Measurement of stress gradients in hydrogenatedmicrocrystalline silicon thin films using Raman spectroscopy

Vincent Paillard a,*, Pascal Puech a, Pere Roca i Cabarrocas b

a Laboratoire de Physique des Solides, associ au CNRS(UMR 5477), Universit Paul Sabatier, 118 route de Narbonne,

F-31062 Toulouse cedex 4, Franceb Laboratoire de Physique des Interfaces et des Couches Minces, associ au CNRS (UMR 7647), Ecole Polytechnique,

F-91128 Palaiseau cedex, France

Abstract

Raman spectroscopy is a powerful tool to measure stress in semiconductors. We show both large stress values and

stress gradients in hydrogenated microcrystalline silicon thin films, by the use of different excitation wavelengths, from

red to near-ultraviolet, allowing us to probe different film depths. For films deposited by standard radio frequency glow

discharge at different substrate temperatures, we find that stress evolves from highly compressive in the bulk of the film,

close to the glass substrate, to tensile near the film free surface. Moreover, the higher the substrate temperature, the

higher the stress gradient. � 2002 Elsevier Science B.V. All rights reserved.

PACS: 78.30.A; 78.20; 07.10.L

1. Introduction

For several years, hydrogenated microcrystal-line silicon (l-Si:H) produced by plasma-enhancedchemical vapor deposition (PECVD) has drawn aconsiderable interest in thin film applications suchas thin film transistors and solar cells. While pre-serving the use of low cost substrates as well as aperfect compatibility with the common hydroge-nated amorphous silicon (a-Si:H) technology, this

material should offer better electronic propertiesthan a-Si:H. Many articles have shown that l-Si:Hstructure and properties depend on many factorssuch as the preparation conditions [1], the sub-strate nature [2] or the deposition temperature [3].The attention is commonly focused on the amountand the type of Si–H bonds, or the fraction andsize of the crystallites dispersed in the amor-phous matrix. However, to our knowledge, the roleof residual stress in l-Si:H films was rarely con-sidered, although recent reports establish a clearcorrelation between the stress and the electronmobility in different silicon based devices [4–6].

In this work, we study PECVD l-Si:H films byusing Raman spectrometry, which is highly sensi-tive to stress [7]. We show that depending on thepreparation conditions, the samples can be under

Journal of Non-Crystalline Solids 299–302 (2002) 280–283

www.elsevier.com/locate/jnoncrysol

* Corresponding author.

E-mail address: vincent.paillard@lpst.ups-tlse.fr (V. Pail-

lard).

0022-3093/02/$ - see front matter � 2002 Elsevier Science B.V. All rights reserved.

PII: S0022 -3093 (02 )00936 -5

high tensile or compressive stress. Even more im-portant, by varying the excitation laser wavelength,we obtained a stress profile changing from highlycompressive at the glass substrate/silicon layer in-terface to tensile near the film surface. The experi-ments were performed on specific samples toeliminate any influence on the phonon peak otherthan the stress contribution.

2. Film preparation

The l-Si:H films on Corning glass substrateswere prepared by the standard RF glow dischargedecomposition of silane and hydrogen in a ca-pacitively coupled monochamber reactor. Thepreparation conditions were kept constant, withthe exception of the substrate temperature rangingfrom 150 to 250 �C. Silane and hydrogen flowswere fixed at 3 and 150 sccm, respectively, for atotal pressure of 1.6 Torr. The rf (13.56 MHz)power density was set at 250 mWcm�2. The de-position rate was in the 0.16–0.20 nm s�1 range. Anextra film was deposited at 250 �C with a smallersilane flow rate (1.5 sccm) in order to increase itscrystalline fraction. For further details on thepreparation conditions and the electrical proper-ties of the films, see Ref. [3].

3. Results and discussion

Both thickness and crystalline fractions of thebulk and subsurface layers (Table 1) were ex-tracted from the analysis of the ellipsometry data

in a multi-phase model [8]. Using the Debye–Scherrer formula, the crystallite size was estimatedfrom X-ray diffractograms, which also evidenced a(2 2 0) preferential orientation of grains [3]. Thesize being always larger than 14 nm, the Ramanshift can be fully attributed to stress, since theoptical mode redshift due to phonon confinementis negligible in that case [9]. The stress was deducedfrom Raman spectra, according to the followingequation [10]:

rðMPaÞ ¼ �250Dx ðcm�1Þ: ð1ÞIn Eq. (1), r is the in-plane stress andDx ¼ xs � x0. In this expression, x0 is the opticalphonon wave number of the stress free singlecrystal and xs is the wave number of the stressedsample. Hence, negative or positive stress valuescorrespond to compressive or tensile stress, re-spectively. Several excitation radiations, providedby Arþ and Krþ lasers, were used to investigatedifferent volumes in order to measure a stressgradient in the films. The different wavelengthswere 676, 647, 568, 530, 488 and 364 nm. Ramanspectra as a function of a few excitation wave-lengths are shown in Fig. 1. The spectrum of thesingle crystal reference is given for each wave-length to determine accurately the shift Dx forstress measurements from Eq. (1). The spectrawere fitted using two bands to take account ofboth amorphous and crystal contributions.Though the stress should not change the phononpeak shape [10], an asymmetric Lorentz functionwas used to take account of a reduced phononcorrelation length in defective grains. In the case ofthe red excitation lines, the fit results were also

Table 1

Structural properties of the microcrystalline silicon films as a function of the substrate temperature

Sample Ts (�C) Rd (nm/s) d (nm) Bulk layer Subsurface

Thick (nm) Fc (%) Fv (%) Thick (nm) Fc (%) Fv (%)

80507 150 0.16 14.7 1006 74 12 11 84 16

80506 200 0.16 16.8 980 72 9 10 88 12

80505 250 0.16 18.0 980 63 10 7 91 8

80616 250 0.08 23.3 642 93 7 12 90 10

The last sample was produced at higher hydrogen dilution, contrary to the three others deposited under the same plasma conditions.

For such samples, increasing the substrate temperature has for consequences (i) the increase of the grain size d, deduced from X-ray

diffraction, and (ii) the decrease of both the crystalline and void fractions of the bulk layer (deduced from spectroscopic ellipsometry).

This is compensated by an increase in the amorphous fraction (the complement to 100%), not shown in the table.

V. Paillard et al. / Journal of Non-Crystalline Solids 299-302 (2002) 280–283 281

used to estimate the average crystalline fraction (inthe film volume) by standard methods [11]. Wehave also tried to draw the crystalline fractionprofile and correlate it to the stress profile. Wewere however unable to obtain accurate results,because of the big difference in the Raman scat-tering cross-sections of amorphous and crystallinesilicon as a function of the wavelength. For in-stance, the data on amorphous silicon changedramatically depending on the hydrogen content[12], while the grain size or stress can affect theRaman scattering cross-section of crystalline sili-con [13].

Table 2 summarizes the stress measurements.The stress is given as a function of the scatteringdepth d(k) given by

dðkÞ ¼ 1=½2aðkÞ�: ð2ÞAs mentioned previously, the absorption coeffi-cient aðkÞ is deduced from the spectroscopic el-lipsometry measurements. The resulting probedepth is certainly not exact, but sufficient to clearlyevidence the stress gradient.

It is obvious that all samples are under highcompressive stress near the substrate (large scat-

tering depth using the 676 nm excitation radiation,Fig. 1(a)), which decreases to finally reverse to atensile stress near the surface (near-UV line, Fig.1(c)). In the film volume, the compressive stressdecreases with decreasing substrate temperature(Fig. 1(a)–(c)), which can be correlated with theincrease in the void fraction (see Table 1). Noticethat a compressive stress value as high as 1 GPa inthe film volume may be a low limit value, since theRaman spectrum is the integration over the wholescattering volume starting from the surface, wherethe laser light is focused. Only the UV line, whichprobes the first 10 nm from the surface, gives alocal measure of stress.

4. Summary

In conclusion, we have shown, for the first timeto our knowledge, the existence of large stressgradients in l-Si:H films produced by PECVD. Itis remarkable to find both large stress values andstress gradient. These factors should now be con-sidered when investigating the electrical properties

Fig. 1. Raman spectra taken at different excitation wavelengths on the microcristalline silicon films. Increasing the excitation wave-

length allows probing deeper regions of the sample.

282 V. Paillard et al. / Journal of Non-Crystalline Solids 299-302 (2002) 280–283

of l-Si:H films, and should thus receive the sameconsideration than other factors such as the crys-talline fraction or the grain size.

Acknowledgements

Thanks to the CNRS and the ADEME (Agencede l’Environnement et de la Maııtrise de l’Energie)for specific grant (ECODEV program).

References

[1] S. Kumar, D.K. Pandya, K.L. Chopra, J. Appl. Phys. 63

(1988) 1497.

[2] P. Roca i Cabarrocas, N. Layadi, T. Heitz, B. Dr�eevillon, I.Solomon, Appl. Phys. Lett. 66 (1995) 3609.

[3] S. Hamma, P. Roca i Cabarrocas, Solar Energy Mater.

Solar Cells 69 (2001) 217.

[4] S.M. Hu, J. Appl. Phys. 70 (1991) R53.

[5] R.B. Wehrspohn, S.C. Deane, I.D. Frensh, I. Gale, J.

Hewett, M.J. Powell, J.R. Robertson, J. Appl. Phys. 87

(2000) 144.

[6] T. Iida, T. Itoh, D. Noguchi, Y. Takanashi, Y. Takano, Y.

Kanda, J. Appl. Phys. 89 (2001) 2109.

[7] I. De Wolf, Semicond. Sci. Technol. 11 (1996) 139.

[8] P. Roca i Cabarrocas, S. Hamma, A. Hadjadj, J. Ber-

tomeu, J. Andreu, Appl. Phys. Lett. 69 (1996) 529.

[9] V. Paillard, P. Puech, M.A. Laguna, R. Carles, B. Kohn,

F. Huisken, J. Appl. Phys. 86 (1999) 1921.

[10] E. Anastassakis, in: J. Kassavod (Ed.), Proc. of the 4th Int.

School ISPPM, Varna, Bulgaria, 1985, p. 28.

[11] B. Demauduit, C. Bourgerette, V. Paillard, P. Puech, B.

Caussat, J. Phys. IV France 9 (1999) 1091.

[12] D. Bermejo, M. Cardona, J. Non-Cryst. Solids 32 (1979)

405.

[13] C. Ossadnik, S. Veprek, I. Gregora, Thin Solid Films 337

(1999) 148.

Table 2

Raman shifts and stress values (rounded) in the microcrystalline films deposited at different substrate temperatures or silane dilution

dRaman 80616 80505 80506 80507

Dx (cm�1) r (MPa) Dx (cm�1) r (MPa) Dx (cm�1) r (MPa) Dx (cm�1) r (MPa)

700 4.20� 0.12 �1050� 30

500 3.16� 0.12 �790� 30 3.00� 0.12 �750� 30 1.31� 0.12 �350� 30

400 2.61� 0.12 �650� 30

350 2.27� 0.12 �570� 30 0.54� 0.12 �150� 30

280 2.39� 0.12 �600� 30

250 1.96� 0.2 �490� 50

160 �1.03� 0.2 260� 50 �0.38� 0.2 100� 50 �1.23� 0.2 300� 50

120 0.32� 0.2 �80�100 �1.56� 0.2 390� 50 �1.59� 0.2 400� 50 �1.95� 0.2 500� 50

80 �1.38� 0.2 340� 50

60 �2.37� 0.2 590� 50 �2.60� 0.2 650� 50

50 �2.22� 0.2 560�10 �1.10� 0.3 180� 75 �0.70� 0.3 180� 75 �1.41� 0.3 350� 75 �1.80� 0.3 450� 75

Negative (positive) values correspond to compressive (tensile) stress. The Raman probe depth is calculated from Eq. (2) using the

ellipsometry data. The errors on dR is �60 nm for of the red radiations, �15 nm for the green–blue radiations, and �5 nm in the UV,

respectively.

V. Paillard et al. / Journal of Non-Crystalline Solids 299-302 (2002) 280–283 283