University of Ljubljana
`Faculty for mathematics and physics
`Department of physics
`
`Seminar
`
`MEMS ACCELEROMETERS
`
`Author: Matej Andrejašiˇc
`
`Mentor: doc. dr. Igor Poberaj
`
`Marec 2008
`
`Abstract:
`
`MEMS accelerometers are one of the simplest but also most applicable micro-electromechanical
`systems. They became indispensable in automobile industry, computer and audio-video tech-
`nology. This seminar presents MEMS technology as a highly developing industry. Special
`attention is given to the capacitor accelerometers, how do they work and their applications. The
`seminar closes with quite extensively described MEMS fabrication.
`
`1
`
`Page 1 of 17
`
`HAPTIC EX2009
`
`

`

`1 Introduction
`
`An accelerometer is an electromechanical device that measures acceleration forces. These
`forces may be static, like the constant force of gravity pulling at our feet, or they could be
`dynamic - caused by moving or vibrating the accelerometer. There are many types of ac-
`celerometers developed and reported in the literature. The vast majority is based on piezo-
`electric crystals, but they are too big and to clumsy. People tried to develop something smaller,
`that could increase applicability and started searching in the field of microelectronics. They
`developed MEMS (micro electromechanical systems) accelerometers.
`The first micro machined accelerometer was designed in 1979 at Stanford University, but it
`took over 15 years before such devices became accepted mainstream products for large volume
`applications [1].
`In the 1990s MEMS accelerometers revolutionised the automotive-airbag-
`system industry. Since then they have enabled unique features and applications ranging from
`hard-disk protection on laptops to game controllers. More recently, the same sensor-core tech-
`nology has become available in fully integrated, full-featured devices suitable for industrial
`applications [2].
`Micro machined accelerometers are a highly enabling technology with a huge commercial
`potential. They provide lower power, compact and robust sensing. Multiple sensors are often
`combined to provide multi-axis sensing and more accurate data [3].
`
`2 MEMS technology
`
`What could link an inkjet printer head, a video projector DLP system, a disposable bio-analysis
`chip and an airbag crash sensor - yes, they are all MEMS, but what is MEMS? Micro Electro
`Mechanical Systems or MEMS is a term coined around 1989 by Prof. R. Howe [2] and others to
`describe an emerging research field, where mechanical elements, like cantilevers or membranes,
`had been manufactured at a scale more akin to microelectronics circuit than to lathe machining.
`It appears that these devices share the presence of features below 100µm that are not ma-
`chined using standard machining but using other techniques globally called micro-fabrication
`technology. Of course, this simple definition would also include microelectronics, but there
`is a characteristic that electronic circuits do not share with MEMS. While electronic circuits
`are inherently solid and compact structures, MEMS have holes, cavity, channels, cantilevers,
`membranes, etc, and, in some way, imitate ‘mechanical’ parts. The emphasis on MEMS based
`on silicon is clearly a result of the vast knowledge on silicon material and on silicon based
`microfabrication gained by decades of research in microelectronics. And again, even when
`MEMS are based on silicon, microelectronics process needs to be adapted to cater for thicker
`layer deposition, deeper etching and to introduce special steps to free the mechanical structures.
`MEMS needs a completely different set of mind, where next to electronics, mechanical and ma-
`terial knowledge plays a fundamental role. Then, many more MEMS are not based on silicon
`and can be manufactured in polymer, in glass, in quartz or even in metals...[2].
`The development of a MEMS component has a cost that should not be misevaluated and
`the technology has the possibility to bring unique benefits. The reasons that prompt the use of
`MEMS technology are for example miniaturization of existing devices, development of new de-
`
`2
`
`Page 2 of 17
`
`

`

`vices based on principles that do not work at larger scale, development of new tools to interact
`with the micro-world. Miniaturization reduces cost by decreasing material consumption. It also
`increases applicability by reducing mass and size allowing to place the MEMS in places where
`a traditional system doesn’t fit. A typical example is brought by the accelerometer developed
`as a replacement for traditional airbag triggering sensor also used in digital cameras to help sta-
`bilize the image or even in the contact-less game controller integrated in the latest handphones.
`Another advantage that MEMS can bring relates with the system integration. Instead of having
`a series of external components (sensor, inductor...) connected by wire or soldered to a printed
`circuit board, the MEMS on silicon can be integrated directly with the electronics [2]. These
`so called smart integrated MEMS already include data acquisition, filtering, data storage, com-
`munication, interfacing and networking [4]. As we see, MEMS technology not only makes the
`things smaller but often makes them better.
`The MEMS component currently on the market can be broadly divided in six categories
`(Table 2.1), where next to the well-known pressure and inertia sensors produced by different
`manufacturer like Motorola, Analog Devices, Sensonor or Delphi we have many other products.
`The micro-fluidic application are best known for the inkjet printer head popularized by Hewlett
`
`Product category
`
`Pressure sensor
`
`Inertia sensor
`
`Microfluidics
`bioMEMS
`
`/
`
`Examples
`Manifold pressure (MAP),
`blood pressure..
`Accelerometer, gyroscope, crash sensor...
`Inkjet printer nozzle, micro-bio-analysis sys-
`tems, DNA chips...
`
`tire pressure,
`
`.
`
`Micro-mirror array for projection (DLP),
`micro-grating array for projection (GLV),
`Optical MEMS /
`optical fibre switch, adaptive optics...
`MOEMS
`High Q-inductor, switches, antenna, filter..
`RF MEMS
`Relays, microphone, data storage, toys...
`Others
`Table 2.1; MEMS products examples. The MEMS component currently on the market can be broadly divided in
`six categories [2].
`
`.
`
`Packard, but they also include the growing bioMEMS market with micro analysis system like
`the capillary electrophoresis system from Agilent or the DNA chips. Optical MEMS (MOEMS)
`includes the component for the fibre optic telecommunication like the switch based on a moving
`mirror produced by Sercalo. Moreover MOEMS deals with the now rather successful optical
`projection system that is competing with the LCD (liquid crystal display) projector. RF (radio
`frequency) MEMS is also emerging as viable MEMS market. Next to passive components like
`high-Q inductors produced on the IC surface to replace the hybridized component as proposed
`by company MEMSCAP we find RF switches and soon micromechanical filters. But the list
`does not end here and we can find micromachined relays (MMR) produced for example by
`Omron, HDD (hard disk drive) read/write head and actuator or even toys, like the autonomous
`micro-robot EMRoS produced by EPSON [2].
`
`3
`
`Page 3 of 17
`
`

`

`3 MEMS accelerometers
`3.1 The basics
`There are many different ways to make an accelerometer. Some accelerometers use the piezo-
`electric effect - they contain microscopic crystal structures that get stressed by accelerative
`forces, which causes a voltage to be generated. Another way to do it is by sensing changes in
`capacitance [3]. This seminar is focused on the latter.
`Capacitive interfaces have several attractive features. In most micromachining technolo-
`gies no or minimal additional processing is needed. Capacitors can operate both as sensors and
`actuators. They have excellent sensitivity and the transduction mechanism is intrinsically insen-
`sitive to temperature. Capacitive sensing is independent of the base material and relies on the
`variation of capacitance when the geometry of a capacitor is changing. Neglecting the fringing
`effect near the edges, the parallel-plate capacitance is [4]:
`
`(1)
`
`,
`
`1 d
`
`= A
`
`A d
`
`C0 = 0
`
`where A = 0 A and A is the area of the electrodes, d the distance between them and the per-
`mittivity of the material separating them. A change in any of these parameters will be measured
`as a change of capacitance and variation of each of the three variables has been used in MEMS
`sensing. For example, whereas chemical or humidity sensor may be based on a change of ,
`accelerometers have been based on a change in d or in A. If the dielectric in the capacitor is air,
`capacitive sensing is essentially independent of temperature but contrary to piezoresitivity, ca-
`pacitive sensing requires complex readout electronics. Still the sensitivity of the method can be
`very large and, for example, Analog Device used for his range of accelerometer a comb capac-
`itor having a suspended electrode with varying gap. Measurement showed that the integrated
`electronics circuit could resolve a change of the gap distance of only 20pm, a mere 1/5th of the
`silicon inter-atomic distance [2].
`Typical MEMS accelerometer is composed of movable proof mass with plates that is at-
`tached through a mechanical suspension system to a reference frame, as shown in Figure 3.1.
`Movable plates and fixed outer plates represent capacitors. The deflection of proof mass is
`measured using the capacitance difference [4]. The free-space (air) capacitances between the
`movable plate and two stationary outer plates C1 and C2 are functions of the corresponding
`displacements x1 and x2:
`
`= A
`
`= C0 + ∆C.
`
`(2)
`
`1 x
`
`C2 = A
`
`= C0 − ∆C,
`
`= A
`
`1 x
`
`C1 = A
`
`1
`1
`d − x
`d + x
`1
`2
`If the acceleration is zero, the capacitances C1 and C2 are equal because x1 = x2. The proof
`mass displacement x results due to acceleration. If x (cid:54)= 0, the capacitance difference is found
`to be
`C2 − C1 = 2∆C = 2 A
`x
`d2 − x2 .
`Measuring ∆C, one finds the displacement x by solving the nonlinear algebraic equation
`∆Cx2 + Ax − ∆Cd2 = 0.
`
`(3)
`
`(4)
`
`4
`
`Page 4 of 17
`
`

`

`This equation can be simplified. For small displacements, the term ∆Cx2 is negligible. Thus,
`∆Cx2 can be omitted. Then, from
`
`x ≈ d2
`∆C
`C0
` A
`one concludes that the displacement is approximately proportional to the capacitance difference
`∆C [4].
`
`∆C = d
`
`(5)
`
`Figure 3.1; Accelerometer structure. Proof mass is attached through springs (kS: spring constant) at substrate. It
`can move only up and down. Movable and fixed plates construct capacitors [4].
`As one can see in the Figure 3.1, every sensor has a lot of capacitor sets. All upper capac-
`itors are wired parallel for an overall capacitance C1 and like wise all lower ones for overall
`capacitance C2, otherwise capacitance difference would be negligible to detect. Equation 5
`now doesn’t hold true just for one pair of capacitors, but for all system. Let’s see now how
`does a simplified electric circuit, that measures capacitance change, look like (Figure 3.2). As
`an example we will describe Analog Devices accelerometer ADXL05 [5], that has 46 pairs of
`capacitors. Sensor’s fixed plates are driven by 1MHz square waves with voltage amplitude V0
`coming out of oscillator. Phases of the square waves that drives upper and lower fixed plates
`differs for 180◦. One can picture to himself this hole system as a simple voltage divider whose
`output goes forward through buffer and demodulator. First of all we are interested in voltage
`output Vx, that is actually the voltage of the proof mass. It holds true that
`(Vx + V0)C1 + (Vx − V0)C2 = 0
`and if we use Equations 2 and 5 we get for voltage output
`C2 − C1
`C2 + C1
`
`Vx = V0
`
`(6)
`
`(7)
`
`V0.
`
`x d
`
`=
`
`5
`
`Page 5 of 17
`
`

`

`Vx is square wave with the right amplitude proportional to acceleration. We also can’t just
`simply use this output signal, because it is weak and noisy [6]. When there is no acceleration
`(a1 = 0), the proof mass doesn’t move, and therefore, the voltage output is zero. If we accelerate
`the sensor (a1 > 0), the voltage output Vx changes proportional to alternating voltage input V0
`(Equation 7). To avoid signal attenuation, we read Vx with voltage follower (buffer), therefore
`signal Vy is actually Vx multiplied by 1. If we inverse the acceleration (a1 < 0), signals Vx
`and Vy get negative sign. Demodulator then gives us the sign of the acceleration, because it
`multiplies the input signal Vy with the square waves V0 coming from oscillator. Now we finally
`have voltage output Vout with the right sign of acceleration and the right amplitude.
`
`Figure 3.2; a) Electric circuit that measures acceleration through capacitor changes. b) If acceleration is zero,
`voltage output is also zero. c)→ e) When acceleration isn’t zero, we get with the voltage follower square wave
`with the right amplitude and after demodulator voltage output Vout with the right amplitude and the right sign
`[5,6,7].
`
`For an ideal spring, according to Hook’s law, the spring exhibit a restoring force FS which
`is proportional to the displacement x. Thus, FS = kSx, where kS is the spring constant. From
`Newton’s second law of motion, neglecting the air friction (which is negligibly small), the
`following differential equation results ma = md2x/dt2 = kSx [4]. Thus, the acceleration, as a
`function of the displacement, is
`
`kS
`m
`Then, making use of Equation 7, the acceleration is found to be proportional to voltage output
`
`a =
`
`x.
`
`(8)
`
`a =
`
`kSd
`mV0
`
`Vx.
`
`6
`
`(9)
`
`Page 6 of 17
`
`

`

`There are some interesting facts and numbers we can state. The mass of the proof mass
`mentioned above is approximately 0.1µg, the smallest detectable capacitance change is ≈ 20aF
`and gaps between capacitor plates are approximately 1.3µm [8].
`This was the simplest example of one axis accelerometer. It’s capacitance changes due to
`changes of distance d between capacitor plates (Equation 1). If one includes sets of capacitors
`turned in perpendicular directions, one can get two axis or even three axis accelerometer (Figure
`3.3).
`
`Figure 3.3; a) 3D accelerometer structure. It has three different sensors for x-/y-/z-axis acceleration and three
`different electronic circuitry for each axis [9]. b) 3D accelerometer structure without electronics. All three sensors
`are linked with the same proof mass [4].
`
`In the last few years scientists came up with some new ideas that can be used at MEMS
`accelerometers. Dr. Richard Waters (Space and Naval Warfare Systems Center San Diego -
`SPAWAR) was able to envision an accelerometer based on Fabry-Perot interferometer technol-
`ogy that could offer equal or greater performance at a lower cost than those currently in use.
`"In fact, it was a device that achieved world-record sensitivity right out of the box," said Brad
`Chisum, formerly of SPAWAR [10].
`Let’s see some of the important parameters accelerometer have. First and foremost, one
`must choose between an accelerometer with analog output or digital output. Then there is
`number of axis and measurement range. A ±1.5g accelerometer will be more than enough for
`gravity measurements, ±2g to measure the motion of a car and at least ±5g or more for a project
`that experiences very sudden starts or stops. Then we have sensitivity and bandwidth. Band-
`width is the frequency we use to measure changes in acceleration. Frequency of the oscillator
`has to be a lot bigger than bandwidth frequency, because electronic circuit must read changes in
`capacitance faster than acceleration changes and demodulator needs a certain number of cycles
`before it calculates output [5,7].
`Since MEMS accelerometers are used in many systems, noise characteristics of these de-
`√
`√
`vices are also very important. Analog Devices ADXL05 has voltage noise density typically
`Hz, newer ADXL202E 200µg/
`Hz [5,7]. Like we see from the unit, voltage
`around 500µg/
`noise changes with inverse square root of the bandwidth. Faster we have to read accelerometer
`changes (vibration in compare with car driving), worse accuracy we get. The noise characteris-
`tics will influence the performance of the accelerometers especially when operating at lower g
`conditions, since there is smaller output signal. One can conclude that there are three primary
`
`7
`
`Page 7 of 17
`
`

`

`noise sources in a typical MEMS accelerometer measurement: from the mechanical vibration
`of the springs, from the signal conditioning circuitry and from the measurement system itself. If
`one measure the noise characteristics of analog device MEMS accelerometer (Analog Devices
`ADXL190) operating at 0g, +1g and −1g (Figure 3.3), one can conclude from the results that
`MEMS accelerometer noise sources have 1/f -type noise characteristics at low frequencies and
`white Gaussian noise at high frequencies. The magnitude of the noise PSD (power spectrum
`density) at ±1g are slightly higher than the magnitude of noise PSD at 0g. One can conclude
`that the additional magnitude is caused by the mechanical vibration of the springs when the
`device are at ±1g [11].
`
`Figure 3.3; Figure shows the noise characteristics of MEMS accelerometer Analog Devices ADXL190 operating
`at 0g. The total noise power spectral density (PSD) of the accelerometer being measured is plotted together with
`the noise PSD of the measurement system (MS) [11].
`
`3.2 Applications
`Accelerometers are being incorporated into more and more personal electronic devices such as
`media players and gaming devices. In particular, more and more smartphones (such as Apple’s
`iPhone and the Nokia N95) are incorporating accelerometers for step counters, user interface
`control, and switching between portrait and landscape modes. They use accelerometers as a
`tilt sensor for tagging the orientation to photos taken with the built-in camera. The Nokia 5500
`sport features a 3D accelerometer that can be used for tap gestures, for example to change
`to next song by tapping through clothing when the device is in a pocket. Camcorders use
`accelerometers for image stabilization. Still cameras use accelerometers for anti-blur capturing.
`Some digital cameras, such as Canon’s PowerShot and Ixus range contain accelerometers to
`determine the orientation of the photo being taken and also for rotating the current picture when
`viewing [12].
`Accelerometers are also being used in new contactless game controller or mouse. IBM and
`Apple have recently started using accelerometers in their laptops to protect hard drives from
`damage. If you accidentally drop the laptop, the accelerometer detects the sudden freefall, and
`
`8
`
`Page 8 of 17
`
`

`

`switches the hard drive off so the heads don’t crash on the platters [2].
`In a similar fashion, high g accelerometers are the industry standard way of detecting car
`crashes and deploying airbags at just the right time. They are used to detect the rapid nega-
`tive acceleration of the vehicle to determine when a collision has occurred. They also have
`a built-in self-test feature, where a micro-actuator will simulate the effect of deceleration and
`allow checking the integrity of the system every time you start up the engine. Recently the
`gyroscopes (they rely on a mechanical structure that is driven into resonance and excites a sec-
`ondary oscillation in either the same structure or in a second one, due to the Coriolis force [13])
`made their apparition for anti-skidding system and also for navigation unit. The widespread use
`of accelerometers in the automotive industry has pushed their cost down dramatically [2, 14].
`Accelerometers have also found real-time applications in controlling and monitoring mili-
`tary and aerospace systems. Smart weapon systems (direct and indirect fire; aviation-launched
`and ship-launched missiles, rockets, projectiles and sub munitions) are among these applica-
`tions [14]. Some MEMS sensors have already been used in satellite. The development of micro
`(less than 100kg) and nano (about 10kg) satellites is bringing the mass and volume advantage
`of MEMS to good use [2].
`
`4 MEMS fabrication
`
`Micro-fabrication is the set of technologies used to manufacture structures with micrometric
`features. This task can unfortunately not rely on the traditional fabrication techniques such
`as milling, drilling, turning, forging and casting because of the scale. The fabrication tech-
`niques had thus to come from another source. As MEMS devices have about the same feature
`size as integrated circuits (IC), MEMS fabrication technology quickly took inspiration from
`microelectronics. Techniques like photolithography, thin film deposition by chemical vapor de-
`position (CVD) or physical vapor deposition (PVD), thin film growth by oxidation and epitaxy,
`doping by ion implantation or diffusion, wet etching, dry etching, etc have all been adopted by
`the MEMS technologists. Moreover, MEMS also grounded many unique fabrication techniques
`that we will describe in this seminar like bulk micromachining, surface micromachining, deep
`reactive ion etching (DRIE), etc [2].
`In general, MEMS fabrication tries to use batch process to benefit from the same economy
`of scale that is so successful in reducing the cost of ICs. As such, a typical fabrication process
`starts with a wafer (silicon, polymer, glass...) that may play an active role in the final device or
`may only be a substrate on which the MEMS is built. This wafer is processed in a succession
`of processes (Table 4.1) that add, modify or remove materials along precise patterns [2].
`
`Additive process Modifying process Substractive process
`Evaporation
`Oxydation
`Wet etching
`Sputtering
`Doping
`Dry etching
`CVD
`Annealing
`Sacrificial etching
`Spin-coating
`UV exposure
`Development
`...
`...
`...
`Table 4.1; Process classification [2]. We will explain some of them in this seminar.
`
`9
`
`Page 9 of 17
`
`

`

`The problem of patterning a material (or making layout) is generally split in two distinct
`steps: first, deposition and patterning of a surrogate layer that can be easily modified locally. In
`the most common process called photo-patterning, the surrogate layer used is a special polymer
`(called a photoresist) which is sensitive to UV-photon action (Figure 4.1).
`
`Figure 4.1; Photo-patterning. The photoresist is first coated on the substrate as a thin-film. Then it is exposed to
`UV radiation through a mask. The mask has clear and opaque regions according to the desired pattern, the clear
`regions allowing the photoresist to be exposed to UV radiation and modifying it locally. After development the
`surrogate layer patterned over the whole surface of the wafer can be used for pattern transfer [2].
`
`Now we have to transfer the pattern to the material of interest. There are two main tech-
`niques that can be used to transfer the pattern: lithography and lift-off (Figure 4.2). Combi-
`nation of photo-patterning and lithography is known as photolithography and is nowadays the
`most common techniques for micro-fabrication, lying at the roots of the IC revolution [2].
`This is how the basics of MEMS or at least patterned wafers, that will be used in further
`process, are made. Technologically very important and also quite expensive step in process is
`packaging. It can present even more than 50% of final product cost [2]. Lets now look in detail
`at some materials and some processes or techniques that can be used during MEMS process.
`We already mentioned some above.
`
`Figure 4.2; Pattern transfer by lithography and lift-off. In lithography the patterned layer allows exposing locally
`the underlying material. The exposed material is then etched physically or chemically before we finally remove
`the protective layer. For lift-off, we deposit the material on top of the patterned layer. Complete removal of this
`layer (called a sacrificial layer) leaves the material only in the open regions of the pattern [2].
`
`10
`
`Page 10 of 17
`
`

`

`4.1 The MEMS materials
`The choice of a good material for MEMS application is no more based like in microelectronics
`on carrier mobility, but on more mechanical aspect: small or controllable internal stress, low
`processing temperature, compatibility with other materials, possibility to obtain thick layer,
`patterning possibilities...[2].
`From microelectronics’ root MEMS has retained the predominant use of silicon and its
`compounds, silicon (di)oxide (SiO2) and silicon nitride (Six Ny). It is an excellent mechani-
`cal material. Silicon is almost as strong but lighter than steel, has large critical stress and no
`elasticity limit at room temperature as it is a perfect crystal ensuring that it will recover from
`large strain. Unfortunately it is brittle and this may pose problem in handling wafer, but it is
`rarely a source of failure for MEMS components. For sensing application silicon has a large
`piezoresistive coefficient, and for optical MEMS it is transparent at the common telecommu-
`nication wavelengths. Silicon nitride is even stronger than silicon and can be deposited in thin
`layer with an excellent control of stress to produce 1µm thick membrane of several cm2. There
`is also silicon carbide (SiC) in use. SiC has unique thermal properties (albeit not yet on par with
`diamond) and has been used in high temperature sensor [2].
`But silicon and its derivative are not the only choice for MEMS, many other materials are
`also used because they posses some unique properties. For example, quartz crystal (strong
`piezoelectric effect), glass (forms tight bond with silicon, bio-compatibility), polymers (biodegrad-
`ability and bioabsorbability, versatility, thermoplastic property), metals (conductivity, ability to
`be grown in thin-films),...[2].
`
`4.2 Bulk micromachining
`Bulk micromachining refers to the formation of micro structures by removal of materials from
`bulk substrates. We said that bulk substrate in wafer form can be silicon, glass, quartz, crys-
`talline Ge, SiC, GaAs, GaP or InP. The methods commonly used to remove excess material are
`wet and dry etching, allowing varying degree of control on the profile of the final structure [2].
`
`4.2.1 Isotropic and anisotropic wet etching
`
`Wet etching is obtained by immersing the material in a chemical bath that dissolves the surfaces
`not covered by a protective layer. The main advantages of the technique are that it can be quick,
`uniform, very selective and cheap. The etching rate and the resulting profile depend on the
`material, the chemical, the temperature of the bath, the presence of agitation, and the etch stop
`technique used if any. We have isotropic and anisotropic etching. Isotropic etching happens
`when the chemical etches the bulk material at the same rate in all directions, while anisotropic
`etching happens when different etching rate exists along different directions [2].
`For substrates made of homogeneous and amorphous material, like glass, isotropic wet etch-
`ing is usually observed. However, for crystalline materials, e.g. silicon, the etching is either
`isotropic or anisotropic, depending on the type of chemical used. In general, isotropic etch-
`ers are acids, while anisotropic etchers are alkaline bases. Figure 4.3 compares isotropic and
`anisotropic wet etching of silicon. The top-left inset shows isotropic etching of silicon when
`
`11
`
`Page 11 of 17
`
`

`

`the bath is agitated ensuring that fresh chemical constantly reaches the bottom of the trench
`and resulting in a truly isotropic etch. Isotropic wet etching is used for thin layer or when the
`rounded profile is interesting, to obtain channels for fluids for example. For silicon, the etcher
`can be HNA, which is a mixture of hydrofluoric acid (HF), nitric acid (HNO3), and acetic acid
`(CH3COOH). The isotropic etching rate for silicon can reach 80µm/min. Etching under the
`mask edge is unavoidable with isotropic wet etching [2].
`
`Figure 4.3; It is impossible to obtain etching in only one direction. This is commonly quantified by estimating the
`overetch (w/d), that is the lateral etch with respect to the vertical etch. This parameter may range between 1 for
`isotropic etching to about 0.01 for very anisotropic etch [2].
`
`To achieve anisotropic etching of silicon we usually need the right choice of chemicals. The
`most common are potassium hydroxide (KOH), tetramethyl ammonium hydroxide (TMAH)
`and ethylene diamine pyrocatechol (EDP). The etching anisotropy has its roots in the different
`etch rates appearing for different crystal planes because they have different density of electrons.
`Experiments with etching of silicon have shown that some planes act as etch stoppers as etch-
`ing rates along directions perpendicular to these planes are substantially lower than about other
`directions. So we can get trapezoid-like cavity, almost vertical walls... Etch velocity range
`between 1µm/min to only 2.5nm/min in different crystal planes [2]. With different combi-
`nations of wafer orientations and mask patterns, very sophisticated structures such as cavities,
`grooves, cantilevers, through holes and bridges can be fabricated. The chemical used during
`anisotropic etching are usually strong alkaline bases and requires a hard masking material that
`can withstand the solution without decomposing or peeling. In general a non-organic thin-film
`is used, for example, silicon oxide mask is commonly used with TMAH, while silicon nitride is
`generally used with KOH [2].
`
`4.2.2 Dry etching
`
`Dry etching is a series of methods where the solid substrate surface is etched by gaseous
`species. The etching can be conducted physically by ion bombardment (ion etching or sput-
`tering and ion-beam milling), chemically through a chemical reaction occurring at the solid
`surface (plasma etching or radical etching), or by mechanisms combining both physical and
`chemical effects (reactive ion etching or RIE). Usually the etching is more anisotropic and ver-
`tical when the etching is more physical, while it is more selective and isotropic when it is more
`
`12
`
`Page 12 of 17
`
`

`

`chemical. Most of these methods are used in microelectronics but MEMS necessitates deeper
`etching (> 5µm) [2].
`Typical values for aspect ratio for features (Figure 4.4) range between 1 (isotropic etch) and
`50, for very anisotropic etching like the DRIE process [2].
`
`Figure 4.4; We usually define an aspect ratio for features (h/wr) and for holes (h/wh). Most technologies give
`better results with features than with holes - but generally with only a small difference [2].
`
`To improve the aspect ratio of the etching, several techniques have been developed, usually
`trying to increase the anisotropy by protecting the sidewalls during etching. For example we
`can continuously deposit polymer on the sidewall during the etch or even better, by alternate
`steps of etching, grow oxide layer on the sidewall [2].
`
`4.2.3 Wafer bonding
`
`Wafer bonding is an assembly technique where two or more precisely aligned wafers are bonded
`together. This method is at the frontier between a fabrication method and a packaging method
`and belong both to front-end and back-end process. Wafer bonding has the potential to simplify
`fabrication method because structures can be patterned on two or more wafers and after bond-
`ing they will be part of the same device, without the need for complex multi-layer fabrication
`process. Of course epoxy bonding can be used to bond wafers together but much better MEMS
`techniques do exist. The most commonly used MEMS bonding methods is probably anodic
`bonding which is mainly used to bond silicon wafers with glass wafers. The technique work by
`applying a high voltage to the stacked wafers that induce migration of ion from glass to silicon,
`allowing a strong field assisted bond to form. This technique is commonly used to fabricate
`sensors allowing for example to obtain cavities with controlled pressure for pressure sensor. At
`the same time, the glass wafer provides wafer level packaging, protecting sensitive parts before
`back-end process [2].
`
`4.3 Surface micromachining
`Unlike bulk micromachining in which microstructures are formed by etching into the bulk sub-
`strate, surface micromachining builds up structures by adding materials, layer by layer, on the
`surface of the substrate. The thin film layers are typically 15µm thick [2], some acting as struc-
`tural layer and others as sacrificial layer. Dry etching is usually used to define the shape of the
`structure and supporting layers, and a final wet etching step releases them from the substrate by
`removing the supporting sacrificial layer. A typical surface micromachining process sequence
`to build a micro bridge is shown in Figure 4.5.
`
`13
`
`Page 13 of 17
`
`

`

`Figure 4.5; Basic process sequence of surface micromachining. Phosphosilicate glass (PSG) is first deposited and
`patterned to form the sacrificial layer. Then a structural layer of polysilicon is added and also patterned. Finally,
`the PSG sacrificial layer is etched away and the polysilicon bridge is released [2].
`
`4.3.1 Thin-film fabrication
`
`Common thin-film fabrication techniques are the same as those used in microelectronics fabri-
`cation like oxidation (in dry or in wet oxyge