Method Manifesting a Wide Process Window and Using Hexafluoropropane or Other Hydrofluoropropanes to Selectively Etch Oxide

The invention relates generally to etching in semiconductor processing. In particular, the invention relates to plasma etching of silicon oxide layers, preferably in a process producing high selectivity to silicon nitride or other non-oxide materials and not etch stop and exhibiting a wide process window.

The technology of fabricating semiconductor integrated circuits continues to advance in the number of transistors, capacitors, or other electronic devices that can be fabricated on a single integrated circuit chip. This increasing level of integration is being accomplished in large part by decreasing the minimum feature sizes. Furthermore, advanced processes are being used which are much more tolerant to minute inaccuracies during processing. However, these processes often make extraordinary demands upon the chemistry of the etching process. Oxide etching has presented some of the most difficult demands. Oxide etching refers to the etching of layers of silicon dioxide, silica glass such as BPSG, and related oxide materials that serve as electrical insulators. Advanced integrated circuits contain multiple wiring layers separated from the silicon substrate and from each other by respective oxide layers. Small contact or via holes need to be etched through each of the oxide layers.

An example of an advanced oxide etching process is a self-aligned contact (SAC) process. An example of a SAC structure for two MOS transistors is illustrated in the cross-sectional view of FIG. 1. Two polysilicon lines 10, 12 are deposited and defined over a silicon substrate 14. Each polysilicon line 10, 12 forms a gate structure intended to operate as a gate electrode for a respective MOS transistor. The polysilicon lines 10, 12 act as a mask for the ion implantation of a p-type or n-type dopant into a source region 16 for both of the MOS transistors. Unillustrated drain regions are also formed to complete the principal portions of the MOS transistors. An LPCVD process is used to coat a thin conformal layer 18 of silicon nitride (Si.sub.3 N.sub.4) on the exposed silicon substrate 14 and on the polysilicon lines 10, 12. A narrow portion 20 of the silicon nitride layer 18 is formed over the silicon substrate 14 in a gap 22 between the nitride-covered gate structures 10, 12. This narrow portion 20 is removed by a post nitride etch following the oxide etch to expose the underlying silicon 14 for contacting. The gap 22 is made as small as possible consistent with subsequent processing in order to increase the integration level.

An oxide layer 24 is then deposited, usually by plasma-enhanced chemical vapor deposition (PECVD), to act as an interlevel dielectric. Considerations of dielectric breakdown with normal operating voltages limit the minimum thickness of the oxide layer 24 to between 0.5 .mu.m and 1 .mu.m. A photolithographic step including depositing and photographically defining a photoresist mask 25 followed by an oxide etch forms a contact hole 26 extending to the narrow silicon nitride portion 20 above the silicon source region 16. Following the post nitride etch to remove the narrow silicon nitride portion 20, the contact hole 26 is filled with aluminum or other conductor to form a plug electrically connecting the source region 16 of the two MOS transistors to the wiring level above the dielectric layer 24. The silicon nitride layer 18 acts as an electrical insulator relative to the aluminum plug to isolate the plug from the polysilicon lines 10, 12.

The SAC oxide etch process must satisfy several difficult requirements. The contact hole 26 should be as narrow as possible to increase the integration level, but the oxide thickness is relatively fixed at a significantly larger length. As a result, the contact hole 26 has a high aspect ratio of depth to width. A high aspect ratio can be accomplished only with a highly anisotropic etch, with the wall slope being greater than, for example, 85.degree. and preferably close to 90.degree..

In view of the large number of structures on a wafer and the variations in oxide thickness, it is highly desirable that the oxide etch be highly selective to silicon nitride, that is, that the etch process etch oxide 24 much more quickly than the underlying silicon nitride 18. The contact hole 26 can then be over etched, for example by 100% of the design depth, to accommodate non-uniformities or process variations, thus assuring that the contact hole reaches the bottom nitride portion 20 over the silicon source region 16. But if the etching manifests high selectivity, there is little etching of the silicon nitride so the source region 16 can be made relatively thin.

If the gap 22 is made very small, various considerations may limit the width of the contact hole 26 to be greater than the size of the gap 22. Also, there may be some uncontrolled variations in the position of the contact hole 26. With a nitride selective etch, the contact hole 26 can overlap the polysilicon lines 10, 12, and small variations of the location of the contact hole 26 can be accommodated while contact to the silicon is still assured. A SAC etch is usually also selective to silicon.

As illustrated, the width of the contact hole 26 is about the same as that of the gap 22 between the nitride-covered polysilicon lines 10, 12, but the photolithographic variations cause the contact hole 26 to be offset from the gap 22 and to expose a corner 27 of the nitride layer. Alternatively, the width of the contact hole 26 may be made significantly larger than the width of the gap 22 so that two nitride corners 27 are exposed. Since the nitride corners 27 are exposed the longest to the oxide etch and the acute corner geometry favors etching, nitride corner loss is often the most critical selectivity issue in contact or via etching. The etch process is subject to other constraints, such as the selectivity to the photoresist 25, which is prone to form facets 28. If the facets 28 extend to the underlying oxide 24, the resolution of the photolithography is degraded. However, nitride comer loss is generally considered to be the most demanding selectivity requirement in a SAC process.

Another difficult oxide etch technique not necessarily involving nitrides is a bi-level contact. A single etch is used to simultaneously etch through an upper oxide inter-level layer to a thin polysilicon line underlying the upper oxide layer and also etch through both the upper and a lower oxide inter-level layer to another polysilicon line underlying the lower oxide layer. This technique requires very high selectivity to silicon to avoid etching through the upper polysilicon line while the lower oxide layer is being etched through.

It is now known that reasonably good oxide etch processes can be achieved by using a fluorocarbon or hydrofluorocarbon etching gas, such as the respective types CF.sub.4 or CHF.sub.3 or higher-order compounds of the two types. These two types of etchants may be referred to as a hydrogen-free fluorocarbon and a hydrofluorocarbon although the common terminology includes both hydrogen-free fluorocarbons and hydrofluorocarbons as fluorocarbons. Fluorocarbons formed in linear chains are referred to as fluoroalkanes, using standard organic chemistry nomenclature. Under the proper conditions, the fluorocarbon forms a low-fluorine polymer on the silicon and the nitride but not on the oxide. Thereby, the oxide is etched but the silicon and nitride are not. However, if the fluorine content of the etching gas is too high, the fluorine can etch the underlying silicon or nitride and the selectivity is lost. It is believed that CF.sub.x radicals selectively etch oxide over silicon or nitride, but F radicals etch silicon and nitride as well.

The polymer introduces a further problem of etch stop. In narrow deep holes being etched, that is, holes of high aspect ratio, excess polymer is formed over the oxide walls and floor, and etching may stop at a point before the nitride floor is reached. After etch stop, further exposure to the etching plasma does not deepen the hole. Etch stop can also occur in narrow, deep trenches. The critical dimension for a trench is its width, while for a circular hole it is its diameter. Etch stop depends not only upon the aspect ratio of the hole but also upon its critical dimension. As the critical dimension is being pushed to below 0.5 .mu.m and as aspect ratios are being pushed well above two, etch stop has become a serious problem.

Thus, it is seen that advanced oxide etching must satisfy numerous difficult requirements. Further, in commercial applications, the oxide etch rate must be moderately high for reasons of economy in the use of expensive equipment and clean-room floor space. Several approaches have been suggested for addressing these problems. The success of these approaches must be evaluated in terms of a yet further problem of uniformity or reproducibility. Often a process can be finely optimized to satisfy the various requirements of selectivity, etch rate, etc. However, the optimized etching parameters may produce the required results only for a narrow range of the parameters; that is, the process window is narrow. A narrow process window may mean that random fluctuations or uncontrollable drift may bring the process out of specification from one wafer to another, and especially from day to day. Examples of such variations are reaching thermal steady state after the start of a run or after shut down for maintenance, aging of a chamber part, and replacement of a seasoned chamber part with a new part.

Also, in view of the large number of steps involved in advanced circuitry, etching rates need to be uniform to a few percent over the wafer. If the process window is too narrow, the conditions at the center of the wafer may differ sufficiently from those at the wafer edge to produce significantly different results. Hence, a wide process window is required for a commercially feasible process.

In the past, the most typical etchant gases have been CF.sub.4 and CHF.sub.3. The hydrofluorocarbon CHF.sub.3 has been generally assigned a role of promoting polymerization. Higher-order fluorocarbons have been suggested, and some have been used commercially. Yanagida in U.S. Pat. No. 5,338,399 suggests the use of cyclic fluorocarbons, especially saturated hydrogen-free fluorocarbons such as c-C.sub.3 F.sub.6 and c-C.sub.4 F.sub.8 which have a monomeric composition of --CF.sub.2 --, that is, a C/F ratio of 1/2. Yanagida's results were obtained in a magnetically enhanced capacitively coupled reactor operating at a relatively high pressure of 266 milliTorr, and he observed enhanced selectivity at low temperatures. Yanagida has also disclosed the use of hexafluoropropylene in Japanese Laid-Open Patent Applications 61-133630, 3-276626, and 4-346428.

Tahara et al. in U.S. Pat. No. 5,356,515 suggest the additional use of carbon monoxide (CO) in a similar reactor in combination with CF.sub.4 or CHF.sub.3 in order to achieve high selectivity. They ascribe the effectiveness of CO to its reducing the CF.sub.2 radicals which would otherwise form a polymer on the SiO.sub.2 walls of the hole being etched.

Arleo et al. in U.S. Pat. No. 5,176,790 disclose a low-density plasma process for etching vias through an oxide layer to an underlying metal layer. Among several embodiments is a process utilizing high-order linear fluorocarbons and hydrofluorocarbons C.sub.x H.sub.y F.sub.z, wherein x is 3 to 6, y is 0 to 3, and z is 2x-y+2. He suggests some examples including C.sub.3 F.sub.8, C.sub.3 HF.sub.7, C.sub.3 H.sub.2 F.sub.6, and C.sub.3 H.sub.3 F.sub.5. Arleo et al. favor pressures in the range of 10 to 20 milliTorr although a wider pressure range is suggested.

Marks et al. in U.S. patent application, Ser. No. 07/826,310, filed Jan. 24, 1992 disclose a plasma process for etching contact holes through an oxide layer to an underlying silicon layer. One of the embodiments of the process uses an etching gas of SiF.sub.4 and one of the high-order hydrofluoroalkanes mentioned by Arleo et al. Marks et al. favor operating at a lower pressure of between 1 and 30 milliTorr, although pressures up to 200 milliTorr are included for some applications. Marks et al. in U.S. Pat. No. 5,423,945 disclose a process for selective etching of oxide over nitride using C.sub.2 F.sub.6 at a pressure of 2 to 30 milliTorr, although CF.sub.4 and C.sub.3 F.sub.8 are mentioned as also preferred. This etching was performed in an inductively coupled high-density plasma reactor with a silicon top electrode. No argon is mentioned.

Recently, high-density plasma (HDP) oxide etchers have become available. Although several methods are available for generating a high-density plasma, the commercially most important configuration uses an RF coil to inductively couple energy into the source region to principally generate the plasma while the pedestal supporting the wafer is a cathode electrode to which RF energy is also applied and which thus biases the wafer relative to the plasma. An example of an HDP oxide etch reactor is the Centura HDP Oxide Etcher available from Applied Materials, Inc. of Santa Clara, Calif. and described by Rice et al. in U.S. Pat. No. 5,477,975. Although HDP etchers offer substantially higher etch rates because of the high plasma density, HDP etchers offer at least two other important, more fundamental advantages.

A first advantage of inductively coupling RF power to generate the plasma is that the bias power applied to the pedestal electrode supporting the wafer can be varied independently of the source power. The bias power determines the sheath voltage adjacent to the wafer and thus determines the energy of ions extracted from the plasma to the wafer. We now know that the typical type of oxide plasma etching, called reactive ion etching, requires both fluorine radicals generated by the plasma and energetic ions impinging the wafer and activating the etching chemistry. The energetic ions may be argon ions used as a carrier gas for the fluorocarbon etching gas. However, if the etch is to stop on silicon, the argon ion energy needs to be reduced to prevent damage to the silicon. A small bias power accomplishes this even though a large source power is generating an intense plasma.

A second advantage of a high-density plasma is that a substantial fraction of the atoms in the plasma are ionized, an ionization density of at least 10.sup.11 /cm.sup.3 being one definition of a high-density plasma. The ionization density tends to be strongly peaked in a source region near the inductive coils and remote from the wafer, and it tends to fall precipitously toward the wafer. Ionized radicals in the plasma are then accelerated across the bias-controlled plasma sheath so that they arrive at the wafer with a forward directed velocity distribution. This anisotropy in the etchant flux promotes anisotropic etching and prevents microloading effects at the bottom of high-aspect ratio holes. However, the directional advantage can be obtained only if the plasma is formed in a low pressure gas. If the source region is separated from the wafer by a distance significantly longer than a mean free path of the ionized radicals, which varies inversely with pressure, then the radicals are likely to collide at least once on their way to the wafer, both randomizing the flux and reducing the likelihood of the radical remaining ionized.

For silicon selectivity, Rice et al. rely upon an etching gas composition of C.sub.2 F.sub.6. Later developments with the same chamber have utilized C.sub.3 F.sub.8 and C.sub.4 F.sub.8, which have the higher C/F ratios desired by Yanagida. Rice et al. also rely upon a hot silicon surface for scavenging fluorine from the plasma. The reduction of fluorine results in high selectivity to silicon. The temperature of the silicon-based scavenger needs to be elevated to above about 200.degree. C. to activate the scavenging, and the temperature of other portions of the chamber need to be controlled relative to the scavenger temperature. Marks et al. describe similar silicon-based scavenging to achieve selectivity to nitride in the aforementioned U.S. Patent application.

Another inductively coupled HDP oxide etch chamber is the IPS Oxide Etcher, also available from Applied Materials and described by Collins et al. in U.S. patent application, Ser. No. 08/733,544, filed Oct. 21, 1996. As shown in FIG. 2, a wafer 30 to be processed is supported on a cathode pedestal 32 supplied with RF power from a first RF power supply 34. A silicon ring 36 surrounds the pedestal 32 and is controllably heated by an array of heater lamps 38. A grounded silicon wall 40 surrounds the plasma processing area. A silicon roof 42 overlies plasma processing area, and lamps 44 and water cooling channels 46 control its temperature. The temperature-controlled silicon ring 36 and silicon roof 42 may be used to scavenge fluorine from the fluorocarbon plasma. Processing gas is supplied from one or more bottom gas feeds 44 through a bank of mass flow controllers 46 under the control of a system controller 48. The process recipe is stored in the system controller 48 in magnetic or semiconductor memory. An unillustrated vacuum pumping system connected to a pumping channel 52 around the lower portion of the chamber maintains the chamber at a preselected pressure.

In the used configuration, the silicon roof 42 is grounded, but its semiconductor resistivity and thickness are chosen to pass generally axial RF magnetic fields produced by an inner inductive coil stack 56 and an outer inductive coil stack 58 powered by respective RF power supplies 60, 62.

An example of the need for wide process windows involves the mass flow controllers 46, which typically have a resolution of about 1 sccm (standard cubic centimeter per minute). Some of the prior art recipes use C.sub.4 F.sub.8 in an HDP etch reactor to achieve high selectivity. However, the process window for these processes have been observed to be in the neighborhood of .+-.1 sccm. Obviously, such a process is difficult to control in a commercial environment. In more general terms, the low resolution of the mass flow controllers argues against any process using only a few sccm of any gas.

It is desired to provide a process for oxide etching that has a wide process window. It is also desired that the process provide other favorable characteristics, such as high selectivity to nitride, a high etch rate, no etch stop, and good verticality. It is further desired that the process be usable in a high-density plasma reactor.

The invention may be summarized as an oxide etch process using an active etching gas including either hexafluoropropane and closely related hydrofluoropropanes, preferably in combination with a higher flow of an inactive gas such as argon. These gases promote polymer formation and etching selectivity over nitride, silicon, and other non-oxide materials while also not being prone to etch stop in narrow holes. Also preferably, the etching is performed in a high-density plasma reactor.

Combinations of these gases with other hydrofluorocarbons can be used to achieve optimized performance, for example, balancing selectivity, no etch stop, and a wide process window for various design features with either high or low aspect ratios.

FIG. 1 is a cross-sectional view of a self-aligned contact (SAC) structure in a silicon integrated circuit.

FIG. 2 is a cross-sectional view of a high-density plasma reactor.

FIG. 3 is a sectioned orthographic view of a self-aligned local-interconnection structure.

FIG. 4 is a cross-sectional view of the structure of FIG. 3 taken along section line 4--4.

FIG. 5 is a sectioned orthographic view of a dual-damascene structure.

This invention relies upon the belief that the gas composition and polymer formation can be chosen and controlled for a wide process window and optimized selectivity and characteristics window in oxide etching, particularly at lower pressures (<100 milliTorr) in a high-density plasma. The control according to the invention is achieved by the use of one or more of hydrofluoropropanes, particularly C.sub.3 H.sub.2 F.sub.6 and closely related hydrofluorocarbons.

As observed above, fluorocarbon radicals CF.sub.x promote selective etching of silica, while fluorine radicals F are less selective in etching not only silica but also silicon nitride and silicon. Therefore, for high selectivity to nitride or silicon, the amount of fluorine radical should be minimized. Fluorocarbon or hydrofluorocarbon gases with F/C equal to about 2 are preferred. Gases with high F content (e.g. CF.sub.4, C.sub.2 F.sub.6 etc.) require other mechanism to effectively scavenge the excessive F without causing other side effects. Hydrogen containing gases are often used to scavenge F. However, hydrogen may cause etching profile taper before enough H is introduced. Silicon material can also be used to scavenge fluorine radicals (Rice et al.), however, at least one silicon surface is often kept at high temperature to actively react with F, forming SiF.sub.4 gas, which is promptly pumped out of the chamber.

Furthermore, the fluorocarbon or hydrofluorocarbon gas in a plasma has to polymerize enough to deposit on and protect non-oxide surfaces such as silicon nitride and silicon, and yet it should not form too much polymer such as to cause etch stop in narrow holes in the oxide being etched. These countervailing requirements dictate precise control on polymer formation in the plasma reactor. For gases with small monomers of fluorocarbon, such as CF.sub.4, C.sub.2 F.sub.6 or CHF.sub.3, high flow and pressure are needed for the polymerization. And these gases generally have high fluorine content, which can destroy selectivity easily if polymer protection is not strong or not optimized. On the other hand, gases with large monomers, such as C.sub.4 F.sub.8, are very easy to polymerize, perhaps too easy. The control of polymerization rate using C.sub.4 F.sub.8 as the polymer former often challenges the limit of hardware controllability. As a result, the selectivity is obtained only in a narrow window of process parameters.

We believe that various fluoropropanes compounds take the middle ground. The 3-carbon fluorinated carbon or hydrocarbon compounds having the general formula C.sub.3 H.sub.x F.sub.y, where x is 0 to 3, y is 5 to 8. This class includes hexafluoropropylene (C.sub.3 F.sub.6). All these materials contain either carbon and fluorine; or carbon, fluorine and hydrogen. Examples are C.sub.3 F.sub.8, C.sub.3 HF.sub.7, C.sub.3 H.sub.2 F.sub.6, C.sub.3 H.sub.3 F.sub.5, C.sub.3 F.sub.6, etc. Especially preferred are the hydrofluoropropanes C.sub.3 H.sub.x F.sub.8-x, especially where x is 1 to 3. These provide additional hydrogen for a low-fluorine polymer. A popular fluoropropane etching gas is octafluoropropane (C.sub.3 F.sub.8). Previous development with it has been performed at higher pressures in low-density plasmas, and the reported selectivity is low.

The work defining the invention is divided into two parts. In an initial part, preliminary tests were performed for a varieties of related chemistries, showing the viability of fluoropropane and fluoropropylene chemistries. These results are summarized in TABLE 1. Then, more extensive tests were performed to more tightly establish the process recipe. This second part narrowed the choice of chemistry to C.sub.3 H.sub.2 F.sub.6 and the related hydrofluorocarbons C.sub.3 HF.sub.7 and C.sub.3 H.sub.3 F.sub.5 for the critical final etch.

Various preliminary process examples related to the invention were tested using structures representative of the SAC structure of FIG. 1. These results are reported in the parent patent application, incorporated herein by reference in its entirety, but those results are mostly repeated here. Two different structures are used. In one type, used in Examples 1-4, 6, 9, 10, silicon-nitride covered islands are formed over a substrate layer of silicon dioxide to allow for a measurement of over etching of the oxide. In these structures, the contact holes have a width of 0.6 .mu.m width, which is larger than the gap between the nitride islands. The depth of the oxide to the bottom of the gap is 1.7 .mu.m producing an aspect ratio of 2.8. These latter structures can still test the SAC process requirements since etching of the upper comer of the silicon nitride layer is the most severe test of selectivity to nitride. The other type of structure more closely resembles the SAC structure of FIG. 1 with silicon nitride at the bottom of the hole. The contact hole has a width at its top of between 0.35 and 0.4 .mu.m, and has one side in the gap between the silicon nitride islands and the other side on top of an island. The depth of the contact hole to the bottom of the gap is 1 .mu.m producing an aspect ratio of at least 2.5. For these structures, a separate corresponding patterned oxide wafer was used to calculate the etch rate and selectivity.

In all the examples, except where specifically noted, the silicon roof was held at 200.degree. C., and the silicon ring was held at 270.degree. C. The pedestal was chilled to 10 to 14.degree. C., and a helium thermal contact gas was held at 5 to 7 Torr between the wafer and the pedestal. Etching times were in the range of 85 to 280 seconds. The process conditions for these experiments are summarized in TABLE 1. Only the total power to the two ceiling coils is given since the total power is believed to more directly affect the plasma chemistry. The division of power between the ceiling coils is used to increase uniformity.

One compound related to octafluoropropane is hexafluoropropylene (C.sub.3 F.sub.6). This compound is a unsaturated analog of C.sub.3 F.sub.8 and has one double carbon bond. Yanagida has reported in the aforecited U.S. patent the cyclic analog hexafluoropropane c-C.sub.3 F.sub.6 as an oxide etching gas and has promoted the use of hexafluoropropylene in the aforecited Japanese applications.

In the first example, hexafluoropropylene was used as the fluorocarbon with about 18 times as much argon with a pressure of 40 milliTorr. An oxide covered substrate was used in this and the following three examples to test over etching. After the etching, polymer layers covered the top and side wall of the nitride, providing protection of the nitride so there was no nitride loss. Nonetheless, no etch stopping was observed.

Other fluoropropylenes such as hydrofluoropropylenes are expected to provide similar results.

Two other fluorocarbons useful in oxide etching are heptafluoropropane and hexafluoropropane (C.sub.3 HF.sub.7 and C.sub.3 H.sub.2 F.sub.6). Each of these can be decomposed into three monomers --CF.sub.2 --with the heptafluoropropane additionally producing HF and the hexafluoropropane additionally producing H.sub.2. Both HF and H.sub.2 are volatile, and H.sub.2 has been used as an additive for selective fluorocarbon etching of oxide.

In the second example, heptafluoropropane was used as the fluorocarbon with about 22 times as much argon. The isomer CHF.sub.2 --CF.sub.2 --CF.sub.3 was used, but the choice of isomer is not believed to be important. The pressure remained the same. There was no visible nitride loss.

The third example used the same etching gas as did the second example. However, in the third example as compared to the second example, the fluorocarbon flow was increased, the pressure was increased, and the source power was reduced. Each of these changes would tend in the same direction to increase etch stop. Nonetheless, the results were nearly indistinguishable between the two examples with no etch stop and no observable nitride corner loss.

The fourth example also used heptafluoropropane with conditions somewhat intermediate those of the third and fourth examples. There was no etch stop, and only minimal etching of the nitride comer occurred.

The above three examples demonstrate that changes in the process parameters do not greatly affect the selectivity, and etch stopping does not occur under any of these conditions. That is, in the demonstrated process the process window is wide. The process window for the ratio of the flow rates of the argon to the fluoropropane was demonstrated to extend between about 18 and 23.

In the fifth example, hexafluoropropane (C.sub.3 H.sub.2 F.sub.6) was used as the principal etching gas. The isomer CF.sub.3 --CHF--CHF was used, but the use of other isomers is not expected to greatly affect the process. The conditions were similar to the C.sub.3 HF.sub.7 examples, but the pressure was reduced by half. Again, no comer nitride loss and no etch stop was observed. The oxide etch rate measured on a planar sample was 644 nm/min.

The sixth example was a two-step process. In the first step, C.sub.3 F.sub.8 was used as a hydrogen-free fluorocarbon for a 100 second etch. In the second step, the power was reduced, and C.sub.3 H.sub.2 F.sub.6 was used as the fluorocarbon for a 130 second etch. This sample had a thicker oxide layer. The first etch step is considered to be faster but less selective. Selectivity is supplied by the second step. This example used a silicon dioxide substrate to test the over etch. Nitride comer loss was less than 20 nm.

Some further examples were tested in which octafluoropropane (C.sub.3 F.sub.8) by itself or in combination with varying amounts of the polymer-former difluoromethane (CH.sub.2 F.sub.2) was added to the etching gas.

The seventh example used C.sub.3 F.sub.8 as the only fluorocarbon, and no CH.sub.2 F.sub.2 was used. The argon flow was 22 times that of the C.sub.3 F.sub.8. The etch rate was 950 nm/min, higher than with the hydrofluoropropanes, and the nitride corner selectivity was about 32. This example also forms the baseline for the addition of the CH.sub.2 F.sub.2.

The eighth example was similar to the seventh example, but no CH.sub.2 F.sub.2 was used, and the pressure and argon flow were somewhat higher. The net etch rate was 1085 nm/min, higher than with the hydrofluoropropanes, and the nitride corner selectivity was 27, not significantly different from the seventh example.

The seventh and eighth example show that a wide process window is available with C.sub.3 F.sub.8 under these conditions.

The first through eighth examples all show a satisfactorily selective oxide etch resistant to etch stop with fluoropropylene and various fluoropropanes. For an optimized etch process, these gases can be additively combined in experimentally determined proportions.

The ninth example was a two step process, intended to increase the effective etch rate for a thicker oxide. In the first 70 s etching step a minimal amount of CH.sub.2 F.sub.2 was used while in the second 100 s etching step more CH.sub.2 F.sub.2 than C.sub.3 F.sub.8. Polymer protection is considered more necessary in the final stages of etching, particularly when a significant over etch is performed. A silicon dioxide substrate was used in this example to enable a measurement of over etching. The over etching produced a V-shaped groove in the oxide layer, and the nitride corner loss was less than 40 nm.

The tenth example used process conditions similar to the ninth example, but the source power was somewhat lower in the second step, the C.sub.3 F.sub.8 flow was lower, and the pressure was higher. The nitride corner loss was less than 20 nm.

The ninth and tenth examples used the combination of octafluoropropane and a hydrofluoroalkane, specifically difluoromethane, CH.sub.2 F.sub.2, to produce a satisfactorily selective oxide etch. The combination of an fluoropropane or fluoropropylene with a lower-order hydrofluoroalkane provides additional process flexibility and room for optimization.

The above processes can be modified by the addition of carbon monoxide, nitrogen, or oxygen, all of which are known to enhance selectivity and increase the etch stop margin. However, in the disclosed examples no carbon monoxide is used. An insubstantial amount of carbon monoxide or other additive is defined as less than 20% of the fluorocarbon content.

The above examples show that all of the principal etching gases hexafluoropropylene, octafluoropropane, heptafluoropropane, and hexafluoropropane provide effective oxide etching, particularly in a low-pressure, high-density plasma.

The experimental pressures were all below 100 milliTorr with most of them below 50 milliTorr. Similar results are expected down to at least 10 milliTorr and possibly lower. The low pressure is particularly effective in a high-density plasma with bias power within a factor of two or four of the source power in achieving a highly anisotropic etch without etch stop in a high-aspect ratio hole.

All of the examples used a high argon content of at least 10 times that of the fluoropropane and, in some cases, of the combined fluorocarbon and lower-order hydrofluoroalkane. The high argon flow is believed to promote the anisotropic etch, particularly in a high-density plasma and with strong bias power and to prevent etch stop. Similar results are expected for a ratio of inert gas to active, fluorocarbon gas down to about 2 or 3. A ratio of more than 40 is generally impractical because of limited vacuum pumping speed in the face of a required concentration of the active etching gas.

Other lower-order hydrofluoroalkanes than difluoromethane are believed to provide similar control of the process, enabling a wider process window.

Although all the examples used a relatively high-temperature scavenger, good effects are expected for a low-temperature process. Nonetheless, the scavenger is expected to be beneficial in many circumstances. Alternatively, the fluorine can be scavenged by RF biasing a silicon counter electrode to thereby activate the silicon or by injecting silane or similar Si-containing gases into the chamber.

As stated above, after completion of the initial experiments establishing the results for the recipes of TABLE 1, the search intensified for more precise recipes for use in the IPS chamber of FIG. 2. Any process involves a large number of variables, many of which can be traded off against each other in arriving at an optimized process. We selected several points within the capability of the IPS chamber to determine general trends, particularly for selectivity and etch stop with a reasonably high oxide etch rate being required. The considerations included in arriving at the tentative recipes of TABLE 1 were also pursued. After the general trends for variations of a single variable were established, process results were determined by varying one process parameter after another in a zig-zag path across the map of process parameters. We determined that a two-step process produced the best commercial result. The first step is based on C.sub.3 F.sub.6, the second step on C.sub.3 H.sub.2 F.sub.6. The first step is intended to quickly etch deep into the oxide with a vertical profile but without exposing the nitride. As a result, no selectivity to nitride is required. The second step exposes the nitride and requires a high selectivity to nitride. For the structures tested, we found that the C.sub.3 H.sub.2 F.sub.6 chemistry produced a wider process window than either C.sub.3 F.sub.6 or C.sub.3 F.sub.8 chemistry. Although it is possible that these other chemistries can be improved, initial indications are that C.sub.3 F.sub.6 produces a workable acceptable process flow window of .+-.1 sccm while the flow window for C.sub.3 F.sub.8 seems to be narrower. By comparison, the tested examples presented below show a process flow window for C.sub.3 H.sub.2 F.sub.6 of about .+-.5 sccm or greater.

We established a baseline process for one structure as summarized in the process parameters of TABLE 2. The structure closely resembles the SAC structure of FIG. 1. The contact hole has a width of about 0.4 .mu.m. The contact hole is positioned over one of the shoulders of a nitride-covered gate structure with the result that the bottom of the contact hole, in the area of the gap, had a minimum dimension of about 0.15 .mu.m or less. The oxide thickness above the nitride-covered gate structure was about 0.6 .mu.m. The aspect ratio of the hole next to the nitride shoulder can be well over 3 depending upon the registration of the photolithography.

The baseline process produces an oxide etch rate on this test structure of about 0.9 .mu.m per minute. There is no significant nitride etching and no observable etch stop. The selectivity to nitride is estimated to be more than 30:1 and certainly more than 20:1. These results apply to both the center and the edge of the wafer.

The critical second step of the baseline process has several characteristics. It uses relatively more bias power than source power. The amount of argon is substantially more than the hydrofluoropropane, by at least a factor of two. The pressure is low, much lower than the anticipated maximum of 50 milliTorr, somewhat lower than 15 milliTorr, and even lower than 10 milliTorr. The lower limit of the pressure should be in the range of 1 to 4 milliTorr.

The described process remains relatively simple in providing a commercially viable process with a wide process window for highly demanding structures.

The process window was then evaluated. The recipe of TABLE 2 was followed except that the C.sub.3 H.sub.2 F.sub.6 flow was adjusted to fall at either 24, 32, or 34 sccm. The results are all positive. Nitride selectivity remains very good, above 20:1, and there is no etch stop. The C.sub.3 H.sub.2 F.sub.6 etch time of 40 s is intended to produce a 30% over etch. In another process window test, the C.sub.3 H.sub.2 F.sub.6 etch time was extended to a 60% over etch, that is, an etch time of 60 s. No changes are observed. That is, the selectivity to nitride is sufficiently high to allow a substantial over etch.

Rather than separately establishing a process window for each of the important process variables, a seven variable two-level DOE (design of experiment) orthogonal matrix was designed to test the process window when all seven important variable change randomly. Out of these eight test conditions, the most and least selective process conditions are tabulated in TABLE 3 for the second step. The process conditions of the first step remain essentially the same.

An inspection of the resulting two SEMS showed that the least selective conditions result in some nitride removal, but limited to an acceptable amount where the nitride selectivity is at least 20:1. No etch stop is observed for the most selective conditions.

Another series of tests were performed using a structure having a higher and narrower gap in the nitride and a thicker oxide. The contact hole had a width of about 0.6 .mu.m, the oxide height above the nitride-covered gate structures was about 1.3 .mu.m, the nitride shoulder was about 0.5 .mu.m high, and the gap between the nitride was about 0.2 .mu.m. The effective aspect ratio was thus between 3 and 9. These test structures were fabricated upon a silicon dioxide test structure to test over etching. As a result, the total oxide etch depth was about 2.3 .mu.m instead of the 1.8 .mu.m that would have resulted from stopping on underlying nitride or silicon. Accordingly, both baseline etch times were extended to 80s, and the baseline flow of C.sub.3 H.sub.2 F.sub.6 was increased to 34 sccm. The baseline recipe is tabulated in TABLE 4.

The baseline etch rate is about 0.9 .mu.m/min, no nitride corner loss was observed, and no etch stop was observed even with 0.5 .mu.m over etch.

About this baseline, several process window tests were made. Typically, one important process parameter was varied in two directions from the baseline value while the other process parameters were maintained at the baseline. The varied values are shown in the last two columns of TABLE 4. For all the following variations, no significant change was observed for the variation of a single process parameter. Both inner and outer source powers were together proportionately varied .+-.10%, that is, total source power of 1080 W and 1320 W. The C.sub.3 H.sub.2 F.sub.6 flow was varied -6 sccm and +3 sccm, that is, to flows of 24 and 34 sccm. These were variations of -20% and +10%. The chamber pressure was varied .+-.1 milliTorr, that is, .+-.121/2%. The windows for hydrofluoropropane flow, for source power, and for pressure were substantially above a minimum design window of .+-.5%. The roof temperature was varied -20.degree. C., that is, to 180.degree. C. The chiller temperature was varied to -20.degree. C. The temperature of the silicon ring was varied to 320.degree. C.

Another set of samples were tested having the general structure illustrated in the sectioned orthographic view of FIG. 3 and the cross-sectional view of FIG. 4 taken along section line 4--4 of FIG. 3. This structure is intended for a self-aligned local interconnect (SALI). In selected areas over a substrate 70 are formed lower parts of multiple gate structures 71, as shown only in FIG. 4. A thin layer 72 of silicon nitride is then deposited which forms a conformal coating to the gate structures 71. An oxide layer 74 having a thickness, for example, of 1 .mu.m is deposited over the nitride layer 72 and is usually planarized by one method or another. Following the deposition of a photoresist layer and its definition into the required mask shape, an oxide etch is performed to create a T-shaped trench 75 and circular or square via or contact holes 76, all extending through the oxide layer 74. The nitride-covered gate structures 71 are formed both at the ends of the arms of the T-shaped trench 75, including the arm sectioned in FIG. 3, and perhaps at other points along the trench 75. On the other hand, the contact holes 76 usually extend all the way down to the nitride layer 72. This structure requires that the oxide etch be highly selective to nitride, particularly at the corners of the conformal nitride layer 72, but also at the planar bottom 77 of the T-shaped trench 75 as well as the bottoms of the contact holes 76. But also, etch stop must be avoided, especially in the contact holes 76.

This geometry presents a severe problem in finding an oxide etch with such characteristics. Etching chemistries can often be optimized based upon the aspect ratio of the hole being etched. For circular and square holes, the aspect ratio is simply the ratio of the depth to the width of the hole. On the other hand, the width of a linear trench is substantially smaller than its linear length. Nonetheless, an effective aspect ratio can be posited which is somewhat smaller than the one calculated from the width since the side effects provide more ready access to the bottom of the trench. However, in the T-shaped trench 75, the effective aspect ratio varies from a relatively high value along the middle portion of the arms of the T to a smaller value at the intersection of the arms. Furthermore, the aspect ratio varies significantly between the contact holes 76 and the T-shaped trench 75. Accordingly, it is difficult to optimize a process for such different geometries if the process has only a narrow process window.

Despite these difficulties, a baseline recipe was established for the SALI structure that conforms closely to that of TABLE 4 except that the C.sub.3 F.sub.6 flow is increased to 26 sccm, the C.sub.3 H.sub.2 F.sub.6 flow is decreased to 32sccm, and the two etching times were decreased to 25 s and 35 s. The SEMs show good selectivity to nitride in all portions of the T-shaped trench and in the contact hole. No etch stop is observed in any part of the trench of in the contact holes.

Yet another set of samples were tested in which the silicon nitride was replaced by silicon oxynitride. Silicon dioxide has a lower dielectric constant and optical refractive index than does silicon nitride. The oxynitride has a lower oxygen content than silicon dioxide and a lower nitrogen content than silicon nitride. Substituting some oxygen for nitrogen to produce oxynitride allows a controlled variation of these quantities. However, the oxynitride composition presents a difficult situation for selectively etching silicon oxide over silicon oxynitride since the oxynitride more closely than nitride resembles oxide. The tested structure also has a challenging geometry of a gap between the nitride islands of about 0.2 .mu.m, and the contact hole above the gap is only slightly larger at 0.22 .mu.m. The total oxide depth is 0.8 .mu.m so that the aspect ratio is at least 3.6. Nonetheless, the chemistry of the present invention was successfully applied to this structure. A baseline recipe was established that is close to that of TABLE 4 except that the C.sub.3 F.sub.6 flow is increased to 26 sccm, the C.sub.3 H.sub.2 F.sub.6 flow is increased to 36 sccm, and the two etching times are decreased to 25 s and 40 s. The selectivity to oxynitride is observed to be reduced from the other examples, but it is still acceptable. No etch stop is observed.

A dual-damascene structure, illustrated in sectioned orthographic view in FIG. 5, is being considered for advanced chips, particularly for use with copper metallization, which is very difficult to etch by a dry plasma process. Over a substrate 80 is formed a lower thin nitride layer 82, a first-level oxide layer 84, having a thickness of, for example, 1 .mu.m and an upper thin nitride layer 86, having a thickness of, for example, 0.15 .mu.m. A photolithographic step is then performed to create circular holes 88 in the upper nitride layer 86 having a diameter of, for example 0.8 .mu.m. The etching in this photolithography is selective to oxide so that at this point the holes 88 do not extend into the first-level oxide layer 84. Then a second-level oxide layer 90 is deposited over the partially etched structure to a thickness of, for example, 1.4 .mu.m. A photoresist mask is then deposited and defined into the shape of a trench 92 having a width of, for example, 1.2 .mu.m. A dual-damascene oxide etch is then performed to both form the trench 92 and to extend the lower via holes 88 down to the lower nitride layer 82. This etch must stop at the upper nitride layer 86 at the floor 94 of the trench 92 and at the lower nitride layer 82 at the bottom 96 of the via holes 88. The selectivity to nitride in both the relatively open trench floor 94 and at the corners 98 of the contact holes 88 is particularly crucial since these areas are exposed to the etching plasma while the via holes 88 are being etched. Of course, etch stop in the narrow via holes 88 must be avoided.

Following the etching of the dual-damascene structure, a single metallization operation fills both the via holes 88 and the trench 92. The metallization, particularly of copper material, can be performed to overlie the top 100 of the second-level oxide layer 90. Chemical mechanical polishing is then performed which stops on oxide. Thereby, the metallization is restricted on the top of the wafer to the trench 92. No etching of the copper metallization is thereby required.

The dual-damascene structure requires selectivity to nitride in two widely differing geometries, very high selectivity for the upper nitride layer 86, and no etch stop in a very deep structure. Nonetheless, the chemistry of the invention has been successfully applied to this structure. A baseline recipe was developed as tabulated in TABLE 5, which is relatively close to that of TABLE 2.

Selectivity throughout the structure is adequate, and no etch stop is observed.

Variations of the dual-damascene structure include other non-oxide materials in place of one or the other or both of the nitride layers 82, 86, for example, silicon, polysilicon, silicide, metal, etc. Nonetheless, the selectivity provided by the chemistry of the invention is believed to be applicable to these non-oxide materials as well.

Although all the above examples use C.sub.3 F.sub.6 in a first etching step, it is understood that the second etching step using C.sub.3 H.sub.2 F.sub.6 is the critical one requiring high selectivity to nitride and operating in a narrow, deep geometry favoring etch stop. Other chemistries are possible in the first step if they provide reasonable anisotropy. However, it is also possible to use C.sub.3 H.sub.2 F.sub.6 for the entire oxide etch.

The invention is not limited to the baseline recipes detailed above since different chambers and different structures may require somewhat different but closely related recipes. The examples however show the advantage of several process conditions: (1) the excess of argon over the hydrofluorocarbon, even to a factor of 3 and more; (2) a low pressure of less than 50 milliTorr, preferably less than 20 milliTorr, and most preferably less than 10 milliTorr, down to 1 milliTorr; (3) a higher bias power than source power; (4) and the use of a silicon-based scavenger. It is expected that some of these conditions can be violated in particular cases.

The preliminary results of TABLE 1 show that C.sub.3 HF.sub.7 produces etching results very similar to those of C.sub.3 H.sub.2 F.sub.6, and preliminary indications are that its process window is relatively wide. Since nearly the same chemistry applies to C.sub.3 HF.sub.7 as to C.sub.3 H.sub.2 F.sub.6, it is believed that an optimized baseline recipe can be easily established for C.sub.3 HF.sub.7 for at least some structures. The trend established by the pair of C.sub.3 H.sub.2 F.sub.6 and C.sub.3 HF.sub.7 indicate that the neighboring monomer pentafluoropropane (C.sub.3 H.sub.3 F.sub.5) should show beneficial results as well.

The preliminary results of TABLE 1 show that combinations of hydrofluorocarbons can be used in optimized proportions to achieve better overall performance. For example, CH.sub.2 F.sub.2 can be added to C.sub.3 H.sub.2 F.sub.6 or C.sub.3 HF.sub.7 to increase the amount of polymerization and hence of selectivity. Other hydrofluorocarbons or fluorocarbons such as C.sub.3 F.sub.6, C.sub.3 F.sub.8, or C.sub.4 F.sub.8 can be used for this and other processing considerations. Even the combination of C.sub.3 H.sub.2 F.sub.6 with C.sub.3 HF.sub.7 or C.sub.3 H.sub.3 F.sub.5 can be optimized for specific etching characteristics. It is believed that the addition of a hydrofluorocarbon adds hydrogen to the chemistry while a hydrogen-free fluorocarbon removes hydrogen.

As described above, it is believed that a selective oxide etch depends upon a polymer forming on the non-oxide surface to which the etch should be selective. Hence, the same chemistries are applicable to etching oxide over a number of different materials, including silicon nitride, silicon, metals such as aluminum and copper, and silicides such as TiSi.sub.x, WSi.sub.x and others. As mentioned above for TABLE 3, silicon oxynitride can be formed in proportions such that the same chemistry applies because of the differential content of oxygen.

Thus it is seen that this new chemistry is very effectively and economically applied to oxide etching of difficult structures.