NIPPON STEEL TECHNICAL REPORT No. 83 JANUARY 2001
Cleaning Technology of Silicon Wafers
Abstract
ich UEMURA
Teun HAIBARA
‘eshiro MOR
High quality wafers are in demand, as increasingly high integration of VLSIs
became a commercial practice. This is especially true with highly clean wafers with
little metallic impurities and particulate and organic matters on the surface. For
producing the highly clean wafers, it is necessary to establish wafer surface clean-
ing technologies through experimental and theoretical analysis of the behavior of
the surface impurities. This paper explains adhesion mechanisms of metallic and
particulate impurities to silicon wafer surfaces and proposes some cleaning meth-
ods.
1. Introduction
Increasing integration density of LSIs (large scale integration cir-
cuits) call for higher quality silicon wafers. Higher quality wafers
‘mean the superior quality in crystal precision, shaping quality, and
surface quality. Problems concerning chip size increase and manv-
acturing cost increase must be considered. Practical application of
'300-mm wafors has been discussed in recent years. Extremely high
‘surface cleanliness is required of the 300-mm wafers” (reer to Fig.
1), Specifications for organic impurities have begun tobe prescribed
‘as well as traditional impurities such as metals and particles. These
‘impurities are generated from wafer manufacturing equipment,
chemicals, human bodies clean room materials, dust-free garments
and reaction products, and are often deposited on the wafer surfaces
‘They cause interconnect failures and crystal defecs, then degrade
Deine)
Fig. 1 Relationship between device design rule and required wafer
surface cleanness"
“1” Testnieal Development Bureau
“61
the device performance. For this reason, the surface cleanliness of
silicon wafers is continuously improved.
"To meet these requirements, itis important to theoretically ana-
|yze the phenomena tha are taking place on the silicon wafer sur-
face and to perform the process design on the basis of the results of
the theoretical analyses. This report especially deals withthe adhe-
sion mechanisms of metas and particles tothe wafer surface, and in-
‘woduces new cleaning methods besed on our theoretical examination,
2, Adsorption mechanism of Metals to Silicon Wa-
fer Surface
Since iron on silicon wafer surface causes crystal defects and
degrades device performance, as much oft should be removed as is
possible. Henley etal” indicated that an iron contamination level of
10" em” is required fora gate oxide thickness of 8 nm. Like iron,
‘other metals (€., nickel, chromium, copper, zine, and sodium) de-
{grade the device performance (or change the transistor threshold
Values, induce erystal defects to degrade the p-n junction leakage
and pate oxide integrity, or accelerate oxidation). They must be re-
‘duced to meet shrinking device sizes. To remove these metals or to
contol their adsorption in the cleaning Solution, it s necessary to
understand their behavior in the cleaning solution.
"Two major mechanisms are present for metal adsorption inthe
leaning solution. One is chemical absorption, as is represented by
the adsorption of metals to native oxide (SiO, in an alkaline solu-
tion. The other is electrochemical adsorption to bare Si in an acidic
solution, To understand these mechanisms, we conducted ionic equi-NIPPON STEEL TECHNICAL REPORT No. 83 JANUARY 2001
[ibrium analysis and electrochemical analysis, The results ofthese
stugles are discussed below.
2.1 Tonic equilibrium analysis
‘An NH,OH+H,0,+H1,0 mixture (APM), also called “Standard
Cleaning solation I (SC-I)" which i a representative alkaline clean-
{ng solution, has such an extremely excellent particle removal capa-
bility that itis widely vsed®. However, if metal ions exist inthe clean-
{ng solution, they adsorb on the wafer surfae®. Especially, iron, when
present even ina trace amount less than parts per billion (pp), de-
‘grades the electrical performance such as recombination lifetime,
It poses a serious problem when itis caried over from the previous
processor when low-purty chemicals are used.
Metal adsorption in the SC-1 cleaning solution was traditionally
‘explained using metal oxide formation enthalpy or a complex ion
‘model, but was rarely discussed quantitatively. An analysis of the
issue was tied according tothe chemicel equilirium theory”.
Inthe SC-1 cleaning solution, formation of native oxide (Si0,)
by hydrogen peroxide and etching of SiO, by alkalis proceed atthe
same time, For this reason, the overall wafer thickness is slowly re-
‘duced, but a certain thickness of SiO, exists on the wafer surface,
inrespective ofthe time. In this case ifthe etching ais slow enough
1nd the metal adsorption and desorption reactions are fast enough, &
‘certain metal ion adsorption equilibrium holds between SiO, and the
solution. Actually, since te etching rate of SiO, by the $C-I clean
{ng solution is afew mono-layers per minute” while the adsorbed
amount of metals reaches @ constant value within one second after
‘wafer immersion, this equilibrium may be considered to hold. The
relationship between the metal on concentration in the solution and
the adsorbed amount ata given temperature is called “adsorption
isotherm’. Fig. 2 shows adsorption isotherms fora 1:1:5 mixture of
29% ammonium hydroxide, 31% hydrogen peroxide, and de-ion-
Jed water, andthe temperature of 80°C,
Here, the ionic equilibrium in the solution is focused on. For 2
‘metal jon denoted by Mf, the following chemical equilibrium holds
inthe SC- cleaning solution:
M+ OH'= M(H)
M(OH)" + OH" = M(OH),"*
M(OH), “+ OH= M(OH),"*
‘The existing proportion of metal hydroxide complexes denoted by
.M! to M(OH),**is a function of pH and can be theoretically caleu-
lated if each equilibrium constant is given",
‘The results of calculation for Fe in a concentration of 1x 10"
‘mol/l. (about 05 ppb) are shown in Fig. 3), Since the pH of the
‘SC-I cleaning solution is about pH 10.6 (for 1:1:5 proportion), the
Fig. 2 Adsorption isotherms of Fe", NP, and Zn in SC-1 solution
(138, 80°C)
+62
nr \\
eseiotnoit
ISN
25810 7eatt=
ASA
creme
Fig. 3 pH dependence of Fe hydroxide complexes
isting proportion of hydroxide complexes is Fe(OH), > Fe(OH),
> Fe(OH), > Fe(OH? > Fe in decreasing order. From tis sul
however, it eannot be eietyconluded ta he sored spies in
the SC- leaning sluion s FOH), Ifthe reaction iin chemical
equilicum, th adsorbed emount depends on the prac the con-
entation and adsorption equbrium constant ofeach complex (if
complex exists in a small amount bt ely large in equilibrium
Constant, it can become a main adsorbed species). A further study
was mad o identify the adsorbed species.
Fe(OH), a hydroxide complex of Fe witoot charges can dis
solve nan aqueous solution to eran concentration, but becomes
sold and precipitates when texcedsthe erie eancentation d=
termined by the solubility product. The dissolved species and solid
species are denoted by FH), (aq) and Fe(OH), (5) respecively
{oq ands sand fr aghoots and sol, respectively), The results of
{hecaleltion including the formation af the sali species se shown
inFigs. 306) and (c) When the Fe concentration reaches bout 5
10" mol, Fe(OH), 5) sas to form, When Fe is ded to higher
concentration, the concentration ofthe other dissolved hydroxide
complexes saturates in the pH range of 41011
"The caleslated concentration of Fe(OH) aq) a pH = 10.6, (ot
responding tothe pH of te SC lenin ston, is shown in Fig
4), The measured adsorption isotherm of Fes shown in Fig (5)
‘The star of formation of Fe(OH) almost coincides with the start
of saurstion ofthe adsorbed amount of Fe, This means that the
Adsorbed species is one ofthe dissolved hydroxide species
Fe(OH) aq) ater than the solid species Fe(OH), (). The finding
thatthe sole species isnot he adsorbed species supported by the
mmessoroment by tal-refiestion x-ray Muorescene spectroscopy
COXRE.
Tig § shows the TXRF angle-scan measurement results fora
wafer produced inthe region where the sli species Fe(OH) xNIPPON STEEL TECHNICAL REPORT No. 83 JANUARY 2001
i a
= ol
i
|
t
i
i
a Fe xem a)
Fig. 4 Comparison of calculated values of FOF), (aq) concentration
{@) and measured values of Fe adsorbed amount (b) tthe pH
or 10s
eye)
‘Gancae sete)
Fig. 5 Comparison of TXRF angle-sean measurement results of wafers
with Fe adsorption in Fe(OH), (5) formation region and non-
Fe(OH), (6) formation reglon
formed and for a wafer produced inthe region where the Fe(OH)(s)
{snot formed, The verical axis indicates normalized maximum val-
ves. The TXRF angle-scan profile depends on the depth distribution
of analyte elements. The dissolved species and solid species are con-
sidered to adsorb inthe bulk of SiO, in film form and on the surface
(of SiO, in particle form, respectively. Ite slid species adsorbs, the
angle-Scan profile should be diferent from that obtained wien the
dissolved species adsorbs. In actual data (Fig. 5), both profiles al-
‘most agree with each other. This may be taken to mean thatthe solid
species does not adsorb,
‘The above-mentioned experiments show thatthe adsorbed spe-
cies is one ofthe dissolved hydroxides. When Fig. 3 is looked at here
again, the pH dependence ofthe existing proportion ofthe hydrox-
ide complexes varies with each other. Ifthe pH dependence ofthe
adsorbed amount is measured and compared with te calculated val-
tes, the adsorbed species can be identified. By adding tetramethy-
lammonium hydroxide (N(CH,) OF, TMAH) tothe SC-1 cleaning
solution, the pH was changed in the range of 10.5 ro 12, and the pH
dependence of Fe adsorption was investigated. The results are shown
in Fig. 6.
‘The slope inthis pH range was ~0.56 in linear approximation.
‘The slopes expected from the chemical equilibrium calculation are
given for each dissolved species in Table 1. When x = 3, the caleu-
lated value agreed best withthe measured valve. This means that the
main adsorbed species isthe neutral hydroxide complex Fe(OH)(aq)
dissolved inthe SC-I cleaning solution. The greatest likelihood of,
neural hydroxide complexes o adsorb is supported by the theoret-
-63-
ot
Fig. 6 pH dependence of adsorbed amount of Fe, Ni and Zn
‘Table 1 Comparison of pH dependence for adsorbed amount and for
hydroxide complex concentration
‘eesured value | Caleoledvaloe wit expen MOA) 00)
(Slope at pH= 1055-12) we! [x2 [xed | xe
036 | 338 | -238 | -138 | 038 | Oa
si ost | 27g | 1.78 | -078 | 4022 | 1.22
zo 4013 | -268 | -1463 | -063 | +037 | +197
cal calculation proposed by RO. James etal. for metal adsorption.
[Ni and Zn were also studied. These metal ions are known to react
‘with ammonia to form extremely stable ammine complexes. The
‘existing proportions of various complex species calculated by con-
sidering the formation of the ammine complexes are shown in Figs.
“Tand 8, From these diagrams, it can be seen that most of Ni and Zn
exist as aramine complexes inthe SC-1 cleaning solution (pH = 106.
‘Table 2 shows the critical concentration, calculated from the solu-
bility product of Ni(OH),(s) and Za(OHD,(s), at which these species
‘stato precipitate It is evident that the Ni and Zn contaminant con-
centrations never reach the critical concentration forthe formation
m Yes
Fig. 7 pH dependence of Ni hydroxide and amine complex
foncentrations
oe
cone )
Fig. 8 pH dependence of Zn hydroxide and ammine complex
‘concentrationsNIPPON STEEL TECHNICAL REPORT No. 83 JANUARY 2001
‘Table? Ammine and hydroxide complex concentrations in ammonia-
‘eal solution
TS oe oi
spina sgn by MOM.)
'EMINH(ag) | EM(OH G9) | EM(OH),)
qootty | cmottey | emote
NE | bao 710" =I
za | bein sci <0!
ofthe respective solid species in usual, This isthe eason why the
_dsorpion isotherms of Ni and Zn do not level off Fe does in Fig.
2.
‘By measuring the pH dependence of the sorbed amount for Ni
and Zn in the same manne as Fe, it maybe possible to identity the
adsorbed species, The measured values are shown in Fg. 6, andthe
calculated values are given together with those for Fe in Table 1. For
[isthe calculated values agreed best wih the measured values when
1n=2. This suggests that the adsorbed species isthe neueal hydrox-
{de complex Ni(OH), (aq) dissolved in the SC-1 cleaning slution as
noted for Fe. Zn des not show any match between calculated and
‘measured values. This may be explained by formation ofthe poly-
‘clear complex Za,(OHD,” inhibitor effet of TMAH, or satura-
tion ofthe adsorbed amount, buts identified yet.
‘According othe above results, itisblieved thatthe adsorbed
species are not ammine complexes but hydroxide complexes also
for Ni and Za, To substantiate ithe adsorption behavior of Nin a
solution, which doesnot contain ammonia, was investigated The
‘I of the system was adjusted with potassium hydroxide (KOH) in
place of ammonia, andthe pH dependence ofthe adsorbed amount
‘vas measured. The esl are also shown in Fig. 6. he adsorbed
‘mount was larger by mare than one order of magnitude than when
‘ammonia was contained. This i probably because a large amount of
hydroxide complex, which is considered to he the adsorption spe-
cles, was formed. Te adsorbed amount, however, ineeass only by
‘one or to orders of magnitude (egans si to nine orders of magai-
tude expected from Fig. 7) and exhibits no pH dependence. This
may be because the adsorption sits are ssturated
From the above discussion, it was found thatthe adsorbed spe-
cies in SC-1 solution i neutral hydroxide complex dissolved inthe
solution. This finding canbe puto use in developing methods for
reévcing meal adsorption inthe SC- cleaning soltion. For example,
lowering the pH to reduce the existing proporion of Fe(OH, (ag)
may prevent the Fe adsorption. Or adding a chelating agent that
forms a stable complex with Fe(OFD, (aq canals be effective.
2.2 Metal adhesion mechanism in acd solution
Acid cleaning solutions ae excellent in meu removal charac-
terse, but the adesion of Cu and Au is considered problematic in
solutions, such asa Slaton of ited hydrofluoric acid". These
adhesion behavior ae explained by using the ionization tendency.
Given the conduction type, resistivity and anion effect, however, it
is ficul to explain them simply by the ionization bebavioe alone
‘To suppor this thinking, th adhesion behavior of metals in acidic
solutions by electrochemical technique was analyzed’ The results
ar introduced here. Firs, the relationship between the pH of the
solution and the metal adhesion amount is described. In this ease,
the recombination lifetime was examined by the b-PCD method to