Enhanced discrete DMOS power trench gate oxide growth
08/01/2002
By Debra S. Woolsey, Fairchild Semiconductor, West Jordan, Utah
Overview
The influence of selectively implanted fluorine in the silicon substrate prior to furnace gate oxidation is experimentally shown to increase thickness, create an improved Vbd in DMOS capacitors, and allow for the simultaneous growth of gate oxides at various thicknesses.
As the semiconductor industry pushes ahead with increasingly smaller microchips, the need for a dependable, high-quality gate oxide becomes a crucial area of concern. In particular, when constructing a MOS trench transistor, the ability to create a selectively grown quality oxide that is more resistant to voltage breakdown (Vbd) is essential to device reliability. Alternative ways must be found to grow these oxides as the industry presses for higher quality.
In this study, the double-diffused metal oxide semiconductor (DMOS) capacitor's trench gate oxide is constructed by implanting a shallow concentration of fluorine ions, or BF2, at a 0° angle into the bottom region of the silicon substrate's trench where a thicker oxide is desirable. While the bottom of a trench does not greatly impact the die's performance, it is most often a source of premature voltage breakdown and significantly contributes to the trench capacitance. As the silicon oxidizes, the bottom of the implanted trench grows at an accelerated rate, while simultaneously growing a conventional oxide on the sidewalls (Figs. 1 and 2).
 Figure 1. Cross section of a DMOS power trench gate grown at 950°C in a Thermco furnace with a fluorine dose of 4.5 x 1015 ions/cm2. |
Implanting selected regions of the silicon substrate with BF2 or fluorine creates an environment where the growth rate improves in the implanted regions 2-3x above normal during the standard furnace dry oxidation process. At shallow implant depths, the data indicate that most of the fluorine is liberated during the oxidation, and thereby removed from the wafer. Liberating the fluorine ions from the substrate decreases the risk of creating trap charges in the bulk oxide that come from high doses of fluorine ions in the silicon substrate. Thus, thick oxide in selective regions of the gate oxide can remain and be of high quality. Once the oxide is grown, it is possible to deduce the growth trends as they pertain to time, temperature, species concentration, depth of implantation, and how that growth affects the results seen at electrical testing.
 Figure 2. Cross section of a DMOS power trench MOSFET. |
Experiment
A series of samples were created using production-quality, 600μm thick, <100> boron-doped, 150mm dia., p-type, silicon wafers. A 1μm-deep trench was etched into the silicon DMOS substrate with a critical dimension of 0.5μm (Figs. 1 and 2). The substrates were implanted with fluorine ions, or BF2, at a 0° angle in an Eaton GSD, chosen for its ability to implant standard production-quality silicon substrates at a variety of depths and dose concentrations, along with a range of tilt and twist angles. A 0° angle was chosen for its selected capability to precisely implant an area at the bottom of a power trench with nominal disturbance to the surrounding sidewall regions.
A 0° angle is also found to increase the implanted junction depth at the 5.0 x 1015 ions/cm2 range by 20% over the standard 7° angle used in industry. The 0° angle, however, does not have the 10% shift from its peak position created by the 7° angle most often used in manufacturing today [1]. The wafers were tested between implant energies ranging from 5-60keV, corresponding to depths ranging from 50-1500Å, with doses from 1.0 x 1014 to 4.5 x 1015 ions/cm2.
A series of splits were processed at gate oxide deposition using a conventional horizontal, atmospheric, Thermco furnace. The recipes used an argon ramp and anneal with dry oxidation temperatures ranging from 800-1175°C. It is essential to the gate process to anneal the damage caused by etching and the subsequent ion implant prior to the oxidizing step, to avoid growing silicon dioxide in damaged areas. Growing an oxide in a damaged area creates a poor oxide/silicon interface and degrades the gate oxide integrity (GOI).
A second experimental matrix was run using the same basic parameters as the previous one with the exception of a hydrogen anneal prior to oxidation in an Applied Materials rapid thermal process (RTP) tool. The flash RTP anneal replaces the traditional Thermco high-temperature, time-consuming furnace anneal and produces an annealed hydrogen-terminated silicon surface prior to oxidation. The hydrogen-terminated surface eliminates the 10-25Å of poor-quality native oxide normally obtained as newly cleaned silicon is exposed to air. The RTP was run for 60 sec at 240torr and 1000°C; the parameters were chosen not only for their ability to create hydrogen-terminated silicon, but also to eliminate the diffusion of previous fluorine or BF2 implant ions in the silicon during the anneal.
The wafers were then processed in the diffusion furnace, growing a dry 300Å gate oxide, at varied temperatures ranging from 800-1125°C. The purpose of this matrix was to collect enough samples in order to allow comparisons of the fluorine effects on the oxide films.
Oxidation mechanism for fluorine and BF2
The Deal and Grove model is the dry silicon oxidation mechanism generally accepted in industry for pure oxygen growth of greater than 30nm. They postulated that the oxidation rate is determined by a combination of two separate processes. Equation 1 consists of the actual chemical reaction of oxygen growth on a silicon substrate and Eqn. 2 integrates the diffusion of oxygen through the previously grown oxide film [2, 3].
 |
null
 |
null
In the above equations, k = the first-order rate constant for the oxide growth, where temperature is a function of k; C* = the oxide layer on the surface of the silicon substrate; N = the number of oxygen molecules enveloped into a given volume of the oxide layer; h = the gas phase transport coefficient of oxygen; and Deff = the diffusion coefficient of oxygen in SiO2 [2, 3]. The linear parabolic rate law of Deal and Grove is combined in Eqn. 3:
 |
null
where x = the oxide thickness, and F = the total flux of oxygen molecules that migrate through the oxide. F is enhanced with the presence of fluorine ions causing the rate of the reaction to increase.
Fluorine ions are deposited into the silicon by ion implanting F- or BF2 into the substrate prior to the oxidation. Using the Deal and Grove model and the experimental results, a mechanism to explain the enhanced growth can be postulated.
As the temperature in the furnace rises, the silicon reacts with the F- ions to form a Si-F bond. In the case of BF2, the fluorine dissociates itself from the boron to attack the silicon. Si-F bonds together more readily due to the greater electronegativity of the F- ion. The binding energy of Si-Si is far lower (42.2kcal/mol) than that of the Si-F molecule (129.3kcal/mol). Furthermore, the silicon atom bonded with the fluorine ion is a more positively charged molecule due to the large electronegativity difference between fluorine and silicon. The positively charged Si-F molecule pulls the negatively charged O-2 ion toward the silicon surface at a faster rate, thereby acting as a catalyst to form the SiO2 molecule.
While both Si-Si and Si-F react with the oxygen, it has been determined that the Si-F molecules, as opposed to the Si-Si molecules, appear to be more highly reactive with oxygen, thus causing a thicker oxide [4, 5]. This holds true as can be seen in Fig. 2; the larger the implanted dose, the greater the oxidation thickness. In essence, as the dose increases, there are more Si-F molecules available to react with the oxygen.
The molecules' packing densities may clarify one of the secondary causes of enhanced growth. It has been shown that SiO2 grown in a Si-F enhanced silicon substrate has somewhat lower packing density than that in a pure silicon substrate [4]. The lower packing density permits more space or greater accessibility for oxygen to diffuse through the silicon substrate and react with the Si-F molecules.
It has been speculated that the trap creation mechanism responsible for stress-induced leakage current is hydrogen related. The incorporation of fluorine atoms replaces weak Si-H bonds with strong Si-F bonds and can subsequently improve the GOI. Nevertheless, high quantities of fluorine trapped in the silicon are also known to deteriorate GOI by creating trap charge in the bulk oxide as well as at the interface. This explains why, in past experiments, a medium dose fluorine implant yielded a greater improvement [6].
 Figure 3. The ellipsometer measured oxide vs. temperature for fluorine and BF2 at a dose of 4.5 x 1015 cm2. |
Results
Greater thickness is achieved with the implantation of the BF2 molecule vs. the fluorine radical (Fig. 3). Data suggest the difference in the increase in thickness is due to the size of the molecule. The rate of the oxidation reaction should not increase with the rate of dissociation of the BF2 molecule. (The BF2 molecule's limiting reagent is not the dissociation of BF2 to make F- ions but the fluorine's ability to oxidize. Therefore the BF2 does not have an advantage over the pure F- radicals being deposited/implanted into the substrate.) However, the BF2 molecule creates more surface damage at the time of implant, allowing the oxygen to more easily react with silicon at the interface. A Thermowave was used to determine the extent of the damage created by the fluorine and BF2 implants.
The theory is proven in the second matrix where the damage was RTP hydrogen annealed prior to the oxidation. In these tests, the BF2 showed no greater growth rate than that of the fluorine. The theory that surface damage does play an important role in the oxidation growth is therefore substantiated. There is a direct correlation between the amount of damage created and the increase in thickness for these two implanted species; however, with the increase in surface damage more voltage breakdown is apparent. The increased breakdown occurs when oxide is grown in damaged areas of the silicon.
Experiments confirm that wafers run at temperatures between 900-950°C have the maximum growth potential for increased oxide thickness (Fig. 3). As temperatures increase from 950-1100°C, however, enough energy has been provided to start breaking the F-Si bonds, thus creating free radical F- ions, which decreases the amount of available positively charged F-Si molecules. As the concentration of F-Si bonds decreases, the rate of oxidation also decreases. At temperatures exceeding 1100°C, the data indicate that the fluorine is competing with the oxidation growth. At a higher energy level and at a greater concentration of fluorine, it is postulated that an unstable and volatile SiF4 molecule is created instead of the Si-F molecule, which will attack the Si-O-Si bonds [4].
 Figure 4. Vbd measurements run on wafers oxidized and annealed at 950°C. |
Figure 4 illustrates the same consistent increase of thickness at 900-950°C for all fluorine implanted doses between 4.5 x 1014-4.5 x 1015 deposited at energies between 15 and 30keV. It has been proven experimentally that there is a nominal effect on the thickness when the energy is increased from 5keV to 60keV as long as the dose is kept constant and in the range of 1.0 x 1014-4.5 x 1015 (Figs. 4 and 5).
 Figure 5. The ellipsometer measured oxide thickness vs. energy for fluorine-implanted wafers at various temperatures. |
It is apparent, however, that when energy is increased past 60keV, the implant's depth becomes too great (>1400Å) to be effective during 300Å oxide growth. Fluorine deposited at depths >300Å becomes trapped in silicon, creating an electron-rich area, or trap charges, in the bulk oxide, which are detrimental to the DMOS capacitor. Bench tests to determine the amount of trap charge show that fluorine implanted at <200Å or 10keV has no greater trap charges than that of normal nonimplanted gate oxides. This leads to the theory that at least some of the fluorine ions had to be liberated from the gate oxide during oxidation. Interestingly, data suggest BF2 was not liberated from the oxide, so it likely caused the poor voltage breakdown, but not enough data has been obtained to prove this theory.
When analyzing the electrical effect of the enhanced oxide for Vbd and the electrical thickness (Vcrit), an interesting trend appeared (Figs. 4 and 6). In areas where the fluorine was implanted with lower doses (in the 1.0 x 1014 range), Vbd showed no measurable difference in areas where the fluorine was implanted, and there were no significant electrical or Vbd differences, even though the ellipsometer indicated a slight physical increase in thickness.
When the dose was increased to 4.5 x 1014, however, Vbd decreased by 14% from the control tests 13 and 14, with a nominal amount of electrical thickness decrease, <3%. The ellipsometer measurements at the 4.5 x 1014 range, however, confirm an oxide increase of ~10%. These data suggest that at lower concentrations of F-, where growth potential is low, oxide grown in damaged areas caused by the implant produces a poor-quality oxide. It is interesting to note that as the dose increases from 4.5 x 1014 to 4.5 x 1015 ions/cm2, Vbd improves. In test 10, Vbd increased an astounding 15% to 42.5V on average at 5keV or 50Å in depth. The electrical thickness increased to about 400Å, a 33% enhancement. The ellipsometer showed a greater increased thickness of ~530Å, resulting in a 76.7% increase over the control tests.
Conclusion
Data clearly show a systematic increase in dry oxide growth is correlated with the increase of implanted fluorine and BF2 between doses of 1 x 1014 and 4.5 x 1015 ions/cm2. The optimal oxide growth temperature in a Thermco atmospheric horizontal furnace for these species is between 900-950°C. Further, it has been shown that BF2 will grow oxide at a faster rate than fluorine due to the heavy damage it creates; however, with the increase in damage, poor Vbd values result.
When growing thin gate oxides with fluorine implanted at extremely shallow depths of 50-200Å, the optimal temperature has been experimentally shown to increase oxide growth by 76.6% at a dose of 4.5 x 1015, while showing a Vbd improvement of 15% over traditionally grown thin oxides. The data imply that at least some of the fluorine is being liberated during the SiO2 growth due to the absence of trap charge in the bulk oxide, which is known to deteriorate the oxide integrity when high doses of fluorine are present.
Using a selective fluorine-enhanced oxide process in selected areas, a thicker oxide in conjunction with a thinner oxide can be grown in areas not implanted without sacrificing quality.
References
1. V. Raineri et al., Semiconductor Science and Technology, Vol. 5, 1007-1012, 1990.
2. T.K. Whidden, P. Thanikasalam, M.J. Rack, D.K. Ferry, Journal Vacuum Science Technology, B13(4), July/Aug. 1995.
3. B.E. Deal, A.S. Grove, Journal of Applied Physics, 36, 3770, 1965.
4. M. Morita, T. Kubo, T. Ishihara, M. Hirose, Appl. Phys. Lett., 45, 1312, 1984.
5. D. Kouvatsos et al., Electrochemical Society, 138, 1752, 1991.
6. C. Chen et al., 5th Intl. Symp. Plasma Process-Induced Damage, 121-4, 2000.
Debra Woolsey is a process development engineer at Fairchild Semiconductor, 3333 W. 9000 S., West Jordan, UT 84088; ph 801/562-7079, fax 801/562-7337, e-mail [email protected].