Modeling and Optimization of the Step Coverage of Tungsten LPCVD in Trenches and Contact Holes

A model is presented to calculate the step coverage of blanket tungsten low pressure chemical vapor deposition (W-LPCVD) from tungs ten hexafluoride (WF6). The model can calculate tungsten growth in trenches and circular contact holes, in the case of the WF6 reduction by H2, Sill4, or both. The step coverage model predictions have been verified experimentally by scanning electron microscopy (SEM). We found that the predictions of the step coverage model for the Ha reduct ion of WF 8 are very accurate, if the partial pressures of the reactants at the inlet of the trench or contact hole are known. To get these reactant inlet partial pressures, we used a reactor model which calculates the surface partial pressures of all the reactants. These calculated surface partial pressures are used as input for our step coverage model. In this study we showed that thermodiffusion plays a very important role in the actual surface partial pressure. In the case where Sill4 was present in the gas mixture trends are predicted very well but the absolute values predicted by the step coverage model are too high. The partial pressure of HF, which is a by-product of the Ha reduction reaction, may be very high inside trenches or contact holes, especially just before closing of the trench or contact hole. We found no influence of the calculated HF partial pressure on the step coverage. Differences between step coverage in trenches and contact holes, as predicted by the step coverage model, were found to agree with the experiments. It is shown that the combinat ion of the step coverage and reactor model is very useful in the optimization towards high step coverage, high throughput, and low WF6 flow. We found a perfect step coverage (no void formation) in a 2 ~m wide and 10 i~m deep (2 x 10 i~m) trench using an average WF6 flow of only 35 sccm, at a growth rate of 150 nm/min. In general, it is shown that the reduction ofWF 6 by SiH~ offers no advantages over the reduction by H2 as far as step coverage is concerned. With the increasing degree of complexity in integrated circuits the aspect ratio of contacts and vias also increases. These geometries require deposition techniques capable of filling submicron high aspect ratio contact holes without void formation. Blanket tungsten, deposited by LPCVD and subsequent back etching (Fig. la), is widely used to fill these contacts (1, 2). A second possibility is to fill these contacts using a selective deposition scheme (3-11), in order to avoid the need of back etching (see Fig. lb). An advantage of selective over blanket deposition of tungsten is the economic use of WF6. However, selective deposition of tungsten is only possible when all the contacts are equal in depth (12), which is not always the case (see Fig. 2). Other disadvantages of selective deposition of tungsten are the sensitivity to pretreatments and the nonreproducibility. To deposit tungsten two reducing agents are widely used. First the H2 reduction reaction of WF~ with the following overall reaction WF6 + 3H2 ---> W + 6HF [1] Second, the Sill4 reduction reaction of WF6 2WF6 + 3Sill4 -> 2W + 3SiF4 + 6H2 [2] The H2 reduction reaction has shown its excellent step coverage (1, 2), but it has some disadvantages compared to the Sill4 reduction reaction such as low and very temperature dependent growth rate (3-5) and rough layers (1, 2, 13). 1 Deceased. The Sill4 reduction reaction however has a high and nearly temperature independent growth rate and results in layers with a small grain size (1, 11, 13, 14). The reactivity of WF8 with silicon in the case of the Sill4 reduction reaction is much less than the H2 reduction reaction. The WF6 react ion with silicon results in the undersirable gouging and encroachment of silicon. The problem of step coverage has been evaluated in the case of physical vapor deposition (PVD) (24). In that case, . . . . . . . . . . . . . " . . . . . -