Planarization

This invention relates to etching of thin films used for semiconductor manufacturing and to improvements in planarization obtained with such films.

Planarization in semiconductor processing involves deposition of a sacrificial layer of material that is then etched back to provide a surface that is smoother than the surface on which the sacrificial layer is deposited. This is illustrated in FIG. 1, in which a substrate 13 has a plurality of metal or other electrical conducting lines 15, 17 and 19 at spaced apart positions. Another layer of metal lines (not shown) is to be deposited above this structure in fabrication of an integrated circuit, and the two sets of lines must be electrically insulated from each other. A layer of an oxide of silicon 21, which will serve as an electrical insulator, is then deposited conformally on the substrate 13 and on the metal lines 15, 17 and 19 so that the metal lines are fully covered with insulating material. Ideally, the oxide of silicon layer 21 would also provide a smooth, planar exposed surface at the top of this structure so that planarization would not be necessary. In practice, the exposed surface of the oxide of silicon layer 21 is "bumpy" wherever a metal line, such as 15, 17 or 19, lies directly underneath. To remedy this and to provide a smoother surface, a sacrificial layer 23 of another insulating material. such as spin-on-glass (SOG), is deposited over the oxide of silicon to provide a smoother exposed surface, and the exposed surface is etched back to a position such as a plane LL indicated in FIG. 1.

Where only the SOG material 23 is exposed, the etch rate can be well controlled because the surface is homogeneous. However, as soon as portions of the underlying oxide of silicon layer 21 are exposed, the homogeneous layer becomes instead a patterned structure with two (or more) distinct materials being etched back. The etch rates of the SOG may be considerably influenced by the amount of the exposed underlying oxide.

Planarization of a patterned structure containing disparate materials, such as the oxide of silicon and SOG layer exposed at the plane in FIG. 1, has been studied by many workers. One goal is improving the quality of such planarized surfaces. Problems arise because of the strong dependence of etch rate of spin-on-glass on certain parameters such as (1) the relative surface area of the oxide of silicon being exposed and etched and (2) the relative percentage the etchant gases such as CHF.sub.3 and C.sub.2 F.sub.6.

Where a patterned structure such as 11 shown in FIG. 1 is developed, containing exposed surfaces of two or more semiconductor processing materials, maintenance of acceptable planarization is difficult at best. The two materials may have very different etch rates for the type of etch chemistry used. During etchback of patterned structures exposed surfaces, such as (1) a stoichiometric silicon dioxide (SD) material and (2) an SOG material, may be plasma etched simultaneously. The spin-on glass may be either a siloxane or a silicate based glass. As seen from FIG. 1, a patterned wafer has SOG covering its surface prior to etchback. In the initial stages of the etch process, the SOG etch rate is fairly constant. However, as the etch proceeds, the underlying oxide is exposed. This exposed oxide tends to increases the etch rate of the SOG adjacent to the exposed oxide. The oxide etch rate remains unchanged during the entire etch process. This concept could be understood by defining a parameter called selectivity given by

As can be seen from the above discussion, the etch selectivity varies as the SOG etch rate depending on the amount of oxide exposed. In order to maintain good planarization, it is very important to etch the SOG and the oxide at the same rate. In other words, the selectivity during the entire etch process should remain constant and close to unity.

Planarization can be significantly degraded if the SOG etch rate is higher than the etch rate of the oxide (S for a patterned wafer >1). The quality of planarization of a surface developed in a semiconductor processing material is often measured by analyzing the dimensions of a gap in an otherwise-planar surface, as illustrated schematically in FIG. 2. The ratio R of the horizontal width A of the gap to the vertical depth B is a quantitative measure of quality of planarization. The higher the ratio R the better the planarization. For good planarization, the width of a gap relative to its depth should be very large and R>>1. As the value of R decreases, subsequently formed layers may have poor step coverage and may thereby leave behind undesirable stringers after the etch process is terminated.

Many workers in semiconductor processing have sought solutions for this selectivity problem. Vines and Gupta, in "Interlevel Dielectric Planarization With Spin-on Glass Films", VMIC Conf. Proc., 1986, pp. 506-515, discuss the state of the art of interlevel planarization in 1986, noting that a spin-on glass etch rate which is independent of the amount of etchback is desirable for accurate transfer of the smooth topography of a sacrificial layer to an underlying oxide layer.

Hausamann and Mokrisch, in "The Dependence Of Oxide And Spin-on Glass Etchrates On Their Area Ratio", VMIC Confer. Proc., 1988, pp. 293-298, noted the appearance of an unexpected dependence of SOG etch rate on oxide exposure from a plasma enhanced chemical vapor deposition (PECVD) film during etchback for planarization. This observed effect was probably due to oxygen loading. The simultaneous presence of oxide of silicon and SOG on patterned wafers was simulated with unpatterned oxide and SOG wafers and the addition of oxygen as a process gas to the etch chemistry. As can be seen from their results shown in FIG. 3, increasing the amount of oxygen increases the SOG etch rate, whereas the oxide etch rate remains almost unchanged. The authors suggest use of a two-step process. In the first step, approximately 60 percent of the SOG thickness would be removed. In the second step, the oxide and the remaining SOG would be etched with a lower selectivity etchant.

Bogle-Rohwer and Nulty, in "SOG/PSG Etchback Planarization Process", Proc. S.P.I.E., vol. 1392, 1991, pp. 280-290, describe the effect of use of CHF.sub.3, CF.sub.4, C.sub.2 F.sub.6, SF.sub.6 and Ar gases as etchant components for simultaneous etchback of SOG and SD material. Processes having initial selectivity S(SOG;SD)=1.0 were not successful in planarizing structures over patterned wafers. Use of pure CHF.sub.3 or pure Ar as the etchant process gas produced no significant etching of the SOG. Use of fluorine-rich gases, such as CF.sub.4, C.sub.2 F.sub.6 and SF.sub.6, in various mixtures produced adequate etching, although planarization was still a problem.

One persistent problem here is that the SOG etch rate r(SOG) increases with increased exposure of the oxide adjacent to the etch site. Therefore, the SOG etch rate r(SOG) and the selectivity S for any localized region on a wafer may not be constant throughout an etch process. This phenomenon is often referred to as micro-loading. Conventionally, micro-loading has been compensated for by using an SOG material and an etchant that provide a selectivity S(SOG;SD) as low as 0.5 for unpatterned wafers. With a selectivity less than unity on these unpatterned wafers, the SOG etch rate is made deliberately less then the oxide etch rate. This compensates for the increase in SOG etch rate seen on patterned wafers with exposure of the underlying oxide. Thus, a selectivity less than 1 on unpatterned wafers translates to a selectivity close to 1 on patterned wafers. These selectivities are attained by using a fluorine-containing etchant gas, such as CHF.sub.3 combined with a small fraction of CF.sub.4, C.sub.2 F.sub.6 or SF.sub.6, as the process gas. Improvements in control, using this approach, were demonstrated by Bogle-Rohwer and Nulty, op. cit. The effect of the micro-loading problem is reduced somewhat.

Another problem occurs with this approach. Polymer formation within the process gas is enhanced in a fluorine-deficient environment, such as use of a pure CHF.sub.3 etchant, and polymer formation degrades the etch chamber cleanliness and degrades the uniformity of the etch process.

Another persistent problem is that the selectivity S(SOG;SD) for these two materials varies with the fraction f(CHF.sub.3) of CHF.sub.3 present in the etchant gas. This etchant component is a preferred etchant. However, the selectivity S for stoichiometric silicon dioxide SD decreases dramatically with increasing fraction of CHF.sub.3 present.

These problems have prevented achievement of acceptable planarization for a patterned semiconductor structure. What is needed is an approach to planarization of topographies encountered in semiconductor processing that: (1) provides relative insensitivity of SOG etch rates to modest percentage changes in the amount of exposed oxide area and (2) improves planarization and the controllability and reproducibility of the planarization process.

These needs are met by the invention, which provides a method for improvement of planarization of patterned structures having exposed surfaces of two or more materials using an etchback process. This has been achieved in the case where the two exposed materials are an oxide of silicon and SOG by modifying the composition of the oxide film. The oxide film is enriched with silicon with an average composition represented by SiO.sub.x with x<2. A stoichiometric oxide would correspond to x=2. The etchant process gas preferably includes CHF.sub.3 and C.sub.2 F.sub.6.

By providing an oxygen-deficient, silicon-rich oxide (SRO), the sensitivity of etch rate of spin-on-glass to the exposed amount of the oxide is reduced and the quality and smoothness of a planarized layer are improved. Using a silicon-rich oxide also makes the selectivity less dependent on the relative percentage of the fluorine-richness of the etch chemistry.

Another benefit of the invention is that the silicon-richness of the oxide of silicon and the fluorine content of the etchant gases can be independently varied to achieve a wide range of selectivities. This technique extends to use of other spin-on-materials (SOM), such as polyimide or photoresist material, in place of the SOG. These features significantly widen the process window for the etchback process.

These and other advantages of the present invention will become apparent to those skilled in the art upon a reading of the following specification of the invention and a study of the several figures of the drawing.

FIG. 1 is a schematic sectional view of a portion of a multi-layer integrated circuit to which planarization is applied.

FIG. 2 is a schematic sectional view of typical multi-layer semiconductor processing materials to which the invention is applicable.

FIG. 3 is a graphical view of etch rates of SOG and oxide of silicon versus flow rate of oxygen supplementing the etchant process gas.

FIG. 4 is a graphical view of SOG: oxide selectivity ratios versus fraction of CHF.sub.3 present in the etchant process gas, for varying relative amounts of silicon in a silicon-rich oxide, used according to the invention, and in conventional silicon dioxide.

FIGS. 5A and 5B are graphical views of measured spectroscopic peak intensity of Si--H and Si--O stretching vibrations for a silicon-rich oxide and a stoichiometric oxide, respectively.

FIGS. 6A and 6B are schematic views of etchback of patterned, multi-layer semiconductor processing materials with the invention applied.

We have discovered that use of a silicon-rich, hence oxygen-deficient, oxide (SRO) and SOG in a patterned surface will (1) compensate for the micro-loading effects, (2) decrease the strong dependence of selectivity S on the fraction of CHF.sub.3 present in the etchant gas and (3) widen the process window for the etchback process. We have experimentally determined the selectivity, as a function of f(CHF.sub.3) for four oxides of silicon, including SD, and have plotted these results in FIG. 4. These results were obtained using unpatterned wafers of stoichiometric oxide (SiO.sub.x with x=2) and unpatterned wafers of SRO (x<2), of thickness 5,000 .ANG., both formed as PECVD films, with a film of SOG of thickness 4,500 .ANG. adjacent to each such wafer, using a GCA 650 Triode Etcher. The process gas or etchant was a mixture of CHF.sub.3 and C.sub.2 F.sub.6, total pressure was 225 millitorr, applied power was 300 Watts RIE, and the process gas component flow rates (measured in sccm) for the two components were CHF.sub.3 :C.sub.2 F.sub.6 =150:30, 155:25, 160:20, 165:15, 170:10 and 175:5. As the CHF.sub.3 :C.sub.2 F.sub.6 ratio increases, the relative amount of fluorine for this combination decreases.

Curve A of FIG. 4 corresponds to (stoichiometric) silicon dioxide SiO.sub.x with x=2. Curves B, C and D in FIG. 4 correspond to oxides of silicon with increasing silicon richness in going from curve B to curve C to curve D. Note that the selectivity S(SOG;SRO) is much less sensitive to changes in the fraction f(CHF.sub.3) than is S(SOG;SD), where the oxygen fraction of the oxide of silicon is reduced. Also note that, for f(CHF.sub.3).ltoreq.0.85, selectivity S(SOG;SRO) is approximately independent of the relative amounts of Si and O present in the oxide of silicon material. Different values of selectivity S(SOG;SRO) can be obtained: (1) by changing The relative amount of oxygen present in the SRO material while keeping the fraction f(CHF.sub.3) constant; (2) by changing the fraction f(CHF.sub.3) present in the etchant or process gas while keeping the relative amount of oxygen present in the SRO material constant. A fraction f(CHF.sub.3) less than 1.0 is preferred. The results shown in FIG. 4 are qualitatively similar where the spin-on-glass (SOG) is replaced by a spin-on-material (SOM), such as polyimide or photoresist material.

The relative amounts of Si and O present in the three non-stoichiometric SRO materials used for FIG. 4 are most easily characterized by measurement of the refractive index RI of the OS material for a selected wavelength of light, such as .lambda.=0.633 .mu.m. We have also characterized the relative silicon content in an oxide of silicon by spectroscopic techniques, using the relative spectroscopic absorption peak intensities for an Si--O--Si stretching vibration and an Si--H stretching vibration shown in FIGS. 5A and 5B, respectively, to determine the relative amounts of Si and O present. In silicon dioxide, the Si--H absorption peak is zero, relative to a background absorption line BL that is subtracted out. In a non-stoichiometric oxide of silicon such as SRO, unsaturated Si bonds are present that can accept H atoms and thus produce characteristic Si--H absorption. The silicon richness coefficient CSR is defined as:

where I is the measured absorption peak of the indicated stretching vibration. Table 1 exhibits the selectivity S, the refractive index RI and the silicon richness coefficient CSR for f(CHF.sub.3 )=0.9 for the results of FIG. 4.

The results of FIG. 3 provide confirmation that control of the amount of local free oxygen adjacent to the etch site is a key to control of selectivity and of the quality of planarization for a patterned structure containing SOG and an oxide of silicon. Where silicon-rich oxide is used according to the invention, the etch rate r(SOG) does not vary so dramatically with oxygen released from etching (FIG. 3) or with the percentage of CHF.sub.3 present in the etchant (FIG. 4).

Improvements in planarization using the invention have been quantified by performing etchback experiments on the configuration shown in FIG. 2, where the relevant layer dimensions are initial oxide thickness h.sub.1 =5000 .ANG., initial SOG thickness h.sub.2 =2800 .ANG., and inter-metal (IM) spacing h.sub.3 =1.1 .mu.m, 1.2 .mu.m and 1.7 .mu.m. The ratio R=A/B was measured for oxide of silicon material with CSR=0 and with CSR=0.02 after etchback of 4,500 .ANG. of SOG, and the results are illustrated in Table 2. The figure of merit

is a measure of improvement of planarization, using the invention. The planarization improvement is greater for relatively narrow spaces with small inter-metal spacings (for example, h.sub.3 =1.1 or 1.2 .mu.m)than for larger inter-metal spacings, and the dimensionless figure of merit .eta. appears to increase monotonically with decreasing spacing h.sub.3.

FIGS. 6A and 6B, which are analogous to FIG. 2, schematically compare planarization of a patterned structure containing exposed surfaces of SOG and an oxide of silicon that is either stoichiometric (FIG. 6A) or silicon-rich (FIG. 6B). The improvement in planarization is substantial.

The temperature at which plasma etching is performed may vary from 50.degree. C. to as high as 250.degree. C., and the silicon richness coefficient CSR may be as high as desired. For example, the silicon-rich oxide 21 shown in FIG. 1 may be replaced by amorphous silicon, with no oxygen present in the insulator material. This invention is also applicable to other SOM, such as polyimide or photoresist materials, used instead of SOG for the sacrificial layer.

While this invention has been described in terms of several preferred embodiments, it is contemplated that alterations, modifications and permutations thereof will become apparent to those skilled in the art upon a reading of the specification and study of the drawings. It is intended that the appended claims include all such alterations, modifications and permutations as fall within the spirit and scope of the present invention.