CMOS and Bipolar Fabrication Process Using Selective Epitaxial Growth Scalable to Below 0.5 Micron

1. Field of the Invention

This invention relates to semiconductor device fabrication methods and, more particularly, to a CMOS and bipolar fabrication process using selective epitaxial growth.

2. Description of the Related Art

As device geometries get smaller, well tried technologies become inadequate for isolating devices of the same or opposite conductivity type. For example, the widely used LOCOS (isoplanar) scheme requires too much silicon area for geometries below one micron, due to the bird's beak encroachment. In addition, latch-up considerations in CMOS prevent putting opposite type devices in very close proximity unless more complex processing is added.

Current state-of-the-art one to two micron level approaches to CMOS isolation include the use of trenches and epitaxial buried layers. Trench isolation has the disadvantage of requiring very complex and costly processing, and it requires some other type of oxide isolation for the majority of the chip's surface (typically LOCOS). Additionally. MOS transistors cannot be set directly against a trench wall because of degradation of device characteristics, thus increasing the area consumed by one transistor. Epitaxial buried layer isolation, while somewhat effective, still has a lower limit of approximately 2.5 micrometers for PMOS to NMOS spacing due to junction breakdown and punch-through.

Recently. CMOS isolation by selective epitaxial growth (SEG) has been proposed. In one method, not necessarily in the prior art, a silicon substrate is etched to form openings in the substrate, and insulators are formed on the side walls of the openings. Thereafter, the substrate is masked and doped to a chosen conductivity type, and an epitaxial layer is grown to fill the openings. A final LOCOS isolation then is performed. The disadvantages of this method are the requirement of etching into the silicon surface and the requirement of LOCOS isolation with the inherent bird's beak encroachment.

In another technique, also not necessarily in the prior art, a silicon dioxide layer is formed over a silicon substrate, and the silicon dioxide layer is etched for forming openings extending to the substrate. The openings are filled by growing epitaxial layers having a selected conductivity type (e.g., N-type) on the substrate. This creates doped wells in which devices of a particular type may be constructed. The wells then are covered by thin thermal oxide layers to protect them from later process steps. The process then is repeated to form wells having an opposite conductivity type (e.g., P-type). Thereafter, the thin oxide layers over the previously formed wells are stripped. Although this method does not require LOCOS isolation, it must be implemented with multiple SEG steps.

The present invention is directed to a relatively simple front-end process for isolating semiconductor devices wherein the process requires only one SEG step, provides twin buried layers which may be controlled independently, and uses a simple thermal oxidation step for isolation between devices. Active P and N regions are separately formed with self-aligned implants, and with fewer masking steps than conventional techniques. Unlike devices isolated by trenches, the isolation width can be any dimension above the resolution of the patterning tool for MOS devices of the same type, and the resolution of the patterning tool plus one registration tolerance for opposite type devices. Thus, if 0.5 micron is the resolution of the patterning tool, 0.5 micron isolation between same-type devices is achieved, and the process is scalable well into the submicron region. At submicron levels, no other presently known technique can provide a minimum device pitch of twice the resolution of the patterning tool with a planarized surface without substantially more complex processing.

For bipolar devices, the method according to the present invention minimizes collector-base capacitance through the use of oxide isolated collectors while also using a bipolar buried layer, whereas the independent P+ and N+ buried layers minimize the collector-to-substrate capacitance. Punch-through problems are eliminated because of full oxide isolation between devices. Finally, the front end of this process is compatible with the most advanced CMOS and bipolar backend device fabrication processes, such as silicided source-drain-gate areas and contacts/posts overlapping on field oxide for MOS devices, and a polyemitter or a poly-base and emitter for bipolar devices.

In one embodiment of the present invention, a silicon dioxide layer is formed over a silicon substrate. The silicon dioxide layer is etched for forming separate collector and base/emitter regions for a bipolar device, and PMOS and NMOS regions for corresponding PMOS and NMOS devices. Buried layer implants are performed, and an epitaxial layer is grown over the exposed portions of the silicon substrate. The silicon dioxide walls between the devices provide full dielectric isolation between the devices as well as between the collector and base/emitter regions of the bipolar device. Nonetheless, the oxide wall between the collector and base/emitter of the bipolar device is sufficiently small to allow the buried layer implants to join under the wall for forming a conventional buried layer for the bipolar device. Because of the oxide walls, the minimum distance between devices may be 0.5 microns or less.

FIGS. 1-5 illustrate the method according to the present invention for forming CMOS and bipolar devices using selective epitaxial growth.

FIG. 1 shows one embodiment of a silicon substrate 10 after undergoing preliminary processing according to the present invention. The substrate material is P-type 100 with 100 flat orientation and resistivity greater than 10 ohm-cm. The flat orientation is preferable to achieve the best selectively grown epitaxial layers with minimal faceting and stress. The orientation also helps to reduce fixed charges along the vertical sidewalls of the oxide isolation regions.

First, a silicon dioxide layer 14 is thermally grown to a thickness of from approximately 0.8 to 1.5 microns by placing substrate 10 in an oxygen environment at 1000.degree. C. for approximately 300 minutes, making certain that the oxidation cycle includes conventional denuded zone formation steps. This ensures that the selectively grown epitaxial layer will not have stacking faults along the horizontal surface. Thereafter, a photoresist layer 18 is deposited and developed for forming the openings shown. The portions of silicon dioxide layer 14 beneath the openings in photoresist layer 18 then are etched vertically to the surface of substrate 10 by Reactive Ion Etching (RIE) to form openings 22, 24, and 26. Openings 22 and 24, together with an oxide wall 30 therebetween, define a bipolar region, and opening 26 defines a PMOS region.

Using photoresist layer 18 as a mask, arsenic or arsenic and phosphorous ions are implanted in succession without tilting the wafers in the implanter. In this embodiment, all implants prior to epitaxial growth are performed without wafer tilt. The self-aligned implant in openings 22 and 24 is used to form a buried layer and an N-well for the bipolar device. Opening 22 is used for constructing the collector of the bipolar device, and opening 24 is used to construct the base and emitter of the bipolar device. The self-aligned implant in opening 26 is used to form a buried layer and an N-well for the PMOS device. The concentration of arsenic is approximately 5.times.10.sup.15 to 1.times.10.sup.16 atoms/cm.sup.2, and it is implanted with an energy of approximately 100 KeV. The concentration of phosphorous is approximately 5.times.10.sup.13 to 5.times.10.sup.14 atoms/cm.sup.2, and it is implanted with an energy of approximately 150 KeV. The arsenic implant ensures low resistance of the implanted regions, whereas the faster diffusing phosphorous species ensures formation of the N-wells for the PMOS and bipolar devices.

Next, photoresist layer 18 is removed, and an anneal is performed at 1000.degree. C. in an inert atmosphere, such as nitrogen, for approximately 60 minutes. This eliminates the implant damage prior to SEG. The anneal also creates side diffusion of the phosphorous and arsenic implants to ensure that the buried layers beneath openings 22 and 24 join under oxide wall 30. Preferably, oxide wall 30 has a width of 0.5 microns (e.g., definable by E-beam) or less. If oxide wall 30 is wider, a longer arsenic diffusion may be required prior to epitaxial growth to ensure that the doped regions beneath openings 22 and 24 join.

An alternative to successive arsenic and phosphorous implants is to implant the arsenic, remove photoresist layer 18, anneal and diffuse the arsenic, and then implant phosphorous, taking advantage of the masking of silicon dioxide layer 14 to selectively implant only in the exposed silicon regions.

As shown in FIG. 2, a photoresist layer 40 is deposited and developed. The exposed portions of silicon dioxide layer 14 beneath the openings in photoresist layer 40 are vertically etched to the surface of substrate 10 by RIE to form openings 41. 42, and 43. Then, boron is implanted to a dose of approximately 2.times.10.sup.13 to 2.times.10.sup.14 atoms/cm.sup.2 at an energy of approximately 120 KeV and photoresist layer 40 is removed. The implant in openings 41 and 42 is used to form a guard ring around the bipolar device. The bipolar device is separated from the guard ring by oxide walls 44 and 45. The implant in opening 43 is used to ensure a continuous P-well for an NMOS structure after the process is complete. The PMOS device is isolated from the NMOS device by an oxide wall 50, thus providing full dielectric isolation between opposite type MOS transistors. An added benefit of the boron implant is the lowering of substrate resistance R.sub.S which is beneficial in latch-up suppression.

Although the P and N regions are separately formed, only two masking steps have been required. This eliminates one masking step from conventional techniques wherein a mask is required for active region definition and for each implant. Additionally, the process according to the present invention avoids the necessity of dealing with a photoresist mask in peaks and valleys of the active regions.

The minimum distance between two adjacent NMOS devices is 0.5 microns or the resolution of the patterning tool. The distance between two adjacent PMOS devices is also 0.5 microns, since they can share the same N-well. If a PMOS device or another bipolar device is directly adjacent to the bipolar device (i.e., no guard ring). wall 45 preferably is wider by 0.4 to 0.6 microns than wall 30 to ensure isolation of the PMOS or the second bipolar device from the adjacent bipolar device. The minimum width for oxide wall 50, which forms the isolation region between the NMOS and PMOS devices, is the larger of (a) the minimum resolution of the patterning tool plus one registration tolerance, or (b) two registration tolerances. Thus, in the future, when both registration tolerances and resolution are reduced, this technology can be scaled below 0.5 microns, with the limit for isolation width then being an acceptable value of threshold voltage of the oxide isolation region.

By separating the regions of arsenic and boron implants, the bipolar collector-substrate capacitance is substantially reduced. The adjustment of this capacitance is by boron and phosphorous implant doses, the width of oxide wall 44, and total heat treatment included in the process. By allowing a wider separation between NMOS and NPN bipolar devices (assuming an NMOS device is located to the left of the bipolar device). this capacitance can be the absolute minimum. On the other hand, some designs may not require an NPN bipolar device next to an NMOS device, thus ensuring a low collector-to-substrate capacitance.

As shown in FIG. 3, undoped epitaxial silicon is selectively grown to the thickness of the remaining portions of silicon dioxide layer 14, 24. 26, 41. 42, and 43. This may be accomplished by a five minute in situ hydrogen bake at 1000.degree. C. and 25 torr, followed by SEG deposition at 850.degree.-950.degree. C. and 25 torr in a hydrogen dichlorosilane and hydrogen chloride ambient to minimize faceting. This forms a collector region 51 in opening 22, a base/emitter region 52 in opening 24, a guard ring region 53 in opening 41 and 42, a PMOS region 54 in opening 26, and an NMOS region 55 in opening 43. Although guard ring region 53 is shown as separate regions in cross section, it is actually a continuous region encircling collector region 51 and base/emitter region 52. Thereafter, a thin silicon dioxide layer 48 is grown to a thickness of approximately 200-300 angstroms to alleviate the stress at the epi-isolation oxide interface. Oxide layer 48 also is used as a sacrificial oxide for gate oxide integrity improvement.

Proceeding to FIG. 4, a photoresist layer (not shown) is deposited and developed for exposing collector region 51, base/emitter region 52, and PMOS region 54. The exposed regions then are implanted with phosphorous to a concentration of approximately 1.times.10.sup.12 to 4.times.10.sup.12 atoms/cm.sup.2 at an energy of 150 KeV for setting the N-well impurity profile at the surfaces of the P-channel devices and bipolar devices if necessary. The photoresist layer is removed, and a new photoresist layer 56 is deposited and developed to expose guard ring region 53 and NMOS region 55. The exposed regions then are implanted with boron for setting the impurity profile for the bipolar guard ring and the P-well for the N-channel devices. In some cases, photoresist layer 56 may not be necessary, and a blanket V.sub.T implant may be all that is required to set the correct P-channel and N-channel threshold voltages and to dope the bipolar guard rings.

Next, substrate 10 is placed in an inert environment at 1050.degree. C. for approximately 60 minutes so that the substrate buried layers join with their respective N-well and P-well surface implants. After the drive-in, a V.sub.T implant is performed with BF.sub.2 to a concentration of 1.times.10.sup.12 to 3.times.10.sup.12 atoms/cm.sup.2. The thin sacrificial oxide layer 48 then is removed, and a gate oxide layer 60 having a thickness of 120-150 angstroms is grown, as shown in FIG. 5. This thin oxidation is the only oxidation step that the impurities will see, and thus segregation and depletion of boron on the sidewalls of isolation regions are minimized. No birds-beak encroachment is encountered in this process; thus, defined active and isolation dimensions will be the true electrical dimensions of the device and can be 0.5 micron or even less.

After the gate oxidation step, conventional gate material deposition and definition follow. The gate material can be polysilicon, silicide, or a combination of the two. For 0.5 micron devices. P+ polysilicon for P-channels and N+ polysilicon for N-channels are desirable. This is accomplished by doping the polysilicon at the time of the source/drain implants. The two types of polysilicon can be shorted by a silicide or a metal strap. At the time of the source/drain implantation, contacts to the N-well and substrate can be implemented, although they are not shown in these figures. These contacts are presumed to be in planes perpendicular to the plane of the cross sections. Sidewall spacers on gates are formed, and source-drain-gate silicidation is performed. Posts or unguarded contacts can be used for minimizing the interconnect real estate.

A guard ring should be formed for each bipolar transistor, but the precision of the present process allows the guard ring to be accurately spaced to provide very high performance.

After gate oxide layer is grown, the base is implanted, and base/emitter contacts (e.g., polysilicon buried contacts) are formed. After gate oxide layer 48 is grown, and before the N-well mask and implant, a collector sink mask, phosphorous implant (in collector region 5) to a concentration of 1.times.10.sup.15 to 1.times.10.sup.16 atoms/cm.sup.2 and a drivein may be performed for performance improvement.

While the above is a complete description of a preferred embodiment of the present invention, various modifications may be employed. For example, bipolar devices may be omitted. Consequently, the scope of the invention should not be limited except as properly described in the claims.