Process for reliable ultrathin oxynitride formation

[0002] This application is a divisional under Rule 1.53(b) of application Ser. No. 08/708,428, now U.S. Pat. No. 5,939,763, filed Sep. 5, 1996. Application Ser. No. 08/708,428 is hereby incorporated by reference in its entirety into the subject application.

[0003] This invention relates to processes for the formation of ultra-thin dielectric layers for use as gate or tunnel oxides employed in integrated circuits.

[0004] The trend in integrated circuits is toward higher performance, higher speed, and lower cost. Correspondingly, device dimensions and feature sizes are shrinking for all types of integrated circuit technology. This trend necessitates the use of ultra-thin dielectrics in the fabrication of such devices as Metal-Oxide-Semiconductor (MOS) transistors and floating gate memory elements.

[0005] MOS transistors are comprised of highly doped source and drain regions in a silicon substrate, and a conducting gate electrode is situated between the source and drain but separated from the substrate by a thin gate dielectric layer. When an appropriate voltage is applied to the gate electrode, a conducting channel is created between the source and drain. Shorter channels, shallower source and drain junctions, and thinner gate dielectrics are critical to achieving smaller and faster MOS devices.

[0006] Certain Electrically Erasable Programmable Read-Only Memory (EEPROM) elements utilize a two layer polysilicon structure comprising an electrically disconnected polysilicon gate electrode, referred to as “floating gate”, and a second control transistor gate above the floating gate and more removed from the substrate. The floating gate, which retains electrical charge for a long time period unless altered by an external energy source, is charged or discharged by quantum mechanical tunneling of electrons through very thin dielectrics known as “tunnel oxides”. The threshold voltage of the control transistor differs for the charged and uncharged states of the floating gate.

[0007] Presently, ultra thin dielectrics less than 100 Angstroms thick, usually of high quality SiO₂, are utilized as MOS gate dielectrics (commonly called gate oxides), and as tunnel oxides in floating gate EEPROM memory elements. Reliability and reproducibility of these ultra-thin oxides can be adversely affected by many factors including lack of thickness control, poor interface structure, high defect density, and impurity diffusion through the oxides. These factors can seriously degrade device performance.

[0008] Diffusion of impurities, particularly boron, through thin oxides is a major problem in processing technology. In Complementary MOS (CMOS) technology, many front end processing steps such as polysilicon gate deposition can be performed simultaneously for the NMOS and PMOS devices of CMOS circuits; however, the dopant implantation steps are performed separately, since different dopants are required. Arsenic and phosphorous, donor-type materials which provide free electrons as charge carriers, are most often used to dope the gate and source/drain regions of the NMOS devices. Boron, an acceptor-type material which provides free holes as charge carriers, is the most often used dopant for PMOS devices. Boron from the doped polysilicon gate has a much higher diffusion rate through the gate oxide layer than do arsenic or phosphorus, and can cause severe degradation of PMOS device characteristics. A concentration of charged boron ions within the gate oxide degrades the insulating characteristics of the oxide, causing gate oxide rupture at sufficiently high concentration. Additionally, boron charge within the gate oxide results in a shift of the transistor threshold voltage V_T. The magnitude of this shift is a function of the concentration of diffused boron ions times the depth of their penetration into the oxide. For ultra-thin gate oxides, boron can diffuse completely through the gate oxide into the underlying substrate, causing even more severe threshold shift problems. Similar problems with boron diffusion are evidenced for the very thin tunnel oxides used in floating gate memory elements of EEPROMS. The resulting degradation in oxide breakdown characteristics lowers the number of possible program erase cycles.

[0009] Poor interface structure between a Si substrate and an SiO₂layer results largely from strain caused by lattice mismatch between Si and SiO₂. One consequence of this is the formation of interface states during high electric field stress or during exposure to high energy radiation such as x-rays. These interface states cause degradation of transistor turn-on characteristics.

[0010] Incorporation of nitrogen into the thin oxide layer has been shown to inhibit boron diffusion and to improve the Si—SiO₂interfacial structure. Specifically, a nitrogen concentration profile having a double peaked structure with a peak of nitrogen at the Si—SiO₂interface and a peak at the SiO₂surface adjacent the polysilicon gate in MOSFET's, and having a low nitrogen concentration therebetween, has been shown to effectively impede boron diffusion from the doped polysilicon gate and to maintain oxide integrity. Additionally, incorporation of nitrogen at the Si—SiO₂interface has been shown to relax the interfacial strain and improve the immunity of the oxides to interface state generation under high field stress.

[0011] Several methods for forming a nitrided oxide layer have been used. The first of these is referred to as the Nitrided Oxide (NO) method, which is described by M. Moslehi et al in J Electrochem Soc: Solid State Science and Technology,Vol 132, No. 9, September 1985, pp 2189-2197, which is hereby incorporated by reference. This method comprises growing a thin thermal oxide on the Si substrate which is then annealed in an ammonia (NH₃) atmosphere to incorporate nitrogen into the oxide. Furnace anneal was initially utilized, but most recently, Rapid Thermal Anneal (RTA) has been used as an alternative. Using the NO method, peaks in nitrogen concentration are seen at the Si—SiO₂interface, hereafter referred to as the “interface”, and at the SiO₂surface adjacent the polysilicon gate in MOSFET's, hereafter referred to as the “oxide surface”. The nitrogen concentration within the oxide film increases monotonically with nitridation time. Thin oxides fabricated using the NO method exhibit improved resistance to boron penetration, as well as improved Si—SiO₂interfacial characteristics and low defect densities. However, decomposition of NH₃during the nitridation process also results in incorporation of hydrogen into the SiO₂layer. Si—H bonds and Si—OH bonds form, causing a large increase in electron and hole trapping and a high density of fixed charges, which result in threshold voltage instability for MOSFET's and degradation of breakdown endurance for MOSFET's and EEPROMs.

[0012] A second method, known as the reOxidized Nitrided Oxide (ONO) method, is described by T. Hori et al in IEEE Transactions on Electron Devices,Vol. 36, No. 2 February 1989, pp 340-350, also hereby incorporated by reference. The ONO method adds an additional high-temperature (800-1200° C.) oxidation step after the ammonia nitidation of the NO method. The hydrogen incorporated into the oxide layer during the ammonia nitridation is reduced by the oxygen present during the subsequent oxidation step, and diffuses out at the high oxidation temperature. As reoxidation proceeds, the hydrogen concentration in the film is found to decrease monotonically, with the rate of decrease depending on the reoxidation temperature and on the nitrogen peak concentration. The hydrogen concentration approaches a minimum value approximately equal to the hydrogen levels found in thermally grown oxide. A more heavily nitrided surface layer is thought to act as a higher barrier for oxygen diffusion, making the reoxidation process slower. The reduction in hydrogen concentration is shown to proportionately reduce the electron charge trapping evidenced in the nitrided oxides.

[0013] A disadvantage of the ONO method is the relatively narrow process window for achievement of optimum oxide quality. Over-reoxidation has been shown to actually degrade oxide electrical qualities. A further disadvantage of the NO and ONO processes is the high level of nitrogen in the bulk of the oxide. The bulk nitrogen concentration, which can be as high as 5-10×10²⁰atoms/cc, weakens the dielectric and degrades its breakdown characteristics.

[0014] Another method of formation of an oxynitride layer utilizes an anneal in N₂O ambients. Two variations of this method have been used:

[0015] 1. Formation on a Si substrate of a thermal SiO₂layer in oxygen ambient, followed by anneal in N₂O, which is described by A. Uchiyama et al in IEDM Technical Digest, IEEE,1990, pp 425-428, hereby incorporated by reference, and

[0016] 2. Growing of a thin silicon oxynitride layer directly on the Si substrate by high temperature exposure of the Si substrate to a pure N₂O ambient, described by H. Hwang et al in Appl Phys Lett57 (10), Sep. 3, 1990, pp 1010-1011, which is hereby incorporated by reference.

[0017] Dielectric layers formed by both of these variations exhibit a nitrogen peak at the Si—SiO₂interface, and relatively small amounts of nitrogen incorporated into the oxide bulk. By way of example, a nitrogen peak concentration of 2-3×10²¹/cc and a nitrogen concentration in the oxide bulk of approximately 10¹⁸/cc have been measured for a thermal oxide annealed at 1100 degrees Centigrade in N₂O. Compared with control thermal oxides, these oxynitrides show significant reduction in interface state generation under high field stress, and lowered electron trapping. They are also shown to act as a barrier for inhibiting boron penetration into the Si substrate. The relatively low nitrogen levels in the oxide bulk yield favorable oxide breakdown characteristics.

[0018] For ultra-thin silicon oxynitride dielectric layer growth, the oxidation of Si directly in an N₂O ambient (the second variation of the above cited N₂O method), has the added advantage of a suppressed growth rate. The growth rate of silicon oxynitride in pure N₂O ambient at 1100° C. using an RTP has been measured as 1.2 Å/second. By comparison, the growth rate of oxide in an O₂ambient for the same processing conditions is 10 Å/second. Simultaneous nitrogen incorporation with oxide growth results in gradual formation of an interfacial silicon oxynitride (SiO_xN_y) layer which acts as an oxidant diffusion barrier. The suppressed oxidation rate provides good thickness control even in the ultra-thin range (<60 A).

[0019] A major problem with ultrathin oxides formed with N₂O ambients is the absence of any nitrogen-rich layer at the oxide surface, as reported by H. Hwang et al, in IEDM Technical Digest, IEEE,1990, pg. 424. Accordingly, no barrier exists to prevent boron from penetrating into the oxide, even if the nitrogen peak at the Si surface is effective in preventing boron penetration into the substrate. Furthermore, studies have shown that boron has diffused into the substrate for N₂O-based oxynitrides, indicating that their Si—SiO₂interface nitrogen peak concentration is below the optimal level for blocking boron diffusion.

[0020] Another prior method of nitridation of a thermally grown SiO₂layer, by either furnace or rapid thermal exposure directly to a nitric oxide (NO) ambient, has been reported very recently. The rapid thermal method is described by M. Bhat et al, in IEDM Technical Digest, IEEE,1994, pp 329-332, which is hereby incorporated by reference. The depth profile, as measured by Secondary Ion Mass Spectrometry (SIMS), of nitrogen incorporated into the oxide is similar in shape to that of a thermal SiO₂annealed in N₂O, and has an interface peak nitrogen concentration as high as 10²²/cc for anneal at 1000 degrees Centigrade. This peak value is nearly 2 orders of magnitude higher than that seen by the authors for an N₂O annealed oxide under similar processing conditions. The enhanced interfacial nitrogen peak also provides a highly self-limiting oxynitride growth due to the barrier properties of incorporated nitrogen to diffusion of oxidants. The thickness of the nitrogen-rich interface oxynitride layer saturates at a value of approximately 3 Å. The oxynitrides produced by exposure of thermal SiO₂to NO, while having higher interface nitrogen peak levels than those produced in N₂O, share the problem of lacking a surface nitrogen barrier to prevent boron diffusion into the oxide layer itself.

[0021] It has been concluded from kinetic studies described by P. Tobin et al in VLSI Tech. Sympos.,1993, pp 51-52, which is hereby incorporated by reference, that NO is the critical species producing interfacial nitrogen pileup during oxynitridation of thermal oxide in N₂O. Heating of the N₂O causes its decomposition by the reactions:

[0022] N₂O→N₂+O, where the atomic O recombines into O₂, and

[0023] N₂O+O→2NO

[0024] It has been estimated that at 950° C., the N₂O is fully decomposed before the N₂O reaches the wafer, and the composition of the oxynitridation ambient is 64.3%N₂, 31.0%O₂, and 4.7%NO. Thus, the formation of a nitrogen interfacial peak by N₂O anneal depends on the indirect, thermodynamically unfavorable dissociation reaction of N₂O to NO. In contrast, the favorable, direct reaction of NO with Si is thought to produce the enhancement of interface nitrogen peak levels for NO-annealed oxides.

[0025] Still another method of incorporating nitrogen into a thin oxide layer is by ion implantation of nitrogen, described by Haddad et al in IEEE Electron Device Letters,Vol. EDL-8, No. 2, February 1987, pp 58-60, which has been utilized to provide a two-peaked nitrogen structure. Whereas this method can be effective for inhibiting boron diffusion and improving interface state generation and charge-to-breakdown values, it has numerous drawbacks. Ion implantation is expensive, and incorporating it into the process during oxide growth involves major redesign of the standard CMOS manufacturing process. Additionally, the process windows for optimal implant dose and energy are narrow, to avoid damage to the dielectric structure while still improving breakdown characteristics.

[0026] We have provided an improved process for forming an ultra thin silicon oxynitride dielectric layer on a Si substrate with improved thickness control, electrical characteristics, and resistance to boron penetration into the oxide as well as into the Si substrate. In this process a thin nitrogen-rich silicon oxynitride layer is grown on the substrate, then further processing provides an oxynitride layer with a nitrogen peak at the silicon-dielectric interface, and a peak at the dielectric surface.

[0027] An object of this invention is to provide an improved process for forming an ultra thin dielectric layer on Si, and an improved ultra thin dielectric layer formed by this process.

[0028] Another object of this invention is to provide a process for forming an ultra thin silicon oxynitride layer on Si which is resistant to boron penetration, and a silicon oxynitride layer formed by this process.

[0029] Another object of this invention is to provide a process for forming an ultra thin silicon oxynitride layer on Si which has a low density of electron traps, and a silicon oxynitride layer formed by this process.

[0030] A further object of this invention is to provide a process for forming an ultra thin silicon oxynitride layer on Si with favorable oxide breakdown characteristics, and a silicon oxynitride layer formed by this process.

[0031] A further object of this invention is to provide a process for forming an ultra thin silicon oxynitride layer on Si with a peak in nitrogen concentration at the oxide-silicon interface and a peak in nitrogen concentration at the oxide surface, and a silicon oxynitride layer formed by this process.

[0032] A further object is to provide a process for forming an ultra thin silicon oxynitride layer on Si which has precise thickness control, and a silicon oxynitride layer formed by this process.

[0033] A further object is to provide a process for forming an ultra thin silicon oxynitride layer on Si which includes forming a self-limiting nitrogen-rich layer and thereafter forming an oxide beneath the nitrogen-rich layer with a second nitrogen-rich layer forming at the Si interface.

[0034] A further object is to provide a process for forming an ultra thin oxynitride layer on Si which does not utilize hydrogen-bearing species, and an oxynitride layer formed by this process.

[0035] A further object is to provide a process for forming an improved ultra thin oxynitride layer on Si which is compatible with existing semiconductor manufacturing process flow.

[0036] A further object is to provide a process for forming an improved ultra thin oxynitride layer on Si which does not substantially add to manufacturing costs.

[0037] FIG. 1 is a process flow diagram for a preferred embodiment of the invention.

[0038] FIG. 2ais a cross sectional view of the preferred oxynitride layer structure.

[0039] FIG. 2bis a schematic representation of the nitrogen concentration profile in the oxynitride layer of FIG. 2a.

[0040] FIG. 2cis a cross sectional view of the preferred oxynitride layer structure utilized in a MOS transistor structure.

[0041] FIG. 2dis a schematic representation of the nitrogen concentration profile in the oxynitride layer of FIG. 2c.

[0042] FIG. 2eis a schematic representation of the concentration profile of dopant species in the MOS transistor structure of FIG. 2c.

[0043] FIG. 3 is a SIMS profile of nitrogen concentration vs. depth below oxide surface, for a first sample processed according to this invention.

[0044] FIG. 4 is a SIMS profile of nitrogen concentration vs. depth below oxide surface, for a second sample processed according to this invention.

[0045] FIG. 5 is a prior art SIMS profile of nitrogen concentration vs. depth below oxide surface, for sample with thermal oxide annealed in N₂O.

[0046] With reference to FIG. 1, a preferred process flow embodiment of our invention is shown. The entire process sequence has been performed in a Reactive Thermal Processing (RTP) system, such as the Rapid Thermal Reactor manufactured by PAPRO. In step 1, a silicon wafer is provided with a clean silicon surface on which to grow the dielectric layer. In step 2, the wafer is immersed in a one atmosphere pressure of nitric oxide (NO) gas at elevated temperatures in the range of 800° C. to 1150° C. for a time period in the range of 10 to 3000 seconds. In step 4, the NO flow is turned off, the NO is pumped out, and one atmosphere pressure of N₂O is introduced. The wafer is annealed in the N₂O at temperatures in the range of 800° C. to 1150° C. for a time period in the range of 10 to 3000 seconds. Other embodiments of the invention may utilize other oxidizing ambients in place of N₂O, oxygen or steam by way of example. In step 6, the N₂O is pumped out, and an NO ambient is again introduced when additional nitrogen incorporation at the oxide/silicon interface is desired. In that case, the wafer is again annealed in NO at temperatures in the range of 800° to 1150° C. for a time period in the range of 10 to 3000 seconds.

[0047] With reference to FIG. 2a,the preferred structure for the oxynitride layer on the Si substrate is shown. Substrate 8 is annealed in NO and N₂O gases as described in the process flow, forming dielectric oxynitride layer 10 on surface 11. Upper surface 12 of oxynitride layer 10 would be adjacent to any subsequently formed conducting gate for a MOS transistor. Surface region 14, and interface region 16 of dielectric 10 have elevated nitrogen concentrations, as represented by nitrogen concentration profile 20 in FIG. 2b.Nitrogen concentration surface peak 22 and nitrogen concentration interface peak 24 provide barriers to impurity diffusion through dielectric 10, as well as improving structure of the Si-dielectric interface at surface 11. In FIG. 2c,the preferred structure for oxynitride layer 10 on Si substrate 8 is shown, with a conducting gate 13 such as heavily doped polysilicon above the oxynitride layer 10. Nitrogen surface peak 22 in region 14 provides a barrier to impurity diffusion of impurities 23 from polysilicon gate 13 into oxynitride 10. FIG. 2eschematically illustrates the abrupt concentration profile of a dopant species of a transistor according to this invention in which the oxynitride layer of this invention is seen to preclude the diffusion of the high concentration dopant from the conductive polysilicon gate 13 into or across the oxynitride layer 10, as well as serving as a barrier for diffusion of dopant species from or into the underlying Si 8.

[0048] Table I summarizes N₂O anneal conditions and thickness data for oxide grown on Si substrates processed according to the process flow of FIG. 1. The data is obtained from SIMS analysis, with a low energy, low fluence Cs⁺ ion beam used to collect depth profiles of CsO⁺, CsN⁺, and CsSi⁺ molecular clusters. The position of the oxide/Si interface is defined as the depth at which the CsO signal drops by one decade. In all calculations and depth profiles the first few data points are excluded to eliminate the SIMS blast-through artifact. Unfortunately, this artifact prevents obtaining accurate data points at the surface of the oxide layer. SIMS analysis is described in Semiconductor Material and Device Characterization,D. K Schroder, John Wiley and Sons, 1990, Ch. 10.

[0049] For the experiments of Table I, substrate samples I and II have both been first rapidly thermally annealed in 1 atmosphere NO for 50 seconds, at a temperature of 1000° C. Then sample I is provided 100 seconds rapid thermal anneal in 1 atmosphere N₂O at 1050° C. Measured oxide thickness on sample I is 45 Å, corresponding to a growth rate of 0.45 Å/sec. Sample II is provided 200 seconds rapid thermal anneal in 1 atmosphere N₂O at 1100° C. Measured oxide thickness on sample II is 110 Å, corresponding to a growth rate of 0.55 Å/sec. These growth rates are approximately a factor of two lower than those observed in the prior art for oxidation in an N₂O ambient without initial NO anneal.

[0050] For each of the samples, there is considerable oxide growth during the N₂O anneal. This growth indicates that, during the N₂O anneal, oxidizing species penetrate through the nitrogen-rich interface layer 16 formed during the initial NO anneal, since basic oxidation kinetic theory states that the oxidation reactions occur directly at the Si surface. Thus, the oxidation mechanism must involve the diffusion of atomic or molecular oxygen, which are products of the dissociation reaction of N₂O, through the nitrogen-rich interface region. Although the oxide growth occurs, it is quite slow, and therefore the inventive process provides precise control of the oxide thickness.

[0051] With reference to FIG. 3, a SIMS depth profile through the oxide layer and the Si—SiO₂interface is shown for Sample I, but the data very near (i.e., within the first 10 Angstroms beneath) the oxide surface is excluded because of the blast through artifact. A peak nitrogen concentration of approximately 7×10²⁰atoms/cc is seen near the Si—SiO₂interface, at a depth of approximately 30 Å. This peak concentration corresponds to approximately 1 atomic% nitrogen.

[0052] With reference to FIG. 4, a SIMS depth profile through the oxide layer and the Si—SiO₂interface is shown for Sample II, again with the surface data excluded. A peak nitrogen concentration of approximately 6×10²⁰atoms/cc is seen near the Si—SiO₂interface, at a depth of approximately 90 Å. This peak concentration corresponds to approximately 0.9 atomic% nitrogen. For the samples of Table I, the hydrogen atom concentration in said silicon oxynitride layer is less than 10¹⁸atoms/cc. Also, the bulk nitrogen concentration in said layer is less than 10¹⁸atoms/cc.

[0053] FIG. 5 shows a SIMS depth profile from the reference by Bhat et al, cited above, for a thermally grown SiO₂layer of 104 Å thickness rapidly thermally annealed in N₂O at 1000° C. for 100 seconds. The surface data is excluded, as in FIGS. 3 and 4. It is seen that the position and magnitude of the nitrogen peak concentration closely match with those for substrate II after 200 seconds rapid thermal anneal in N₂O at 1100° C., as seen from FIG. 4. The nitrogen incorporation in the bulk of the oxide and at the SiO₂—Si interface for the inventive process of an NO anneal followed by N₂O anneal is quite similar to that for the prior art thermal SiO₂annealed in N₂O. However, in the inventive process, since the oxygen species from the N₂O dissociation reaction must penetrate the nitrogen-rich layer formed during NO anneal to grow the oxide layer at the Si surface, it is believed that the nitrogen-rich layer must be pushed up and remain at the surface of the oxide layer, even though it cannot be resolved by SIMS due to the SIMS blast-through artifact. The resultant structure will provide the desired double-peaked nitrogen profile for blocking of boron penetration into the oxide and substrate, and for yielding improved Si-oxide interface characteristics. The low nitrogen concentration in the oxide bulk will avoid degradation of oxide breakdown characteristics.

[0054] When higher interface nitrogen concentration is required, tailoring of the interface peak nitrogen concentration is accomplished by further anneal in NO ambient following the N₂O anneal.

[0055] Our invention, in addition to providing the preferred nitrogen profile in the ultra thin oxide layers, involves process modifications which are easily incorporated into existing semiconductor manufacturing processes, and they add insubstantially to the cost of the process. Our invention provides precise oxide thickness control without narrowing the process window due to the extremely slow oxide growth rate. Also, the preferred embodiment of the inventive process is performed without introduction of any hydrogen species, and avoids hydrogen-induced charge trapping.

[0056] Although our preferred process utilizes rapid thermal processing in the temperature range 800° C. to 1150° C., and ambient atmospheric pressures of NO and N₂O, it is not essential that this exact method and parameters be used. Other embodiments may utilize furnace anneals for all the processes, and other oxidizing ambients such as O₂or steam may be used in place of N₂O.

[0057] It is not our intention to limit our invention to the preferred embodiment, but rather the scope of our invention should be construed in view of our claims. With this in mind,