPs process step for which the contamination

was introduced ........................................

3.4

41

3.5 Spatial distribution of faults in the microresonator .................... 41

4.1 Appendix A - The complete MCNC MUMPs process described in the PREDITOR

format ............................................. 51

viii

Chapter 1

Introduction

Microelectromechanical systems (MEMS) are miniature electromechanicM sensor and actuator sys-

tems developed from the mature batch-fabricated processes of VLSI technologies. MEMS have wide

applications such as miniature inertial measurement units, biochemical analysis on a chip, arrayed

micromanipulation of parts, optical displays and micro-probes for neural recording. The current

and increasing success of MEMS stems from its promise of better performance, low manufacturing

cost, miniaturization and its capacity for integration with electronic circuits.

The low cost of MEMS, as with the integrated circuit (IC), is attributed to the amortization

of capital cost over millions of individual batch-fabricated devices. This has substantially reduced

the cost of sensors and actuators by several orders of magnitude. For example, Analog Devices’

MEMS accelerometer used extensively in airbag systems sells for about $5 a unit when sold in large

quantities. An accelerometer of similar quality manufactured from a competing technology sells for

more than $1000 [5].

The smaller and reduced-mass characteristics of MEMS structures allow higher operating fre-

quencies and lower power consumptions. These properties make MEMS ideal for many portable

and remote applications. For example, MEMS-based pressure sensors are being used to moni-

tor truck tire pressure in order to extend tire life, while optical gratings are being developed for

heads-up color displays [5].

The ability to use the same manufacturing technology to integrate MEMS and electronic systems

adds another dimension of utility. Single-chip systems of electronics and mechanical sensors are

being produced and envisioned for a variety of applications. Texas Instruments, for example, has

developed a micromirror projection display consisting of a two-dimensional array of aluminum

mirrors, each individually controlled by on-chip RAM cells.

By the early 1980s, progress in microelectronics had reduced the cost of a microprocessor to less

than that of a typical silicon sensor; today the peripheral functions of sensing and actuation con-

tinue to represent the principal bottleneck in the application of microelectronics to many emerging

systems, not only in terms of cost but also in terms of reliability and accuracy as well. It is thus

evident that continued progress in MEMS is likely to exert considerable leverage in the microelec-

tronics industry beyond the direct markets for sensors and actuators. Accelerometers [1], pressure

sensors [2] and thermal ink-jet printheads [3] are the most matured MEMS applications accounting

for almost all of the commercial MEMS market of around $1 billion in 1994. The MEMS market

is projected to reach between $12 and $14 billion by the year 2000 [5]. The technology trend is

towards higher degrees of integration that result in greater capabilities by MEMS-based products to

asses and manipulate their surrounding environment. Realizing these market projections requires

that the open problems in design methodology and process and package technology be confronted

and adequately addressed.

There is a desperate need for computer-aided design (CAD) tools that shorten the design and de-

velopment time for MEMS-based products involving physical interactions between mechanical, elec-

trostatic, magnetic, thermal, fluidic, and optical domains° Current design cycles for microsystems1

are measured in years [6]. Success in this area depends greatly on new design methodologies that

allow complex microsystems of mixed domains (mechanical, electrical, thermal, etc.) to be hier-

archically represented and simulated. Improvements in packaging and process technology will also

be required to support the mixed-system applications required by MEMS integration. Currently,

most MEMS are made from commonly used IC fabrication techniques. Process and equipment

development specifically for MEMS will expand the design space and result in more opportunities

for completely integrated systems.

Advances in the above areas alone will not fuel MEMS growth. Continued success for MEMS will

require cost-effective methods of manufacturing. Similar to digital ICs, success in this area must

include a testing methodology that allows products to be economically tested while ensuring high

quality and reliability. This is especially important in applications where MEMS are integral parts of

safety-critical systems such as automotive airbag systems that utilize Analog Devices’ ADXL-series

of accelerometer sensors. MEMS testing methodologies must be developed in concert with new

design, packaging, and process technologies. Most industrial practices focus on partially exercising

the functionality of MEMS by performing certain electrical measurements. However, there is a need

to create formal links between failure modes (i.e. electrical measurements to identify failures) and

the underlying physical causes. These correlations will allow accurate modeling of complex effects

that can be used in fault model generation, fault diagnosis and the development of efficient testing

1Microsystems is a generic term for all micro-sensors, micro-actuators and other self-contained~ miniaturizedsub-systems that possess additional processing capabilities.

combdrive

shuttle

movingcombs flexurefixed anchorscombs

(a) (b)Figure 1.1: (a) A 8EM and (b) top view of a surface-micromachined comb-drive microresonator.

techniques for MEMS. Such links are also critical for providing important guidelines for modifying

process control parameters that result in fast yield improvements.

We have chosen a folded-flexure comb-drive microresonator2 as our research vehicle because it

possesses many of the basic structures (beams, joints, springs, etc.) that many researchers believe

will form the core pri~niti~es of a MEMS design library [14, 15, 16]. The hope is that the analysis

performed with the resonator will be applicable to many classes of MEMS. The resonator structure

(see Figure 1.1a.) belongs to a class of MEMS known as surface-micromachined MEMS. Prototype

surface micromachining processes are available from MCNC (Multi-user MEMS Processes service

(MUMPs) [17]), Analog Devices’ iMEMS process, and from Sandia NationM Labs. We have selected

the MUMPs process for the contaminations simulations of the resonator due to its open availability.

In this process, thin films are sequentially deposited and patterned on top of the substrate. The

movable MEMS structure is created when a "release" step is used to etch away a sacrificial layer.

The seminal paper on the resonator is given in [18]. Analytic models of the resonator~s pertinent

2We will refer Go the folded-flexure comb-drive microresonator simply as the "resonator".

characteristics can be found in [19, 20, 21]. The resonator structure is a mature case study in the

design of "suspended MEMS’~ which are now used in commercial accelerometers [22, 23], gyro-

scopes (soon), and micromirror optical beam steering [29]. Future commercial applications are

resonator-based oscillators [26], IF mixers, high-Q IF filters for communications~ micromechanical

filters [25], micromachined resonant pressure sensors [24], and microstages for probe-based data

storage [30].

A detailed description of the structure and operation of the resonator is provided in chapter 3.

Here, we give a general overview of the resonator. A top view of the resonator is shown in Fig-

ure 1.lb. It is a mass-spring-damper system made only from the first and second polysilicon layers.

A majority of the resonator’s structure is suspended above the wafer surface. It is connected to

the wafer only at the four anchor locations shown. The shuttle ~nass is supported by symmetrical

folded flexures that are designed to be compliant in the x direction and rigid in the y direction.

Each comb drive is formed from interdigitated fingers on either side of the shuttle mass. Between

every pair of fixed comb fingers is a moving comb finger that is attached to the shuttle. Such an ar-

rangement of fixed and movable beams produces an array of differential parallel plate capacitances.

The resonator is primarily used for sensing acceleration. The shuttle mass moves under applied

acceleration (in the x direction) causing a change in the area of overlap between the fixed combs

and moving combs which in turn results in a capacitance change. The change of capacitance is

then sensed by signal processing circuitry connected to the resonator. The folded flexure provides

a restoring force to the shuttle movement.

Complexity of physical failures is a major obstacle in the extensive use of MEMS-based devices.

It is imperative to test these devices thoroughly for all the possible failures to achieve high quality.

Figure 1.2 illustrates an example of a complex failure called stiction. Stiction is defined as adhesion

of the microstructure to adjacent surfaces [42]. It is a result of surface tension forces (mostly due to

moisture) overpowering the stiffness of the comb fingers. Figure 1.2b shows the "stuck-down" beams

that lie flat on the substrate due to stiction. Figure 1.2c shows the "tipped" beams which get tilted

towards the die surface due to surface tension. Both, stuck-down and tipped beams result in loss

of resonator functionality. Complex failure mechanisms like these degrade the reliability of MEMS-

based devices. Such effects need to be modeled accurately for fault simulation and test generation.

In [8], we have identified a failure mode to differentiate between the above two physical failures.

The information obtained in this way is critical for MEMS testability and yield improvement.

Success of any testing methodology is highly dependent on the fault models employed. Fault mod-

els that do not "cover" real defective behavior can reduce defect coverage and degrade test quality.

The first step of our work towards development of a comprehensive MEMS testing methodology

focuses on fault model generation for MEMS. MEMS fault models, unlike their digital and analog

counterparts, must explicitly consider the impact of defects on the micromechanical structures. Our

approach centers on the inductive generation of the possible faulty behaviors from realistic con-

tamination simulations. Moreover, the simulations are performed on the microresonator, a generic

MEM structure that we believe contains the design primitives for many classes of MEMS.

Our experience is that a major cause of faulty behavior in MEMS is due to particulate contam-

inations that occur during various steps of fabrication. Contaminations can cause a significant

perturbation in the structural and material properties of the microstructure [7]. A formal assess-

ment of both the possible defective structures and the corresponding faulty behaviors on MEMS

design primitives will lead to the formation of effective MEMS fault models. Such fault models

6

Shuttle Mass

Normal Beam

Stuck-down beam

(b)

Tipped beam

(c)Figure 1.2: Effect of stiction on surface-micromachined microresonator.

1. Deposit Nitride2. Depesit.sw (poly0)3. Lithe.Spin on (po~y0-rnask)4. Lithe.expose (poty0-mask)5. Li~o.develop (poly0-mask)

Process recipe

Process Simulation

Structure Extraction

ElectromechanicalSimulation

Parameter Analysis

Contaminations~ Design Layout

CARAMEL

DefectiveStructure

’x-ysplit ~1

]zx

’~-------~ Mesh Model

Top view of mechanical mesh

Figure 1.3: MEMS contamination analysis performed using CARAMEL.

will undoubtedly lead to methods for fault grading, test generation, design for testability (DFT),

and design for fault avoidance. The fault modeling process can also be used to form links between

defects and faulty behaviors. Such links would aid in diagnosis by facilitating the identification of

process steps that are likely to produce the observed faulty behavior [11]. Thus, our approach to

developing a complete MEMS testing methodology relies on a fault inductive approach that :

1. Completely characterizes the faulty behavior of MEMS primitives.

2. Supports the generation of fault models for simulation and test generation.

3. Provides techniques for testable and reliable MEMS design.

We have developed a MEMS contamination analysis tool called CARAMEL (Contamination And

Reliability Analysis of MicroElectromechanical Layout) for analyzing the impact of contamination

particulates on the properties of microelectromechanical layout. The core of CARAMEL is based

on the IC contamination analysis tool CODEF [12]. CARAMEL is an integral component of our

MEMS fault model generation methodology (see Figure 1.3). CARAMEL requires three inputs:

1. Design definition: This is a layout of the design in the Caltech Intermediate Form (CIF).

2. Process definition: This includes a sequence of process steps with all the required details

such as deposition thickness, etching rate, etching time etc.

3. Contamination definition: This includes geometrical and material characteristics of the

particulate contamination, its location in the MEMS layout, and its process step of introduc-

tion.

CARAMEL performs process simulation and creates a three-dimensional representation of the

defective microelectromechanical layout. It then extracts a mesh representation from the defective

layout whose form is completely compatible with the mechanical simulator ABAQUS [13]. Me-

chanical simulation of the mesh then allows us to link the contamination of concern to a defective

structure and a faulty behavior. Observed faulty behaviors are then classified and used to form

models at the next level of abstraction. Monte Carlo iteration around the flow of Figure 1.3 provides

a mechanism for creating realistic fault models for MEMS.

The rest of the report is organized as follows. Chapter 2 presents prior work in the area of

MEMS testing. Chapter 3 describes our tool CARAMEL and illustrates its use in the fault model

generation flow illustrated in Figure 1.3. Finally, chapter 4 presents our conclusions and outlines

areas of future work.

10

Chapter 2

Related Work

Most research work in the MEMS area centers on design, technology, and packaging problems and

not testing. However, there have been a small number of researchers that are concerned with

MEMS testability. In this chapter, we present the open issues in MEMS testing and discuss the

important work done in this area.

2.1 Open Problems in MEMS Testing

The test problem is complicated by systems that contain a large number and large variety of

component or modules. Thus, MEMS inherently poses a new challenge to the testing community

due to its multi-domain operational characteristics. The testing problem associated with MEMS has

created serious limitations to the high-volume commercial production of microsystems. Described

below are the main difficulties in MEMS testing [38]:

11

¯ Accessibility, Controllability, and Observability: There is limited electrical access (I/O

connections) to most MEMS-based devices. This prevents internal nodes access which in turn

makes controlling certain internal parameters and observing test responses difficult.

¯ Diverse Function: Different modules within the microsystem (i.e. sensors, actuators, analog

or digital signal processing circuits) require completely new approaches to testing. Unlike

digital and analog circuits, there are no known fault models and test algorithms designated

especially for the microelectromechanical structures found in microsystems.

¯ Interference: In highly integrated microsystems, various sub-systems are in immediate prox-

imity and therefore, can adversely affect each other. For example, heat dissipation of a digital

signal processor may result in the thermal expansion of the microstructure used in the sensor

circuitry. Such a change in the sensor geometry results in faulty sensor output.

¯ Test Environment: A system under test typically requires some form of initialization. This

is inherently difficult for systems with sensors and actuators where special environmental

conditions are required.

¯ Packaging Influence: Packaging of a high-density multi-module system can cause me-

chanical stress that may change the MEMS functionality. For example, in case of a bulk-

micromachined pressure sensor, any stress developed on the diaphragm due to packaging can

create undesired voltage fluctuations at the output.

¯ Complex Failure Mechanisms: A comprehensive MEMS testing methodology requires

a complete knowledge of all the possible failure mechanisms. Problems such as damping,

stiction, cross-axis sensitivity, over-load protection and in-process survivability need to be

12

analyzed and understood. These effects are quite complex and difficult to model accurately

for test generation purposes.

The research performed so far in the area of MEMS testing addresses a subset of the above issues.

In the next few sections, we describe research activities related to various aspects of MEMS testing.

2.2 Fault Modeling and Simulation of MEMS

Fault models are assumptions about how physical defects affect the behavior of the unit under test.

In the case of MEMS, where physical failure mechanisms are much more complex due to presence

of mixed domains, developing fault models becomes a difficult challenge.

Lubraszewski et. al [35] provide an extensive overview of the issues and possible solutions for the

problems related to MEMS fault modeling, simulation, test generation, design for testability (DFT),

and built-in self test (BIST). They also described their MEMS testing environment termed CAT

(Compttter-Aided Testing). They use an electrical schematic to model MEMS structures. MEMS

defects are then modeled using the concept of mutants and saboteurs. Mutants use analog-like

faults that model non-electrical (mechanical, optical, etc.) defects. The modeled defects normally

cause parametric changes in the electrical model representation. Saboteurs, on the other hand, cause

components to be removed or added to the electrical model. They are used to represent defects

that add or remove components of the microstructure. However, the accuracy of such modeling is

very much dependent on the accuracy of the electrical representation of the system. Inaccuracies

in the fault model lead to poor-quality testing techniques. Moreover, experiments involving the

fabrication and analysis of real faulty microsystems is needed to validate the modeling accuracy of

13

mutants and saboteurs. The paper proposes to perform fault simulation by simply modifying the

fault-free microsystem description by instantiating mutant or saboteur descriptions of an HDL-A

fault model library. A drawback of such an approach is that there is no methodology for selecting

the probable mutant/saboteur faults.

Vermeiren et. al [37] stress the importance of the MEMS model and its relationship to defects

of the MEMS structure. They too use an electrical ~nodel to represent mechanical and electricM

components of MEMS. However, the model is constructed in a such way so as to allow the accurate

modeling of a wide variety of MEMS defects. They have focused on the construction of a simulation

model that allows fault injection in order to perform fault simulation. Experiments are performed on

two microsystems: a resonant silicon beam force sensor and a miniature opto-electric transformer.

Some of the requirements for fault models and fault simulation listed in [37] include:

Simultaneous development of the functional and behavioral models of microsystems at higher

levels of abstraction along with the development of simulation models for faulty MEMS be-

havior.

Meeting the requirement of underlying physical laws by simulation models in the presence of

injected faults.

Consideration of "side effects" while developing simulation models. These side effects may

become dominant under the influence of faults. For example, consider an accelerometer with

impeded shuttle movement. A voltage applied across the comb drives fails to move the shuttle

mass resulting in a heating side effect. This may lead to thermal expansion of the beams which

may affect beam stiffness and therefore, the resonant frequency of the resonator.

14

¯ Systematic methods for fault list reduction (similar to Analog fault simulation) through iden-

tification of equivalent or dominant faults.

Most of the on-going research in MEMS fault modeling and simulation builds upon the approaches

used in analog fault modeling and simulation. However, a large variety of physical, chemical

and other effects complicate fault model generation for MEMS, making a purely analog approach

inadequate.

2.3 Test Generation for MEMS

Most test generation approaches proposed for microsystems are similar to those applied to analog

circuits. Here, we discuss some important test generation approaches explored in the previous work.

The approach described in [35] is based on the sensitivity-guided search process originally de-

scribed in [36]. The sensitivity of an operational parameter to a given component is defined as the

ratio of change in the former to the change in the latter. The fault model applied here uses the

mutant/saboteur concept described earlier. Test generation begins with the selection of an initial

test stimulus that corresponds to the mid-range of input voltage values. A fault simulation is then

performed to estimate the fault coverage. Authors do not state how the estimate is obtained. The

undetected faults are re-simulated using test patterns that are higher and lower than the initial

voltage value. Faults which remain undetected by the new tests are partitioned into two new lists;

one containing faults for which the sensitivity of the measuring parameter has increased for higher

input stimuli, and ~nother containing those faults for which the sensitivity of the measuring pa-

rameters has increased for lower input stimuli. The p~per does not discuss the case in which the

15

sensitivity of the measuring parameter remains unchanged. The procedure is recursively applied

until all faults are detected. The rationale for using such a fault partitioning approach is not clear

from the paper. Moreover, the test set produced is not guaranteed to be optimal. Also, the use of

saboteurs can lead to highly complicated transfer functions for the system which in turn, may result

in an intense fault simulation and test generation process. It is necessary to examine the viabil-

ity of this approach through systematic analysis of the complexity of the operations implemented.

Finally, the use of saboteurs and mutants to model MEMS defects remains suspect.

Another test generation approach is presented by Olbrich and others in [38]. They also use an elec-

trical schematic to model both the electrical and mechanical components of a bulk-micromachined

accelerometer. Similar to digital fault modeling, they model defects using stuck-at, bridging, and

stuck-open faults. Carefully placed resistors are a~ided to their electrical schematic to model these

faults. An additional level of complexity is added by allowing the parameter values of these "fault"

resistors to vary. Exhaustive Hspice simulations are used to determine which voltage and current

nodes are good candidates for observing faulty behavior. These simulations require sweeping the

values of the fault resistors from 1K~ to 1GFt -- the extremes of the value range attempts to mimic

shorts and opens, respectively. Voltage and current are measured at various nodes in the circuit

for every value of the fault resistor. However, taking such measurements from a packaged sensor is

practically impossible. Measurement windows for ranges of voltage and current for which a circuit

is defined to be faulty or fault-free are established somewhat arbitrarily. It is not clear how many

fault-free simulations are conducted, i.e., whether it is fully exhaustive. The authors state that

fault masking may occur, that is, the combination of a fault and allowable resistor tolerances can

cause fault-free measurements. The boundaries that define faulty behavior are set "intuitively". A

"grey zone" is used to further push the fault boundaries out for safety reasons. This seems to be

16

counter intuitive because it appears that more "out-of-spec" behavior is tolerated when the fault

boundaries are extended. The authors claim that the faults under consideration come from the

analysis of accelerometer failures. Using statistics of field and manufacturing failures, they choose

which faults to simulate according to their actual occurrence in the field. From this information,

the location of the most sensitive nodes to monitor accelerometer temperature are identified. But,

the method used to monitor accelerometer temperature is not discussed by the authors. They have

also presented the voltage values at different nodes under the influence of faults. These values

facilitate identification of observable faulty behaviors at the node of concern. Unlike the testing

approaches given in [39, 41~ 40], this approach allows on-line monitoring. However, the paper does

not give any detailed explanation of why and how the faults under consideration are chosen.

Test generation for MEMS is not a straight forward extension of digital or analog test techniques.

Presence of multiple domains like mechanical, chemical, optical etc. complicates the test generation.

Also, the accuracy and effectiveness of test generation is highly subject to the accuracy of the fault

models applied.

2.4 BIST and Fault Tolerance for MEMS

In this section, we describe the research performed in the areas of built-in self test (BIST) and

fault tolerance for MEMS. Similar to MEMS fault simulation and fault modeling, ideas presented

in these areas also leverage techniques from analog test. There are two broad classes of analog

BIST that can be identified:

17

piezoresistors.__ framevoltage excitation /~ capfor self test / \ I I . , .

[ J ’---- shuttle mass

deflection directionshuttle ma.ss ~ piezoresistor

(b/Figure 2.1: (a) Cross section of piezoresistive accelerometer die and (b) motion of doubly supportedmass under acceleration. (Cap is not shown for clarity.)

1. Spatial Redundancy." These techniques require extra circuitry to perform the self test.

These include specific test/monitoring cells [43, 44], analog test buses and circuit duplication.

2. Analytical l:tedundancy: These include analytic redundancy techniques such as parametric

analysis, frequency chaxacteristics [46], reliability indicators [45], signal delay measurements,

etc.

Most BIST approaches for MEMS use a combination of the above two approaches. In [39], Allen

and his colleagues have described a BIST approach for a piezoresistive accelerometer. Figure 2.1a

shows a crossection of the piezoresistive accelerometer. The motion of the silicon mass under the

acceleration is depicted in Figure 2.lb. The following relation was used to compute the voltage

required to subject the silicon-mass to a desired acceleration.

eA V’~gelectro = 19.6mxo2 (2.1)

where g~tect, o is the silicon-mass acceleration, A is the electrode area, V is the applied voltage, m

is the mass of the shuttle, e is the dielectric constant of the damping media and Xo is the initial

position the movable silicon mass. The electrostatic force applied to set the desired acceleration

gelectro depends on voltage and geometry of the shuttle and is suitable for calibration purposes. The

above equation is valid only for small deflections (<3%) of the silicon mass. During test, an applied

18

voltage is varied in DC step from ÷20V to -20V and the sensor output is measured. At every

applied voltage, shuttle mass acceleration and the sensitivity (V/g) of the device is computed

verify whether the device is functional. Structural design modifications are suggested for protecting

the device from excessive force. The self-testing feature provides two benefits. The first is that the

user can confirm that the shuttle-mass is free to respond in critical operating conditions. Secondly,

the self-test is a force applied to the mass, which cannot be differentiated from an acceleration force.

Hence, using a known electrostatic force, the accelerometer can be calibrated over temperature.

Built-in self test approaches for micromachined accelerometers described in [39, 47] rely on ap-

plication of electrostatic forces to the silicon mass. However, electrostatic forces are fundamentally

weak. Therefore, very high voltages are required to achieve full-scale electrostatic actuation in ac-

celerometers. This is the main drawback in using electrostatic actuation for implementing built-in

self test for MEMS-based devices.

In [48], Pourahmadi and others have proposed a thermal self-test mechanism for accelerometers

(with 50G full-scale deflection). This technique uses a differential thermal expansion to provide

actuation. It is claimed that the BIST structure used is virtually insensitive to ambient temperature

variations and has no effect whatsoever on the thermal performance of the sensor. Figure 2.2a shows

a schematic of accelerometer with thermal self test. The self-test feature uses an actuation beam

that is thermally isolated from the seismic mass (suspended silicon) by a gap of 8 microns.

ion-implanted heating resistor is situated near the middle of the central beam. When the voltage

pulse is applied to the heat resistor, the temperature of the actuation beam increases. As a result,

it expands and creates a force on the seismic mass that causes a downward movement. (See

Figure 2.2b). The piezoresistive sense elements are built in such a way that they remain unaffected

19

downward deflectionof seismic massbuckledPiezoresistive Actuation Beam

sense ele/nent / actuator beam

Actuation heaterSeismic~ ~------~----~

(a) (b)

Figure 2.2: (a) Schematic of accelerometer with thermal self test and (b) deflection of seismic massdue to thermal expansion of the actuator beam.

by the voltage pulse and the temperature change at the actuation beam. The sense element detects

the downward motion of seismic mass as an acceleration signal. The mechanism, though simple

in theory, is difficult to implement. However, no other performance aspects of the sensor itself are

compromised by the actuator structure and mechanism. The authors do not comment about the

time required for thermal actuation of the seismic mass and also the the time required to restore

the mass to its normal position. We believe that this could be a potential performance issue in

such a BIST technique.

Most BIST approaches for MEMS-based inertial measurement devices use actuation of the me-

chanical part for test. None of the previous work provide an analysis of BIST structures’ potential

impact on performance. Also, all the practical BIST approaches are off-line, that is, they require

devices to switch to a test mode where normal operation must be suspended. Such mode-switching

may not be desirable in many MEMS applications. It should also be noted that most BIST ap-

proaches perform an indirect measurement of the mechanical movement of the microstructure. The

electrical/thermal circuitry involved in applying self test may hide (for instance, due to electrical

or thermal noise) or not uncover (due to inadequate self test) the non-electrical/mechanical faults

2O

in the system. The researchers in this area provide no discussion on this topic. Thus, the BIST

approaches described above may not guarantee 100% fault coverage.

2.5 Summary

In this chapter, we have presented an overview of the research in MEMS testing. The list of

references attached at the end of the report contains some more interesting papers in this area

[41, 44, 49, 50, 51]. Following are our observations :

1o Most testing techniques are highly design specific. As a result, they may work well with one

design but not with others. There is a need to develop a comprehensive testing methodology

which can be applicable to all instances of a class of MEMS structures like inertial sensors.

The long-term objective of our research work is to develop such a testing methodology.

2. The success of any testing technique depends on the fault models applied. None of the papers

described here have general guidelines for developing fault models. Fault models discussed

so far are highly design specific and are in general very difficult to apply. Also, they are not

guaranteed to cover all the possible defects in the design. We plan to address this concern by

developing realistic fault models that can be applied to a class of MEMS designs at a higher

level of abstraction.

3. Most BIST techniques estimate the mechanical movement of the microstructure by perform-

ing electrical measurements. In practice, such measurements are difficult, if not impossible~

to perform because of the presence of noise. Setting the range for fault-free behavior, as in

analog testing, is a difficult task. Mistakes in setting such thresholds may result in informa-

21

tion loss that degrades test quality. This problem can be tackled by analyzing various failure

mechanisms and the resulting deviations in the final functionality. Knowledge of such correla-

tions will lead to more effective BIST techniques (like modification of structural and material

properties, selection of better measurement techniques) and aid in MEMS fault diagnosis.

Therefore, our work in MEMS testing first centers on exploring a wide range of defects and

the faulty MEMS behaviors associated with them°

22

Chapter 3

CARAMEL

In this chapter, we first describe the normal operation of the surface micromachined microres-

onator. We then describe our CAD tool CARAMEL (Contamination And l~eliability Analysis of

MicroElectromechanical Layout) and its application to the development of realistic fault models for

MEMS. Finally, simulation results obtained using CARAMEL for the microresonator are discussed

in detail.

3.1 Micromachined Microresonator

The folded-flexure comb-drive microresonator that we consider for our analysis can be fabricated

using a simplified version of the three-layer polysilicon MUMPs process. (Appendix A provides

complete listing of the MUMPs process steps.) The top view of a resonator shown in Figure 1.1b

is repeated in Figure 3. la. The resonator is a mass-spring-damper system made only from the first

and second polysilicon layers. A majority of the resonator’s structure is suspended above the wafer

23

surface. It is connected to the wafer only at the four anchor locations shown. The shuttle mass

is supported by symmetrical folded flexures that are designed to be compliant in the x direction

and rigid in the y direction. The shuttle can be electrostatically actuated in the x direction by

applying a voltage across the comb drives as shown in Figure 3.1a. Each comb drive is formed from

interdigitated fingers on either side of the shuttle mass.

Lateral motion in the x and y directions are modeled by the second order equations:

Fe,x = rnx~ + Bx~ + kxx (3.1)

Fe,y = my~) + Bye] + kyy (3.2)

where F~,~ and F~,y are the electrostatic forces generated by the comb drives in the x and y

directions, respectively. The parameter values for the effective masses (rex and my), damping coef-

ficients (Bx and By), and spring constants (ks and ky) are calculated from the material properties

and geometries of the resonator°

The parameters [32] that define the resonator’s operation are described in Table 3.1 and illustrated

in Figures 3. lb, c, and d. These parameters along with properties of the materials are used to model

the operation of the resonator. The effective masses rn~ and my are given by

1 12= + + (3.3)8my = m~ + ~mt + mb (3.4)

where ms is the shuttle mass, and mt and mb are the total masses for the all the truss sections and

long beams~ respectively. ~om [20], the damping coefficiem in the x direction is

24

combdrive

movingcombsfixedcombs

shuRlemass

x

folded

anchors

(a)

(d)

Figure 3.1: (a) Top view of the microresonator and pertinent dimensional parameters of the (b)comb drive, (c) shuttle, and (d) folded-flexure.

where # is the viscosity of air, d is the spacer gap, 5 is the penetration depth of airflow above the

resonator structure, g is the gap between comb fingers, and As, At, Ab, and Ac are the surface

areas for the shuttle, truss beams, flexure beams and comb-finger sidewalls, respectively.

The linear equations for the flexure spring constants for the x and y directions are taken from

[19]

kz = L~ 4L2t + 41o~LtLb + 36(~2L~ (3.6)

2Etwat 8L~t + 8aLtLb + a2L~ky = L~ 4L~ + lOc~LtLb + 5(~2L~ (3.7)

where E is Young’s modulus for polysilicon, t is the polysilicon thickness and c~ = (Wt/Wb)3. For

the special case of wc = g = t = d, we have from [21] the force generated by each of the comb drives

Fe,z ~ 1.12eoNt-V2 (3.8)g

25

Parameter Parametername description

Lb length of flexure beam

Wb

LtWt

Lsy

Wsa

Wsy

LcgXo

TE

PN

ms

mb

m$

width of flexure beamlength of truss beamwidth of truss beamlength of shuttle yokewidth of shuttle axlewidth of shuttle yokelength of comb fingersgap between comb fingerscomb finger overlapthickness of resonatorYoung’s modulus of polydensity of polynumber of comb fingersshuttle masslong beam masstruss section mass

Typical

value82 ttm2 t~m12 #m4 #m50 #m18 #m11 #m20 t~m2 #m10 #m2 #m165 GPa2330 Kg/m3

151.61e-11 Kg6.11e-12 Kg3.73e-13 Kg

Table 3.1: Resonator parameters (and typical values) that define the modeled operation of the

microresonator.

where eo is the permitivity of air and V is the voltage applied to the N-finger comb drives. The

above relationships define the resonator’s maximum deflection Xmax and resonant frequency Ix [32].

Xmax - (3.10)

However, it should be noted that the above mathematical models have limited ~ccuracy because

they do not account for beam compression and axle bending effects.

3.2 Resonator Application: Accelerometer

In this section, we describe the use of resonator in designing an accelerometero Figure 3.2a shows

the basic sensing operation of a typical accelerometer. It consists of a shuttle mass which is designed

26

to move with respect to its package under applied acceleration in the y direction (See Figure 3.2).

There is an array of parallel comb fingers attached to both sides of the shuttle mass1. Between

every pair of fixed combs is a moving comb. Such an arrangement of fixed and movable combs

produces an array of differential parallel-plate capacitors° The capacitance C, for every pair of

moving-fixed parallel beams is given by the simple equation :

AeC- d (3.11)

where A is the area of parallel plates (i.e. beam overlap), e is the permitivity of the spacer between

the combs, and d is the distance between adjacent combs.

Thus, when the shuttle moves in y-direction as shown in Figure 3.2a, the distance d between

the parallel beams changes, resulting in a corresponding change in the capacitance. Observe in

Figure 3.2a, the capacitance for one pair of parallel combs increases to C + AC and decreases

for the other pair to C - AC. Similarly, if the shuttle moves in the x-direction, overlap area A

of the comb fingers which again results in a change of capacitance. Thus, shuttle movement can

be converted to an equivalent change in capacitance. A simple circuit for sensing the capacitance

change is shown in Figure 3.2b. This circuit can sense capacitance changes and produce a linear

change in the output voltage. Two pulses of the same amplitude but 180° out of phase are applied

to the potential divider arrangement. Initially, when the shuttle is stationary, the opposite voltages

applied to the fixed beams produce an output of zero volts. However, with the slightest shuttle

movement, the capacitance balance is disturbed and a non-zero output voltage Vb results. A

demodulator is used to produce a DC voltage Vclem representing the magnitude and direction (+ or

-) of the acceleration. One of the driving signals fed to the potential divider arrangement is used

1For illustration purpose we have shown only one pair of differential capacitance in the Figure 3.2.

27

by the demodulator as a reference voltage.

The shuttle senses acceleration and converts it to a corresponding change in the differential

capacitance. Sensing circuitry then converts this capacitance change to a corresponding linear

change in output voltage. This is the basic principle behind accelerometer operation. As shown

in Figure 3.2, a typical accelerometer consists of a microresonator and some surrounding electrical

circuitry. There are three distinct regions identified on the microresonator. The shuttle is a proof

mass that moves under acceleration, combs are the parallel beams (fixed and movable) that form

differential capacitances, and the tether beams are the long, thin beams that apply a restoring

force to the displaced shuttle displacement.

The microresonator is also used in other applications such as micromachined resonant pressure

sensors [24], microelectromechanical filters [25], oscillators [26], and various signal processing cir-

cuits [27, 28].

3.3 CARAMEL Implementation

CARAMEL is a process simulator that maps particle contaminations to defective microstructures.

It is built around the tool CODEF, which is a contamination-to-defect-to-fault mapper for pure

electrical layouts [12]. Described next are the three phases of CARAMEL’s operation.

3.3.1 Process Simulation

This phase of CARAMEL maps spot contaminations to layout defects and is implemented using

a modified version of CODEF. CODEF determines the impact of a particle on the contaminated

region of the IC layout. It accepts layout information, ~ process description, and contamination

28

Shuttle Mass

Shuttle at rest

Tether beams AppliedFixed beam ,~ "~ Acceleration

~m Moving beam ~~

~ Anchors ~~ c-~c

acceleration

Demodulator DC Voltage

(b)Figure 3.2: Accelerometer basics: (a) Sensing operation of a typical accelerometer. (b) Signal sensingand processing circuitry for transducing a mass deflection into an output voltage.

29

PolyO silicon nitride

PSG--~

PolylL

micromechanicalstructure

anchor

polyO

anchor cut

polylstructure

releasedstructure

(b)

Figure 3.3: Abbreviated flow for MCNC’s MultiUser MEMS Process service. (a) Cross-sectional

view. (b) Top view (layout).

statistics for each processing step of interest. CODEF allows for the exact characteristics of the

contamination particle to be simulated including the particle’s size, density, conductivity, and fab-

rication step of introduction. Given the contamination parameters, it simulates all the fabrication

steps and creates a three-dimensional (3D) structure of the defective device. CODEF, in its unmod-

ified form, is used to analyze the effects of particles on an electrical IC layout. It utilizes a circuit

extractor which traverses the 3D structure to create a SPICE netlist. Defective circuit behavior

resulting from the contaminations can then be analyzed through SPICE simulations.

To make use of CODEF for MEMS contamination analysis, we have defined a complete MUMPs

fabrication process as a sequence of steps in the PREDITOR format [31]. The MUMPs process

is used to form micromechanical structures composed of thin films formed on the surface of the

substrate. These thin-film microstructures are called surface micromachined MEMS. Because the

resonator is a single-polysilicon structure, fabrication of the resonator utilizes only a subset of

3O

the complete three-polysilicon MUMPs process. This simplified version of the MUMPs process is

described in [32] and is shown in Figure 3.3. First, a low-stress silicon nitride is deposited on the sil-

icon substrate to provide electrical isolation between microstructureso An electrical interconnection

layer of polysilicon is then deposited and patterned. Next, a 2 #m-thick layer of phosphosilicate

glass (PSG) is deposited. PSG acts as sacrificial spacer layer for the microstructures. After con-

tact cuts are made in PSG, a 2 ttm-thick layer of polysilicon is deposited and patterned to form

a microstructure. A final wet etch in hydrofluoric acid (HF) dissolves the PSG and releases the

microstructure so that it is free to move. Contact cuts in the PSG are anchor points that fix the

microstructure to the substrate.

In CARAMEL, we have modified CODEF to handle the MUMPs "release" step (i.e. the HF etch

of PSG) as described above. Adding this step, which is not part of any standard CMOS process,

allows us to perform MUMPs process simulations using realistic contaminations.

3.3.2 Structure Extraction

The structure extraction phase produces a three-dimensional (3D) mesh representation of the de-

fective structure generated by CODEF. Mechanical simulation of the mesh (using finite element

analysis (FEA) tools like ABAQUS [13]) allows detailed analysis of the effects of contaminations

on the mechanical structure.

The process simulator creates a 3D representation of the resonator structure consisting of hierar-

chically connected layers of m~terial known ~s the Chip Data Base (CDB) [33]. CDB is created from

the MUMPS process flow and the CIF layout of the resonator. In addition, contamination particles

may be introduced at random locations in the process flow~ under control of the process simulator,

31

which can alter the final 3D structure of the resonator structure. CARAMEL creates a separate

"mesh database" from the CDB to represent the MEMS structure as a set of three-dimensional

(3D) corner-stitched rectangular regions [34]. Each region of the mesh consists of a set of material

elements. However, the mesh database differs from a traditional corner-stitched database because

elements in the former are allowed to be less than maximal width. This is a necessary feature since

the mesh used for mechanical simulation requires that each region have a single neighbor or no

neighbor along each of its edges. Each element also has associated data that describes its material

characteristics, its position in the z direction, and a flag that indicates if the base of the element is

anchored.

The MEMS meshing phase:

1. Creates a corner-stitched database containing the regions to be simulated.

2. Determines the connectivity of the regions.

3. "Splits" the mesh database to ensure that each region has only one or zero neighbors in the

x, y, and z directions.

4. Identifies the elements and generate node numbers for all the vertices of every region. (This

is required by the FEA tool for mechanical simulation.)

5. Generates a final mesh input model for FEA tool.

Mesh Database Creation

The mesh database for the resonator consists only of polysilicon one (POLY1) and defect material,

i.e., the material of the particle contamination. The mesh is created by examining every region

32

edge~--~

(b)

z split

Figure 3.4: Illustration of element splitting for mechanical mesh construction: (a) A set of regionsbefore split and (b) the same regions after x-y and z splits.

of CODEF’s CDB. Each element of POLY1 or defect material is added to the mesh database

along with a flag indicating if the element is fixed to some other material or contains free space

underneath. The flag information is used later to determine which elements are free to move.

Determining Element Connectivity

Understanding the behavior of the resonator requires mechanical simulation of its moving parts.

The moving parts of the resonator includes the mass shuttle and everything connected to the shuttle.

Because a particle contamination can alter the topography of the resonator, the resulting defective

structure must be derived from the CDB of elements. In this phase, CARAMEL determines all the

connected elements from a a user-specified element on the resonator’s layout. A simple algorithm is

employed that marks the user-specified element and continues to mark neighboring elements until

all connected regions are marked.

Element Splitting

The mechanical mesh required by FEA tools must be constructed so that adjacent regions meet

only at region vertices (nodes). Caramel must meet this adjacency constraint not only in the x and

y directions but also in the z-direction.

33

x-y splitting: In a standard corner-stitched database, a region edge can have multiple

neighboring regions or a partial neighbor as shown in Figure 3.4a. Such a situation violates

the constraint described above for the mechanical mesh. The x-y splitting operation modifies

an element so that every region edge has a single neighboring element or no neighbor at all.

Splitting is performed by analyzing the edge neighbors of every region in the CDB. For each

edge that violates the constraint, the current region is split at the neighbor’s edge. (See

Figure 3.4.)

¯ z splitting: In an analogous way, elements may need to be split in the z direction as well.

In this case, adjacent regions that have elements of differing heights must be split in the

z direction so that the resulting elements share nodes. This is accomplished using a simple

algorithm that compares the top and bottom coordinates of an element with all its neighboring

elements. If an adjacent element is contained in the current element, the current element is

split. (See Figure 3.4.) There is one exception to the z-split operation when the element

concern is fixed. Typically, split elements inherit identical properties (material, conductivity,

etc.) from the original element. In the case of fixed elements, only the newly-created bottom-

half element is fixed.

The result of CARAMEL’s extraction phase is a mechanical mesh that is directly compatible

with the FEA tool ABAQUS. The resulting mechanical mesh captures the impact of the particle

contamination on both the electrical and mechanical properties of the resonator. It should be

noted here that CARAMEL is a general tool and is therefore not restricted to the resonator but is

applicable to any generic MEMS layout. It is therefore possible to analyze the impact of particle

contaminations on a wide variety of MEMS topologies. In the next section, we examine the impact

34

of 721 contaminations on the structure and behavior of the resonator.

3.4 Simulation Results

CARAMEL was used to perform 721 unique contamination simulations of the surface microma-

chined comb-drive resonator shown in Figure 1.1. We analyzed contaminations of various sizes and

material properties. In addition, contaminations were placed at many different resonator locations

and were introduced at many different steps of the MUMPs process. Out of 721 contaminations, 263

were physically in contact with the resonator structure. The effect of each of these 263 contamina-

tions was observed at each phase of CARAMEL’s operation. The mechanical simulator ABAQUS

was also used to compute the resonant frequency of each of the mechanical meshes generated by

CARAMEL. The resonant frequency is analyzed because it is one of the crucial parameters for

determining if the resonator is functioning properly [8]. In Table 3.2, we provide details of five

representative cases of contamination analysis. Note that the first entry of Table 3.2 gives the

resonant frequency for the defect-free resonator. For each of the five defects considered, we provide

the location of the contamination (Table 3.2), the resulting 3D defective structure (Figures 3.5a,

b, c and Figures 3.6a, b), top view of the mechanical mesh (Figures 3.5a, b, c and Figures 3.6a, b),

and the resulting resonant frequency reported by ABAQUS (Table 3.2).

Comb Defect: Figure 3.5a shows the impact of a contamination located between adjacent

comb fingers. The process simulation phase of CARAMEL reveals that the contamination

welds together the two normally-moving fingers. The fixed fingers transforms the two comb

drives into a single structure thereby changing the mesh model required for mechanical simu-

lation. Mechanical simulation of a defect-free resonator requires only the shuttle and flexure.

35

Fixed combs are also simulated

(a)

Mesh changes on the flexure

(b)

Additional anchor

Figure 3.5: Three-dimensional representations of defective resonators and their corresponding me-chanical mesh for particle contaminations located (a) between comb fingers, (b) on a beam, (c) under the shuttle.

36

Contamination Diameter Density Added after Resonant % change inlocation size (#m) (kg/m °) step # frequency fz (Hz)

NoneCombBeamShuttle-1OutsideShuttle-2

2.41.52.02.01.5

23302330233023302330

Polyl depositionPSG depositionPoty0 PR StripPoly0 depositionPSG deposition

697729011668646883286977269721

+29:2%-1.61%+26.6%o%0%

Table 3.2: Five representative cases of defect analysis using CARAMEL.

Mesh remains unchanged

(a)

Mesh changes on the shuttle

(b)

Figure 3.6: Three-dimensional representations of defective resonators and their corresponding me-

chanical mesh for particle contaminations located (a) outside of the resonator’s active area and (b)

on the shuttle.

37

The result obtained from mechanical simulation shows a 33% increase in resonant frequency;

an indication that this defect has caused a catastrophic failure.

¯ Beam Defect: The impact of a defect affecting a beam of the folded-flexure is illustrated in

Figure 3.5b. Such a contamination results in a slightly heavier beam. The increased mass of

the beam decreases the resonant frequency by 1.61% . Note the meshing complexity of the

area surrounding the defect’s location. The increased "meshing" is a reflection of the number

of splits required to accurately represent the region containing the defect.

¯ Shuttle-1 Defect: A contamination that becomes lodged between the resonator and the

substrate acts as an anchor. CARAMEL handles such cases by defining an anchor element at

the contamination location during the extraction phase. The meshing phase discovers that

the contaminant is connected to the suspended structure and the substrate surface. The

resulting mesh used in the mechanical simulation therefore has an extra anchor. The impact

of such a defect is shown in Figure 3.5c. Simulation of the corresponding mesh indicates that

the resonant frequency has increased which is quite intuitive since the resonator is now fixed

to the substrate at an additional location.

¯ Outside Defect: Contaminations lying outside the structure are treated like the others,

but during the "connectivity" phase of meshing, it is determined that the contaminant is

not connected to the structure of interest. Meshes for such defects are generated but are

not simulated. For these cases the nominal values of resonant frequency are reported. An

example of this type of harmless defect is illustrated in Figure 3.6a.

¯ Shuttle-2 Defect: Figure 3.6b shows another interesting case where the contamination

becomes totally encapsulated by the shuttle mass. Process simulation indicates that a small

38

bump is formed on the shuttle surface. The creation of such a bump changes the meshing of

the affected region evidenced by the dense meshing surrounding the defect area. Mechanical

simulation reveals a very small (almost none) decrease in resonant frequency due to the

additional mass.

The five analyzed cases demonstrate CARAMEL’s ability to model and simulate a variety of

contamination effects on the MEMS microstructure. The tool can be effectively used to generate

several thousand contamination simulations under a Monte-Carlo mode of operation. In Monte

Carlo mode, contaminations are introduced into the process at random steps, with random sizes,

and at random locations in the layout. Such analysis can produce the full spectrum of MEMS

failure modes. The observed deviations in behavior can then be systematically categorized under

various fault classes. These fault classes will then be mapped to appropriate fault models at the

higher level of abstraction. In other words, such an analysis provides a technique for formulating

behavioral models for defective beams, gaps, shuttles, etc.. We have performed 721 simulations

using CARAMEL to illustrate its Monte-Carlo mode of operation. Process simulation showed that

only 263 contaminants were in physical contact with the microstructure. Mechanical simulation

showed that 65 of the 263 simulations produced defective structures that significantly altered the

resonant frequency of the resonator.

Table 3.3 presents a coarse fault categorization of the 263 defects. The three broad fault classes

are identified as catastrophic, parametric, and harmless. Catastrophic defects are those defects

that cause more than a 30% change in the resonant frequency fx. Defects that cause a frequency

deviation between 5% and 30% are termed parametric. Defects are termed harmless if they have

none or negligible affect on

39

Number of DeviationsDefect type occurrences in f~ CommentsCatastrophic 47 Af:~ > 30% Resonator functionality is

completely destroyed.Parametric 18 5% _< Af~ _< 30% Resonator still performs but

with large deviations fromthe desired functionality.

Harmless 198 ~f~ < 5% No impact on the functionality.

Table 3.3: Coarse fault categorization based on deviation in resonant frequency fx.

More details about the defects are provided in Table 3.4. It shows the occurrences of defects with

respect to its process step of introduction. This information indicates which MUMPs process steps

are most vulnerable to particle contaminations. It can be seen that step 8 (i.e. Etch Poly0 resist)

is highly prone to catastrophic failures. This result is quite intuitive. Poly0 forms the base of the

resonator while the "suspended’ structure is built from Polyl. Hence, any contamination occurring

on Poly0 can bind the two layers and impede resonator movement. This situation often leads to

catastrophic failures. Thus, the Poly0 deposition and development phases of fabrication are quite

vulnerable to particle contaminations. As indicated in Table 3.4, contaminations introduced during

this phase (steps 3 to 12) lead to over 60% of the catastrophic failures observed.

Table 3.5 shows the spatial distribution of the contaminations over the resonator layout that

caused catastrophic and parametric failures. It can be observed that the comb fingers are the most

defect-prone region in the resonator causing around 43% of the total catastrophic faults. Defects

affecting the combs have resulted in significant changes in the resonant frequency. The shuttle, due

to its relatively large area, is quite susceptible to being "hit" by a contamination; however for the

same reason the changes in the shuttle do not affect the resonator operation. The smaller surface

area of the folded-flexure beams, on the other hand, is less likely to be affected. However, when

the beams are impacted, their behavior is significantly disturbed.

4O

Process Number Of % of occurrences Catastrophic Parametricstep occurrences causing faults faults faults

131 (Initialize)2 (Nitride deposit)3 (Poly0 deposit)4 (Litho spinon Poly0)5 (Litho expose Poly0)6 (Litho develop Poly0)7 (Etch Poly0)8 (Etch Poly0 resist)9 (PSG deposit)12 (Litho develop PSG)13 (Etch PSG)19 (Etch Anchorl resist)20 (Polyl deposit)21 (Litho spinon Polyl)22 (Litho expose Polyl)23 (Litho develop Polyl)24 (Etch Polyl)25 (Etch Polyl resist)48 (PSG release)

916334414221331292622193824

30.844.468.833.366.710025.085.71000.00.00.012.913.811.64.610.515.84.2

41122

4431

Total defects 263 -- 47 18

Table 3.4: Defect categorization based on the MUMPs process step for which the contamination

was introduced.

ContaminationlocationShuttleComb fingersFolded flexure

Catastrophicfaults (%)31.942.625.5

Parametricfaults (%)33.316.750.0

Totalfaults (%)32.335.432.3

Table 3.5: Spatial distribution of faults in the microresonator.

41

It is also important to note that all the catastrophic failures due to shuttle contaminations occur

before deposition of the Polyl layer. These conta~ninations are lodged between the Polyl and Poly0

layers creating an anchor effect. Conversely, contaminations occurring on the shuttle after/during

Polyl deposition do not cause appreciable change in the resonant frequency. Thus, contamination

impact on resonator behavior is a function of both the location and process step in which it is

introduced.

Following are the main observations that can be drawn from the simulation results :

1. Particle contaminations have a large impact of the resonator functionality and the spectrum

of faulty behaviors can be discovered through contamination and FEM simulations.

2. A systematic classification of the faulty behavior can be made by mapping various fault

classifications to a higher level of abstraction.

3. Impact of particle contamination is a function of both the location and process step in which

it is introduced.

4. The comb drive is the most defect prone region of the resonator and the Poly0 deposition

process is the most vulnerable to a particle contamination that leads to catastrophic faults.

42

Chapter 4

Conclusions and Future Work

Our long term goal is to develop a comprehensive MEMS testing methodology. The first step in

achieving this goal requires the generation of effective and accurate fault models for MEMS. Our

approach towards developing realistic fault models has resulted in a systematic analysis of a large

spectrum of faulty MEMS behaviors that can be further mapped to fault models at a higher level

of abstraction° This was accomplished by:

Performing process contamination simulations to produce a large variety of defective three-

dimensional microstructures.

Mapping the effects of structural damages to faulty MEMS behavior.

Performing systematic classification of faulty MEMS behavior.

We have automated the contamination analysis process for microelectromechanical layout using

our tool CARAMEL. CARAMEL can be effectively used to directly investigate the faulty behavior

of defective MEMS structures. Here, we have illustrated CARAMEL’s use on a folded-flexure res-

43

onator. Simulation of 263 different defects indicates that spot contaminations can have a significant

impact on the behavior of the resonator. Moreover, the results show that the impact of a particular

defect is highly dependent upon where it is located in the resonator’s layout and when it is intro-

duced into the manufacturing process. The results obtained were classified under three broad fault

classes: catastrophic, parametric, and harmless. The simulation runs show that CARAMEL can

be effectively used to generate fault models for MEMS.

We plan to conduct a large number of random process simulations using real contamination

data at every step of the manufacturing process in order to produce a large spectrum of defective

MEMS structures. Low-level mechanical simulations will be performed to categorize these defective

structures into a smaller set of faulty behavior classes. These fault classes will form the basis of

MEMS fault models and serve as our first step in developing a comprehensive testing methodology

for such systems.

There are several open research areas that we plan to address concerning the development of a

MEMS testing methodology.

Process Simulation: CARAMEL uses the process simulator CODEF to simulate MEMS

designs. The accuracy of the process simulation is somewhat limited since it does not account

for the manufacturing characteristics inherent to MEMS processes. The preditor format used

for defining the process sequence should be modified to capture the manufacturing details crit-

ical to micromachining. For example, the type of etching (isotropic, directional, anisotropic,

etc.), the variation of etching rates with respect to plane orientation, selectivity, and as-

pect ratios are quite crucial for bulk micromachining and need to be accounted for. What

is required is a process simulator that can simulate both bulk- and surface-micromachining

44

processes accurately. The new process simulator should be able to identify and simulate the

effects from stiction, buckling, etc. that are present in the microstructures, in addition to the

problems caused by particulates.

¯ Structure extraction: Our testing methodology depends on gaining a comprehensive un-

derstanding of the effects of spot contaminations. Process simulations of random contamina-

tions will create a large variety of perturbed structures. In order to classify these defective

structures into an appropriate fault model, we have developed CARAMEL that automatically

derives (when possible) the structural and material parameters that define behavior. Cur-

rently, we are addressing the following issues to improve CARAMEL for MEMS fault model

generation:

1. Mesh Optimization: Presently, the mechanical mesh generated by CARAMEL is not

optimized in any way. It contains many artifacts from the original process simulation

that add to the size and complexity of the mesh. We plan to optimize the mesh in order

to accelerate the mechanical simulation phase.

2. Automatic Fault Categorization: We plan to have CARAMEL sort through the

large amount of simulated data and extract the unique faults. This will reduce the

amount of mechanical simulation required for fault categorization.

3. Simulation Complexity: We are currently simulating static structures that do not

change during the course of simulation. Static structures do not capture a number of

interesting and possibly important defects. For example, a particle located between but

not touching a comb finger may limit the motion of the resonator. This and other types

of interactions will require more knowledge of the material properties of the contaminants

45

but can be modeled using our approach.

¯ Fault Model Verification: We plan to verify the accuracy of the developed fault models

using actual fabricated systems. Our approach will use defective devices to measure and

compare actual and predicted faulty behaviors. For example, a number of contaminations

can be introduces during the structure release process and their impact can be observed by

measuring the device parameters. Also, the surface micromachined microstructures can be

manually pressed down to cause stiction. The effect of stiction can then be measured by

evaluating the faulty device.

¯ Test Methodology Grading: Given an accurate fault model, one can measure the effective-

ness of any given test methodology. Presumably, the fault model will provide an enumeration

of possible faulty behaviors along with their likelihood of occurrence. Coverage figures for

current MEMS testing methods can then be determined through systematic fault simulations.

¯ Test Methodology Development: Any possible shortcomings in current testing method-

ologies will be exposed by test methodology grading. Even if shortcomings do not exist,

knowledge about MEMS faulty behavior may lead to more effective test or design techniques

that reduce cost and increase quality.

¯ Design For Testability Techniques: Results obtained by performing contamination sim-

ulations indicate the most defect-prone parts of the MEMS design. Such guidelines can be

used to make appropriate structural design changes to make defects less likely or less harmful.

For example, a comb finger gap can be made wider to reduce the effect of contaminations

trapped between fixed and moving comb fingers. CARAMEL can also be used to compare

and contrast different MEMS designs based on process contamination simulations.

46

¯ Analytical Fault Model Generation: We obtain a large spectrum of possible defects

and corresponding faulty MEMS behavior using CARAMEL. These observations can then be

related to analytical models to define the faults at a higher level of abstraction. For example,

the shuttle-2 defect shown in Table 3.2 can be defined as a small increase in shuttle mass ms

and therefore in ro, x as shown in equation a.a. The change in mz will in turn cause change in

resonant frequency fx according to equation 3.9. Thus, shuttle-2 defect can be represented

by equation 3.9 with increased rex. Such an analysis will be complicated for the defects like

comb defects where the analytical model itself needs to be changed.

¯ MEMS Diagnosis: Finally, more effective testing methodologies will undoubtedly lead to

better diagnosis. We plan to create formal links between observed faulty behavior, testing

methods, and process contaminations for diagnostic purposes.

47

Bibliography

[1] C. Spangler and C.J. Kemp, "A Smart Automotive Accelerometer with On-chip Airbag DeploymentCircuits", Technical Digest on Solid-State Sensor and Actuator Workshop, pp. 211-214, June 1996.

[2] C. Ajluni, "Pressure Sensors Strive to Stay on Top", Electronic Design, pp. 67-74, Oct. 1994.

[3] C.C. Betty "A Chronology of Thermal Ink-jet Structures", Technical Digest on Solid-State Sensor andActuator Workshop, pp. 200-204, June 1996.

[4] R. Thomas, "Micromachining: Assessing the Market for an Emerging Technology", Microsystem Tech-nology News, Dec. 1997.

[5] G.K. Fedder, "Future Trends in MicroElectroMechanical Systems (MEMS)", Memo., ECE Department,Carnegie Mellon University, Sept. 1996.

[6] R.E. Harr, "Design for Large Scale Integration of Mixed Technology Microdevices", http://eto.sysplan.com/ETO/CompCAD, R 8J D Program of DARPA, Electronics Technology Office, 3701, North FairfaxDrive, Arlington, VA 22203-1714, 1996.

[7] A. Kolpekwar, R.D. Blanton "Development of MEMS Testing Methodology", In the Proc. of Interna-tional Test Conference, pp. 923-931, Washington D.C., Nov. 1997.

[8] A. Kolpekwar, R.D. Blanton and D. Woodilla "Failure Modes for Stiction in Surface-MicromachinedMEMS", In the Proc. of International Test Conference, Washington D.C., Nov. 1998.

[9] A. Kolpekwar, C. Kellen and R.D. Blanton, "Fault Model Generation for MEMS Using CARAMEL",In the Proc. of International Test Conference, Washington D.C., Nov. 1998.

[10] A. Kolpekwar, C. Kellen and R.D. Blanton, "Fault Model Generation for MEMS", In the Proc. of FirstInternational Conference on Modeling and Simulation of Microsystems, Semiconductors, Sensors andActuators, Santa Clara, CA, April 1998.

[11] J. Khare, W. Maly and N. Tiday, "Fault Characterization of Standard Cell Libraries Using InductiveContamination Analysis (ICA)", In the Proc. of the 14th VLSI Test Symposium, pp. 405-413, April1997.

[12] J. Khare and W. Maly, "From Contamination to Defects, Faults and Yield Loss", Kluwer AcademicPublishers, Boston, 1996.

[13] ABAQUS Web Page, http://www.hks.com, Hibbit, Karlsson & Sorensen, Inc., 1080 Main Street,Pawucket, RI 02860.

[14] R.D. Blanton, G.K. Fedder and T. Mukherjee, "Hierarchical Design and Test of MEMS", MicrosystemTechnology News, Spring 1998.

[15] G.K. Fedder, S. Iyer and T. Mukherjee, "Automated Optimal Synthesis of Microresonators", TechnicalDigest of IEEE International Conference on Solid State Sensors and Actuators, Vol. 2, pp. 1109-1112,June 1997.

48

[16] T. Mukherjee and G.K. Fedder, "Structured Design of Microelectromechanical Systems", In the Proc.of the 3~th Design Automation Conference, pp. 687-685, June 1997.

[17] D.A. Koester, R. Mahadevan and K.W. Markus, "Multi-user MEMS Processes (MUMPS) Introductionand Design Rules", http://mems.mcnc.org/mumps.html, Oct. 1994.

[18] W.C. Tang, T. C.-C. Nguyen, M.W. Judy and R.T. Howe, "Electrostatic Comb Drive of Lateral Polysil-icon Resonators", Sensors and Actuators A, Vol. 21, Nos. 1-3, pp. 328-331, Feb. 1990.

[19] G.K. Fedder, "Simulation of Microelectromechanical Systems", PhD thesis, University of California atBerkeley, Sept. 1994.

[20] Z. Zhang and W.C. Tang, "Viscous Air Damping in Laterally Driven Microresonators", In the Proc. ofthe IEEE Micro Electro Mechanical Systems Workshop, pp. 199-204, Jan. 1994.

[21] W.A. Johnson and L.Ko Warne, ’%lectrophysics of Micromechanical Comb Acutators", Journal ofMicroelectromechanical Systems, Vol. 4, No. 1, pp. 49-59, March 1995.

[22] Analog Devices Inc.,"ADXL Series Accelerometer Datasheets", http://www.analog.com, One Technol-ogy Way, P.O. Box 9106, Norwood, MA 02062-9106, 1996.

[23] Motorola,"MMAS40G10D Accelerometer Datasheet", http://mot-sps.com/senseon/ad.html, Semicon-ductor Products Sector, 6501, William Cannon Drive, Austin, TX, 1996o

[24] C.J. Waltham, et al, "A Laterally Driven Micromachined Resonant Pressure Sensor", Transducers’gS,pp. 586-589, Stockholm, June 1995.

[25] L. Lin, C.T.-C. Nguyen, R.T. Howe and A.P. Pissano, " Microelectromechanical Filters for SignalProcessing", In the Proc. of IEEE Micro Electro Mechanical Systems, pp. 226-231, Feb. 1992.

[26] C.T.-C. Nguyen and R. T. Howe, "CMOS Micromechanical Resonator Oscillator", In the Proc. of IEEEInternational Electron Devices Meeting, pp. 199-202, 1993.

[27] P~.T. Howe and C.T.-C Nguyen, "Micromechanical resonators for Frequency References and SignalProcessing", In the Proc. of IEEE International Electron Devices Meeting, San Francisco, CA, Dec.1994.

[28] C.T.-C. Nguyen, "Micromechanical Signal Processors", Ph.D. Thesis, University of Berkeley, 1994.

[29] M. A. Mignardi, "Digital Micromirror Array for Projection TV", Solid State Technology, Vol. 37, No.7, pp. 63-64, July 1994.

[30] G.K. Fedder, S. Santhanam, M.L. Reed, S. Eagle, D. Guillou, M.S.C. Lu and L.I~. Carley, "LaminatedHigh-Aspect-Ratio Microstructures in a Conventional CMOS Process", In the Proc. of the IEEE MicroElectro Mechanical Systems Workshop, pp. 13-18, Feb. 1996.

[31] D.M.H. Walker, C.S. Kellen and A.J. Strojwa& "The PREDITOR Process Editor and Statistical Sim-ulator" In the Proc. of the 1991 International Workshop on VLSI Process and Device Modeling, pp.120-123, May 1991.

[32] G.K. Fedder and T. Mukherjee, "Physical Design for Surface-Micromachined MEMS", In the Proc. ofthe 5th ACM/SIGDA Physical Design Workshop, pp. 53-60, April 1996.

[33] D.M.H. Walker, C.S. Kellen, D.M. Svoboda, and A.J. Strojwas, "The CDB/HCDB SemiconductorWafer Representation Server", IEEE Transactions on Computer-Aided Design o/Integrated Circuitsand Systems, Vol. 12, No.2, pp. 283-95, Feb. 1993.

[34] J. K. Ousterhout,"Corner Stitching: A Data-Structuring Technique for VLSI ", IEEE Transactions onComputer-Aided Design of Integrated Circuits and Systems, Vol. 3, No. 1, pp. 87-100, Jan. 1984.

49

[35] M. Lubaszewski, E.F. Corn, and B. Courtois, "Microsystem Testing: an Approach and Open Problems’~,In the Proc. of Design and Test in Europe, Paris, Feb. 1998.

[36] N. Nagi, A. Chatterjee, A. Balidava. J.A. Abraham, "Fault based Automatic Test Generator for LinearAnalog Circuits", IEEE/ACM International Computer-Aided Design Conference, Santa Clara, USA,1993.

[37] W. Vermeiren, B. Straube, A. Holubek, "Defect-Oriented Experiments in Fault Modeling and FaultSimulation of Microsystem Components", In the Proc. of European Design and Test Conference, March1996.

[38] T. Olbrich, A. Richardson, W. Vermeiren, B. Straube, "Integrating Testability into Microsystems",Microsystem Technologies, pp. 72-79, Feb. 1997.

[39] H.V. Allen, S.C. Terry and D.W. DeBruin, "Accelerometer Systems with Self-testable Features", Sensorsand actuators A.20, pp. 153-161, 1989.

[40] L. Zimmermann, "Airbag Application: A Microsystem including a Silicon Capacitive Accelerometer,CMOS Switched-Capacitor Electronics and True Self-Test Capability", Sensors and actuators A 46-47,pp. 190-195, 1995.

[41] J. Bryzek, "New Technologies for Silicon Accelerometers Enable Automotive Applications", CMOSSwitched-Capacitor Electronics and True Self-Test Capability", Lucas Nova Sensors SAE 920474, pp.25-32.

[42] R. Maboudian and R.T. Howe "Critical Review: Stiction in Surface Micromechanical Structures",Journal of Vacuum Science and Technology, August 1996.

[43] R. Harjani and B. Vinnakota, "Analog Circuit Observer Blocks’~ IEEE Transaction on Circuits andSystems II Analog and Digital Signal Processing, Vol. 44 No. 3, pp. 154-163, 1997.

[44] T. Olbrich, D.A. Bradley and A.M.D. Richardson "Built-In Self and Diagnostic for Microsystems", 27thInternational Symposium on Advanced Transpor~ Applications - Aachen Germany, Proc. on dedicatedcon#rence on Mechatronics, pp. 511-519, 1994o

[45] T. Olbrich, A.M.D. Richardson and D.A. Bradley, "BIST & Diagnostics in Safety Critical Microsys-terns using Reliability Indicators", 5th European Synopsium on Reliability o] Electronic Devices FailurePhysics and Analysis, Glasgow, pp. 4-7~ Oct. 1994.

[46] K. Arabi, B. Kaminska and J. Rzeszut, "BIST for D/A and A/D Converters", IEEE Design and Testo] Computers, Vol. 13, No.4, pp. 40-49, 1996.

[47] T. Core, R.S. Payne, D. Quinn, S. Sherman and W.K. Wong, "Integrated Complete, Affordable Ac-celerometer for Air-Bag Applications", In the Proc. o] Sensors Expo, pp. 204-1, Chicago IL, 1991.

[48] F. Purahmadi, L. Christel and K. Petersen, "Silicon accelerometer with new thermal self-test mecha-nism", Technical Digest IEEE Solid-State Sensor and Actuator, SC, USA, June 1992.

[49] T. Olbrich, D. Bradley and A.M.D. Richardson, "Self-Test And Diagnostics for Smart Sensors in Au-tomotive Applications", IEE Colloquium on Automotive Sensors, Solihull, UK, Sept. 1994.

[50] W. Vermeiren, B. Straube and G. Elst, "Microsystem Fault Simulation -- Aspects and Requirements",In the Proc. of Micro System -- 94, Berlin, Germany, Oct. 1994.

[51] C. Kantzsch, M. Chownetz, W. Wolz and K.D. Muller Glaser, "Micro System Design for Testability",Microsystem Technologies --91, Berlin, Germany, 1991.

5O

Step ~ Name Description

34567891011121314151617181920212223242526272829303132333435363738394O4142434445464748

InitializeNitride Deposit

Deposit Poly0Litho SpinOnLitho ExposeLitho DevelopEtch Poly0Etch Poly0 ResistDeposit OxidelLitho SpinOnLitho ExposeLitho DevelopEtch OxidelEtch Dimplel ResistLitho SpinOnLitho ExposeLitho DevelopEtch OxidelEtch Anchorl ResistDeposit PolylLitho SpinOnLitho ExposeLitho DevelopEtch PolylEtch Polyl ResistDeposit Oxide2Litho SpinOnLitho ExposeLitho DevelopEtch Oxide2Etch Oxide2 ResistLitho SpinOnLitho ExposeLitho DevelopEtch Oxidel2Etch Anchor2 ResistDeposit Poly2Litho SpinOnLitho ExposeLitho DevelopEtch Poly2Etch Poly2 ResistDeposit MetalLitho SpinOnLitho ExposeLitho DevelopEtch Metal ResistRelease

Heavy doping of wafer surface using phoshous (POC13)Deposition of 600 nm silicon nitride layer using a low stress,low pressure chemical vapor deposition (LPCVD)Deposition of 500 nm polysilicon film (poly0) using LPCVDCoating of wafer with photoresist (PR)Exposure of PR with CPZ maskPR developmentEtching of poly0 layer with reactive ion etch (RIE)PR stripDeposition of phosphosilicate glass (PSG), oxide1 by LPCVDCoating of wafer with PRExposure of PR with COS maskPR developmentEtching of PSG layer with RIE to form DIMPLESPR stripCoating of wafer with the PRExposure of PR with COF maskPR developmentEtching of PSG layer in RIE to form ANCHORS (anchor1)PR stripDeposition of 2 #m of polysilicon blanket (polyl) using LPCVDCoating of wafer with PRExposure of PR with CPS maskPR developmentEtching of Polyl layer in RIEPR stripDeposition of 0.75/zm of PSGCoating of wafer with PRExposure of PR with COT maskPR developmentEtching of oxide2 with RIE to form polyl-poly2 viasPR stripCoating of wafer with PRExposure of PR with COL maskPR developmentEtching of oxide1 and oxide2 with RIE to form ANCHORS (anchor2)PR stripDeposition of 1.5 #m of polysilicon blanket (poly2)Coating of wafer with PRExposure of PR with CPT maskPR developmentEtching of Poly2 with RIEPR stripDeposition of metal (gold with adhesion layer)Coating of wafer with PRExposure of PR with CCM maskPR developmentRemoving the unwanted metal and PR in a solvent bathRelease of sacrificial PSG layer by immersing in 49% HF solution

Table 4.1: Appendix A - The complete MCNC MUMPs process described in the PREDITORformat. 51