Split-Polysilicon CMOS Process for Multi-Megabit Dynamic Memories Incorporating Stacked Container Capacitor Cells

This invention relates to integrated circuit manufacturing technology and, more specifically, to low-cost processes utilizing a reduced number of mask sets, for manufacturing CMOS dynamic random access memories. The instant process integrates stacked capacitor container-type cells in a "split-polysilicon" process flow in which N-channel and P-channel devices are patterned using separate masks.

The business of producing dynamic random access memory (DRAM) devices is a very competitive, high-volume business. Process efficiency and manufacturability, as well as product quality, reliability, and performance are the key factors that determine the economic success of such a venture.

Each cell within a DRAM device, an individually-addressable location for storing a single bit of digital data, is comprised of two main components: a field-effect access transistor and a capacitor. As component density in memory chips has increased, it has been necessary to at least maintain cell capacitance as cell size shrinks. Each new generation of DRAM devices generally has an integration level that is four times that of the generation which it replaced. Such a quadrupling of device number per chip is normally accompanied by a decrease in device geometries. All DRAM memories of 4-megabit and greater density utilize cell designs having three-dimensional capacitors. Although both trench and stacked capacitor designs have proven serviceable at the 4-megabit level, most manufacturers now seem to favor the stacked capacitor design for its manufacturability and somewhat higher yield.

Most current-generation dynamic random access memories (DRAMs) utilize CMOS technology. Although the term "CMOS" is an acronym for (C)omplementary (M)etal (O)xide (S)emiconductor, the term CMOS is now more loosely applied to any integrated circuit in which both N-channel and P-channel field-effect transistors are used in a complementary fashion. Although CMOS integrated circuit devices are often referred to as "semiconductor" devices, such devices are fabricated from various materials which are either electrically conductive, electrically nonconductive or electrically semiconductive. Silicon, the most commonly used semiconductor material can be made conductive by doping it (introducing an impurity into the silicon crystal structure) with either an element such as boron which has one less valence electron than silicon, or with an element such as phosphorus or arsenic which have one more valence electron than silicon. In the case of boron doping, electron "holes" become the charge carriers and the doped silicon is referred to as positive or P-type silicon. In the case of phosphorus or arsenic doping, the additional electrons become the charge carriers and the doped silicon is referred to as negative or N-type silicon. If a mixture of dopants having opposite conductivity types is used, counter doping will result, and the conductivity type of the most abundant impurity will prevail. Silicon is used either in single-crystal or polycrystalline form. Polycrystalline silicon is referred to hereinafter as "polysilicon" or simply as "poly". Although polysilicon has largely replaced metal for the MOS device gates, the relatively low conductivity of that material (even when heavily doped) has led many semiconductor manufacturers to create a layer of refractory metal silicide on transistor gates in order to decrease sheet resistance and, thereby, increase device speed. In conventional DRAM processes, two additional poly layers are used for the lower and upper cell capacitor plates.

CMOS manufacturing processes generally begin with a lightly-doped P-type or N-type silicon substrate, or lightly-doped epitaxial silicon on a heavily doped substrate. Although P-type silicon is usually chosen as the starting material, selection of N-type silicon as the starting material changes the process very little, the primary difference being that for a given step, dopant types are reversed.

Triple polysilicon layer manufacturing processes for contemporary stacked-capacitor dynamic random access memories require some fourteen to eighteen masking steps. As device geometries shrink, each photolithographic step becomes more costly. In light of the costs associated with masking operations, manufacturing processes which permit a significant reduction in the number of masking operations are greatly preferred.

In 1982, Japanese patent number 57-17164 was issued to Masahide Ogawa. This patent teaches the fabrication of a CMOS integrated circuit by processing N-channel and P-channel devices separately. As with conventional CMOS processes, a single polysilicon layer is used to form both N-channel and P-channel gates. However, N-channel devices are formed first, with unetched polysilicon left in the future P-channel regions until N-channel processing is complete. The mask used to subsequently pattern the P-channel devices is also used to blanket and protect the already-formed N-channel devices. This process is herein referred to as the split-polysilicon CMOS process. The split-polysilicon CMOS process, though largely ignored by semiconductor manufacturers in the U.S. and abroad., has been used extensively by Micron Technology, Inc. of Boise, Idaho as a means to reduce the number of masking steps and, hence, the cost of manufacturing dynamic random access memories.

U.S. Pat. No. 5,134,085 discloses a split-polysilicon CMOS DRAM process which incorporates conventional stacked capacitors. Although conventional stacked capacitors are adequate for 4-megabit and 16-megabit densities, more complex capacitors are required for the 64-megabit density level and beyond. Cylindrically-shaped cell capacitors (commonly referred to as container-type capacitors) have gained favor in 64-megabit DRAM designs because they are relatively easy to manufacture, and also because capacitance may be increased simply by increasing stack height. However, high topography which results generally requires the processing of P-channel devices early in the manufacturing process (i.e., before the container-type capacitors are fabricated). The problem with this approach is that boron trifluoride, the dopant normally used to dope P-channel source/drain regions, diffuses relatively rapidly under elevated temperature conditions. If the doping occurs early in the process, the thermal budget remaining in the process will cause the P-channel dopant to diffuse into the channel regions where it will produce short-channel effects. The most serious of those effects is device leakage when a high signal is applied to the gate.

What is needed, therefore, is a split-polysilicon CMOS DRAM process which incorporates container capacitors in an efficient process flow having a low number of masking steps, and which permits P-channel processing late in the process sequence.

This is a process for fabricating a CMOS dynamic random access memory on an upper stratum of a silicon substrate or wafer. The focus of this invention is the integration of container cell capacitors in a split-polysilicon CMOS manufacturing process flow which permits P-channel source/drain doping subsequent to capacitor formation. In a split-polysilicon process flow, N-channel device gates and P-channel device gates are patterned and etched from the same first conductive (doped polysilicon) layer at different stages of the process flow. In this split-polysilicon process, N-channel gates are patterned and etched from the first polysilicon layer, while an unetched expanse of the first polysilicon layer is left in N-well regions. A main feature of this new process is the deposition and planarization of a thick insulative mold layer subsequent to N-channel device patterning, but prior to P-channel device patterning. Following deposition of the mold layer, storage-node contact openings are etched through the mold layer in the P-type regions within the array. A second conductive layer is then deposited which blankets both N-type regions and P-type regions and lines the storage-node contact openings. Portions of said second conductive layer which are on the upper surface of the mold layer are then removed, leaving those portions of the second conductive layer which line the storage-node contact openings. The mold layer may then be optionally thinned until portions of the second conductive layer which line the storage-node contact openings extend above the surface of the mold layer. After a cell dielectric layer is then formed on the upper surface of the second conductive layer, the dielectric layer is overlaid with a third conductive layer. Portions of the second and third conductive layers which are peripheral to the cell array are then removed. P-channel gates are then patterned from the unetched expanse of said first conductive layer. A P-channel source/drain implant follows.

In one embodiment of the process, portions of the mold layer overlying the N-well region (the regions where P-channel devices will be formed) are removed during the storage-node contact etch. Thus, a low-aspect-ratio etch can be employed to pattern P-channel devices, and a blanket P+ implant may be performed without implanting the P-type impurity into source/drain regions of the N-channel devices.

FIG. 1 is a cross-sectional view of a portion of an in-process dynamic random access memory at a stage of manufacture where N-well regions have been formed in a lightly-doped P-type substrate, silicon dioxide field isolation regions have been formed, a gate dielectric layer has been formed on active areas, and a three-layer sandwich, consisting of a first polysilicon layer which is doped to render it conductive, a silicon dioxide buffer layer and a thick silicon nitride layer, has been formed by sequential deposition steps;

FIG. 2 is a cross-sectional view of the portion of the in-process memory of FIG. 1 following the patterning of the three-layer sandwich to form a series of wordlines in the N-channel region and an unetched expanse in the P-channel region (i.e., the N-well region);

FIG. 3 is a cross-sectional view of the portion of the in-process memory of FIG. 2 following the blanket deposition of a silicon nitride spacer layer;

FIG. 4 is a cross-sectional view of the portion of the in-process memory of FIG. 3 following an anisotropic etch of the silicon nitride spacer layer to form spacers on the sidewalls of the wordlines and on the sidewall of the unetched expanse of the stacked sandwich layers, the blanket deposition of a silicon nitride etch stop layer, the primary purpose of which is to cover active areas, the deposition of a boro-phospho-silicate glass (BPSG) layer, and the planarization of the BPSG layer;

FIG. 5A is a cross-sectional view of the portion of the in-process memory of FIG. 4 following the patterning and etching of the BPSG layer to form storage node contact openings and remove that portion of the BPSG layer which overlies the P-channel region;

FIG. 6A is a cross-sectional view of the portion of the in-process memory of FIG. 5A following the deposition of a second polysilicon layer and chemical-mechanical planarization of the resulting structure;

FIG. 7A is a cross-sectional view of the portion of the in-process memory of FIG. 6A following an etch back of the thick insulative layer, deposition of a cell dielectric layer, deposition of a cell plate layer, and optional deposition of silicon nitride capping layer;

FIG. 8A is a cross-sectional view of the portion of the in-process memory of FIG. 7A following removal of capacitive layers from all but the cell array region;

FIG. 9A is a cross-sectional view of the portion of the in-process memory of FIG. 8A following masking of the N-channel region, and patterning and etching of the unetched expanse of sandwich layers in the P-channel region;

FIG. 10A is a cross-sectional view of the portion of the in-process memory of FIG. 9A following a photoresist strip, deposition of an oxide spacer layer, an anisotropic etch of the oxide spacer layer, and a P-channel source/drain implant. FIG. 5A is a cross-sectional view of the portion of the in-process memory of FIG. 4 following the patterning and etching of the BPSG layer to form storage node contact openings;

FIG. 5B is a cross-sectional view of the portion of the in-process memory of FIG. 4 following the patterning and etching of the BPSG layer to form storage node contact openings;

FIG. 6B is a cross-sectional view of the portion of the in-process memory of FIG. 5B following the deposition of a second polysilicon layer and chemical-mechanical planarization of the resulting structure;

FIG. 7B is a cross-sectional view of the portion of the in-process memory of FIG. 6B following an etch back of the thick insulative layer, deposition of a cell dielectric layer, deposition of a cell plate layer, and optional deposition of silicon nitride capping layer;

FIGS. 8B and 8C are a cross-sectional view of the portion of the in-process memory of FIG. 7B following removal of capacitive layers from all but the cell array region;

FIGS. 9B and 9C are a cross-sectional view of the portion of the in-process memory of FIG. 8B following masking of the N-channel region, and patterning and etching of the unetched expanse of sandwich layers in the P-channel region; and

FIGS. 10B and 10C are a cross-sectional view of the portion of the in-process memory of FIG. 9B following a photoresist strip, deposition of an oxide spacer layer, an anisotropic etch of the oxide spacer layer, and a P-channel source/drain implant.

Referring now to the cross-sectional view of the in-process DRAM circuitry FIG. 1, N-well regions 11 have been formed in a lightly-doped P-type substrate 12, silicon dioxide field isolation regions 13 have been formed, a gate dielectric layer 14 has been formed on active areas 15, and a three-layer stack 16, consisting of a first conductive layer 17, a silicon dioxide buffer layer 18 and a thick silicon nitride layer 19, has been formed by sequential deposition steps. Both N-channel and P-channel transistor gates will be formed from the three-layer stack 16. In a preferred embodiment of the process, the first conductive layer 17 is either doped polycrystalline silicon alone, or doped or undoped polycrystalline silicon that is coated with a refractory metal silicide to achieve reduced sheet resistance.

Referring now to FIG. 2, the three-layer stack 16 of FIG. 1 has been patterned and etched to form a series of wordlines 21 in the N-channel region and an unetched three-layer expanse 22 in the P-channel region (i.e., the N-well region 11). The unetched three-layer expanse 22 is the distinguishing characteristic of a "split-polysilicon" CMOS DRAM process. At this point in the process, N-channel source/drain regions 23 are formed by implanting arsenic.

Referring now to FIG. 3, a silicon nitride spacer layer 31 has been deposited over all in-process circuitry.

Referring now to FIG. 4, the silicon nitride spacer layer 31 has been anisotropically etched to form spacers 41 on the sidewalls of the wordlines 21 and on the sidewalls of the unetched expanse 22. The anisotropic etch is terminated at the point where the spacers 41 cover the edges of all three layers of each remnant of the sandwich 16. Following the formation of spacers a silicon nitride etch-stop layer 42 is deposited over all in-process circuitry. Following the deposition of the etch-stop layer 42, a thick insulative mold layer 43 is deposited on top of the etch-stop layer 42. Boro-phospho-silicate glass is deemed to be the preferred dielectric material from which the mold layer 43 is formed. After the mold layer 43 is deposited, it is planarized. Although a number of planarization techniques are known in the art, chemical mechanical planarization is deemed to be the preferred method. FIG. 4 represents the departing point for two distinct variations of the process. Thus, the process continues with either FIG. 5A or 5B.

Referring now to FIG. 5A, which is the departing point for the second embodiment of the first pair of process variations, mold layer 43 has been patterned with a contact photoresist mask 51, and etched to form storage node contact openings 52, and to also remove those portions of the mold layer 43 that overlie the P-channel (N-well) region, thus creating a recessed region 53 in the mold layer 43 above the unetched expanse 22.

Referring now to FIG. 5B, the mold layer 43 is patterned with a contact photoresist mask 51 and etched to form only storage node contact openings 52.

Referring now to FIG. 6A, a second conductive layer 61 is blanket deposited (i.e., such that it covers all in-process circuitry). As a capacitance enhancing feature, the second conductive layer 61 is preferably doped hemispherical grain (HSG) silicon. Doping may occur in-situ during deposition, or it may occur subsequent to deposition. The polysilicon-lined N-well region and contact openings are then optionally filled with a filler material 62, such as photoresist. The resulting structure is then planarized a second time, using either a plasma etchback or chemical-mechanical planarization in order to remove horizontal portions 63 of the second polysilicon layer 61. The remaining portions of the second polysilicon layer 61 line the recess in the mold layer 43 above the N-well and the storage-node contact openings 51. The latter portions will function as the storage-node capacitor plate 64.

Referring now to FIG. 6B, a second conductive layer 61 is blanket deposited (i.e., such that it covers all in-process circuitry). As a capacitance enhancing feature, the second conductive layer 61 is preferably doped hemispherical grain (HSG) silicon. Doping may occur in-situ during deposition, or it may occur subsequent to deposition. The polysilicon-lined contact openings are then optionally filled with a filler material 62 such as photoresist. The resulting structure is then planarized a second time, using either a plasma etchback or chemical-mechanical planarization in order to remove horizontal portions 63 of the second polysilicon layer 61. The remaining portions of the second polysilicon layer 61 will function as the storage-node capacitor plates 64.

It should be noted that the first planarization step described while referring to FIG. 4 may be eliminated so that the second planarization step described while referring to both FIG. 6A and FIG. 6B may be employed both to planarize the mold layer 43 and to singulate the storage-node capacitor plates 64.

Referring now to FIG. 7A, the filler material 62 has been removed and the mold layer 43 is thinned during an etch-back step until portions of the second conductive layer 61 which line both the recesses above the unetched three layer expanse 22 and the storage-node contact openings 51 extend above the surface of the mold layer 43. A cell dielectric layer 71 is then formed on at least the exposed surfaces of the remnants of the second polysilicon layer 61. In a preferred embodiment of the process, the cell dielectric layer is a silicon dioxide-silicon nitride-silicon dioxide sandwich, and the silicon nitride layer of this sandwich is blanket deposited via chemical vapor deposition. Following the formation of a cell dielectric layer 71, a conductive cell plate layer 72 is formed on top of the cell dielectric layer 71. In a preferred embodiment of the process, the cell plate layer 72 is also a layer of doped polysilicon. A silicon nitride etch stop layer 73 is then deposited on top of the cell plate layer 72.

Referring now to FIG. 7B, the filler material 62 has been removed and the mold layer 43 is thinned during an etchback step until portions of the second polysilicon layer 61 which line the storage-node contact openings 51 extend above the surface of the mold layer 43. A cell dielectric layer 71 is then formed on at least the exposed surfaces of the remnants of the second polysilicon layer 61. In a preferred embodiment of the process, the cell dielectric layer is a silicon dioxide-silicon nitride-silicon dioxide sandwich, and the silicon nitride layer of this sandwich is blanket deposited via chemical vapor deposition. Following the formation of a cell dielectric layer 71, a conductive cell plate layer 72 is formed on top of the cell dielectric layer 71. In a preferred embodiment of the process, the cell plate layer 72 is also a layer of doped polysilicon. A silicon nitride etch stop layer 73 is then deposited on top of the cell plate layer 72. FIG. 7A is the departing point for two distinct variations of the process. Thus the process continues with either FIG. 8B or FIG. 8C.

Referring now to FIG. 8A or 8B, the capacitive layers (the second conductive layer 61, the cell dielectric layer 71, and the cell plate layer 72) are removed from all but the cell array region through the use of a series of selective etch steps.

Referring now to FIG. 8C, the capacitive layers (the second conductive layer 61, the cell dielectric layer 71, and the cell plate layer 72) are removed from all but the cell array region through the use of a series of selective etch steps, thus exposing the mold layer 43. The mold layer 43 is then thinned to a level that exposes the upper surface of the unetched three-layer expanse 22, yet does not expose the source/drain regions 81 of N-channel transistors 82 that are peripheral to the cell array.

Referring to FIG. 9A, following the masking of the entire N-channel region with photoresist, and the masking of the unetched three-layer expanse 22 to provide a mask pattern for P-channel device gates, the in-process circuitry is etched to form those P-channel device gates 91.

Referring to FIG. 9B, following the masking of the entire N-channel region with photoresist, and the masking of the mold layer in the N-well region so as to provide a mask pattern 92 for P-channel device gates, the in-process circuitry is etched to form those P-channel device gates 91. It will be noted that the primary difference between the process of FIG. 9A and 9B at this stage is that patterning of the P-channel devices in FIG. 9B will require an etch through the mold layer 43 before the three-layer expanse 22 is etchable.

Referring to FIG. 9C, following the masking of the entire N-channel region with a photoresist blanket 91, and the masking of the unetched three-layer expanse 22 to provide a photoresist mask pattern 92 for P-channel device gates, the in-process circuitry is etched to form those P-channel device gates 93.

Referring now to FIG. 10A, 10B, or 10C, following a photoresist strip step, a silicon dioxide spacer layer (not shown) is deposited and subsequently anisotropically etched to form spacers 102 on the vertical sidewalls of the P-channel gates 91. It will be noted that non-functional spacers 103 are also formed on the sidewalls of all vertical structures in the N-channel region. Following spacer formation, P-channel source/drain regions 101 are formed with a blanket boron-trifluoride implant.

Although only several embodiments of the new process have been disclosed herein, it will be obvious to those having ordinary skill in the art that modification and changes may be made thereto without departing from the spirit and the scope of the process as hereinafter claimed. For example, although the process is described in the context of an N-well process, a P-well process or a "twin-tub" may be used with equivalent results. The terms "N-type region" and "P-type region" are intended to apply to any of the three cases.