Arthur Sherman - Chemical Vapor Deposition for Microelectronics (779637), страница 6
Текст из файла (страница 6)
In this way, we calculateT at each point up to the upper boundary. Then, with a new estimate of Tavailable, we recalculate k, Wj, and Dj and repeat the procedure.Next, we have to solve for the Yj IS from the species continuity equations,Equation (32). Unfortunately, these equations cannot be integrated by a similar simple point iteration scheme as they are n1athematically Istiff"16 anditerative approaches are unstable. To solve these simultaneous equations, weturn to a perturbation analysis developed by Newman 17 where the equationsare linearized about an initial guess, and the resulting linear equations are solvednumerically.
The solution is then used as the next guess, and the linear equations are resolved. The procedure is repeated until the solution no longer changes.If there are n species, we have n simultaneous linear ordinary differentialequations, which can be solved by well-known techniques.Typically, 7-10 iterations are needed to achieve convergence if an adequatenumber of grid points have been chosen. For problems involving chemicalkinetics, this can be a large number, which leads to a lengthy calculation. Forsome of the cases we calculated, it was necessary to use 3000 grid points overa 3-cm channel height to secure convergence.Once the Yj'S have been calculated, we can recalculate the temperaturesacross the channel. Then, the corrected temperatures can be used to generatea new set of Yj's.
When the T and Yj arrays no longer change, the flow field(u) can be calculated directly, since we can then calculate u as a function ofchannel height.It should be noted that although it may use a large number of points tosolve for the Yj's, a large number is not necessary to obtain accurate representations of T and u.For these calculations, attention was limited to a temperature range of950 to 1350 K, a pressure range of 300 mTorr to 7.6 Torr, and SiH 4 massfractions of 15 to 30% in H2 Under these conditions, the mass fraction ofSi H2 formed was at most on the order of 10-3 , so that the influence of theYj's on the temperature distribution was small.
To a good degree of approximation, the temperature was calculated to be linear. By the same token, theflow field which was easily calculated did not demonstrate any unique behavior.Most of the effort was spent trying to integrate the three simultaneous Yequations. The Y distributions across the channel for a typical condition areshown in Figure 10. The YSiH2 exhibits a peak near the hot wall and is a fairlyfull profile. This can be attributed to the high diffusion coefficient at thesepressures, which allows the Si H2 to readily diffuse toward the cold wall. Deposition on the cold wall is many times smaller than on the hot wall, as evidencedby the smaller value of dY SiH2 /dy there.Deposition rates as a function of hot wall temperature are presented inFigure 11 with pressure as the parameter, and Figure 12 with mass fraction asthe parameter.
For the temperature range studied here, there is no evidenceof a reduction in the rate of increase in deposition rate as the temperature isincreased. Variations with pressure and mass fraction are as would be expected.00o26Chemical Vapor Deposition for MicroelectronicsIt is interesting to compare the present results with data obtained in a hotwall furnace 18 tube, even though the present calculations are for one hot andone cold wall and a different physical arrangement. For one case, depositionrate was measured at p = 532 mTorr, Y = 2.3% and T = 898°K.
Without runningthe exact case numerically, we can estimate from Figures 11 and 12 a calculatedvalue of 2.5 A/min compared to a measured value of 4 A/min.Having the numerically calculated value on the same order of magnitudeas experimentally-measured values lends credibility to the model being usedespecially since the model has been developed from first principles and involvesno adjustable parameters.T= 1050K, P = 300MT, 15 PERCENT SiH 4 IN H21.00.90.80.7:EuIIJ:"wJ:0.60.5..JW22<!0.4J:u0.30.20.1oo0.10.20.30.40.50.60.70.80.91.0MASS FRACTIONFigure 10: Mass fraction distribution of three species in SiH 4 pyrolysis intoSi H2 and H2 .Fundamentals of Thermal CVD27800·C10,0008,0006,0006,0004,000SiH 4 /H 215% MASS FRACTIONT· TEMPERATURE OF HOT WALL3,0002,0001,0008002~600600~400t-300UJ~a::I:~2000a:C)100806040302010 L..._ _--L7.07.5....L.8.0.J..-_ _- &8.5B.O--'--'-:"_ _-9.510 4/T(-K)Figure 11: Sil icon film growth rate as a function of temperature for differentpressures.28Chemical Vapor Deposition for Microelectronics1,000·C100,000900·C800·C70,00050,00040,000SiH 4/H 2300 mTORRT· TEMPERATURE OF HOT WALLY • MASS FRACTION30,00020,00010,0008,0006,0004,0003,0002,0002~3wt<ia;1,000800600:t:~0a:30CJ2010070503020107.07.58.09.08.5104 /T(·K)9.510.0Figure 12: Silicon film growth rate as function of temperature for differentinitial mass fractions.1.4 FILM MORPHOLOGYThere is a substantial literature describing the structure of thin films grownby evaporation, sputtering, and CVD for a variety of applications.
19 ,20. SuchFundamentals of Thermal CVD29films can initially be grown as amorphous, polycrystalline, or single-crystalmaterials, depending on conditions during growth. Amorphous films can beconverted to polycrystalline by annealing at high temperatures. The physicalnature of the thin film material will play an important role in determining filmproperties, such as electrical resistivity, stress, adhesion, among others. Also,the surface roughness may be important in determining the ease with whichoptical lithography can be used in integrated circuit manufacture.Films grown by chemical vapor deposition are similar to the films describedabove, with one exception.
The CVD process allows for the possibility of gasphase particle nucleation, and the incorporation of such particles in a growingsurface can contribute to a surface roughness. This is one reason that atmospheric CVD is being used less and less, as compared to low-pressure CVD wheresuch gas phase nucleation is less likely.In general, dielectric films such as Si0 2 and Si 3 N 4 grow in CVD processesas amorphous films. Metallic films generally grow as polycrystalline with atypical columnar structure. Figure 13a shows a low-pressure CVD polysiliconfilm as deposited with such a structure.
19 Annealing of metallic films can alterthis columnar structure, again depending on conditions. For undoped polysilicon films, annealing (1 hour, 950°C) leaves the film structure unchanged.However, when the polysilicon film is phosphorus doped, annealing substantiallyalters the grain structure, as shown in Figure 13b.O.26J18Figure 13: Polysilicon film deposited by LPCVD, after Murarka.
1930Chemical Vapor Deposition for MicroelectronicsCVD tantalum silicide and aluminum 21 go down as polycrystalline films,as shown by the TEMs in Figures 14 and 15. In general, any of the metallicfilms will be polycrystalline. The crystal size, however, will depend on processing history and/or heat treatment.
As noted earlier, surface roughness mayoccur when these films are deposited. In particular, CVD aluminum grown in ahot tube 21 tends to exhibit a rough surface, as shown in an SEM in Figure 16.Figure 14: SEM of CVD TaSi 2 on oxide.Figure 15: TEM of aluminum thin film deposited in LPCVD hot tube furnace.After Levy?1 (Reprinted by permission of the publisher, The ElectrochemicalSociety, Inc.)Fundamentals of Thermal CVD31Figure 16: SEM showing a rough surface of LPCVD aluminum grown in a hottube furnace. After Levy.
211.5 LABORATORY THERMAL CVD REACTORSIn this section we will review the various types of CVD reactors scientistsand engineers have used for the development of thermal CVD processes. Thiswill be distinct from the commercial reactors used for production which willbe covered in a later chapter. A similar review of reactors for development ofplasma-enhanced CVD processes will be made at the end of the next chapter.We will cover the so-called cold wall systems for either single or multiple wafersfirst, followed by a discussion of continuous belt systems.
Finally, we will review the hot wall reactor approach.1.5.1 Cold Wall Systems-Single WaferWhen we speak of a cold wall CVD reactor, we refer to a continuous flowsystem where the wafer is kept at the required high temperature, but all othersurfaces bounding on the reacting gases are cold. The objective here is to causethe desired reaction only on the hot wafer and keep all other surfaces free ofdeposits. In practice this is a goal that can only be partially attained. Althoughreactions will proceed more slowly on colder surfaces, they will proceed-andfilms will build up. At the same time the films that form on the colder surfacesmay be more porous than the normal film and may spall off more easily.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
