Diamond growth in and above trenches in silicon

ELSEVIER

Diamond and Related Materials 6 (1997) 1019-1025

DIAMOND AND R[L T[D T[R|AL

Diamond growth in and above trenches in silicon 1 G. Schaarschmidt *, B. Mainz, S. Laufer, T. Werninghaus, D.R.T. Zahn, H.-J. Hinneberg Fakultdt fiir Naturwissenschaften, Technische Universitgit Chemnitz-Zwickau, D-09107 Chemnit:, Germalo' Received 9 September 1996; accepted 18 November 1996

Abstract We have investigated the deposition of diamond lilms onto trench structures in silicon substrates with different aspect ratios (i.e. depth to width ratios}. Deposition has been performed by microwave plasma-assisted CVD preceded by an in situ bias pretreatmenl to support nucleation. For comparison, bias pretreated samples have been coated ex situ by hot filament CVD and a hollow cathode arc discharge technique, respectively. The experimental observations suggest that diamond grows in the trenches with a nearly constant deposition rate until the moment when the diamond film above the trench is completely closed. Cross-sectional Raman spectroscopy has been applied to get a view inside trenches. These measurements have revealed differences in phase purity of diamond which has been deposited at distinct stages of growth. In general, very similar results have been achieved with the three different deposition techniques applied. Differences in film structure can be explained, assuming that diamond growth is controlled by the diffusion of hydrocarbons and atomic hydrogen from the plasma to the bottom side of the trench. © 1997 Elsevier Science S.A. Kej'wordv." Cross-sectional Raman spectroscopy; Diamond growth

1. Introduction Diamond films of high perfection and phase purity can be deposited by means of a number of thermally activated or plasma-assisted low pressure CVD-methods [ 1]. For technological application, the capability of these methods to coat non-plane substrates and topological structures might be crucial. Apart from coating tools with inclined surfaces and lateral dimensions in the range of centimetres, structure sizes on a micrometer scale are also of great interest, as typical of electronic device manufacture. In an earlier paper [2] we reported the deposition of diarnond films by means of microwave-assisted C V D on patterned silicon substrates, A,a essential advantage of the M W C V D - m e t h o d is the possibility to ensure sufficient nucleation by substrate bias pretreatment, especially since it is unsuitable to apply any kind of mechanical pretreatment in the case of microstructures. Nucleation during the bias phase takes place at all locations under the influence of ion impact. Starting from these locations, diamond grows isotropically during subsequent * Corresponding author, c-mail: schaar(q!physik.tu-chemnitz.de ~Paper presented at the 7th European Cor,ference on Diamond, Diamond-like and Related Materials, Tours, 8-13 September 1996. 0925-9635/97/$17.00 (c')1997 Elsevier Science S.A. All rights reserved Pll S0925-9635(96)00769-8

deposition. In this way, steps on the surface can easily be overgrown, while exterior edges are growing round. The top layer overgrows trenches in the silicon substrate that are filled up with diamond to a thickness which corresponds nearly to half the french width independent of trench depth. Thus only trenches with an aspect ratio less than or equal to 0.5 can completely be filled with film material. These experimental observations suggest that diamond grows in the trenches with a constant deposition rate until the mement when the diamond film above the trench is completely closed. It was the objecli~'e of the present contribution to prove the deposition characteristics obtained for M W C V D in comparison to other deposition methods. Furthermole, the diamond quality was systematically assessed by means of Raman spectroscopy, while buried diarr, ond deposits were analysed, too.

2. Experimental details Periodic structures of trenches varying m width were produced in (100) Si-wafers using standard photolflhographic techriqttes and sttbsequent reactive plasma etching in an Ar-Cl2 atraosphere (for details see Ref. [2]). Bias pretreatment of all samples was performed in a

1020

G. SchaarschmMt et al. /' Diamomt aml Reklted Materhds 6 (1997) 1019-1¢)25

r,~icrowave plasma CVD facility which had already been used iu earlier experiments [3]. For a number of samples, after the bias step the process was interrupted and the so-pretreated samples were removed and provided for two other deposition methods: the hot filament technique (HFCVD) and hollow cathode arc discharge (HCACVD) [4,5]. Comparing investigations of trench structures deposited by means of MWCVD ensured that a process interruption with a temporary exposure to air did not influence the results. The deposition conditions of all types of sample preparation are summarized in Table 1. After deposition the samples were broken vertically to the trench direction in order to prepare cross-sections. SEM studies were carried out to assess the results of deposition and to choose appropriate locations for Raman spectroscopy analysis. Raman spectra were recorded in a spectral range of 400 to 2000 cm-~ using a Dilor-XY-Raman spectrometer equipped with a CCD camera for multichannel detection, an optical microscope providing a laser beam focus diameter of about 1 l,tm (la-Raman spectroscopy), and an automated x-y stage. For excitation laser lines 457.9 or 441.6 nm of an Ar ÷ or a He-Cd laser were used, respectively. The laser power on tbe sample surface was approximately 5 mW. After removing the luminescence background, each la-Raman spectrum was curvefitted assuming Lorentzian lineshapes tbr the first-order silicon (~520cm -~) and diamond ( ~ 1 3 3 2 c m -t) phonon contributions and the D-band ( ~ 1350 cm-~) as well as Gaussian lineshapes for the G-band (~ 1560 cm t) and a line centred near 1475 c m t [6]. To obtain reasonable lits in some cases an additional Lorentzian line at 115{)cm t was considered, its contribution being relatively small.

3. Results and discussion The tbcus diameter of the exciting laser of approximately l ~ m limited the spatial resolution. For that reason only trench structures with a relatively consider-

able depth of 5 or 7 lam were used for deposition. Deep trenches show a slightly concave cross-section because of a not optimally adjusted Ar-Cl2-ratio during the preparation of the trenches in silicon. This results in a clear separation of the upper layer from the U-shaped deposit on the trench bottom, since nucleation during bias pretreatment is only possible within the projection of the upper trench edges onto the bottom and the lower region of the walls [2]. Fig. 1 shows a typical structure after deposition. If the individual deposition parameters of each method are adjusted in such a way that qualitatively comparable films grow on the top of the substrate, the SEM cross-sectional images of trench structures are almost identical. Independent of the applied deposition method the thickness of the deposit layer in the trenches in each case corresponds to about half of the trench width between the upper edges as shown by the graph in Fig. 1. Line scans along the cross-sections of the samples were recorded by means of Raman spectroscopy to obtain further information about the deposits [7, 8]. The sample was moved under the laser focus with a minimal step width of 100 nm. Such a procedure also allows to study buried parts of the layer. Moreover, spectra taken from different sites can be related to an evolution of the deposit in time. An essential requirement therefore is breaking the samples in such a way that the substrate and the deposit form a plane. Appropriate sites of analysis had to be selected separately for each sample by means of the SEM-images. Fig. 2 presents the results of two cross-sectional line scans of a MWI type sample. The measurements were carried out along the centre of a trench of 1.8 lain width (Scan2) and along the upper layer on 0ae substrate surface (Scani). The trench was completely overgrown by a 2.6 ~tm thick diamond layer and the deposit in the trench 5 l.tm in depth was considerably separated from the upper layer. As a result of the deconvolution of the Raman spectra the line intensities of the diamond and silicon phonons as well as of the D- and G-bands are presented. Further details were omitted for the sake of better clearness.

Table I Main deposition paranleters of samples Type

Bias pretreatment

MWl HI-"

MWCVD: power !.i kW. T~b = 860C, 23 mbar, 2% CH4 in H2, 500 seem, Ub,~ = - 200 V, duration--15 rain

HCA MW2

MWCVD: power i. I kW, Ts~b = 860C, 40 mbar, 1% CHa in H2, 500 seem,additionally 0.3% He, 120 ppm N2,Ub,~s= --270 V, duration-6 rain

Deposition MWCVD: power !.1 kW, Tsub=730 C. 40 mbar, 1% CH4 in H2, 500 sccm. duration 2...6 h HFCVD: W-filament ~ 0.5 mm, Tn~=2100 C, distance to the substrate =5.5 ram, Ts,,b=93ff'C,40 mbar, 0.75% CH4 in H2, 100 sccm, duration 6...13 h HCACVD:electron current to the substrate, Is =2.3 A; Tsub= 780"C, 25 mbar, 0.7% CH 4 in H2, 500 sccm, duration 6 h MWCVD: power 1.1 kW, T~ub=800~C, 40 mbar, 2% CH 4 in H2, 500 sccm, additionally 0.3% He, 120 ppm N,, duration 8...14 h

G, Sehaarschmidt et aL / Diamond and Related Materials 6 (1997) 1019-1025

7

[]

HF - D5 - 5,8 pm

o

H C A - D5 - 5,9 pm

e

MW1-D5-2,6pm

1021

o*

@ ** ,,

=

,..,

-'"'"

• ""

O

"•

M W 1 - D 7 - 2,1 pm

E

D

,oO.

5

•

MW2-

D7 - 6,6 pm

••

.0,,. u, "

~

~ ~ •e~

.................. l i n e

-"112 .

.

.

.

4

.°°.."" o°'

~

VI

°.

..'""

3

~

.°°'••

e~~.

-" !• ~

o.°"

.°•o.

2

E

°,.o • .

..."

0

m

o •° . . ° ° ° ~

N

o.O

0

~

0

2

~

~

4

6

(

i

8 10 trench width wt [prn]

:

~

m

12

14

16

18

Fig. !. Dependence of determined film thickness tit in trenches upon trench width w,. Sample type, the trench depth of 5 lain (D5) or 7 pm (D7) respectively, and thickness of the upper layer on the silicon surface are indicated in the inset• For comparison the broken line represents a thicknessto-width ratio of 0.5. In the upper part the SEM-image of a cross section is shown representing a MW2 sample.

! 022

G. Schaars('hmidt et ai. / Diamond am/Rehtted Materials 6 ' 1997)/019-1025

3

Scan2

~

2.5

silicon diamond

~

.- o... D-band ..

o...

G-band

+..5 0.9,

o+! o ~ ° ° - ° 4

t

:

-+

-7,°°-

. !

°-~~,~.

i:

:::

|

,

Scan1 3.53 !,

,,,I +1

0,5

0 0

2

4

6

8

10

I ........................ .................................................. ....................... ,.12 14 16 18

relative position [IJm] Fig. 2. Fitting results of Raman spectra taken from a cross-section of sample MW! along scan lines indicated on the SEM-image.

G. Schaarschmidt et al. / Diamond and Related Materials 6 (1997) 1019-1025

The upper layer represents a diamond deposit with a small amount of graphitic phases, which decreases with growing layer thickness. The behaviour of the silicon signal reflects an artefact of such kind of measurements. It shows an abrupt increase at the position of the layerto-substrate interface, a level unequal to zero in the diamond layer, and an imaginary low content of silicon on the film surface. Because of the large aperture of the laser focus the exciting beam is partially refracted towards the silicon interface by the low absorbing diamond. The scattered light originating from different sites is detected by the spectrometer and influences the silicon peak. This effect will be discussed in a forthcoming paper [9]. The line scan along the trench shows two distinctly separated deposits. The altered contributions of D- and G-bands to the spectra indicate a poorer diamond quality of the deposit on the trench bottom and on the underside of the upper layer, grown into the trench. Hereby, both the geometry of the deposits and the absorption of the material must be considered for comparing the absolute intensities of the diamond phonon. As a measure of the diamond quality we consider the ratio of the intensities of the diamond phonon and Dand G-bands. The thickness of the deposit within the trench in Fig. 2 was insufficient to obtain more detailed information. Furthermore the trench was so narrow that the laser beam struck the silicon of the trench walls. In the case of HF and HCA samples, in principle similar line scans were obtained. The content of graphitic phases in the sample coated by hot filament CVD was lower, probably caused by the lower methane concentration. To enable measurements in wide and completely overgrown trenches a further deposition series, MW2, was carried out extended in duration of growth. In this experiment the influence of an increased methane concentration was examined. The corresponding line scans of Raman investigation along two trenches are presented in Fig. 3. The upper side of the surface layer above the trench 2.5 lam in width (Scan2) shows a good diamond quality, which is also emphasized by a FWHM value of the diamond phonon of 5.1 cm -1 (uncorrected to the spectrometer resolution of 3 cm-~). The contribution of graphitic structures increases essentially on the underside of the surface layer. This behaviour gets more evident in the case of a trench 4.8 l.tm in width (Scanl) that closes during deposition some time later. The intensity profiles show clearly that at the top of the sample the diamond quality increases with increasing distance to the bottom. In contrast at the bottom of the trench a decrease of the diamond quality is observed with increasing distance to the bottom. Comparing with sample MWI the absolute intensities have changed dramatically caused by the higher methane concentration. The generally weak signals in the bottom range of the first

1023

trench will probably be assigned to the fact that the deposit is not broken in plane with the silicon upon preparation. The line scans are regarded to reflect the film evolution during deposition. The "iamond signal of the layer on the substrate surface grows, since after the formation of a closed layer the crystallites grow increasingly large and the contribution of grain boundaries is reduced by the van-der-Drift mechanism of evolutionary selection [10]. In the case of the surface layer above a trench this mechanism will only operate after the time a continuous layer has been formed. From the trench bottom and comparably from the upper edges of the trench a deposit starts growing into the trench whose portion of nondiamond phases is continuously increasing, while the surface layer above the trench closes gradually, and the deposit in the trench stops growing. To some extent the three applied deposition methods are based on rather different exciting mechanisms of the gas phase. The fact that all deposition methods result in similar dependencies of the film quality in the trenches can be understood if these results are related to the transport of activated species to the growing surface by diffusion. When the gap on top of the trench gets smaller the exchange of the activated species and of the end products is increasingly impeded. Atomic hydrogen which is responsible for the stabilisation of the diamond surface and for the etching of sp2-bounded carbon has a high recombination probability on surfaces [ 11 ]. Thus the concentration of atomic hydrogen in the trench with the flux density outside the trench being constant is always lower than on the film surface. This concentration decreases in the course of the trench overgrowth, which results in an increased formation of graphitic structures. Since an increase of the methane concentration withh~ the feed gas causes a general decrease in the concentration of atomic hydrogen in the plasma [12], strong dependence of the deposit quality in trenches on the C H 4 admixture is plausible. Furthermore the experiments revealed that the growth rate of the deposit on the bottom and at the upper part of the trench is about the same and similar to that at the free surface. The low deposition efficiency of hydrocarbon on surfaces [13] can be responsible for this averaging of the growth rates in the trench. With increasing deposition time a trench acts as a cavity preventing the outgoing particles fi'om leaving it. Therefore the concentration of hydrocarbon in the trench is high enough not to limit the growth of the deposit. As the growth conditions at all sites in such a cavity are the same, deposition in the trench stops at the deposit thickness equal to half the distance between upper edges. 4. Conclusions

Diamond deposition onto trench structures in silicon varying in width shows specific features in the way the

G. Schaarschmidt et aL ,/Diamond and Rektted Materials 6 (1997) 1019-1025

1024

silicon --e-- diamond

++

Scan2

--<~--D-band --<>--G-band

I

i

It 4

E 3

o t-t-

2

q

1

O a

+0 ¢,.¢,

p

": +++:o+ o 0 9 a+ o o-o'

o.0"~ 1o

'O ,a

6 O

........

:E

~I 0-

F~

Scan1

[.o:2: Q~ _ * * ~ , , ~ o ' J/'o a o*g, o~,..

~-.................. ÷.................. --'-~-

0

2

4

'

,.

.¢'~.~,.

~

t

I

t

6

8

10

~-

12

14

"t~~

16

--

I

4

t

18

20

relative position [pm] Fig. 3. Fitting results o f R a m a n s p e c t r a taken f r o m a cross section o f s a m p l e M W 2 . T w o scans are s h o w n a l o n g the c e n t r e o f t r e n c h e s 2.5 a n d 4.8 p m in width, respectively.

deposits are formed on the surface and in the trench. It was possible to analyse the phase purity of buried diamond layers by means of cross-sectional Raman spectroscopy. Line scans along the sample cross-section reflect the quality of the deposit at different depth and therefore at different moments of the deposition process. Owing to a low deposition efficiency of hydrocarbons

aud their considerable resident time in the trenches all d,:posits grow uniformly there. The concentration of atomic hydrogen decreasing in the trenches results in an increasing deposition of non-diamond carbon. High 6iamond qualities can be obtained in trenches with an ~.~spect ratio of more than 0.5 only at high excess of atomic hydrogen in the discharge.

G. Schaalwchmhit et aL / Diamond and Related Materials 6 (1997) 1019-1025

Acknowledgement The authors would like to thank the staff of the Zentrum for Mikrotechnologie of lhe Technische Universit~it Chemnitz-Zwickau for preparation of the silicon wafers as well as Mr. K.G. Tschersich and Mr. K.-H. Wt~llner-Fischer of IGV of the Forschungszentrum JiJlich for the deposition carried out by means of HFCVD and the Graduiertenkolleg "DiJnne Schichtcn t~nd nichtkristalline Materialien" for financial support.

References [ 1] P.K. Bachmann and W. van Enckevort, D&mond Relat. Mtiter., 1 (1992) 1021. [2] A,Fl6ter, G. Schaarschmidt, B. Mainz, S. Laufer, S. Deutschmann and H.-J. Hinneberg, Diamond Relat. Mater.. 4 (1995) 930.

1025

[3] A. Fl6ter, B. Mainz, J. Stiegler, U. Faike, S. Schulze, S. Deutschmann and G. Schaarschmidt, Diamond Relat. Muter.. 3 (1994) 1097. [4] K.-H. Wtillner-Fischer, Report Jiil-3154. (1995) Forschungszentrum Jtilich, 52425 Jtilich, Germany. [5] J. Stiegler, B. Mainz, S. Laufer, T. Weber and G. Schaarschmidt, Diamond Relat. Mater., 3 (1994) 1235. [6] P.K. Bachmann and D.U. Wiechert, Diamond Relat. Mater., 1 (1992) 422. [7] T. Werninghaus, M. Friedrich and D.R.T. Zahn, Phys. Star. Sol. (A), 154 (1996) 269. [8] T. Werninghaus, S. Laufer, H.-J. Hinneberg and D.R.T. Zahn, in K.V. Ravi and J.P. Dismukes (eds.), Electrochem. Soc. Proc. Vol. 95-4, p. 151. [9] T. Werninghaus and D.R.T. Zahn, Cross-Sectional Raman Spectroscopy of Wide-Gap Semiconductors, to be published. [10] C Wild, N. Herres and P. Koidl, J. Appl. Phys., 68 (1990) 973. [ 11 ] B.J. Wood and H. Wise, J. Phys. Chem., 65 ( 1961 ) 1976. [12] M. Chenevier, J.C. Cubertafon, A. Campargue and J.P. Booth, Diamond Relat. Mater., 3 (1994) 587. [13] M.A. Cappelli and M.H. Loh, Diamond Relat. Muter., 3 (1994) 417.