Evidence of annealing effects on a high-density Si/SiO2 interfacial layer
S. D. Kosowsky and P. S. Pershan
Harvard University, Cambridge, Massachusetts 02138

K. S. Krisch, J. Bevk, M. L. Green, D. Brasen, and L. C. Feldmana)
Bell Laboratories, Lucent Technologies, Murray Hill, New Jersey 07974

P. K. Roy
Bell Laboratories, Lucent Technologies, Orlando, Florida 32819

Received 17 April 1996; accepted for publication 9 April 1997 Thermally grown Si 001 /SiO2 samples were studied by x-ray reﬂectivity. Fits of model electron density proﬁles to the data reveal the existence of an interfacial layer at the Si/SiO2 interface up to 15-Å-thick, with density higher than either the crystalline Si or the main oxide layer. This density of the layer is reduced by a postoxidation anneal. © 1997 American Institute of Physics. S0003-6951 97 02823-4 With the continual drive to scale down the dimensions of metal-oxide-semiconductor MOS devices, the structure of the silicon/silicon oxide interface is becoming increasingly important. It has long been expected that the interfacial 5–10 Å of SiO2 has anomalous characteristics.1–3 Current 0.35 m gate-length devices use 70–80-Å-thick gate oxides, and for these devices, anomalies in the interfacial structures on the order of 5 Å4,5 or larger6 are a signiﬁcant fraction of the oxide and may affect device characteristics.6 Effects of interfacial microstructures on mobility degradation and oxide breakdown are expected to become increasingly important in the performance of these submicron devices.7 Nevertheless, the density proﬁle for MOS structure has still not been fully determined. Studies of the as-grown Si/SiO2 interface have reported a layer at that interface, whose structure6,8–11 and stoichiometry1,12 differ from that of the surrounding crystalline Si and amorphous SiO2 regions. This interfacial oxide region has been shown to be Si-rich, compared to SiO2, 1,8,11,13 but the electron density proﬁle there is still unknown. Furthermore, in the case of devices, it is important to know how this structure evolves with annealing, which, for example, has been shown to change the silicon oxide density.14 In this letter we report the electron density proﬁles extracted from x-ray reﬂectivity studies of thermally grown Si 001 /SiO2 samples. These proﬁles contain the highdensity layer at the crystal/oxide interface that can be anticipated by some other studies of oxidized Si 001 surfaces. The extent of the layer depends on the growth environment, and the layer is affected by thermal annealing. The oxides used in this study were grown via rapid thermal oxidation at 1000 °C for 10–200 s to create a range of oxide thicknesses. Three groups of samples were studied see Table I . The ﬁrst group, with thicknesses from about 50 to 300 Å, was formed in a 760 Torr O2 ambient, and the second group, with thicknesses 35–70 Å, in a 40 Torr O2 environment. Two of the samples from the ﬁrst group were subsequently annealed and comprise group three. The anneals were done with a ﬂow of prepuriﬁed Ar gas, at 1000 °C for 1 h. X-ray data were taken on a two-circle spectrometer usa

ing a Rigaku rotating anode source of Cu K 1.54 Å radiation and a Ge 111 monochromator to create a highly collimated, narrow beam. The measured x-ray reﬂectivity R(q) is essentially a Fourier transform of the spatial derivative of the sample density in the perpendicular (z) direction, averaged over the lateral (x,y) directions. Speciﬁcally, for q q c , where q is the wave vector transfer along the surface normal15 R q Rf q 1
sub

d

z iqz 2 e dz , dz

1

Present address: Dept. of Physics and Astronomy, Vanderbilt U., Nashville, TN 37235.

where R f (q) is the Fresnel function for the bulk substrate, q c is the critical angle (q c 0.0316 Å 1 ),and sub is the electron density well inside the bulk Si substrate 2.33 gm/cm3 ). 16 The data were actually analyzed using an exact formalism,17 valid even at the critical angle. However, for discussion, we present the kinematic approximation Eq. 1 to make the physics more transparent. A typical reﬂectivity data set for an as-grown sample is shown along with ﬁts in Fig. 1 a , with the corresponding real space models in Fig. 1 b . The R/R f falls below 1.0 at the smallest q values because the incident beam does not fall entirely within the sample boundaries. To ﬁt the data, we use a slab model with ﬁtted parameters that represent the bulk substrate, the oxide overlayer layer 2 and a dense interfacial layer layer 1 , where each layer above the bulk substrate is described by a density n and a layer thickness d n . This is the solid line in Figs. 1 a and 1 b . Note that in Eq. 1 , R/R f is sensitive to (1/ sub)d /dz i.e., n / Si 1), hence the sensitivity to small density differences among the layers. Two of the other real space models which were attempted, and which have fewer parameters, are shown as broken lines in Fig. 1 b , with the corresponding ﬁts in Fig. 1 a . These models omit the dense interfacial layer (d 1 0), and a visual inspection of the ﬁts rules out both these models for the as-grown samples. The x-ray data from sample 2, measured before and after annealing, are shown in Fig. 2 along with ﬁts based on Eq. 1 . The real space models producing these ﬁts are shown in the inset. The pre and postanneal reﬂectivities both show oscillations in q, whose periods are functions of the oxide layer thicknesses, through Eq. 1 . The change in the R/R f caused by annealing clearly demonstrates that the oxide

Appl. Phys. Lett. 70 (23), 9 June 1997 0003-6951/97/70(23)/3119/3/$10.00 © 1997 American Institute of Physics 3119 Downloaded¬07¬Oct¬2010¬to¬128.103.149.52.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright;¬see¬http://apl.aip.org/about/rights_and_permissions

TABLE I. Fitted model parameters for the slab model. The values of indicated in Fig. 3.

1

and

2

are normalized by

Si

2.33 gm/cm3).The range of excess strength values are

1

Sample 760 Torr O2 unannealed 1 2 3 4 5 6 7 8 9 2a 4a

d1 Å 15.7 2.0 14.7 2.0 14.8 2.1 14.2 1.9 14.7 2.3 14.3 2.1 14.7 3.1 12.7 3.0 9.6 3.1 8.7 6/ 8.7 12.8 4.2

d2 Å 272.6 138.9 101.2 70.2 59.7 40.0 55.4 41.0 25.3 154.6 87.6 0.8 0.9 1.0 0.9 1.2 1.3 1.4 1.1 1.1 1.3 1.2

1
2

Si

/

Si

Excess strength Å 0.36 0.27 0.28 0.33 0.29 0.33 0.19 0.13 0.11 0.01 0.12

d Ellipsometry Å 290 153 115 82.8 72.7 53.5 67.5 53.5 32.9 166 99.9

40 Torr O2 unannealed 760 Torr O2 annealed

0.023 0.004 0.018 0.004 0.019 0.005 0.023 0.005 0.019 0.006 0.023 0.006 0.013 0.005 0.111 0.005 0.012 0.006 0.001 0.008/ 0.001 0.009 0.005

0.984 0.981 0.980 0.990 0.977 0.991 0.987 0.980 0.989 0.958 0.968

0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.004 0.005 0.003

overlayer has changed. However, the overall thickness of the oxide as determined through the x-ray scattering, d 1 d 2 , corresponds well with ellipsometry data taken on these samples. For all of the unannealed samples, grown either in 760 or 40 Torr O2 environments, the additional layer is required to produce a satisfactory ﬁt to the measured data. A summary of the best ﬁtted model parameters is found in Table I. Ideally, the interfacial widths, thicknesses, and densities of the model would be determined from ﬁts to the data through Eq.

1 . The interfacial widths of the additional layer are set equal to 0, since without large q data, the ﬁts are not sensitive to the detailed shape of the proﬁle. Also given in the table is the product d 1 ( 1 / Si 1),which is the integrated electron density excess or ‘‘excess scattering strength,’’ of the layer. It characterizes the extent of the dense, interfacial layer and, to ﬁrst order, its importance to the x-ray scattering structure, since it is a leading term in the expansion of Eq. 1 . Furthermore, in the ﬁtting, small changes in the ﬁtting parameter d 1 can be compensated by an opposite change in

FIG. 1. a Typical x-ray reﬂectivity data from an unannealed sample sample 4 , with ﬁts corresponding to the real space models of b . The single oxide layer models broken lines produce poor ﬁts, whereas the two-layer model solid line ﬁts the data well.

FIG. 2. a and c Comparison of the reﬂectivities before samples 2, 4 and after samples 2a, 4a annealing. Annealing causes signiﬁcant changes in the reﬂectivities, indicating signiﬁcant changes in the samples. b and d Real space proﬁles producing ﬁts in a and c , respectively : unannealed — and annealed ---- .

3120 Appl. Phys. Lett., Vol. 70, No. 23, 9 June 1997 Kosowsky et al. Downloaded¬07¬Oct¬2010¬to¬128.103.149.52.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright;¬see¬http://apl.aip.org/about/rights_and_permissions

FIG. 3. Excess scattering strength d 1 ( 1 / Si 1) in the interfacial layer as a function of oxide thickness. 760 Torr O2, 40 Torr O2, annealed. The arrows indicate the change from before to after annealing.

the parameter 1 . Physical interpretation in terms of the excess scattering strength avoids uncertainties associated with this thickness–density interdependence. The values for the excess strength deduced from the ﬁts are displayed in Fig. 3 with the range indicated by the bars the symbol identiﬁes the group . The unannealed samples grown in a 760 Torr O2 environment have the largest values of excess strength, averaging about 0.3 Å. The excess strength values exhibit no thickness dependence over the range studied. The unannealed samples grown in 40 Torr O2 show a middle range of excess strength, around 0.1–0.2 Å. The annealed samples have excess densities of 0.14 Å, half the unannealed values. In fact, the x-ray reﬂectivity data for the annealed samples can be reasonably well ﬁt without any interfacial layer at all zero excess density . This is presumably because the lower values of 1 Si make the interface and hence the layer more transparent to x-rays. The ellipsometry thickness data, however, requires that the additional oxide be present, and so supports the small, but ﬁnite excess density values. Theoretical studies of the oxidation of silicon have suggested the presence of an intermediate layer between the Si 001 and SiO2. 5,18,19 While the bulk of the oxide is a stable amorphous SiO2 layer,18 the region near the Si 001 interface may be a transition layer, possibly stabilized by high stresses which are known to exist within the oxides20,21 and especially at the Si/SiO2 interface.14 Alternatively, the layer could be a region of higher density strained silicon, where the lattice is compressed slightly in the vertical. In either case, annealing, which is well known to reduce stresses, would affect the Si/SiO2 interfacial structure. High-resolution core level spectroscopy studies on native oxides have shown that annealing shifts Si/SiO2 spectra away from interfacial, substochiometric states towards higher Si 4 states.12 This is consistent with the annealing-induced reduction in excess strength of the interfacial layer reported here. For the unannealed samples, the ﬁtted density of the main oxide layer 2 is independent of the oxide thickness and the O2 partial pressure during growth, and is unaffected by the layer 1 parameters. The density is affected by annealing, however, as the annealed samples show an average oxide density ( 2 ) reduction of 2.3% as compared to the unan-

nealed samples. Furthermore, the values of the main oxide density 2 both annealed and unannealed agree well with values deduced indirectly through ellipsometry measurements of the index of refraction of thin thermally grown SiO2 ﬁlms.14 Finally, we mention that the oxide–vacuum interface has a root mean square roughness of 3.5–5.1 Å, in agreement with similar oxide measurements.15 In conclusion, we have used x-ray reﬂectivity to probe the Si/SiO2 interface, revealing an interfacial layer with density higher than either crystalline Si or the main oxide layer. Quantitative values obtained for the excess scattering strength, a measure of the extent of this layer, is a function of the O2 partial pressure of the growth environment in the as-grown oxide. Postoxidation annealing reduces the excess strength of this interfacial layer, and additionally reduces the density of the main oxide layer. This study could be extended to include wider ranges of ambient oxygen partial pressures as well as thicker oxides, to explore possible trends. The authors thank K. Evans-Lutterodt, D. Monroe, and T. Sorsch for valuable discussions. One author S.D.K. gratefully acknowledges the support of Bell Laboratories, Lucent Technologies. This work was supported by Bell Laboratories, Lucent Technologies and by Harvard Materials Research Laboratory under Grant No. NSF-DMR-89-20490.

F. J. Himpsel, F. R. McFeely, A. Taleb-Ibrahimi, and J. A. Yarmoff, Phys. Rev. B 38, 6084 1988 . 2 A. Ourmazd, D. W. Taylor, J. A. Rentschler, and J. Bevk, Phys. Rev. Lett. 59, 213 1987 . 3 F. Rochet, M. Froment, C. D’Anterroches, H. Roulet, and G. Dufour, Philos. Mag. B 59, 339 1989 . 4 T. A. Rabedeau, I. M. Tidswell, P. S. Pershan, J. Bevk, and B. S. Freer, Appl. Phys. Lett. 59, 3422 1993 . 5 A. M. Stoneham, C. R. M. Grovenor, and A. Cerezo, Philos. Mag. B 60, 189 1989 . 6 E. Hasegawa, A. Ishitani, K. Akimoto, M. Tsukiji, and N. Ohta, J. Electrochem. Soc. 142, 273 1995 . 7 T. Ohmi, M. Miyashita, M. Itano, T. Imaoka, and I. Kawanabe, IEEE Trans. Electron Devices 39, 537 1992 . 8 E. Taft and L. Cordes, J. Electrochem. Soc. 126, 131 1979 . 9 T. A. Rabedeau, I. M. Tidswell, P. S. Pershan, J. Bevk, and B. S. Freer, Appl. Phys. Lett. 59, 706 1991 . 10 G. Renaud, P. H. Fuoss, A. Ourmazd, J. Bevk, and B. S. Freer, Appl. Phys. Lett. 59, 706 1991 . 11 N. W. Cheung, L. C. Feldman, P. C. Silverman, and I. Stensgaard, Appl. Phys. Lett. 35, 859 1979 . 12 F. J. Himpsel, D. A. Lapiano-Smith, J. F. Morar, and J. Bevk, Local Bonding at SiO2 /Siinterfaces Electrochemical Society, New York, 1992 . 13 E. P. Gusev, H. C. Lu, T. Gustafsson, and E. Garfunkel, Phys. Rev. B 52, 1759 1995 . 14 K. Taniguchi, M. Tanaka, C. Hamaguchi, and K. Imai, J. Appl. Phys. 67, 2195 1990 . 15 I. M. Tidswell, B. M. Ocko, P. S. Pershan, S. R. Wasserman, G. M. Whitesides, and J. D. Axe, Phys. Rev. B 41, 111 1990 . 16 The x-ray data actually determine the electron density proﬁle. However, for compounds of the light elements such as Si and O , electron density and mass density are proportional independent of composition to well within the error bars of this experiment. 17 L. G. Parratt, Phys. Rev. 131, 359 1954 . 18 C. J. Soﬁeld and A. M. Stoneham, Semicond. Sci. Technol. 10, 215 1995 . 19 I. Ohdomari, H. Akatsu, Y. Yamakoshi, and K. Kishimoto, J. Appl. Phys. 62, 3751 1988 . 20 L. M. Landsberger and W. A. Tiller, Appl. Phys. Lett. 51, 1416 1987 . 21 J. Fitch, G. Lucovsky, E. Kobeda, and E. Irene, J. Vac. Sci. Technol. B 7, 153 1989 .

1

Appl. Phys. Lett., Vol. 70, No. 23, 9 June 1997 Kosowsky et al. 3121 Downloaded¬07¬Oct¬2010¬to¬128.103.149.52.¬Redistribution¬subject¬to¬AIP¬license¬or¬copyright;¬see¬http://apl.aip.org/about/rights_and_permissions