Overcoming Gating Factor

1. Gate Dielectric Control METRO L O G Y Overcoming the Gating Factor Inline Characterization of Nitride Gate Dielectric Films, with Prediction of Threshold Voltage James Chapman and Terry Letourneau, Micron Technologies Kwame Eason, Torsten Kaack, Xiafang Zhang, and Michael Slessor, KLA-Tencor CorporationInline electrical characterization is well-suited for studying and monitoring nitride dielectric films without requiring full wafer processing.Introduction Experiment The semiconductor industry strongly relies In this study, nitride oxides were produced on high on its ability to continuously scale devicequality p-types both after oxide on the plasma nitride feature size to increase performance and gate oxide and after nitration on Si (100) wafers. Inline reduce power consumption as well as cost.electrical measurements were performed using the One of the many challenges in CMOS KLA-Tencor Quantox and KLA-Tencor UV-1280SE. scaling is the continued increase in the gateThe measurement sequence within dielectric formation dielectric capacitance per unit area. This isprocess is illustrated in Figure 1. The measurement accomplished by either reducing the gate principles of Corona-Oxide-Si (COS) technology are dielectric thickness or increasing the gatehighly analogous to MOS C-V.8 The Quantox system is dielectric constant (εr). Presently, the gatebased on combining three non-contacting technologies: dielectric is silicon dioxide (SiO2), but in charged corona, vibrating Kelvin probe and a pulsed the ultra-thin gate oxide regime, utilizationlight source, as shown in Figure 2. Charged corona ions of pure SiO2 is increasingly difficult due toprovide biasing, and emulate the functions of the MOS high gate leakage (Ig), oxide non-uniformity,electrical contact. The Quantox EOT parameter surface roughness, and boron penetration (GateTox™) is determined from measured dielectric from the p+ polysilicon electrodes. Thecapacitance. The capacitance is determined from dQ/dV nitridation of SiO2 has been successfullyin accumulation in the COS system.9 The capacitance is shown to improved device performance and converted to thickness using εr = 3.9. In an actual tool commercialization.1-7 application, some second order corrections can beapplied to acquire data to account for semiconductor A key device performance metric is thresh- old voltage matching for NMOS and PMOS transistors. The PMOS, long channel GateAnnealPolysilicon threshold voltage (long Pch Vt) is utilized Oxidation Deposition to characterize the effectiveness of the boron(Base OX) (B) penetration resistance of the dielectric; however, a significant drawback of transistor characterization is the need for costly and SiON time-consuming processing. This workSi Si describes the correlation of long Pch Vt to1 23 inline electric and optical parameters obtained from the KLA-Tencor Quantox™ and UV-1280SE tools, respectively.Figure 1. Steps in generating the nitrided oxide film. UV-1280SE mea-surements taken at “1” and “3”, Quantox measurements taken at “3”.12 1 Spring 2003 Yield Management Solutions 2. MET R OL O G YThe experiment is designed to have theCorona Bias, Kelvin Probe,Surface Photovoltage, nitridation process as the major excur- QVSurfSPVsion mode for the purpose of evaluating+8kV LIGHT Mechanical Oscillatorthis concept. The Pchannel Vt is chosen Corona Source, Kelvin ProbeElectronicsTransientDetectionas the end of line monitor due to its CO3-, H30+ sensitivity to B penetration resultingSPVVSurf OXIDEfrom the poly doping and source/drain P SILICONformation steps. The long Pchanneltransistor is chosen because the Vt is 1. 2.3.primary controlled by the gate, unlike Apply Q Corona BiasMeasure V S (=V OX + ψ)Stop vibration, flash light, the short channel device, where the turn Measure each ∆ Q Probe vibration drivesand measure SPV AC current: dψon characteristics are heavy influenced I ≈Cdt by the drain voltage (phenomena knownI = V S - V kp dC as drain induced barrier lowering10). dt This concept is illustrated in Figure 3. 4. RepeatThe inline characterization parametersare affected by the physical thickness, Figure 2. Quantox COS measurement theor y.composition, and quality (or leakage) ofthe film. Table 1 highlights the impact of film charac- band-bending. The tunneling voltage (Vtunnel) parameterteristics that result in a positive shift in the Vt, and the is used to monitor the high-field leakage properties of corresponding response of the inline parameters. The the oxide. All dielectrics eventually reach a point where, physical interpretations of these parameters are easily as more and more charge is applied, the voltage across related to physical thickness, dielectric quality (or the dielectric reaches the maximum sustainable volt- resistance to gate leakage) and nitrogen content. age, defined as Vtunnel. Vtunnel provides a good indication of the oxide integrity and quality in a manner similar The predicted Vt is a model built on linear combinations to more traditional soft-breakdown measurements. of inline parameters. This approach (the model) uses aorder Taylor Series expansion of the functional responses The inline electrical measurements were done on moni-of Vt to the inline parameters. The model is developed tor wafers, with one wafer per lot, five sites per wafer. using SAS JMP4 software and has the form of The end of line electrical data comes from two to twelve probed wafers per lot, nine sites per wafer. The lot aver- Y = ax1 + bx2 + cx3 ... ages are used for correlation in this study.where Y is Vt; a, b, c are coefficients; and x1, x2, and x3are inline parameters. The fit of the predicted data to the Results and discussion The equation for threshold voltage is provided in Equation 1.10 The major contributions to the Vt are PolyPoly film capacitance (Cox), bulk Si band bending (YB) and substrate doping (NA). The later two (YB and NA) are GateGate controlled primarily by near surface doping of channel. The inline measurements are not sensitive to variations in surface doping of channel due to absence of any Vt adjusted doping on inline samples.Si SubstrateSi SubstrateLong ChannelShort Channel [ Vt = öms – Qeff Cox]+ 2ΨB Gate Controlled Drain Controlled √4åsiqNAΨB+ Figure 3. Schematic highlighting the requirement of monitoring long CoxPchannel devices for B penetration resistance. Variations in nitridation Equation 1. Textbook calculation of Vt for MOS transistor, from SZE 10 .will impact the degree of B penetration resistance. Spring 2003Yield Management Solutions 13 3. ME T R O LOG Y Causes for Physical Dielectric Nitrogenmodels range in total inline parameters, the least being PMOS V1 toThicknessQuality Content two and the most being six (i.e. the model is based on increase (↑) (↑) (↑) two to six inline parameters). GateTox (EOT) ↑ ↓ Device -Vtunnel↑↑↓Model 12 3 ρox ↑↑ Optimize 10.9880.795 0.839 Dit ↑Adjust R2 0.9710.549 0.645 Reflectivity↓ ↓Optimize 20.9520.986 0.976Adjust R2 0.8750.970 0.948 Table 1. Parameter response table variations in gate dielectric result-General Function0.9520.986 0.976 ing in a PMOS V t increase.Adjust R2 0.8750.970 0.948Table 2. Table highlighting the optimization V t models for different actual data is presented in Figure 4. The model isdevices. (Note that a “General Function” is a metric which has reasonably based on 12 observations, which is the lower end ofgood application to all devices). The general function has the same establishing a statistical population. The methodology isparameters but different coefficients from device to device. applied to three devices, each with varying base oxides and transistor process flows. A “super-set” of parametersConclusions Actual by Predicted Plot In this paper, nitrided oxide films have been character- -0.5ized using inline non-contact electrical and optical Vt, S91 (0058 Actual0.55measurements. The correlation obtained between the -0.6 EoL Long Pchannel Vt actual and predicted (based on-0.65 inline parameters) has resulted in R2 > 0.97 for indi- -0.7 vidually optimized models. The individually optimized-0.75 models incorporate ~6 inline parameters. A two para--0.8-0.7 -0.6-0.5 -0.4Vt,Predicted P=0.0002 meter model has been successfully developed for oneRSq=0.99 RMSE=0.0104device with R2 > 0.90 and adjusted R2 > 0.88. TheseSummary of Fitresults support that the inline monitoring is sensitive RSquare 0.986438 RSquare Adj 0.970163 to process variations that impact end of line measure- Root Mean Square Error 0.01045 Mean of Response-0.61093 ments, when nitridation is the primary excursion mode. Observations (or Sum Wgts)12 The correlation obtained between Long Pchannel Vt andSummary of Fit Source DFSum of SquaresMean SquareF Ratiomodel-based Quantox and UV-1280SE measurements Model6 0.039713260.00661960.6119 Error5 0.000546000.000109 Prob>F demonstrates that inline electrical characterization is C. Total 110.040259260.0002well suited for studying and monitoring nitrideddielectric films without requiring full wafer processing. Figure 4. Predicted versus Actual V t for one device. The R 2 fit is > 0.98 and adjusted R 2 > 0.97, with 12 observations.References1.S. Hattangady et al., SPIE Symp. Microelec. Manf. have been determined for application to all data sets.(1998). The criteria for successful model generation is R2 > 2. D.T. Grider, et al. VLSI 1997, p. 47-8. (1997). 0.9, with adjusted R2 > 0.85. Table 2 highlights the 3. Rodder-M, et al., IEDM 1998, p. 623-6. (1998). application of the “general” functional model to all 4. K. Eason et al., 198th ECS Toronto, p195-203 (2000). three data sets. The application of this model means 5. F. Cubaynes. IEEE ASMC 2002, TBP. that the inline parameters comprising the model are6. K. Eason et al., AVS ICMI, p251-3 (2002). constant; however, the coefficients in the models are dif- 7. H.N. Al-Shareef, et al., 198th ECS Toronto, p210-213 ferent for the various devices. The general model func-(2000). tional form is provided in Equation 2. The optimized 8. J. Guan et al., ECS, MA 99-2 p1106 (1999).9. T. G. Miller, Semi. International, July (1995).10. S. M. Sze, Semiconductor Devices Physics andVT = f(SPV,Dit,Tox,∆Ref,Vtun,ρox,Qtotal)Technology, 1985. Equation 2. Super-set of parameters used in optimized model. The “General Function” uses six parameters.14Spring 2003 Yield Management Solutions 4. Upset over 90 nm gate stack control? Quantox XP Inline, independent electricalTake two and accelerate yield as needed. measurement of gate leak- age and capacitance, andIt’s time to alleviate the pain of gate stack control. Because with gate dielectrics correlation to end of linegrowing more complex and measuring less than 20 Å, you’re bound to experiencetransistor parametric tests.problems with film thickness variations, composition, capacitance and leakage. So SpectraFx 100 Ultrawhat’s the remedy? It’s a combination of optical metrology to control thickness precise, reliable opticaland composition, and electrical metrology to control capacitance and leakage.monitoring of extremelyAnd only KLA-Tencor has both. Giving you better gate stack control. While speeding thin gate dielectrics onyour way to higher yields at 90 nm. Now that’s relief. product wafers. For solutions and strategies for gate stack control, please visit us atwww.kla-tencor.com/GateXpress Accelerating Yield ©2002 KLA-Tencor Corporation