Paper JM3 001016 received Aug. 7, 2001; revised manuscript received Dec. 13,2001; accepted for publication Dec. 21, 2001.
e-paride
desvarjet
thelityon
pli-lite
a-ress a
thebyavy
su-
itsn-ca-on.ateing
1 Introduction
The working principle of many microelectromechanical dvices is based on the use of diaphragms as a flexuralusually acting as a passive transducing element. The wrange of devices incorporating flexible diaphragms inclumicromachined pressure sensors, microphones, and aety of microfluidic devices such as micropumps and inkprintheads.
The geometrical tolerance of the diaphragm duringfabrication process, as well as its thermal compatibiwith the rest of the device, can have a significant impactoverall device performance. This is especially true in apcations such as low-pressure sensing or precise picoliquid handling.
Different solutions have been employed in terms of mterials and control of the geometry of the diaphragms. Psure sensors years ago employed thin silicon diaphragmthe pressure sensitive element.1,2 Diaphragms were formedby anisotropically etching exposed silicon areas, withthickness of the diaphragms being controlled eithertimed etching or by etch-stop techniques such as heboron doping or reversep–n junction formation. The flex-
Downloaded from SPIE Digital Library on 02 Feb 201
t,
i-
r
-s
ural element in inkjet printheads and micropumps was ually made of stainless steel,3 glass,4,5 or silicon.6,7
The choice of diaphragm material is dependent oncompatibility with the overall fabrication process. In stadard micromachining technology based on batch fabrition, the two main materials used are glass and silicSilicon is preferred since it offers a wider range of accurmicromachining processes and the possibility of integrat
Fig. 1 Schematic cross section of the precision grinding system forsilicon.
Address all correspondence to S.J.N. Mitchell, Tel:~4428! 90335437; Fax:~4428! 90667023; E-mail: [email protected]
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
Fig. 2 Fabrication flow process for thin-diaphragm test structures.
m
tchtivue
inan
ndiy 5of.
ion.
riam
g-vede.gilesili-ndter
oncono
ngtate50
ss
nes-
iser-
sewa-the
ralls,
electronic circuitry with microelectromechanical syste~MEMS! structures.
In most cases silicon diaphragms are formed using eing, accompanied by etch-stop techniques. One alternato this process is precise grinding of silicon, a techniqthat has been widely used in silicon on insulator~SOI!technology.8 The main advantage of silicon grinding liesits purely physical nature and time-saving efficiency. Asexample, removal of 200mm of silicon using anisotropicetching based on KOH aqueous solution in standard cotions takes 3–4 h whereas using grinding requires onlmin. In the present paper we investigate the viabilityprecision grinding for the formation of silicon diaphragms9
Potential limitations of the process of diaphragm formatas well as techniques to overcome them are explained
2 Silicon Precision Grinding
In this work a Shibayama VG-202MKII precision grindewas used. The system is capable of grinding 150 mm dsubstrates with uniformity of60.5 mm. A schematic crosssection of the precision grinding of silicon is shown in Fiure 1. The wafer holder may be slightly convex or concain shape with a maximum convexity/concavity magnituof 2–3 mm. The silicon wafer is held in place by vacuum
The working mechanism is as follows: the grindinwheel and the silicon wafer are kept in direct contact whboth are rotating. This causes constant removal of thecon material as the result of friction between the diamoteeth and the silicon. A thickness gauge is used to demine the amount of material removed.
The process is purely physical and does not dependparameters such as the temperature or wafer dopingcentration. Precision grinding of silicon proceeds in tw
Downloaded from SPIE Digital Library on 02 Feb 201
-e
-
-
-
stages: coarse grinding followed by fine grinding. Durithe coarse grinding stage, the wafer and grind wheel roat 200–250 rpm, the removal rate of silicon is about 2mm/min, and the wafer thickness tolerance is63 mm. Thefine grinding stage provides an improved wafer thicknetolerance of60.5 mm at a slower removal rate of 20mm/min. In applications in which an optically smooth silicosurface is required, an additional polishing step is necsary.
The most common use of the silicon grinding processfor the removal of a portion of the active wafer in thproduction of SOI substrates for MEMS and high perfomance electronic circuits.8
3 Fabrication of Test Structures
Two types of test structure were used in this work; thobased on bonded wafer pairs or those on single siliconfers. The minimum wafer thickness acceptable bygrinder used in this work was 250mm. To ensure that thisminimum is not exceeded and to preserve the ovestrength and rigidity of test structures with thin diaphragm
Fig. 3 Profile of (a) a 100 mm thick diaphragm and (b) a 25 mm thickconcave diaphragm measured using the Alpha Step 200 (TencorInstruments).
167J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
168 J. Microlith.,
Table 1 Range of bending magnitudes within a wafer for each thickness of 6 mm diam diaphragms.
Wafer A B C D E
Diaphragm thickness (mm) ;25 50 75 100 150
Bending magnitude range (mm) Semiconcave 25.6–58.3 7.5–36.8 6–13.4 3.4–8.2
asinafoon
25m
to
im
ctseferre
dia-
e-yof
m
inea
glinth
. Inanncce
e of
msd
r agmsfor
vicepres-k ofss.theum
inguse
ngeresndro-ac-
ingred
inedtnd-ro-thever,
therbi-
osethe
a process involving the bonding of two silicon wafers wdeveloped. To prevent the bonding process from producdifferential pressure on the diaphragm, a throughhole wetched into the support wafer. The preparation processthe bonded test structures is shown in Figure 2. Silicwafers, 100 mm in diameter, with eithern- or p-type dop-ing were used. The initial thickness of the wafers was 5625mm. The process starts with deposition of a 100 nthick nitride layer on two wafers@Figures 2~a! and 2~b!#.Subsequently the nitride on the first wafer is dry etchedform 2–6 mm diam circular patterns@Figure 2~c!#. Thesecond wafer is patterned with square openings and slarly etched@Figure 2~d!#. The next step consisted of KOHetching of the cavity@Figure 2~e!# in the first wafer and ofthe through holes in the second wafer@Figure 2~f!#. Thedepth of the cavity was within the range of 25–75mm.After stripping the nitride from both wafers@Figures 2~g!and 2~h!#, the wafers were bonded using a silicon direbonding process in such a way that the opening in theond wafer was in the center of the cavity of the first wa@Figure 2~i!#. The test structure was then ready for the pcision grinding experiments@Figure 2~j!#.
The above process was required to enable thinphragms ,100 mm to be produced, however, for diaphragms.100 mm thick it was sufficient to use singlwafers. In that case, 160mm deep cavities were anisotropically etched into the front side of the wafer followed bgrinding from the back of the wafer. The processingsingle wafers is illustrated in Figures 2~a!, 2~c!, 2~e!, and2~g! followed by grinding. This process was used to fordiaphragms 100 and 150mm thick.
4 Results and Discussion
4.1 Diaphragm Bending
It was observed that the grinding process induced bendin the diaphragms. The magnitude of bending was msured using a surface-profiling instrument~Alpha Step!.Bending occurred in the case of both bonded and sinwafers, which precluded the bonding process from bethe cause. The magnitude of bending depended onthickness of the diaphragm and its location on the waferall but one wafer the diaphragms had a convex shapedistortion magnitude, defined as the perpendicular distabetween the top of the diaphragm and the wafer surfa
Microfab., Microsyst., Vol. 1 No. 2, July 2002
Downloaded from SPIE Digital Library on 02 Feb 20
gsr
i-
c-
-
-
g-
ege
de,
and ranged between 3.4 and 60mm for the diaphragmthickness between 150 and 25mm, respectively. Dia-phragms with a collapsed shape were observed on onthe wafers with a diaphragm thickness of 25mm. Figures3~a! and 3~b! show typical profiles of a 100mm thick, 6mm diam diaphragm and a;25 mm thick semiconcave 6mm diam diaphragm, respectively. For 6 mm diaphragwith thickness<50 mm, most of the samples exhibitecracks after grinding.
The range of bending magnitudes within a wafer fomeasured diaphragm thickness for 6 mm diam diaphraare given in Table 1. The range of bending magnitudesdiaphragms 2–6 mm in diameter and 50mm thick is givenin Table 2.
4.2 Bending Mechanism
Two potential aspects of the grinding process and destructure that could cause bending stress were vacuumsure acting on the bottom of the diaphragms and/or laca support for the diaphragms during the grinding proce
In order to investigate the first aspect, the back ofwafer was tightly sealed before grinding so that the vacudid not affect the diaphragms. After grinding the bendstill existed, indicating that vacuum was not the main caof the bending.
In order to verify the second hypothesis the followiexperiment was implemented. Three plain wafers wthinned to 250mm by precision grinding. Since no cavitiewere present, the underlying bulk silicon constantly acontinuously supported the surface during the grinding pcess. After grinding, 6 mm diam cavities were etchedcording to steps~a!, ~c!, ~e!, and~g! in Figure 2. The cavi-ties in each wafer were etched to a different depth usKOH aqueous solution. The cavity depths were measuusing the Alpha Step and the diaphragms were determto be 25, 50, and 150mm thick. After etching the cavities iwas observed that, irrespective of their thickness, no being of the diaphragms occurred. This reinforced the pposal that the bending was due to a lack of support fordiaphragms during the grinding process. Vacuum, howemay enhance diaphragm bending.
The forces that induce bending could have acted eiparallel or perpendicular to the wafer surface or a comnation of both. It was assumed that the distribution of thforces was symmetrical with respect to the center of
Table 2 Range of bending magnitudes for 2–6 mm diam, 50 mm thick diaphragms.
Diaphragm diameter(mm)
6 4 3 2.5 2
Bending magnituderange (mm)
25.6–58.3 8.8–13.3 1.8–5.6 0.8–3.6 0.7–17
10 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
ver-
is-gman-
ofinex-ed-
h it,ngturnm
Prochaska et al.: Investigation of precision grinding . . .
Fig. 4 Illustration of the mechanism of diaphragm bending forma-tion during grinding.
en-
r-
if-th
-afeene
ssr-
dia-
a-vi-w-ndbe
gmsfer
woon,
end-ox-ns
Downloaded from SPIE Digital Library on 02 Feb 201
e
r.
experimental results were in good agreement with an aage ratio of 2.13~30 diaphragms were measured!.
Forces parallel to the surface occur when there is a mmatch in the area of the top and bottom of the diaphracaused by shrinking of one side or/and expansion ofother. Such a mismatch is induced during the processgrinding by nonuniform removal of the silicon materialthe diaphragm region because of the reduction in forceerted by the silicon on the grinding teeth. This is illustratin Figure 4. Due to the increasing flexibility of the diaphragm and because of the lack of support underneatthe diaphragm will deflect, with the largest deflection beiat its center and the smallest near the edges. This inwill cause nonuniform removal of the silicon material frothe diaphragm~more material removed near the edges, lenear the center! and as a consequence will induce a diffeence between the area of the top and bottom of thephragm and result in bending of the diaphragm.
5 Bending Prevention
The most straightforward way in which to prevent diphragm bending is to grind the wafers before etching caties and the formation of a diaphragm. This solution hoever applies only to the cases in which the total cavity adiaphragm thickness is large enough for the wafers tohandled as single ones. In cases in which the diaphraare thin and/or it is necessary to perform silicon wabonding before grinding~see Figure 2!, an easily remov-able support must be provided for the diaphragms. Ttechniques, based on SOI technology and porous silicrespectively, were investigated.
5.1 SOI Technology
SOI technology has been used to prevent diaphragm bing. The process steps are shown in Figure 5. Silicon diide, 0.2–0.3mm thick, was grown on two batches of silicowafers @Figures 5~a! and 5~b!#. Subsequently the wafer
diaphragms because of the bending symmetry.If the forces that induce bending act in the plane perp
dicular to the wafer, the bending magnitudew should begiven by10
w}1
h4 , ~1!
whereh is the diaphragm thickness.Alternatively, if the forces act parallel to the wafer su
face, then the bending magnitudew should be given by10
w}1
h2 . ~2!
In comparing bending magnitudes for diaphragms of dferent thicknesses, it was observed that in most casesdependence followed Eq.~2! reasonably closely which suggests that bending stresses act in the plane of the wEquation~2! would suggest that the bending ratio betwe100 and 150mm thick diaphragms would be 2.25. Th
Fig. 5 SOI technology process for diaphragm bending prevention.
169J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
170 J. Microlith.,
Fig. 6 Apparatus for porous silicon formation and schematic cross section of the cavity region of asilicon wafer.
essyerl-
rees-
d
a-gio
feroe
ric
sthe
ingg
theota-
or-as
tortedto,
s.to
asre-usor-
h 2ions
ia-m-be
teredp,edd
me
were directly bonded@Figure 5~c!# and the oxide removedfrom the back of the wafers. Precision grinding@Figure5~d!# was used to form a SOI layer of desired thickness,w,corresponding to the depth of the cavity and the thicknof the diaphragm. Cavities were etched into the SOI la@Figure 5~e!#. For single wafer test structures, this was folowed by etching from the back@Figure 5~f!# with the oxideacting as an etch-stop layer. If bonded structures arequired, chemical mechanical polishing of the SOI is necsary to ensure a smooth surface at step~d!. Followingbonding@Figure 5~g!#, the excess silicon would be removeby etching to give the structure shown in Figure 5~h!.
No bending should occur, in either single or bonded wfer structures, because in both cases the diaphragm reis supported by silicon during the grinding steps.
An experiment was performed using the single waprocess on a 100mm SOI layer with the cavities etched ta depth of 75mm. Subsequently isotropic etching from thback of the wafer took place using 48% hydrofluoacid:70% nitric acid:glacial acetic acid, 10:25:12~HNA!solution to remove approximately 200mm of the siliconfollowed by KOH etching until the oxide layer wareached. During the etching stages the front surface ofwafer was protected either by a special jig or by coatwith wax. Even without removal of the oxide, no bendinof the diaphragms was observed. Although employingSOI method totally eliminates diaphragm bending it is ncost-effective because it requires an additional silicon wfer. A more cost-effective technique that employs the fmation of porous silicon in the cavities before grinding winvestigated.
Microfab., Microsyst., Vol. 1 No. 2, July 2002
Downloaded from SPIE Digital Library on 02 Feb 201
-
n
5.2 Porous Silicon
A process that employs porous silicon was investigatedprevent diaphragm bending. The fabrication process stawith the dry oxidation of two batches of silicon wafersform a 6–15 nm thick protective oxide. Silicon nitride300–400 nm thick, was then deposited on both batche
On batch 1, the front nitride and oxide was patternedcreate a cavity window@see Figure 2~c!# and the backnitride/oxide removed by dry etching. Porous silicon wformed in the exposed silicon using an electrochemicalaction that converts the exposed bulk silicon into poromaterial. The nitride was subsequently removed usingthophosphoric acid.
Processing of the second batch followed steps~b!, ~d!,~f!, and ~h! in Figure 2, and in step~h! the nitride wasremoved and the underlying oxide left in place.
Wafers from batch 1 were bonded to those from batcusing an aligned silicon direct bonding process. Precisgrinding resulted in the formation of silicon diaphragmabove the porous silicon regions. The final~optional! step isthe removal of porous silicon from underneath the dphragms; the high etch selectivity of porous silicon copared to that of bulk material means that this canachieved with little effect on the diaphragms.
Porous silicon was formed using 1:1:2 HF:ethanol:waand 4:1 HF:ethanol solutions. The wafer was illuminatfrom the back using a 240 W tungsten–halogen lamshown in Figure 6. The current densities applied rangfrom 4 to 9 mA/cm2 and the depth of porous silicon rangebetween 7 and 15mm. After porous silicon formation andbefore bonding, the porous silicon was removed from socavities by a short immersion in aqueous KOH solution~40
Fig. 7 Infrared images of parts of the bonded wafers with (a) poroussilicon removed from the cavities before bonding and (b) cavitiesfilled with porous silicon.
Fig. 8 Profiles of the diaphragms after grinding: (a) 75 mm thickwithout porous silicon and (b) <75 mm thick with porous silicon un-derneath.
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
Fig. 9 Profile of a diaphragm with porous silicon support.
ofe 7beo-
o-as
indowrst
hepli
p-tha
s isthetedt th
n-likehete
hedsstheninia-at
Downloaded from SPIE Digital Library on 02 Feb 201
Fig. 10 Infrared image of the fully and partially bonded diaphragmswith the diaphragm bending magnitude indicated.
inity,
glya-h
sig-ing
byfa-ni-ili-the
thesitee
dia-isbe-nd-mstrat-st-thn
wt. %!. This allowed evaluation, within the same wafer,the effect of porous silicon on diaphragm bending. Figurshows infrared images of parts of a bonded wafer pairfore grinding where four of the cavities are filled with prous silicon and the other four are not.
After the grinding stage and before removal of the prous silicon, the amplitude of bending was measured. It wobserved that the presence of porous silicon during gring strongly suppresses diaphragm bending. Figure 8 shbending profiles of two 6 mm diam diaphragms. The fione @thickness 75mm, Figure 8~a!# was formed withoutporous silicon and the second one@thickness<75 mm, Fig-ure 8~b!# with porous silicon. It can clearly be seen that tuse of a porous silicon support reduced the bending amtude by a factor of 4, from.20 to 5mm. The profile of thediaphragm is also noticeably different, with the unsuported diaphragm having a dome shape and the onewas supported being much flatter. It is believed that thidue to the porous silicon allowing some movement indiaphragm during grinding but the movement is restricwhen the porous silicon becomes compressed againssupporting wafer.
Some profiles of the diaphragms with porous silicon uderneath showed a small tip at the center of the profile,in Figure 9. The existence of the tip in the center of tdiaphragm may be explained by the fact that in the cenof the bottom of the cavities there are openings etcthrough the bottom wafer. During grinding there is lesupport provided for the area of porous silicon aboveopening. Thus the area of the diaphragm above the opewill bend more during grinding than other parts of the dphragm and lead to higher postgrinding distortion in th
-
-s
-
t
e
r
g
area. This is further confirmed by the fact that in sampleswhich the throughhole was toward one side of the cavno tip was noticed in the profile.
It was also observed that the diaphragm quality strondepends on the bonding quality in the vicinity of the diphragm’s edge. Any voids resulting from bonding whiccome into contact with the cavity area usually cause anificant increase in the diaphragm postgrinding bendmagnitude as illustrated in Figure 10.
The bending magnitude of the diaphragms supportedporous silicon depends mainly on two factors: the intercial oxide thickness between the wafer surface and thetride mask, and the structural properties of the porous scon used. Figure 11 illustrates the dependence ofdiaphragm bending magnitude on the two factors.11 Thepresence of an interfacial oxide creates a gap betweensurface of the porous silicon and the surface of the oppowafer. During grinding this allows some deflection of thdiaphragm and consequently postgrind bending of thephragm. Figure 11~a! shows that when the thickness of thinterfacial oxide is reduced the diaphragm bending alsocomes reduces. With no interfacial oxide, diaphragm being is minimal. In comparison to unsupported diaphragbending suppression was up to a factor of 300, demonsing that use of porous silicon can virtually eliminate pogrinding diaphragm bending. Macroporous silicon wipore size 1–3mm was found to be the most effective isuppressing diaphragm bending@Figure 11~b!#. Even dia-
Fig. 11 Effect of (a) interface oxide thickness and (b) type of porous silicon on the diaphragm bendingmagnitude: (i) nanoporous, (ii) mesoporous, (iii) macroporous (pore size 1–3 mm), and (iv)macroporous (pore size 3 to over 10 mm), diaphragm thickness 50 mm (Ref. 11).
171J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
Fig. 12 Dependence of the shift of the Raman phonon line (at ;520cm21) on the position on the diaphragm (the diaphragm edges areshown by the vertical lines) for a porous Si supported diaphragmafter removal of porous Si. The positive shift corresponds to com-pressive stress, which varies from ;13107 Pa at the center of thediaphragm to 1.43108 Pa at the edges.
s
ortpec
use
up,
inro-
mith
ion
fcals-ofl-sned,n.en-
hip,
172 J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
Downloaded from SPIE Digital Library on 02 Feb 201
ed-
d
tive in the microscope and Ar ion laser with a 514 nexcitation line allows Raman imaging of the surface wspatial resolution of;0.6 mm.
6.1.1 Results and discussion
The crystalline silicon Raman spectrum in a phonon regmainly consists of a narrow peak around 520 cm21 with ahalf width of about 3.5 cm21. The spectrum is a result oscattering by long-wavelength transverse optiphonons.12,13 With a state-of-the-art spectrometer it is posible to identify a shift in the Raman band of the order;0.01 cm21.12 Removal of the background baseline folowed by line fitting using a Lorentzian function allowthree components of the Raman spectrum to be determinamely, the intensity, the half width, and the positioThese variations are related to the composition, defect dsity, and magnitude of stress, respectively. A relationsexists between the stress,s, ~in Pa! and the Raman shiftDv ~in cm21!:12,15
Dv52231029s, ~3!
Fig. 13 Dependence of the shift of the Raman phonon line (at ;520cm21) on the position on the diaphragm for a SOI based diaphragm.Virtually no stress is observed.
phragms with thickness 25mm or less supported by porousilicon exhibited no cracks or damage after grinding.
6 Stress Measurements in the Diaphragms
Stress measurements in the both unsupported and suppdiaphragms were performed using Raman and x-ray stroscopies and the results are reported here in Sec. 6.
6.1 Stress Measurements Using RamanSpectroscopy
Raman spectroscopy has recently been successfullyfor stress measurements in silicon.12–17Its main advantagesare its nondestructive character, the simplicity of its setand the short time required for obtaining data.
In the present work Raman spectra were registeredbackscattering geometry using a Renishaw 1000 micRaman system equipped with a Leica microscope andXYZmotorized stage. The use of a 1003 magnification objec-
Fig. 14 Raman spectrum obtained (a) before and (b) after removal of the top surface layer aftergrinding.
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
tresslts
Prochaska et al.: Investigation of precision grinding . . .
Fig. 15 X-ray stress analysis of the 100 mm thick diaphragms.
frenessethe
ion
tedfroilar50
ngia-
ofnow
u-
rveon,re
aftee.faseonsumre
ecref-e
ith
ia-
thetom-m-mal
16.urelec-
elec-lat-theaf-
eenia-aryu-
ken0
ualnto
earaterenforatese ofnsn
Downloaded from SPIE Digital Library on 02 Feb 201
e
,
m
n
d
r
respect to the reference sample indicates compressive sin the diaphragms, which is in agreement with the resuobtained by Raman spectroscopy.
7 Simulation
In order to evaluate the performance of the bent dphragms, three-dimensional~3D! finite element method~FEM! analysis of the diaphragms was carried out usingCFD-FEMSTRESSpackage. The main emphasis was giventhe deterioration in deflection of the bent diaphragms copared to that of an ideal flat diaphragm. The two most comon diaphragm operating modes were simulated: norstress mode and shear stress mode, shown in FigureNormal stress is most commonly encountered in presssensors and in inkjet printheads that use stacked piezoetric actuators. Shear stress mode occurs when a piezotric device is attached on top of the diaphragm due toeral shrinkage during actuation. It was assumed insimulations that the diaphragms have uniform thicknesster grinding.
The simulated diaphragms were 75 and 50mm thick.The graphs in Figure 17 show the percentage ratio betwthe deflection of bent diaphragms and that of a flat dphragm of corresponding thickness. The same boundconditions applied to all the diaphragms in respective simlation modes. The geometry of the diaphragms was tafrom the Alpha Step profile by probing the profile at 8–1points and subsequently applying interpolation. Residbuilt-in stress in the bent diaphragm was not taken iaccount in the simulation.
In comparing curves corresponding to normal and shmodes in Figure 17 it is seen that bending causes gredeterioration in the performance of the diaphragms whused in normal mode than when in shear mode. Also,the same bending magnitude the performance deteriormore with a decrease in diaphragm thickness. In the cas75 mm thick diaphragms it is seen that for small distortio~,10 mm! performance deterioration in terms of deflectio
Fig. 16 Two basic actuation diaphragm modes: (a) normal and (b)shear stress.
whereDv5(vstress2v ref) ~in cm21!, vstressis the peak fre-quency of the phonon band of silicon under stress, andv refis the peak frequency of the phonon band of the stresssilicon wafer. A positive or negative shift in the Ramapeak position corresponds to compressive or tensile strrespectively,12 assuming uniaxial stress only, i.e., within thplane of the wafer. Figure 12 shows the dependence ofshift of the Raman phonon line~at ;520 cm21! on theposition of the diaphragm which corresponds to distributof stress across the diaphragm.
As one can see from Figure 12 porous silicon-suppordiaphragms exhibit compressive stress that decreasesthe edge of the diaphragm towards its center and simresults for unsupported diaphragms. For unsupportedmm diaphragms, stress was in the range of 0.53108– 1.253108 Pa, with most of the stress being relieved duridiaphragm bending. For porous silicon-supported dphragms stress was in the range of 0.753108– 1.83108 Pa before porous silicon removal and in the range13107– 1.43108 Pa after porous silicon removal. Ramaspectra of the diaphragms based on SOI technology, shin Figure 13, show virtually no stress, as expected~thedeviations from 0 shown in Figure 13 are within the accracy of the method!.
Surface damage on the ground diaphragms was obseto consist of phase transformation into amorphous silicwhich is in agreement with that reported in Ref. 16. Figu14 shows Raman spectra for, respectively, diaphragmsgrinding and after subsequent removal of the top surfac
The spectrum in Figure 14~a! shows the existence oSi–I amorphous phase in the top ground layer. This phdisappears after the removal of the top several micrthick surface layer from the diaphragm and the spectrshows silicon crystalline phase as illustrated in Figu14~b!.
6.1.2 Stress measurements using X-rayspectroscopy
Figure 15 illustrates the results obtained using x-ray sptroscopy. Three samples were analyzed: a plain siliconerence sample and two 100mm thick bent diaphragms. Thnegative phase shift of the bent diaphragm samples w
-
Fig. 17 Simulation of the deflection of bent diaphragms actuated innormal and shear modes for (a) 75 and (b) 50 mm thick diaphragms.
173J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
ge-asa-poing
re-m
diauseheur-co
ingporudeessRae o
roulogeremor-ms
oxrp.
s-
bio-rd,
: A
uc-
ma-
f are
ri-
andthe
ofn
ason
al
s-
ionpy,’’
ist-tros-
udys in
e
Prochaska et al.: Investigation of precision grinding . . .
is marginal. The magnitudes of deflection in this ranwere observed in thick~.10 mm! diaphragms, which suggests that no support during grinding is needed in the cof thick diaphragms. In applications in which thinner diphragms are required, the use of SOI technology or arous silicon support layer has enabled diaphragm bendto be reduced to,10 mm at which it will have a minimaleffect on diaphragm performance.
8 Conclusions
Precision grinding of silicon has been demonstrated for pcise formation of silicon diaphragms. Diaphragms 2–6 min diameter and 25–150mm thick were produced. It wasobserved that the process induces bending in thephragms if they are not supported during grinding. Theof SOI technology can virtually eliminate bending since tdiaphragm is always supported by underlying silicon ding the grinding steps, however, the process is less enomical since an additional silicon wafer and a bondstep are required. The use of porous silicon as a suplayer has been shown to significantly reduce the amplitof bending by a factor of up to several hundred. Strmeasurements of the diaphragms were performed usingman and x-ray spectroscopies and indicate the existenccompressive stress of the order of 13107– 13108 Pa inunsupported diaphragms and in those supported by posilicon, whereas the diaphragms based on SOI technoare stress free. Simulations of the bent diaphragms wperformed using 3D FEM analysis. The results for 6 mdiam diaphragms indicate that deterioration of the perfmance, in terms of deflection, is negligible for diaphragwith convex bending of,10 mm.
Acknowledgments
The authors of this article would like to thank RandLaboratories Ltd. for financial support and CFDRC Cofor providing the CFD-FEMSTRESS tool.
References
1. Y. Kazui et al. ‘‘Method for manufacturing diaphragm silicon presure sensor,’’ U.S. Patent No. 4,670,969~1987!.
3. E. Stemme and S. Larsson, ‘‘The piezoelectric capillary injectornew hydrodynamic method for dot pattern generation,’’IEEE Trans.Electron DevicesED-20~1!, 14–19~1973!.
4. K. Petersen, ‘‘Fabrication of an integrated, planar silicon ink-jet strture,’’ IEEE Trans. Electron DevicesED-26~12!, 1918–1920~1979!.
5. A. Olsson, P. Enoksson, G. Stemme, and E. Stemme, ‘‘Microchined flat-walled valveless diffuser pumps,’’J. Microelectromech.Syst.6~2!, 161–166~1997!.
6. T. Laurell, L. Wallman, and J. Nilsson, ‘‘Design and development osilicon microfabricated flow-through dispenser for on-line picolitsample handling,’’J. Micromech. Microeng.9, 369–376~1999!.
7. C. Meinhart and H. Zhang, ‘‘The flow structure inside a microfabcated inkjet printhead,’’J. Microelectromech. Syst.9~1!, 67–75~2000!.
8. K. Sadahiro, ‘‘Method of manufacturing semiconductor substratemethod of manufacturing semiconductor devices composed ofsubstrate,’’ European Patent No. 0460,437~1991!.
9. A. Prochaska, S. J. N. Mitchell, and H. S. Gamble ‘‘Productionsilicon diaphragms by precision grinding,’’ U.K. Patent ApplicatioNo. 0015500.2.
10. J. P. Den Hartog,Advanced Strength of Materials, McGraw–Hill,New York, ~1952!.
174 J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
Downloaded from SPIE Digital Library on 02 Feb 201
e
-
-
-
t
-f
sy
11. A. Prochaska, S. J. N. Mitchell, and H. S. Gamble, ‘‘Porous silicona sacrificial layer during formation of silicon diaphragms by precisigrinding,’’ Proc. Int. MEMS Workshop, Singapore, pp. 743–747~July2001!.
12. I. De Wolf, ‘‘Micro-Raman spectroscopy to study local mechanicstress in silicon integrated circuits,’’Semicond. Sci. Technol.11~2!,139–54~1996!.
13. K. Ajito et al., ‘‘Strain imaging analysis of Si using Raman microcopy,’’ J. Vac. Sci. Technol. A13~3!, 1234–1238~1995!.
14. S. Webster, D. N. Batchelder, and D. A. Smith, ‘‘Submicron resolutmeasurement of stress in silicon by near-field Raman spectroscoAppl. Phys. Lett.72~12!, 1478–1480~1998!.
15. Th. Englert, G. Abstreiter, and J. Pontchara, ‘‘Determination of exing stress in silicon films on sapphire substrate using Raman speccopy,’’ Solid-State Electron.23~1!, 31–33~1980!.
16. Y. Gogotsi, C. Baek, and F. Kirsch, ‘‘Raman microspectroscopy stof processing-induced phase transformations and residual stressilicon,’’ Semicond. Sci. Technol.14, 936–944~1999!.
17. I. Zarudi, and L. C. Zhang, ‘‘Effect of ultraprecision grinding on thmicrostructural change in silicon monocrystals,’’J. Mater. Process.Technol.84, 149–158~1998!.
Andrew Prochaska obtained his MSc de-gree in 1997 from the Technical Universityof Lodz, Poland, and in 2001 was awardeda PhD degree in the area of Silicon Micro-machining and Microtechnology fromQueen’s University Belfast, Northern Ire-land (UK). His interests include solid-stateMEMS technology. He is an author and co-author of several international and nationalpublications and holds one patent.
S. J. N. Mitchell received the BSc andPhD degrees in electrical and electronicengineering from the Queen’s University ofBelfast in 1982 and 1986, respectively. Hisresearch interests are in the developmentof semiconductor process technology andhe has over 70 journal and conference pa-pers published in this field. In recent yearshe has been involved in the developmentof bonding and silicon processing tech-niques for silicon micromachining applica-
tions. Particular emphasis has been on the development of technol-ogy for the fabrication of chemical microanalyzers and microfluidicdevices.
Tatiana S. Perova received her MSc de-gree in Physics in 1969 from the Tajik StateUniversity at Dushanbe (Russia). She com-pleted her PhD in Molecular Physics atLeningrad State University in 1979. Dr.Perova joined the staff of Vavilov State Op-tical Institute (St. Petersburg, Russia) in1979, where she was involved in the char-acterization of condensed matter using far-infrared and Raman spectroscopies. In1998 Dr. Perova took a position at the De-
partment of Electronic and Electrical Engineering of the University ofDublin, Trinity College, where she has been involved in optical char-acterization of liquid crystals and thin films formed on silicon.
Remy N. Maurice has worked as a re-search assistant at the Department of Elec-tronic and Electrical Engineering, Univer-sity of Dublin, Trinity College, sinceFebruary 2000. He has been involved inthe phase transformation and stress analy-sis in silicon structures using micro-Ramanspectroscopy. In 1999, he received hisMSc in condensed matter from the Univer-sity of Lyon, France, after he had com-pleted the MSc degree at ISTIL, an engi-
neering school, in Lyon, in 1998.
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
Prochaska et al.: Investigation of precision grinding . . .
Paul Baine graduated with a BEng degreefrom Queen’s University of Belfast. Hewent on to earn his PhD from Queens Uni-versity in 1997. His PhD thesis dealt withthe fabrication of thin single crystal silicondevices on glass using electrostatic bond-ing. After completion of his PhD, Paul tookresearch posts at the university, expandingto the area of SOI where he has been in-volved in the development of novel tech-niques for the thinning of SOI material. He
has also been involved in the bonding of nonstandard materials,including buried multilayer structures. Paul currently holds the postof Senior Microelectronics engineer at the Northern Ireland Semi-conductor Research Center. His current research activities includeSOI, MEMS, and materials science.
Downloaded from SPIE Digital Library on 02 Feb 201
H. S. Gamble, a professor, has worked onsilicon devices and related technologysince 1966. He set up the first UK univer-sity polysilicon gate process for MOSTs in1973 at Queen’s University, Belfast, andsubmicron gate MOSTs were produced asearly as 1981. He was the first to employrapid thermal diffusion for the production of250 nm deep boron junctions, which hadthe world’s lowest leakage currents at thetime. Bonded SOI substrates are of interest
for bipolar and Smart-power applications. A unique SOI substratesuitable for MMIC applications was developed. CVD of metals isnow being investigated for copper interconnects, barrier layers andfor magnetic devices.
175J. Microlith., Microfab., Microsyst., Vol. 1 No. 2, July 2002
0 to 134.226.1.229. Terms of Use: http://spiedl.org/terms
