Vous êtes sur la page 1sur 15

SensorsandMaterials,Vol. 8, No. 6 (1996)373-387 MYU Tokyo

S &M0252

Physical Principles of Thermal Sensors

SandervanHerwaarden

XensorIntegration P. O. Box 3233,2601DE Delft, theNetherlands

(ReceivedMarch29, 1995;acceptedMarch25, 1996)

Key words:

thermaltransduction,transductionphysics

We give an introduction to the physical principles onwhich silicon themlal sensorsare based. We briefly describe the transductions and conversions involved in these sensors, and different themlal sensorsand the physics on which they are based.

1. The Functioilal

Principle

of Thermal

Sensors

In this paper, we describe themlal sensorsbased on silicon and related technologies, with an electrical output signal. The input signal can be any of the six signal types defined by Middelhoek and Audet,(l) i.e., mechanical, magnetic, chemical, radiant, themlal and electrical signals. In themlal sensors, the transduction of the input signal to the output signal is carried out in two steps. First the input signal is transduced into a themlal signal, and then the themlal signal is transduced into the eleetrical output signal.

1.1 Thermal power

sensors

An important distinction among themlal sensors is that between power sensors and conductance sensors. In power sensorsthe input signal which is measuredby the sensoris themlal power, which is used to generate the electrical power of the sensoroutput signal. Therefore, there is no output signal ifthe input signal is zero. These self-generating sensors have no offset, and need no biasing. They operate in three steps. 1) A nonthemlal signal Cis transducedinto a themlal heat signal P, by the sensor-specific action Q.

P= QC

373

(I)

374

SensorsandMaterials, Vol. 8, No. 6 (1996)

2)

The heat P is converted into a thennal signal temperature difference LlT, by a thennal resistanceR.

 

LlT =RP

(2)

3)

The temperature difference ture difference sensorS.

LlTis transducedinto an electrical voltage U, by a tempera-

 

U =SLlT

(3)

The total transfer ratio UIC of a thennal power sensoris given by

U I C=

QRS

(4)

A psychrometer is an exception to this three-step operation. In this sensor,the nonthennal

signal is transduced directly into a temperature difference.

1.2 Thermal conductance sensors In the case of thennal conductance sensors, the input signal C changes the thennal conductance G between the sensitive area and the ambient. To measure the thennal conductance, the sensoris biased with a hearing power P. The transfer in thesemodulating sensorsis as follows. 1) The nonthennal signal C is transduced into a thennal conductance signal G, by a sensor-specific transduction action Q, Gois the offset of the sensor.

2)

3)

G =QC + Go

(5)

The conductance Gis converted into a temperature difference LlT, by the heating power

P.

LlT= PIG

(6)

The temperaturedifference LlTis transducedinto anelectrical voltage U, by atemperature difference sensorS.

U =SLlT

(7)

For thennal conductance sensors,the total transfer cannot be written in a multiplicative

fonn.

lnstead we find

U =PS/(QC + Go)'

This equation indicates the offset character of modulating

"(8)

sensors. In some thennal

SensorsandMaterials,Vol. 8, No. 6 (1996)

375

conductance sensors,the electrical signal in step 3) bas another form, such as a current or resistance, depending upon the type of temperature difference sensorused. In this paperwe concentrateon the first transductionstep,the sensor-specifictransduction action Q. We usevarious thermal sensorsasexamples. Considering the variety ofthermal

sensorsdescribed in the literature,

third transduction stepsare described elsewhere.(2-4)

a comprehensive summary is not given. The secondand

2.

HeatTransferMechanisms

In most thermal power sensors,the heat is generated,and the transduction takesplace in or on the sensor itself. In some, the transport of nonthermal power to the sensorinvolves heat transfer mechanisms. In thermal conductance sensors,on the other hand, the change in the conductance doe to the nonthermal signal is in the ambient directly around the sensor,not in or on the sensoritself. In this case,heat transfer mechanisms are essential to the operation of the sensor. There arefoor heattransfermechanisms,conduction, convection, radiation and phase transition.(S) Conduction always occurs, and in thermal conductance sensorsthe first transduction step is based on the thermal conduction Gcondbetween the active areaof the sensorand the ambient. The thermal conductance between two parallel surfaceswith areaA separatedby a di stanceD is given by

Gcond= k:AID,

(9)

where I( is the thermal conductivity of the medium presentbetween the surfaces. The sensoraction may be based on the dependenceof the physical signal

- on I( (pressure dependence: vacuüm sensor/Pirani gauge; fluid dependence: thermal

conductivity and overflow sensor),

- on D (plate-spacing dependence:accelerometer), or

- on all three parameters 1(,D and A (thermal properties sensor). Convection, the secondmechanism for thermal conductance sensors,is heat transfer to moving fluids, as in flow sensors. Usually convection is negligible in sensorsother than flow sensors. Although convection conducts heat away from the active area,the physical principle is different from that of conduction, since the heat is not transferred by stationary molecules from neighbor to neighbor, but by a continuous supply of molecules flowing past. In the simplest formula for laminar flow over a flat plate, the heat transfer Gconvis

given by

Gconv= 0.664P,n.33Reo,sI(IL,

(10)

where L is a characteristic length, for instance the length of the sensor, Pr is (the temperature dependent material constant) the Prandtl number, and Re is the Reynolds number, which is dependent on the flow velocity V, L and the kinematic viscosity V (in m2 Is), that is Re = VLlv. The sensoraction cao be basedon the flow velocity dependenceof

376

SensorsandMaterials,Vol. 8, No. 6 (1996)

the Reynolds number (as in flow sensors),or on the pressure dependenceof the viscosity and the conductivity, which are also constituents of the Reynolds number (as in same Pirani gaugesfor vacuum measurementnear atmospheric pressure).

Radiation is heattransfer by electromagnetic waves, either asinfrared radiation (thermal

radiation), or other farms, such as rnicrowaves or magnetic fields which generateheat in a dissipative layer (hysteresis in a magnetic layer, for instance). It is a self-generating effect, and is therefore utilized in thermal power sensors. Given below are two formulae for infrared radiation. ODeis for the heat transfer between parallel plates, which, when bath plates are at around room temperature, is equal to

Gir=ex6W/m2K,

(11)

where eis the ernissivity or absorptivity of one plate. The other plate is assumedto be black with ernissivity e = 1. All materials ernit infrared radiation, but usually it does not contribute significantly to the heat exchange of a thermal sensorwith the ambient. The other formula describes the heat transfer between an infrared sensor with black detecting areaAo (as viewed fiom the object) and temperature To and an object of areaAobj (as seenfiom the sensor), temperature Tobiand ernissivity e. The exchangedpower is

Pir =(eO"ln) (Aobjldl) Ao (TObj4- T04).

(12)

Here, 0" is the Stefan-Boltzmannconstant(56.7 x 10-9W IK4m2),andd is the distance

between the sensorand the object. When a sensorwith a sensitive areaof 1 mm2 at 300 K

is exposed to an object at 301 K with an ernissivity of 55% at asolid angle of 0.1 steradian

(i.e., Aobj/dl = 0.1), the exchanged power is 100 nWo

(for example, the heat generatedby evaporation) is the last mechanism

Phasetransition

of heat transfer. Phasetransitions areoften induced by forced convection, which influences the thermal conductance. This is also a self-generating effect. Sirnilar to radiation, phase transitions generate thermal power (be it negative or positive). A peculiarity of heat

transfer by phasetransition is that in same situations heat can be transferred fiom a cold to

a hot object. Table 1 gives an overview of the heat transfer mechanisms described above, together with typical magnitudes of the heat transfer for different mechanisms. It shows the large variations in magnitude among heat transfer mechanisms. In general, heat transferred by conduction and convection is much greater than that transferred by radiation for room- temperature structures. However, since the power ernitted fiom objects in the farm of

infrared radiation is proportional to the fourth power of absolute temperature, the situation

is different at high temperatures. Freeconvection is seldom significant for rnicrostructures,

sinceit is a size-dependentphenomenon,and is insignificant in comparison with conduction in small structures. However, at very high temperatures(> 300°C) there is significant tree convection even in rnicrostructures. ~Phasetransitions such as evaporation should be avoided in self-generating sensorsused to measure other heat signals, since evaporation can give rise to significant parasitic heating or cooling even in thermal equilibrium.

SensorsandMaterials,Vol. 8, No. 6 (1996)

377

Tab1e1

Overview of heat transfer mechanismsand the magnitudeof heat transfer for typica1500 .urn rnicrosensors.

Type

Characteristicof effect

WIKm2

Remark

Radiation

betweenb1acksurfaces

6

objectsat aroundroomtemperature

Conduction betweensurfaces500.urnapart

500

objectsat around1000°C

air at vacuüm

1

1Papressure

air

50

atmosphericpressure

water

1,200

silicon

300,000

Convection for sensors500.urnlong

 

free convection

-

of ten neg1igib1e in rnicrostructures

forcedconvectionin

air

150

1 mis

flow

forcedconvectionin

water

15,000

1 mis

flow

Phase

evaporationof water

1000-10,000*

with forcedconvection

transition

*Temperature

difference

not required

for phase transition

heat transfer

to occur.

3. Overview of Thermal Sensors

,

Below we discuss the physical transduction principles of various thermal sensorsin more detail. Table 2 gives an overview of the thermal sensorsdiscussedbelow.

In Table 2 the first five sensors are self-generating thermal power sensors, and the

others are modulating thermal conductance sensors,which use an electrical resistance to generaleheat. With the exception ofthe EM-field sensorand the psychrometer, all of these

sensorshave been realized using both silicon and thin-film technology.

3.1

Microcalorimeter

A microcalorimeter measuresthe heat generatedduring chemical reactions. Bataillard

describesvarious applications.<6)A chemical reaction occurs between two solutions in the reaction volume of the sensor (near the active area of the thermal sensor) after they are supplied via two tubes. Altematively, a catalyst or enzyme is immobilized on the active area of the microcalorimeter, which initiates a chemical reaction when a single solution comesinto contact with it. (SeeFig. 1). This setupensuresthat the transfer of reaction heat to the sensor is optimized. It is even possible to measure the heat generated by microorganisms such asbacteria immobilized near the active areaof the microcalorimeter. The heat produced by the bacteria (of the order of 1 pW per bacterium) is a measureof the concentration of nutrients in the solution, such as glucose, or of the growth fale of bacteria

378

SensorsandMaterials,Vol. 8, No. 6 (1996)

Table2

Overviewof thermalsensors,their measurandsandoperationprincip1es.

"-

Sensor

Measurand

üperationprinciple

Microcalorimeter

concentration

heat of

chemical reaction - catalyst/enzyme area

Psychrometer

humidity

heatof

evaporation- wet wick

Infraredsensor

infraredradiation

black-bodyradiation- b1ackcoating

RMS converter

electrical power

Joule healing - heater resistance

EM-field

sensor

EM fields

dissipation - resistive termination

Flow sensor

fluid flow

convection

Vacuumsensor

vacuümpressure

conduction- pressuredependence

Conductivitysensor

fluid type

conduction- typedependence

Mechanical sensor

acceleration

conduction - seismic mass

Thermalpropertjessensor

materialpropertjes conductionandcapacitance

liquid channels [

connectors foT tubes

PGA

Z

Fig. 1.

bonding wires

layer

Liquid microcalorimeterin a flow-injection analysis(HA) setup.

when sufficient glucose is present. In practicè,in anenzyme-basedliquid microcalorimeter foTdetermining the concentration of glucose the transduction to the thermal domain takesplace in the following two steps.(7.8) First, the concentration C (in mol/m3) of glucose in water is converted into a reaction rate M (in mol/s) by bringing the mixture into contact with the enzyme.

SensorsandMaterials,Vol. 8,No. 6 (1996)

379

M=

C(dKldt)

(13)

The chemical conversion efficiency dKldt (in m3/s) dependson the chemically active layer (catalyst or enzyme) and the particular experimental setup. For instance, ODeglucose oxidase molecule can convert a maximum of 1000 glucose molecules per second, but converts less if the solution is not replenished. When the solution is replenished, the

chemical conversion efficiency depends solely on the chemically active layer (a material property) and not on the experimental setup. Next, the reaction rate M of the concentrate is transducedinto heat of reaction P (in W) by the change in enthalpy.

P =M (-AH)

(14)

The enthalpy change -dB is the energy released during the chemical reaction (in J/mol) and is determined experimentally. For enzymatic oxidation of glucose, a value of 80 kj I mol is found. According to eq. (13) a monolayer of glucose oxidase could produce a maximum power of approximately 3 W Im2/mol/m3.

3.2 Psychrometer

In this sensorthe relative humidity of air is determined by measuring the psychrometric temperature reduction i1Tpsych of a wetted thermometer doe to evaporation. This sensoris therefore basedon a physical phasetransition. The first transduction step can be described

by

i1Tpsych= Q (Pdry- PweJIPatm.

(15)

Here Pdryand Pwetare the partial vapor pressuresof water at the air temperature and at the wet-thermometer temperature, and Patmis the atmospheric pressure. The sensor-specific transduction factor Q is of the order of 1700 K.<2) i1Tpsychin eq. (15) cannot simply be converted to the relative humidity, becauseit dependson temperature and humidity, and is

much smaller than the dew-point temperature reduction.

multiplying Q by 30% x PdryI Patm. At room temperature with Patm= 100 kPa and Pdry~ 2

kPa, the sensitivity is typically of the order of 0.1 K I %RH.

Note, that in this sensorthe nonthermal signal is transduceddirectly into a temperature difference. There is no conversion in the thermal domain, as there is in allother thermal sensors. Accurate transfer only occurs if good thermal isolation is maintained, and

therefore design of

an appropriate thermal resistance is necessary anyway. Same other

measures, such as forced convection across the wetted thermometer of an appropriate magnitude are also essential for accurate transduction. No silicon version of this sensoris

known to the author.

An estimate is obtained by

3.3 lnfrared

From the transduction point of view this is a fairly simple sensor. The transduction from radiation to heat is carried out by a black absorber,which can have an efficiency of up

Sensor

380

SensorsandMaterials,Vol. 8, No. 6 (1996)

to 99%. The first transduction step fiom incident radiation density P"inc (in W Im2) to thermal power P is given by

P = QP"inc

Q = aAD'rfilter'

(16)

(17)

The absorptivity a is between 0 and 1, and denotesthe fraction of itifrared radiation power which is absorbedby the black coating. Various black coatings areusedfor silicon itifrared sensors. A simple and efficient method is to usethe silicon oxide and silicon nitride layers produced in all semiconductor production processes.(9.IO)Lenggenhager<9)observed an absorption of the order of 50% for radiation with wavelengths of 7 -14 pm. Using the absorption of PVDF and metal electrodes is another war to implement this method.(II) Porous metal coatings are used to fabricate very black layers.(12)Gold black win absorb more than 99% of radiation over theentire itifrared spectrum. A different method is used at Xensor Integration. The sensoris spin-coated with a black polymer. Using techniques very similar to normallithographic and RIE processes,a pattem in a 5-pm-thick coating is obtained on the wafer, which absorbs about 90% of the itifrared radiation. This layer is pattemed in a similar war to other thin films on silicon, and can withstand further processing ofthe wafer. This is not the casefor coatings such asgold black, which are very vulnerable and cannot be handled once they have been applied. ADis the sensitive area of the sensor (usually the areathat is coated with black), while 'rfilteris the transmission of the filter that is usually incorporated when the sensitive area is encapsulated. This filter canbe broadband,transmitting infrared radiation with wavelengths between 2 and 14 pm, high pass, transmitting wavelengths between 7 and 14 .urn, for detection of objects at room temperature which emit radiation with a typical wavelength of 10pm (intrusion alarm), or band-pass,transmitting, for instance, radiation at a wavelength which is in the absorption band of a gas. Using appropriate filters, a gas sensor can be constructed in which the radiation intensity is measured in a reference path and a path in which the gas mixture onder investigation is present. The difference in intensity is doe to the presenceof the radiation-absorbing gas. This method is used for detection of CO2and CO.

3.4 RMS converter

In the RMS converter, the first transduction step, from electrical to thermal, is simply performed by dissipation in an electrical resistor. Complications arise doe to parasitic thermoelectric Thomson and Peltier effects in DC signals, and skin effects and parasitic capacitances and inductances at high frequencies, which cause differences in the heat actually generated in the heater. These problems have been studied by calibration engineers,(IJ)and win not be discussedhere.

3.5 EM-field

sensor

The EM-field

sensor bas also been studied by calibration engineers. The EM-field

sensor is in between the itifrared sensor and the RMS converter.

detects optical signals with very high frequencies of around 1015Hz. The RMS converter

The itifrared sensor

SensorsandMaterials,Vol. 8, No. 6 (1996)

381

detecselectrical signals with frequencies of up to 109Hz and the EM-field sensormeasures

intermediate frequencies.

signals, as does the infrared sensor.

sensorsare known to the author, but current research is

focused on the termination of waveguides by metro pattems on glassplates, which convert the EM-field energy into heat, that is subsequently detected using a bolometer. Here, the ftrst transduction step, fiom radiant to thermal signals, is carried out using a .specifically designedmetro pattem which yields a lossless,reflectionless termination of the waveguide. The actual design dependson the wave frequency and waveguide geometry.(14)

The EM-field

sensor transduces fiom

radiant to electrical

No semiconductor EM-field

3.6

Flow

sensor

Flow sensors are based on the transfer of heat to moving fluids.

This effectively

sensor and the ambient. For flow

increasesthe overall thermal conductance between the

sensors,the physics of the second and third stepsin

as those foT rol thermal sensors, but the similarity ends here. The physics of the fiTst transduction step, fiom flow to thermal signals, and the encapsulation of the sensor, are much more complicated than those foT most other sensors. Also, the encapsulation has a great influence on the fiTst transduction step, because it influences the type of flow. In laminar flow, the fluid flows along straight lines, and in turbulent flow it flows in irregular pattems and the local flow direction is unrelated to the average flow direction. In

microstructures laminar flow often occurs, although in thermal windmeters, microturbulent flow is also encountered.(2) The dependence of the heat transfer on flow velocity is

the transduction processare as simple

different foTlaminar

and turbulent flow. For laminar flow, the fiTst transduction stepcanbe

approximated by

G = Go+ Hyl/2.

(18)

where G is the total thermal conductance fiom the sensor to the ambient, Gois the offset (the conduction part), and H is the heat transfer normalized by the flow velocity V. The convective part of the thermal conductance is proportional to the square root of the flow velocity. The variabie H includes such parameters as the sensor size, flow history, and fluid characteristics (viscosity, thermal conductivity, temperature). The fluid flow upstream of the sensorcan greatly influence the exact flow pattem over the sensorand its heat transfer characteristics. That is why the encapsulation of the sensor is important (seeFig. 2). The temperature profile ofthe flow sensoralso influences the heat

transfer to the flOW.(2.5.15)

3.7 Vacuum sensor

This sensoris usedto measuregaspressuresbelow atmospheric pressureby measuring the pressure-dependentthermal conductivity of gases. At very low pressures, when the mean distance travelled by a molecule between collisions is much larger than the distance between two surfaces,heat transfer between the surfaces occurs via individual molecules. The rate of heat transfer is therefore proportional to the rate at which molecules hit the surface, which is the absolute pressure. At higher pressures(atmospheric foT surfaces500

SensorsandMaterials,Vol. 8, No. 6 (1996)

383

gagessuch as nitrogen and helium, for a reference pressure Po of 1 Pao For the entire pressurerange, the relation between thermal conductance and pressurebecomes

G =(G"oPICID)/(G"oP+ PoICID),

(20)

where ICID is the thermal conductance between two plates separatedby a distance D for a gas with thermal conductivity IC(seeFig. 3 and eq. (9».(2.16)

3.8

Conductivitysensor

This sensoris similar to the vacuum sensor,since it measuresthe thermal conductance of a gas surrounding the sensor. However, in tros casethe conductance does not depend upon the pressure,but, at atmospheric pressure,upon the gas type (or liquid type) instead, since all gageshave different thermal conductivities. For instance, for air the thermal conductivity is IC=26 mW IKm at room temperature. Hydrogen basthe highest conductiv- ity at more than 180 mW I Km. For helium it is about 150 mW IKm while gagessuch as argon (18 mW I Km) and xenon (6 mW I Km) are even less conductive than air. The conductivity sensorcan be usedin various ways. By measuring the thermal conductance it can be used to determine the type of gas (see Fig. 3), or the composition of a binary gas mixture. For the latter case,a complex formula describesthe thermal conductivity of a gas

10-1

Helium

~-

~

 

~

~ --

~

Nitrogen

~c

Argon

'"

Ü

::J

-0

C

0

()

(ij

E

a;

.c

f-

10-6

10.1

102

Pressure (Pa)

105

Fig. 3.

of 0.5 rnrn versus pressure.

Thennal conductance of different gasesbetween two parallel plates saparatedby a distance

SensorsandMaterials,Vol. 8, No. 6 (1996)

sus ended mass

l

j-\~-:--_/--\J

cc:=

~

==:>-

q 1/2

~

q 1/2

thermo ile

q2=f(a)

~q3

x=L

.

X '

Fig.4.

Schematicdrawingandcrosssectionof athermalaccelerometer.

385

suspended in a closed silicon nitride membrane. This is a variation of the floating- membrane structure.(20)The measurementsare performed in vacuum, so that only conduc- tion in the solid material adds to the measuredconductances(radiation effects are usually

not significant). In this war, the thermalconductance/(" and capacitancecpof

materialshavebeendetermined,suchasthosefor silicon (150 W/Km and700 J/kgK), silicon dioxide(1-1.5 WIKm and730J/kgK), low-stresssiliconnitride SiNl,. (3-3.5 W IKm and700JIkgK), polysilicon(18-30 WIKm and770J/kgK) andaluminumwith 1%

silicon (180-220 WIKm).(2,20-22)

various

SensorsandMaterials,Vol. 8, No. 6 (1996)

References

387

I

S.MiddelhoekandS.A. Audet:Silicon Sensors(AcademicPress,London,1989)p. 6.

2

G. C. M. Meijer andA. W. vanHerwaarden:TherrnalSensors(IOP Publishing,Bristol, 1994).

3

A. W. vanHerwaardenandG. C. M. Meijer: TherrnalSensors,in SemiconductorSensors,ed. S.M. Sze(Wiley, New York, 1994).

4

A. W. vanHerwaarden,D. C. vanDuyn,B. W. vanOudheusdenandP.M. Sarro:Sensorsand ActuatorsA21-23 (1989)621.

5

A. J. Chapman:HeatTransfer,4th ed.(MacMil1an,New York, 1984).

6

P. Bataillard:Trendsin Analytical Chemistry,vol.12,no. 10(1993)387.

7

P. Batai1lard,E. Steffgen, S. Haemmerli,A. Manz and H. M. Widmer: Biosensorsand Bioelectronics8 (1993)89.

8

A. W. van Herwaarden,P. M. Sarro,J. W. GardnerandP. Bataillard:SensorsandActuators A43 (1994)24.

9

R. Lenggenhager:CMOSTherrnoelectricInfraredSensors,PhDdissertation(ETH No. 10744, Zurich 1994)37.

10

M. Müller, R. Gottfried-Gottfried,H. Kück andW. Mokwa: SensorsandActuatorsA41-42

 

(1994)538.

11

S.Bauer,S. Bauer-Gogonea,W. Becker,R. Fettig,B. Ploss,W. RuppelandW. von Münch:

SensorsandActuatorsA37-38 (1993)497.

12

W. Lang,K. Kühl andH. Sandmaier:SensorsandActuatorsA34 (1992)243.

13

M.

Klonz andT. Weimann:IEEE TransInstrurn.MeasIM-40 (1991)350.

14

D. Andreone,L. BrunettiandM. Petrizelli: SensorsandActuatorsA41-42 (1994)114.

15

B. W. vanOudheusden:SensorsandActuatorsA30 (1992)5.

16

A. W. vanHerwaardenandP.M. Sarro:J. Phys.E: Sci.Instrurn.21 (1988)1162.

17

R. B. Bird, W. E. StewartandE.N. Lightfoot: TransportPhenomena(Wiley, New York, 1960)

258.

18

R. Hiratsuka,D. C. van Duyn, T. OtaredianandP. de Vries: Techn.DigestTransducers'91,

(SanFrancisco,CA, 1991)p. 420.

19

U. A. Dauderstät,P.H. S.deVries,R. HiratsukaandP.M. Sarro:SiliconAccelerometersBased on Therrnopiles,Proceedingsof Eurosensorsvm (Toulouse,1994).

20

P. M. Sarro,A. W. van HerwaardenandW. van der Vlist: SensorsandActuatorsA41-42

 

(1994)666.

21

P.M. Sarro,A. W. vanHerwaardenandM. lodice:Therrno-PhysicalPropertiesof Low-Stress Poly- and Mono-Silicon and Silicon Nitride, Proceedingsof Eurosensorsvm (Toulouse,

 

1994).

22

O. PaulandH. Baltes:J.Micromech.Microeng3 (1993)110.