Thin Solid Films 443 (2003) 97–107

( ) Chemical vapor deposition of silicide WSix for high aspect ratio applications

B.Sella *,, A.Sanger¨ ,abac G.Schulze-Icking , K.Pomplun , W.Krautschneider

aInfineon Technologies, Konigsbrucker¨¨ Strasse 180, 01099 Dresden, Germany bInfineon Technologies AG, Balanstrasse 73, 81541 Munich, Germany cTU Hamburg-Harburg, Techn. Electronics, Eissendorfer Strasse 38, 21071 Hamburg, Germany

Received 15 January 2003; received in revised form 24 May 2003; accepted 13 June 2003

Abstract

Chemical vapor deposition of tungsten silicide into high aspect ratio trenches has been investigated using a commercial 8-inch Applied Materials Centura single wafer deposition tool.For an in-depth study of both step coverage and stoichiometry, a ( ) combined chemistryytopography simulator has been developed. reduction of WF6 has been ( ) identified as a suitable chemistry to fill deep trenches with tungsten disilicide, while for WF64 reduction with SiH or ( ) disilane Si26 H fundamental drawbacks have been identified for extreme aspect ratios.In the process range under study, good agreement is observed between the simulated step coverages and those obtained from scanning electron microscope images.The simulations predict a deposition regime in which both good step coverage and a suitable stoichiometry are achieved inside deep trenches. 2003 Elsevier B.V. All rights reserved.

Keywords: Chemical vapor deposition; Deposition process; Nanostructures; Tungsten

1. Introduction decreases, while at the same time their length increases. To still maintain sufficient current into and out of With continuously decreasing groundrules in modern DRAM capacitors, low resistivity plugsyvias are indis- ULSI manufacturing, there exists a strong demand for pensable for future DRAM generations. partially replacing doped polysilicon with metals or Refractory metal silicides exhibit properties that make other low resistivity materials.Refractory metal silicides, them prime candidates for replacing doped polysilicon for instance, are widely used as part of the gate stack where low resistivity is required.Given the right stoi- in dynamic random access memory (DRAM) manufac- chiometry, they are stable up to high temperatures, show turing.Another possible application for low resistivity good adhesion and also a good dry etchability w1x. materials is in the capacitor of deep trench-DRAM or Silicides can be prepared in two different ways.Pure in vias of stacked DRAM cells.Both products suffer metal can either be deposited on and transformed from the fact that the serial resistance of the inner into a silicide via thermal annealing or metal and silicon ( ) ( electrode plug, deep trench-DRAM or vias stacked- can be co-deposited, which immediately to a ) DRAM increases with the square of the inverse ground silicide layer.The latter process is generally preferred rule.This effect is further enhanced since the require- since it leads to lower mechanical stress and a smoother ment of a constant capacitance of 25 fF leads to deeper silicide–silicon interface w2x. trenches (or higher stacks, respectively) in future In this paper, we discuss the chemical vapor deposi- DRAMs.Thus, for shrinking the ground rule, the cross- tion (CVD) of low resistivity tungsten silicide (WSi ) section of the plugs (deep trench) or vias (stacked) x into features with large aspect ratios.While we focus *Corresponding author.Tel.: q1-971-2148-783. on deep trench plugs of trench DRAMs (as manufac- E-mail address: [email protected] (B.Sell ). tured by Infineon Technologies), all results presented in

0040-6090/03/$ - see front matter 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0040-6090(03)00922-2 98 B. Sell et al. / Thin Solid Films 443 (2003) 97–107 this paper also apply to vias of stacked DRAMs with condition is referred to as the ‘reaction-limited’ process only minor modifications. regime.On the other hand, at high process temperatures

Key requirements for a WSix -plug are thermal stability the surface reactions can be so fast that the transport of in contact with polysilicon up to frontend processing precursors to the surface is the deposition rate-limiting temperatures (Tf1400 K), good step coverage in step.This regime is generally referred to as ‘diffusion- trenches with aspect ratios 050:1 and sufficiently uni- limited’ (or equivalently ‘transport-limited’).This latter form stoichiometry along the depth of the deep trench. regime is usually employed for deposition on planar The only thermodynamically stable phase in contact wafers, since under these conditions the process temper- ( with silicon is WSi2 .While silicon-rich WSix -films x) ature has only minor impact on the deposition rate, and 2.3) exhibit smooth film-to-substrate interface w3x, they also because the throughput is maximized.However, for are thermally unstable and segregate the excess silicon wafers with high aspect ratio structures (such as deep as silicon crystallites or increase the thickness of the trenches), rapid surface reactions to precursor underlying polysilicon w4x.Tungsten-rich films (x-2.0), depletion inside deep trenches and, therefore, to a poor on the other hand, are also thermally unstable and lead step coverage.A simplified way to look at surface to the consumption of the underlying polysilicon layer reactions is assuming first-order reactions only.In this during anneal steps later in the process w5x.Thus, the case the surface reaction rate can be expressed by the composition of tungsten silicide layers in frontend pro- sticking probability h for particle–surface collision.A cesses should always be larger than 2 and ideally large sticking probability (hsl) means that precursor approximately 2.3 w6x.Note that slightly less stringent molecules stick to the surface after the first collision.If, limitations apply to vias in stacked-DRAM since those on the other hand, h is very small, it takes an average are not subject to high temperature anneals. of 1yh collisions before precursor molecules stick to In this paper, we demonstrate a procedure to choose the surface.Since particles can only reach the bottom appropriate precursors and process conditions for tung- of a deep trench after many collisions with the sidewalls, sten silicide CVD into structures with extreme aspect it is evident that a good step coverage is only achieved ratios.An optimized process with both good step cov- if h is very small.This is equivalent to the extreme erage and suitable stoichiometry for frontend applica- reaction-limited process regime. tions will be proposed. In a gas the mean free path length l is given by ls This paper is organized as follows: Section 2 describes kTy(62sp), with k being the Boltzmann constant, T the the general requirements on growth kinetics to achieve gas temperature, s the collision cross-section of the good step coverage in high AR structures.Selected molecules and p the total pressure.Inside a deep trench, precursors currently available for WSix -CVD are briefly on the other hand, the mean path length between discussed in Section 3 and evaluated in terms of the collisions is approximated well by the trench diameter, proposed application.Experimental results of the inves- provided the latter is smaller than l. In this so-called tigated deposition process are presented in Section 4 ‘Knudsen regime’ the step coverage of a process has and compared to chemistryytopography simulations on been calculated analytically w8x: the feature scale in Section 5.Experimental and simu- lation results are discussed in Section 6, followed by a t 1 L 3h summary and a conclusion in Section 7. B s with fs (1) tWfW y 2 S cosh fq sinh f 2. Growth kinetics 2L

Generally, CVD can be divided into four steps: gas phase reactions of precursors in the deposition chamber (A), adsorption of reactive molecules at the surface (B), surface diffusion of adsorbed species (C) and finally the surface reaction leading to bulk deposition (D). While each of these steps can be present in a CVD process, it has been shown that surface diffusion (step C) is negligible for tungsten deposition w7x.Furthermore, in the remainder of this section we concentrate on feature scale aspects, assuming that step A has already taken place. Under the above assumptions, steps BqD determine the (local) deposition rate and thus the step coverage of the process.At low temperatures, the deposition rate is Fig.1.Schematic of a layer deposited into a trench and definition of generally limited by the surface reaction rate.This the variables in Eq. (1). B. Sell et al. / Thin Solid Films 443 (2003) 97–107 99

tion has been published, let alone a production-worthy process.Mixed-phased layers are generally deposited ( ) using tungsten hexafluoride WF6 or tungsten carbonyl ( ( ))w x W CO 6 11–13 .In this paper we will only discuss WF6 since it is the most commonly used tungsten precursor in the semiconductor industry.As silicon ( ) ( ) containing precursors, silane SiH426, disilane Si H ( ) and dichlorosilane H22 SiCl , DCS are all commonly used and are, therefore, considered in the following text.

For the CVD process using WF6 and silane or disilane, it has been shown that the deposition rate crucially depends on the occurrence of a radical chain reaction in the gas phase w14x.With a process tempera- ( ) ture above the so-called ‘extinction temperature’ Tex , the radical chain reaction rapidly saturates, and the concentration of reactive gas phase precursors does not Fig.2.Sticking probability h required to achieve a specific step cov- erage for a given aspect ratio as calculated using Eq. (1). depend on T.If, on the other hand, the reactor temper- ature is below Tex, the chain reaction ceases and the deposition rate drops to 0 due to lack of reactive where h is the sticking coefficient and the other quan- precursors w9x.It has been shown, however, that even tities are as explained in Fig.1. ( ) s in the case the surface temperature is well below Tex, Eq. 1 indicates that for a given aspect ratio AR deposition can be achieved by heating the gas inlet (and LyW the step coverage solely depends on the sticking thus initiating the radical chain reaction) w9x.Using this coefficient h.To illustrate this behavior, Fig.2 shows technique, CVD of WSi at temperatures as low as 313 the h required to achieve a specific step coverage for a x s K has already been demonstrated.It is important to note given aspect ratio.If, for instance, a trench with AR that under these conditions the step coverage is still 50 is to be filled with 50% step coverage, the sticking determined by the surface temperature rather than by = y4 coefficient should not exceed 5 10 .Note, however, that of the inlet. that as the deposited film moves inwards, the AR continuously increases, resulting in a lower step cover- 3.1. WF and silane age later on in the process w9x. 6 Shimogaki et al.suggested a model for the gas phase 3. Reaction chemistry reaction between WF6 and silane that is similar to the disilane reaction discussed below w15,16x.According to While the above arguments provide general insights this model the radical chain reaction leads to the into the properties of a suitable deep trench-CVD pro- following gas phase reaction: cess, detailed studies are required to evaluate a specific q ™ q chemistry.In this section we discuss some chemistries WF64SiH WF 53Ž. SiH HF commonly used for planar WSi -CVD with focus on x (2) their applicability for high aspect ratio trench. q q ™ q As follows from Sections 1 and 2, a WSix deep WF6yy Ž.SiH345yySiH WFyy Ž. SiH3 q1 HF trench-CVD process has to simultaneously meet two requirements.On the one hand, the total sticking prob- These reactions describe the successive replacement ability of the precursors has to be lower than approxi- of by SiH36 at thc WF -molecule.Assuming that = y4 ( ) mately 5 10 in order to enable sufficient step deposition is mainly due to the resulting WF6yz SiH3 z coverage.On the other hand, the individual concentra- molecules w17x and that the above reactions have posi- tions and sticking probabilities of the precursors need to tive activation energy, this model correctly predicts an f be such that the composition of WSix is x 2.3 every- increasing Si content of the deposited layer for increas- where inside the trench.In the following text we will ing temperature. discuss several chemistries with respect to these For the SiH46yWF -system, Saito et al.have deter- requirements. mined the total sticking probability h as a function of According to theoretical results w10x, single-phase temperature w9x.The observed behavior is plotted in

WSi2 can be deposited using volatile organo-metallic Fig.3.For this chemistry, h is higher than the required w ( 55) x y4 tungsten precursors like HW2552h -C H where h - 5=10 for all temperatures above room temperature. C55 H represents cyclo-pentadienyl.However, to our Additionally, the described chemical model predicts a knowledge no experimental confirmation of this predic- composition close to xsl for low temperatures.Both 100 B. Sell et al. / Thin Solid Films 443 (2003) 97–107

Again, this pair of reaction results in the reduction of • WF625 , by a Si H -group while the reactive WF 5 , radical is restored. The reactions Eqs. (4) and (5) produce a significant ( ) amount of reactive WFy Si256 H yy precursors which can deposit at the surface.According to this model a higher SiyW ratio of the deposited layer is obtained by increas-

ing the degree of reduction of WF6 .This can be achieved by heating either the substrate or the gas phase.Note,

however, that in contrast to the WF6ysilane chemistry, here the film-forming species contains at least two silicon atoms per tungsten atom, even at low tempera- tures w14x.This favorable behavior is consistent with both, thermodynamic calculations w19x and experimental observations w9x. It has been shown by Kimbara that the minimum Fig.3.Total sticking probabilities as a function of deposition tem- ( ) perature.Data is taken from w9x for the silane and the disilane pro- temperature Tex to sustain the above radical chain cesses and from w21x for the DCS process.The open and closed reactions (and thus the deposition) depends on the symbols represent depositions with and without preheating of precur- surface-to-volume ratio (AyV) of the reactor w20x.Gen- sors, respectively. erally it is observed that increasing AyV also increases w x Tex.According to Ref. 9 this dependence of the properties render this chemistry unsuitable for deep extinction temperature on reactor geometry is caused by trench application, which will, therefore, not be consid- a competing loss of reactive radicals to the reactor walls: ered further in this article.

3.2. WF and disilane 6 • q• ™ WF525525Si H WFŽ. Si H w x While WF6 decomposes at silicon surfaces 18 ,it has been shown experimentally that in the presence of ( ) • ™ sufficiently high concentrations of disilane Si26 H 2 H H2 another reduction path dominates w14x.In this radical w x chain reaction, first proposed by Saito et al. 9 ,WFis6 • ™ ( ) very efficiently reduced by an autocatalytic gas phase 2 F F2 6 reaction.According to current understanding, the radical chain reaction is initiated immediately at the gas inlet w x • q• ™ by a thermal dissociation of WF6 14 . H F HF ™• q• ( ) WF65WF F 3 Once this reaction has occurred, both resulting radi- « cals can initiate a cascade of successive replacements of fluorine atoms by Si25 H -groups.In the case of the If the loss of radicals overcomes their production (Eq. • radical F, this is (3)) the chain reaction ceases and the deposition rate • q ™• q drops to 0 due to lack of reactive precursors w9x. F Si26 H Si 25 H HF Though the above gas phase chemistry model pro-

vides a thorough understanding of Si26 H yWF 6 -based • q ™ q• Si25 H WFyŽ. Si25 H 6yy WFyy125 Ž. Si H 7yy F CVD, it has to be noted that no surface chemistry model (ys1«6)(4) exists for this process.Nevertheless, the experimentally obtained sticking coefficients plotted in Fig.3 show that • Similarly the WF5 , radical produced by the thermal even at room temperature acceptable step coverage can dissociation Eq. (3) can also initiate a reduction chain only be achieved for moderate aspect ratios smaller than reaction: 50:1.Because of this limitation we consider the Si26 H y • q ™ q• WF6 chemistry inappropriate for deep trench-CVD and WF526525Si H WFŽ. Si H H will not discuss it further in this study.Note, however, (5) that for medium aspect ratio trenches this process seems •HqWF ™•WF qHF to be well suited, especially because of its favorable Siy 65 W-ratio of the deposited layers B. Sell et al. / Thin Solid Films 443 (2003) 97–107 101

Table 1 Kinetic parameters for Eq. (7) as given by Cale et al.

q w ( 2 ()bjjg x ( ) ( ) Phase kj moly cm s mTorr Ej kcalymol bj gj K 1ymTorr Si 1.3=1019 90 2 0 0 30 WSi2 3.6=10 120 1 1 1000 4 WSi53 9.5=10 40 0.5 1 0

3.3. WF6 and dichlorosilane peratures might already reveal some problems that can occur at low temperatures. In contrast to the two systems described above, there Even without quantitative analysis it is evident that exists no gas phase reaction model for CVD based on the relative activation energies of the above model favor ( ) ( ) H22 SiCl DCS and WF 6 .For this chemistry it is the deposition of the tungsten-rich W53 Si b phase at apparently assumed that either there is no relevant gas low temperatures.This trend is consistent with experi- phase reaction, or that the gas phase reactions do not mental data that will be shown in Section 4.Since, a vary noticeably in the process range explored so far. SiyW ratio of 0.6 is unacceptable for deep trench-CVD However, Cale et al.have developed a surface chemistry for stability reasons (Section 1), it is apparent that this model which summarizes the present state of knowledge behavior is a major drawback of the DCS chemistry for this system w21x.It accounts for the deposition of (especially compared to disilane).However, very high ( ) the desired WSi2 b phase, but also for the deposition DCS and very low WF6 partial pressures might suffice ( ) ( ) of Si b and a tungsten-rich, meta-stable W53 Si b bulk to counteract this effect. phase.This model, which its developers have imple- A great advantage of the DCSyWF6 chemistry as mented into their feature scale simulator Evolve w22x, compared to silane and disilane is the much lower consists of the following three surface reactions: fluorine and bulk concentration, the low mechanical stress, and the better adhesion w4x.It has ™ ( )q been observed that the F and Cl bulk concentrations H22 SiCl Si b 2HCl increase with deposition temperature, a trend which is opposite to that in most other impurities w24x.However, q ™ ( )q q q ( ) both concentrations exhibit no measurable dependence WF6224H SiCl WSi 2b 8HCl SiF 4SiF 27 of the individual flow rates.

A possible problem for a DCSyWF6 -based deep q ™ ( )q q q trench-CVD process is the experimental observation that 5WF6225311H SiCl WSi b 22HCl 7SiF 42SiF the initial reduction of WF6 is by the silicon–substrate rather than by DCS.This effect leads to a tungsten rich The deposition rate, R,j of these reactions has been parameterized by: layer at the interface that is covered by other phases w18x.While these W-rich layers disappear at high dep- osition temperatures they might cause a problem at low BEBEbgjj s CFCFy Eppj DF temperatures. Rjjk exp DGDGq RT 1 KpF Summarizing the DCSyWF6 -chemistry we can state s ( ) ( ) ( ) ( j Si b , WSi253b ,WSi b that this system offers some advantages low FySi concentrations, low stress, good adhesion) but also ( where k,jjis a prefactor, E the activation energy, pF the includes some drawbacks high W content at low T,W- WF partial pressure and p the DCS partial pressure. rich interface).However, it is hard to decide about this 6D w x bjjand g are the respective reaction orders of DCS and chemistry since the model in Ref. 21 has only been WF6 .The rather unfamiliar factor on the right-hand side calibrated to a narrow temperature range between 733 has been introduced in Ref. w21x to account for the and 813 K and experimental data for low T are sparse. experimental observation of Schmitz et al.that the The only available measurement of the individual stick- w x deposition is inhibited by the reaction product SiF4 7 . ing coefficients in Fig.3 indicates that especially the Note that no inhibition has been observed for HCl which value for WF6 is far too high at 753 K.This implies is produced in large quantities by the above reactions. that still much lower temperatures are required to fill a The kinetic parameters for Eq. (7) are summarized in high aspect ratio feature with uniform stoichiometry. Table 1 w23x. Because of the lack of literature data on the DCSy 4. Experimental

WF6 -chemistry at low temperatures it is quite difficult to decide whether this system is suitable for deep trench- To fill the gap in the experimental data base and to CVD.Still, extrapolating the above model to low tem- determine the individual sticking coefficients of DCS 102 B. Sell et al. / Thin Solid Films 443 (2003) 97–107

However, at temperatures above 750 K the deposition rate levels off to a maximum at 80 nmymin.In this regime, which is dependent on chamber design, total pressure, and the precursor flux, the deposition rate is limited by the transport of reactive species to the surface. The transition between the two regimes is indicated by a vertical line. In order to determine the step coverage in the respec- tive regimes deposition experiments on structured wafers have also been performed.Scanning electron microscope (SEM) images of the top of a deep trench are shown in Fig.5.It is clearly visible that at 873 K only little deposition occurs inside the deep trench while at 693 K a smooth layer with good conformality is deposited. Fig.4.Experimental deposition rate as a function of temperature for While the above measurements yield the total sticking a DCS:WF6 flow ratio of approximately 50:1 and a total chamber ( pressure of 1.2 Torr. coefficient as a function of temperature and thus the step coverage) they reveal nothing about the individual sticking probabilities (and thus the stoichiometry of the and WF6 as a function of temperature we have per- formed the experiments described in the following.The deposited layer).To access these we have performed experimental setup and the results are presented in this additional experiments. section.All experiments presented here have been per- A common tool to measure bulk composition is formed in an 8-inch Applied Materials Centura single Rutherford backscattering (RBS).But, since this tech- wafer tool using WF6 and DCS as precursors and argon nique is both tedious and expensive we also applied the as carrier gas.In order to be consistent with earlier much faster and cheaper method of measuring the measurements w25x and with the above chemistry-model thickness using SEM and the film resistivity using a the wafer temperature was assumed to be 50 K lower four-point probe.Proof that this simple method is suffi- than the measured chuck temperature. cient to determine bulk stoichiometry of WSix -films is Fig.4 shows an Arrhenius plot of the measured shown in Fig.6 where data from Refs. w18,26x as well deposition rate on blanket wafers for a DCS:WF6 flow as our own data are plotted.The composition of all data ratio of approximately 50:1 and a total chamber pressure point marked as ‘experiment’ was determined by RBS of 1.2 Torr. The exponential decrease at low tempera- measurements in this study.Once we calibrated this tures indicates that in this regime the deposition rate is analysis with the obtained RBS results, we used thick- limited by thermally activated reactions. ness and resistivity as preferred method to determine

( ) Fig.5.SEM images of a WSi -layerx deposited in DTs with an aspect ratio of 40:1.Layers were deposited at 693 K left image and at 873 K (right image). B. Sell et al. / Thin Solid Films 443 (2003) 97–107 103

Table 2

Molar masses and bulk densities of species involved in the WF6 reduction by DCS w21x

Si WSi2 WSi53 DCS WF6 ( ) Mj gymol 100 298 ( ) Mk gymol 28 240 1004 ( 3) 6 6 6 rk gym 2.2=10 9.3=10 14.55=10

partial pressure of a molecule pjjits deposition rate R is determined by the surface flux Jj, and the sticking probability hj

MpM R sh J d sh j d (8) jjjrr j Fig.6.Specific resistivity of tungsten silicide layers as a function of ddy2pMRTj s composition.Included in the graph are data from Clark w26x,from w x ( ) Hara et al. 18 as well as new data Experiment and a quadratic fit where Mj and Md represent the precursor and bulk molar to the data.Resistivity after an anneal shows only minor influence on masses, and rd the bulk density.Values for the species composition. considered in this paper are summarized in Table 2.The

sticking coefficients for DCS and WF6 obtained from composition.As it is apparent, the specific resistivity of the above data and also from data by Raupp et al. w25x the as-deposited film depends quadratically on stoichi- are shown in Fig.8. ometry.This relation can thus be used to determine the SiyW ratio of as-deposited filmy from measurements of 5. Simulation the specific resistivity.Note that this strong dependence is lost after thermal annealing where a resistivity of For additional insight into CVD using WF6 and DCS, approximately 65 mV cm is achieved as reported also we have performed extensive simulations based on the by others w27x. surface reaction model by Cale et al. w21x.The purpose Using this technique, the bulk composition is meas- of these studies is to identify process conditions under ured as a function of deposition temperature.The which both a good step coverage and a SiyW ratio of obtained stoichiometries are presented in Fig.7.Accord- f2.3 are achieved along the total length of the deep ing to our data, silicon-rich WSix is deposited at a trench. temperature above 820 K while the silicon content Once calibrated, simulations are performed within steadily decreases to 1.0 at 723 K. minutes and give easy access to the step coverage and From the measured planar deposition rate and stoi- the stoichiometry inside the deep trench which is hard chiometry the individual sticking probabilities of the to measure experimentally.However, it is important to precursors can be obtained from the kinetic gas theory remember that for deep trench-CVD the chemistry model w x as follows: given the surface temperature Ts and the in Ref. 21 needs to be pushed far beyond its validated

( ) Fig.7.Measured bulk composition as a function of deposition tem- Fig.8.Sticking coefficients h of DCS and WF6 as a function of perature for a DCS:WF6 flow ratio of approximately 50:1 and a total temperature, as determined from the deposition rate.Also shown are chamber pressure of 1.2 Torr. the h calculated from data by Raupp et al. w25x. 104 B. Sell et al. / Thin Solid Films 443 (2003) 97–107

Fig.10.Arrhenius plot of the reaction rate of Fig.9. Fig.9.Measured and simulated deposition rate as a function of tem- perature for a DCS:WF6 flow ratio of approximately 50:1 and a total chamber pressure of 1.2 Torr. while the planar deposition rate is simulated quite well, Fig.11 shows that the simulated Si yW ratio as a parameter space.Thus, the simulation results presented function of temperature differs significantly from the below should be considered with appropriate care. experimental data.This indicates that the chemistry Two different types of simulations have been per- model in Ref. w21x is less accurate under the rather formed in this study, namely the deposition on planar extreme process conditions studied here. wafers and into deep trenches.For the simulation of Besides modelling planar deposition rates and SiyW planar deposition we assume that the surface of the ratio we have developed a simulator (Knudsim) which wafer is connected to a reservoir of constant precursor calculates the local deposition rate and stoichiometry partial pressures via a diffusion channel.This channel inside a circular deep trench.Here, the main problem is required to account, for the transport limitation at (and the key to its solution) lies in the extreme aspect, elevated temperatures.For given process (reservoir) ratio of 50:1 or more.While simulators like Evolve give conditions detailed balance requires that the molar loss reliable results for deposition into comparably shallow at the reactive surface (given by Eq. (7), w21x) is equal features w21x they are generally unsuitable for structures to the diffusion flux through the transport channel which with very high aspect ratios.This is due to the fact that is governed by Fick’s first law: an accurate calculation of the transport inside a deep trench requires either, massive CPU power, massive sy JjjjD =n memory, or both. Evolve, for instance, solves the trans- (9) port problem using the view-factor method which involves very large matrices for large aspect ratios.To s Dj vtherml avoid these problems we have exploited the quasi-1D where =nj is the concentration gradient of the molecules, s vtherm y2kTym the average thermal velocity and l the mean free path of the precursors as described in Section 2.Given the length of the depletion zone the precursor partial pressures at the wafer surface are varied until the loss through reactions equals the respective flux towards the surface.Here the collision cross-section can be estimated whereas the depletion length has to be fitted to experimental data. Fig.9 compares the simulated deposition rate to the experimental data shown in Fig.4.It can be seen that the overall agreement with experiment is rather satisfac- tory.For the reaction-limited regime, which is important for deep trench-CVD, an Arrhenius plot of the deposi- tion rate is shown in Fig.10.Taking into account an offset of 50 K between measured chuck temperature and Fig.11.Measured and simulated Si yW ratio x as a function of dep- wafer temperature, this regime is also described reason- osition temperature for a DCS:WF6 flow ratio of approximately 50:1 ably well by the reaction rates given in Ref. w21x.But and a total chamber pressure of 1.2 Torr. B. Sell et al. / Thin Solid Films 443 (2003) 97–107 105

Fig.12.Experimental and simulated structures for three different temperatures.‘A’ represents reaction-limited CVD at 693 K; ‘B’ the transition of reaction to diffusion-limited deposition at 713 K; and ‘C’ the deposition in the diffusion-limited regime at 783 K. geometry of deep trenches and assume that inside the Knudsim simulator to our custom feature scale simulator deep trench the concentration is independent on radius. Topsi. In each timestep the local diameter di of the This is a reasonable approach since under typical CVD trench is obtained from the simulated front (Topsi). conditions today’s trench diameters are orders of mag- Using di, the normal velocity for each front segment is nitude smaller than the mean free path in the gas. calculated using Knudsim. The front is then propagated Assuming this kind of 1D transport the diffusion (Topsi) using the levelset algorithm w28x before the next inside the deep trench is also governed by Fick’s first timestep is computed. law (Eq. (9)) but with the (Knudsen-)diffusion constant Fig.12 shows a comparison of the thus obtained now given by: structures with SEM images for three different temper- f ( ) atures and ‘standard’ precursor concentrations.‘A’ Djjvd,therm 10 shows a reaction limited deposition at 693 K, ‘B’ a where d is the diameter of the deep trench.As in the deposition at 713 K in the transition regime between above case of planar deposition detailed balance requires reaction and diffusion limited and ‘C’ a diffusion limited that the molar loss due to surface reactions is equal to deposition at 783 K.As can be seen the step coverages the diffusion flux.But while the surface reactions occur and overall shapes of the evolved structures agree very after the transport through the depletion zone, inside the well for all temperatures. deep trench transport and deposition occur In order to identify a process with both good step f simultaneously. coverage and an uniform stoichiometry SiyW 2.3 we have performed an optimization of the free parameters In order to obtain the precursor concentrations uj of p , p and T with respect to these properties.Our WF6 and DCS we divide the deep trench into segments DCS WF6 of length l and iteratively solve the equation for the simulations suggest such a regime below 650 K, a DCS steady state of uj,i at each discrete position i. partial pressure of 2–10 Torr and a WF6 partial pressure of several mTorr. ≠u 3 syDj y y q ( ) 0' Ž.Ž.2uj,iju ,iq1 uj,iy1 8Rk u1,2 11 ≠ tl ks1 6. Discussion where the reaction rates Rj are those from Section 3.3. To solve Eq. (11) we use successive overrelaxation until Already the general discussion of growth kinetics convergence is achieved.Given the precursor concentra- yielded that a deep trench-CVD process needs to be ( ) tions uj,i the local deposition rate step coverage and reaction limited in order to have a good step coverage stoichiometry (x) are readily obtained from the reaction inside high aspect ratio structures.Assuming an aspect rates 7. ratio of 50:1 and a targeted 50% step coverage relation In order to study the impact of increasing aspect ratio 1 (or Fig.2 ) indicates that the total sticking coefficient during the process we have coupled the above described h has to be lower than 5=10y4 .For the chemistry 106 B. Sell et al. / Thin Solid Films 443 (2003) 97–107 discussed here this in turn implies a deposition temper- coverage of more than 50% can be deposited below 700 ature well below 700 K as shown in Fig.8. K if the DCS partial pressure is 2–3 orders of magnitude

While very low process temperatures might solve the higher than the WF6 pressure. problem of insufficient step coverage it has been shown experimentally that WF6yDCS-based CVD produces Acknowledgments tungsten-rich layers which are unsuitable for deep trench applications.However, the simulations presented in Sec- The authors would like to thank Stefan Jakschik for tion 5 suggest that the higher deposition rate of W can his great support. be countered by (drastically) increasing the DCS partial pressure while decreasing the WF6 concentration.While References the proposed process conditions are far beyond the calibrated parameter space of the chemistry model in w x w x 1 S.Wolf, R.N.Tauber, Silicon Processing for the VLSI Era, Ref. 21 the prospect of metallic deep trench plugs and vol.1, Lattice Press, Sunset Beach, 1988, p.384. vias has already triggered respective experiments which w2x N.Thomas, P.Suryanarayana, E.Blanquet, C.Vahlas, R. are currently performed. Madar, C.Bernard, J.Electrochem.Soc.140 (2)(1993) 475. Cale et al.have demonstrated that CVD using WF w3x K.C. Saraswat, D.L. Brors, J.A. Fair, K.A. Monnig, R. Beyers, 6 ( ) and DCS is inherently transient in nature w21x.This IEEE Trans.Electron.Dev.ED-30 1983 1497. w4x S.G. Telford, M. Eizenberg, M. Chang, A.K. Sinha, T.R. Gow, means that the composition of the film changes during J.Electrochem.Soc.140 (12)(1993) 3689. the deposition in a trench due to the change in aspect w5x J.R. Chen, Y.K. Fang, S.L. Hsu, Vacuum 37 (3y4)(1987) ratio.Compositional changes during the process are 357. observed already with an aspect ratio of 5:1.When the w6x M. Katiyar, G.S. Samal, R.K. Gupta, Deepak, P.K. Sahoo, V.N. opening of a structure shrinks the Knudsen diffusion of Kulkarni, O.Adetutu, Mater.Res.Soc.Symp.Proc.,Materials the reactants into the trench slows down and the depo- Research Society, 2001, p.K571. w7x J.E.J. Schmitz, W.L.N.v.d. Sluys, A.H. Montree, in: G.C. Smith sition rate decreases.While this effect is correctly (Ed.), Tungsten and Other Advanced Metals for VLSIyULSI modelled by our combined KnudsimqTopsi simulator Applications V, Materials Research Society, 1990, p.117. all above simulation results (with file exception of Fig. w8x Y.Shimogaki, T.Saito, F.Tadokoro, H.Komiyama, J.de Phys. 12) were obtained without accounting for local and IV Colloq.C2 suppl.au J.de Phys.II 1 (1991) 95. transient trench diameters.Thus, the results presented w9x T.Saito, Y.Shimogaki, Y.Egashira, H.Komiyama, K.Suga- here only represent the initial timestep in the process. wara, K.Takahiro, S.Nagata, S.Yamaguchi, Electron.Com- mun.Jpn.,Part 2 78 (10)(1995) 73. If, however, the currently performed experiments turn w10x F.A. Kuznetsov, V.A. Titov, A.N. Golubenko, A.A. Titov, out promising, further refinements of the parameters can Proceedings of SPIE, vol.1783, The International Society for be achieved by fully transient simulations. Optical Engineering, Warsaw, 1992, p.541. For the extremely low sticking probabilities h w11x K.Lai, H.Lamb, Thin Solid Films 370 (2000) 114. required for deep trench-CVD the deposition rate nec- w12x L.H. Kaplan, F.M. D’Heurle, J. Electrochem. Soc. 117 (1970) essarily is very low.For a first-order reaction and a 693. w13x F.Hamelmann, S.Petri, A.Klipp, G.Haindl, J.Hartwich, L. given precursor pressure the deposition rate can be Dreeskornfeld, U.Kleineberg, P.Jutzi, U.Heinzmann, Thin calculated using Eq. (8).For instance, deposition at Solid Films 338 (1999) 70. partial pressure of 1 Torr and a sticking coefficient of w14x Y.Egashira, H.Aita, T.Saito, Y.Shimogaki, H.Komiyama, 5=10y6 does not exceed 10 nmymin.For economic K.Sugawara, Electron.Commun.Jpn.Part 2 79 (1)(1996) 83. reason a production worthy deep trench-CVD process, w x therefore, needs to be either performed in a batch reactor 15 Y.Shimogaki, T.Saito, R.Ikawa, Y.Egashira, K.Sugawara, H.Komiyama, 11th International VLSI Multilevel Conference or in a single wafer tool at higher pressures. (VMIC), 1994, p.496. w16x Y.K.Chae,Y.Egashira, Y.Shimogaki, K.Sugawara, H. 7. Conclusion Komiyama, J.Electrochem.Soc.146 (1999) 1780. w17x Y.K.Chae,Y.Egashira, Y.Shimogaki, K.Sugawara, H. Komiyama, Thin Solid Films 320 (1998) 151. Using experiments and simulations we have identified w x the reduction of WF by DCS as a promising candidate 18 T.Hara, T.Miyamoto, H.Hagiwara, E.I.Bromley, W.R. 6 Harshbarger, J.Electrochem.Soc.137 (9)(1990) 2955. for CVD of WSi2 -layers into deep trenches with aspect w19x E.J.Rode,W.R.Harshbarger, J.Vac.Sci.Technol.B 8 (1) ratios of 050:1.The simulations presented in this paper (1990) 91. are based on a literature surface chemistry model and w20x K.Kimbara (Ed.), Gaseous Combustion Theory, Shokabo yield realistic step coverages over a wide range of Press, 1985, p.120. temperature.Justified by these results we propose a w21x T.S. Cale, J.H. Park, G.B. Raupp, M.K. Jain, B.R. Rogers ( ) CVD process with both good step coverage and suitable Eds. , Advanced Metallization for ULSI Applications, Mate- rials Research Society, Murray Hill, NJ, USA and Tokyo, stoichiometry inside the deep trench.While conclusive Japan, 1992, p.93. experimental confirmation is still missing our simula- w22x T.S. Cale, EVOLVE, developed at Arizona State University tions suggest that stoichiometric films with a step and Motorola Inc. B. Sell et al. / Thin Solid Films 443 (2003) 97–107 107 w23x T.S. Cale, T.H. Gandy, M.K. Jain, G.B. Raupp, M. Govil, A. w26x T.E.Clark,J.Vac.Sci.Technol.B 6 (6)(1988) 1678. Hasper (Eds.), Advanced Metallization for ULSI Applications, w27x H.Yoon, C.Chen, A.Singhal, D.Lopes, G.Miner, S.Hong, Materials Research Society, Murray Hill, 1992, p.101. Electrochemical Society Proceedings, vol.10, 1999, p.249. w24x T.H.T. Wu, R.S. Rosler, B.C. Lamartine, R.B. Gregory, H.G. w28x J.A. Sethian, Cambridge Monographs on Applied and Com- Tompkins, J.Vac.Sci.Technol.B 6 (6)(1988) 1707. putational Mathematics, ISBN 0-521-64204-3. w25x G.B. Raupp, T.S. Cale, M.K. Jain, B. Rogers, D. Srinivas, Thin Solid Films 193y194 (1990) 234.