Thickness dependent glass transition temperature of PECVD low-k dielectric thin films: effect of deposition methods

Microelectronics Journal Microelectronics Journal 33 (2002) 221±227

www.elsevier.com/locate/mejo

Thickness dependent glass transition temperature of PECVD low-k dielectric thin ®lms: effect of deposition methods H. Zhou a, H.K. Kim a, F.G. Shi a,*, B. Zhao b, J. Yota b a

Optoelectronics Packaging and Automation Laboratory, 916 Engineering Tower, The Henry Samueli School of Engineering, University of California, Irvine, CA 92697-2575, USA b Conexant System Inc., 4311 Jamboree Road, Newport Beach, CA 92660, USA Received 27 July 2001; revised 15 September 2001; accepted 26 September 2001

Abstract Low-k dielectric carbon-doped silicon dioxide ®lms created by Plasma Enhanced Chemical Vapor Deposition (PECVD) using a six-station sequential deposition system exhibit different glass transition behavior from ®lms created by PECVD in a single deposition station. The enhanced glass transition temperature (Tg) for the PECVD thin ®lms of a layer consisting of six sub-layer deposited in a six-station sequential deposition system to the Tg for ®lms of a single layer deposited in a single deposition system is traced back to the introduced ®lm interface effect inherent to the different deposition methods. Both types of PECVD thin ®lms range in thickness from 50 to 1255 nm and show an increasing Tg with decreasing ®lm thickness. The observed glass transition behavior for ®lms with six sub-layers can be well explained by a theoretical model of thickness dependent Tg for multiple sub-layers obtained by modifying the currently existing theoretical model for the single layer thickness dependent Tg behavior, which explains the observed thickness dependent Tg for single layer PECVD thin ®lms. q 2002 Elsevier Science Ltd. All rights reserved. Keywords: Deposition methods; Glass transition temperature; Low-k dielectrics; Thickness dependence; PECVD

1. Introduction As the minimum geometry in integrated circuits (IC) continues to shrink, the capacitance between metal lines increases dramatically and causes delay in signal propagation. Therefore, low-k dielectric materials are used in ultra large scale integrated (ULSI) devices to reduce the delay in line-to-line crosstalk [1]. Also with the decrease of device sizes, thinner and thinner dielectric ®lms will be employed for insulation. As expected, the constraints imposed by thin polymer ®lm geometry will lead to the deviation of ®lm properties from that of their bulk counterparts. Recent studies on various low-k dielectric materials have demonstrated that most of the structural, optical, mechanical, electrical, dielectric, and thermal properties are expected to be thickness-dependent [2±8]. Therefore, the thickness dependence of the key properties of dielectric ®lms becomes an important IC issue [9]. Thermal reliability of low-k materials is always a vital factor since repetitive thermal treatments are involved in IC device processing. Good thermal stability of the materials to * Corresponding author. Tel.: 11-949-824-5362; fax: 11-949-824-2541. E-mail address: [email protected] (F.G. Shi).

withstand metallization, anneal, cure and chip-attach temperature is required. The thermal reliability of a low-k polycrystalline polytetra¯uoroethylene (PTFE) and its thickness dependence was studied by Kim et al. [6]. However, the majority of low-k candidates for interlevel dielectrics (ILDs) are amorphous materials and the glass transition temperature (Tg) of amorphous materials has been known to be one of the most important single parameter determining their thermal reliability [10]. The temperature selected for processing the thin ®lms is strongly in¯uenced by proximity to the Tg [11]. Some experimental investigation has been performed on the thickness dependent glass transition temperature of amorphous thin ®lms [11±15]. For ILD applications, the glass transition temperature and its thickness dependence of a low-k amorphous FLAREe 2.0 thin ®lm were investigated by Hsu et al. [2]. A few theoretical studies on the effect of ®lm thickness on the glass transition of thin amorphous ®lms have also been reported [16,17]. The thicknessdependent Tg is often described by the relation suggested by Keddie et al. [12], which only predicts a decrease in Tg. A general model is introduced [4] which is able to describe both a thickness dependent reduction as well as an enhancement of Tg relative to its bulk value in amorphous thin ®lms.

0026-2692/02/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved. PII: S 0026-269 2(01)00147-1

222

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227

Regardless of the much experimental and theoretical effort that has been performed, no research has been performed on the effect of deposition methods on the glass transition temperature of amorphous thin ®lms for microelectronic applications. In the current work, the thickness dependent glass transition temperature (Tg) of carbon doped silicon dioxide ®lms formed by PECVD has been studied for different deposition methods. The carbondoped SiO2 ®lms were deposited on the Si substrate using Tetramethyl-cyclotetrasiloxane (TMCTS) as a precursor, ranging from 50 nm to 1255 nm as a single layer or as a layer consisting of six sub-layer. Optical spectroscopy has been applied for the determination of Tg. The enhanced glass transition temperature (Tg) for the PECVD thin ®lms of a layer consisting of six sub-layer deposited in a six-station sequential deposition system to the Tg for ®lms of a single layer deposited in a single deposition system is traced back to the introduced interface effect inherent to the different deposition methods. Both single layer and six sub-layer thin ®lms by different PECVD deposition methods show increasing Tg with decreasing ®lm thickness. The observed glass transition behavior for ®lms with six sub-layers can be well explained by a theoretical model of thickness dependent Tg for multiple sub-layers obtained by modifying the currently existing theoretical model for the single layer thickness dependent Tg behavior, which explains the observed thickness dependent Tg for single layer PECVD thin ®lms.

reduction as well as an enhancement of Tg in amorphous thin ®lms when the ®lm thickness is decreased was introduced as [4]:      Tg …z† 1 a 21 a 21 exp 2 1 ˆ 1 exp 2 2 …1† Tg …1† 2 z=z0 2 1 z=z0 2 1

2. Theoretical analysis

2.2. Thickness dependent Tg for thin ®lms with multiple sublayers

2.1. Thickness dependent Tg for thin ®lms with a single layer A general model describing both a thickness dependent

where z is the thickness of an amorphous ®lm, Tg(z) and Tg(1) are the thickness-dependent glass transition temperature and the bulk value, respectively, a 1 is a measure of the free surface effect, the substrate effect is represented by a 2, and z0 represents a characteristic length. Thus, the thicknessdependent behavior of Tg is determined by both the ®lmsurface and ®lm-substrate interactions, which can exhibit three typical cases. If the segmental mobility of a material near its free surface is enhanced, a reduction in Tg(z) with respect to its bulk value is often expected and the free surface effect is dominant (Case I shown in Fig. 1). If the ®lm-substrate interaction is suf®ciently strong to extend deep into the material, the free surface effect can be overcompensated and can result in an apparent reduction in the material mobility. In this case, an enhancement in Tg(z) with respect to its bulk value as a function of thickness can be expected, as shown by Case II and demonstrated in Fig. 1 (case II). The earlier mentioned free surface and substrate effects are certainly dependent on ®lm thickness, and the substrate effect may become dominant for thinner ®lms, which may result in a minimum of Tg(z) as a function of thickness, as shown in Case III (Fig. 1).

Eq. (1) was applied to predict a thickness-dependent glass transition temperature for thin ®lms with a single layer

Fig. 1. Models of thickness dependent Tg for ®lms with a single layer and ®lms with multiple sub-layers presented by Eqs. (1) and (2) respectively. The solid line represents Eq. (1) and the dashed line with symbol x represents Eq. (2). Case I represents a dominant surface effect. Case II represents the strong ®lmsubstrate effect. Case III represents the free surface effects and substrate effects are dependent on ®lm thickness, which result in a minimum Tg as a function of thickness. For all the three cases, the Tg for multiple sub-layers is enhanced with respect to Tg for single layer ®lm by an extra interface effect.

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227

223

Fig. 2. Schematic illustration of thickness dependent Tg for both single layer and multiple sub-layers. a 1 represents free surface effect and a 2 represents substrate effect. The interface effect between sub-layers is represented by b . (a) Single layer. (b) Multiple sub-layers.

under various experimental conditions [4]. This relationship can be extended to predict a thickness-dependent glass transition temperature for thin ®lms with multiple sub-layers. Fig. 2(a) shows a single layer thin ®lm deposited on the substrate, with free surface effect denoted as a 1 and substrate effect denoted as a 2. For thin ®lms consisting of multiple sub-layers, it is schematically shown in Fig. 2(b). Assume the ®lm consists of n sub-layers of equal thickness and for each interlayer the thickness is negligible. Then (n 2 1) interface is introduced with the interface effect denoted as b . In the case of two sub-layers with one interface, an extra effect of   b21 Tg …1† exp 2 z=z0 2 1 will be introduced to the thickness dependent glass transition temperature Tg(z). Therefore, for ®lms of n sub-layers with (n 2 1) interface, an effect of 0 1 B b21 C B C …n 2 1†Tg …1† expB2 C @ z=n A 21 z0

3. Experimental The low-k dielectric thin carbon-doped silicon dioxide ®lms were deposited at 4008C on the Si wafers by Plasma-Enhanced Chemical Vapor Deposition (PECVD). The low-k PECVD precursor used in this process is cyclic 1,3,5,7-tetramethyl-cyclotetrasiloxane (TMCTS). TMCTS is prepared through hydrolysis of methyldichlorosilane to ®rstly form a linear siloxane polymer that is endcapped with trimethylsilyl groups (derived from trimethylchlorosilane) according to: …CH3 †Si…H†…Cl2 † 1 …CH3 †3 SiCl ! …CH3 †3 Si±O±‰SiHCH3 ±OŠn ±Si…CH3 †3 …CH3 †3 Si±O±‰SiHCH3 ±OŠn ±Si…CH3 †3

should be introduced to the Tg(z) of the thin ®lm. Thus, the thickness dependent glass transition temperature Tg(z) for ®lms with multiple sub-layers can be predicted by,      Tg …z† 1 a1 2 1 a2 2 1 exp 2 ˆ 1 exp 2 Tg …1† 2 z=z0 2 1 z=z0 2 1 0 1 B b21 C B C 1 …n 2 1† expB2 C @ z=n A 21 z0

tion temperature Tg(z) for ®lms with multiple sub-layers is determined by the ®lm-surface effect, ®lm-substrate effect and a ®lm interface effect, which gives an enhanced Tg with respect to ®lms with a single layer as shown in Fig. 1.

…2†

Eq. (2) indicated that the thickness-dependent glass transi-

! TMCTS 1 other cyclic compounds Samples of similar thickness of both ®lms with a single layer and ®lms with a layer consisting of six sub-layer were studied. The single layer ®lms are deposited in a one-station deposition system, while the ®lms consisting of six sublayers were deposited in a six-station sequential deposition system. Their initial thickness is listed in Table 1 for both single layer and six sub-layer thin ®lms. An optical spectroscopy equipped with a computer controlled hot stage was used to investigate the thickness dependence of Tg for the low-k dielectric PECVD thin ®lms. The accuracy of the temperature control is ^0.18C. The

224

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227

Table 1 The initial thickness for thin ®lm samples of single layer and six sub-layer Single layer initial thickness (nm) Six sub-layer initial thickness (nm)

50 56

76 80

105 105

153 130

samples were ®rst heated up to 3508C at 108C/min and kept at 3508C for 30 min to get rid of the moisture and entrapped residual solvents during curing processes, then were cooled down to the room temperature at 28C/min. After the thermal treatment, the samples were heated up to 4558C at 108C/ min. The ®lm thickness as well as the refractive index was constantly monitored and recorded during the second heating. The glass transition temperature can be clearly identi®ed as a well-de®ned kink in the temperature dependence of the thickness and refractive index shift. The thermal history between successive measurements were maintained the same to ensure the determined value of Tg is reliable. If there is a change of thermal history, the shift in Tg as well as the relaxation phenomenon should be observed. Hence, it is crucial to control the heating and cooling rates during measurements. 4. Results and discussion Fig. 3 shows the thickness and refractive index variation as a function of the temperature for a 105 nm single layer PECVD thin ®lm. The Tg value can be clearly identi®ed as a well-de®ned kink in the temperature dependence of the thickness and refractive index shift. There may be a decrease in the (shear) modulus or rigidity of the amorphous materials that produces this response [11]. For the same initial thickness of 105 nm for both single layer and six sub-layers,

270 288

378 383

1244 1255

the Tg difference as de®ned by the kink in the temperature dependence of the thickness shift is shown in Fig. 4. An important observation, which is readily apparent from the data shown in Figs. 5 and 6, is that for thinner ®lms the Tg values, as determined by the interaction between the two linear regions of thickness or refractive index shift versus temperature, are enhanced. The measured Tg values follow a continuously increasing function as the ®lm thickness is lowered. For single layer carbon-doped PECVD thin ®lms, Fig. 5 indicates that the glass transition temperature ranges from 309, 310, 325, 345, 360, 370 to 3858C for ®lm thickness decreasing from 1244, 378, 270, 153, 106, 76 to 50 nm. From Fig. 6, it is shown that for six sub-layer carbondoped PECVD thin ®lms, as the ®lm thickness changes from 1255, 383, 288, 130, 105, 80 to 56 nm, the glass transition temperature increases from 313, 325, 340, 370, 380, 390 to 4058C. The increase or decrease of the glass transition temperature (Tg) with the decrease of the thickness of the amorphous thin ®lms has been investigated both experimentally and theoretically [2,4,11±17]. The thickness-dependent Tg is often described by the relation suggested by Keddie et al. [12], which only predicts a decrease in Tg. A general model [4] demonstrated before as Eq. (1) is able to describe both a thickness dependent reduction as well as an enhancement of Tg relative to its bulk value in amorphous thin ®lms. For ®lms with multiple sub-layers, a general formula of Eq. (2) has also been developed.

Fig. 3. Glass transition temperature identi®ed as a well-de®ned kink in the temperature dependence of the thickness and refractive index for a single layer carbon-doped PECVD ®lm with an initial thickness of 105 nm.

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227

225

Fig. 4. Comparison of Tg for single layer and six sub-layer carbon-doped PECVD ®lms with similar thickness. (X) Thickness variation as a function of temperature for a single layer carbon-doped PECVD ®lm with an initial thickness of 105 nm. (W) Thickness variation as a function of temperature for a six sub-layer carbon-doped PECVD ®lm with an initial thickness of 105 nm.

The experimental data for ®lms of single layer and six sub-layers was ®tted to the model of Eqs. (1) and (2) respectively as shown in Figs. 5 and 6. Both samples show an increased Tg with the decrease of the ®lm thickness. Because of the large surface area to volume ratio in a thin-®lm geometry, effects of both the release of steric constraints at the free surface and speci®c interaction at the ®lmsubstrate interface may affect the segmental dynamics. When supported on substrate for which a substantial ®lmsubstrate attraction is expected, Tg is expected to increase

with decreasing ®lm thickness. Furthermore, with the decreasing of the ®lm thickness, the fraction of molecules in¯uenced by the substrate increases, thus the dynamics of the whole ®lm is reduced. The resulting large, slow mobility regions are also responsible for the higher glass transition temperature. From the comparison between Figs. 5 and 6, it is also seen that the glass transition temperature of the carbon-doped SiO2 PECVD thin ®lms with six sub-layers exhibits an enhanced glass transition temperature compared to the

Fig. 5. Thickness-dependent glass transition temperature variation for single layer carbon-doped PECVD ®lms. The open circles correspond to the experimental results. The solid line represents the theoretical model as demonstrated by Eq. (1).

226

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227

Fig. 6. Thickness-dependent glass transition temperature variation for six sub-layer carbon-doped PECVD ®lms. The open diamonds correspond to the experimental results. The solid line represents the theoretical model as demonstrated by Eq. (2).

thin ®lms with only a single layer. This could be attributed to the multi-layer deposition process that introduces a ®lm interface effect which gives the thin ®lms with six sublayers a higher density. The enhanced glass transition temperature for the six sub-layer thin ®lm could also be due to the stronger ®lm-substrate effect as indicated by the model, where for six sub-layer thin ®lms the ®lm interface effect enhances the ®lm-substrate effect. Furthermore, for six sub-layer PECVD ®lms, the residence time between each ®lm interface could cause a slower diffusion, thus enhancing the glass transition temperature [18]. The correlation between ®lm thickness, deposition methods and Tg could provide a helpful guideline for thermal design of low-k ILD layers in the future. 5. Conclusions Thermal reliability of PECVD low-k dielectric carbondoped SiO2 thin ®lms has been investigated as a function of thickness for different deposition methods. The enhanced glass transition temperature (Tg) for the PECVD thin ®lms of a layer consisting of six sub-layer deposited in a sixstation sequential deposition system to the Tg for ®lms of a single layer deposited in a single deposition system is traced back to the introduced ®lm interface effect inherent to the different deposition methods. Both types of PECVD thin ®lms show an increasing Tg with decreasing ®lm thickness. This thickness-dependent Tg behavior for the single

layer PECVD thin ®lms is explained by a currently existing theoretical model and was attributed to the ®lm/surface and ®lm/substrate effect. As the ®lm thickness decreases, the ®lm-substrate interaction becomes stronger and the dynamics becomes slower. The enhanced Tg for ®lms with sub-layers is well explained by a newly developed model obtained by modifying the current model for single layer glass transition behavior. The correlation between ®lm thickness, deposition methods and Tg could provide a helpful guideline for thermal design of low-k ILD layers in the future. Acknowledgements Support of this study by the micro program (00-088) of the state of California is greatly acknowledged. References [1] The semiconductor technology roadmap for semiconductors, 1999. [2] D.T. Hsu, H.K. Kim, F.G. Shi, B. Zhao, M. Brongo, P. Schilling, S.Q. Wang, ILD thermal stability in deep-submicron technologies: from thin to ultrathin dielectric ®lms, Proceedings of the SPIE Conference on Multilevel Interconnect Technology III 3883 (1999) 60±67. [3] D.T. Hsu, H.K. Kim, F.G. Shi, B. Zhao, M. Brongo, Thickness dependent dielectric properties of low-k materials: a theoretical model. Proceedings of Fourth International Symposium of Low and High Dielectric Constant Materials: Materials Science, Processing, and Reliability Issues, Seattle, WA, May 2000, 99-7 2000, pp. 62±68.

H. Zhou et al. / Microelectronics Journal 33 (2002) 221±227 [4] D.T. Hsu, F.G. Shi, B. Zhao, M. Brongo, Theory for the thickness dependent glass transition temperature of amorphous polymer thin ®lms. In: Proceedings of Fourth International Symposium of Low and High Dielectric Constant Materials: Materials Science, Processing, and Reliability Issues, Seattle, WA, May 2000, 99-7 2000, pp. 53±61. [5] J. Wang, H.K. Kim, F.G. Shi, B. Zhao, T.G. Nieh, Thickness dependence of morphology and mechanical properties of on-wafer low-k PTFE dielectric ®lms, Thin Solid Films 377±378 (2000) 413±417. [6] H.K. Kim, F.G. Shi, Thickness-dependent thermal reliability of lowdielectric constant polycrystalline PTFE submicron dielectric thin ®lms, Microelectron. J. 32 (3) (2001) 215±219. [7] H.K. Kim, F.G. Shi, B. Zhao, M. Brongo, Low-k dielectrics for ULSI multilevel interconnections: thickness dependent electrical and dielectric properties. Conference Record of the 2000 IEEE International Symposium on Electrical Insulation, Piscataway, NJ, USA 2000, pp. 269±271. [8] J. Wang, F.G. Shi, T.G. Nieh, B. Zhao, M. Brongo, S. Qu, T. Rosenmayer, Thickness dependence of elastic modulus and hardness of on-wafer low-k ultrathin poly-tetra¯uoroethylene ®lms, Scripta Mater. 42 (2000) 687±694. [9] R.D. Miller, Device physics- In search of low-k dielectrics, Science 286 (1999) 421±423. [10] J.E. Mark, Physical properties of polymers, second ed, American Chemical Society, Washington, DC, 1993.

227

[11] D.S. Fryer, P.F. Nealey, J.J. de Pablo, Thermal probe measurements of the glass transition temperature for ultrathin polymer ®lms as a function of thickness, Macromolecules 33 (17) (2000) 6439±6447. [12] J.L. Keddie, R.A.L. Jones, R.A. Cory, Size-dependent depression of the glass transition temperature in polymer ®lms, Europhys. Lett. 27 (1) (1994) 59±64. [13] J. Mattsson, J.A. Forrest, L. Borjeson, Quantifying glass transition behavior in ultrathin free-standing polymer ®lms, Phys. Rev. E. 62 (4) (2000) 5187±5200. [14] J.H. Van Zanten, W.E. Wallace, W.L. Wu, Effect of strongly favorable substrate interactions on the thermal properties of ultrathin polymer ®lms, Phys. Rev. E. 53 (3) (1996) R2053±R2056. [15] J.L. Keddie, R.A.L. Jones, R.A. Cory, Interface and surface effects on the glass-transition temperature in thin polymer ®lms, Faraday Discus. 98 (1994) 219±230. [16] J.H. Kim, J. Jang, W.C. Zin, Estimation of the thickness dependence of the glass transition temperature in various thin polymer ®lms, Langmuir 16 (9) (2000) 4064±4067. [17] J.A. Torres, P.F. Nealey, J.J. de Pablo, Molecular simulation of ultrathin polymeric ®lms near the glass transition, Phys. Rev. Lett. 85 (15) (2000) 3221±3224. [18] X. Zheng, M.H. Rafailovich, J. Sokolov, Y. Strzhemechny, S.A. Schwarz, B.B. Sauer, M. Rubinstein, Long-range effects on polymer diffusion induced by a bounding interface, Phys. Rev. Lett. 79 (2) (1997) 241±244.