phys. stat. sol. () 5, No. 5, 5 (8) / DOI./pss.77787 A multi-sensor study of Cl ething of polyrystalline Si physia pss www.pss-.om urrent topis in solid state physis Pete I. Klimeky and Fred L. Terry, Jr. * Department of EECS, University of Mihigan, Ann Arbor, MI, USA Reeived June 7, revised 7 Otober 7, aepted Deember 7 Published online 7 Marh 8 PACS 5.77.Bn, 85..-e * Corresponding author: e-mail fredty@umih.edu Cl hemistries are the basis for ething of polyrystalline Si and other ondutive gate materials in Si CMOS integrated iruit fabriation. It is now well-known that reombination of atomi Cl neutrals on the hamber walls influenes the eth rate and thus leads to manufaturing reproduibility problems. In this work, we make use of multiple real-time measurements to improve the understanding of the physial mehanisms for this effet. In partiular, real-time spetrosopi ellipsometry is used as both a poly-si eth rate monitor and as a virtual SiCl flow rate sensor. This aids in the quantitative interpretation of the optial emission spetrosopy data. 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Introdution The wall-state of reative ion eth tools is known to strongly affet eth results, and previous studies show that neutral speies onentrations vary with hamber seasoning [, ]. In this paper, we will show data from simultaneous, real-time, in situ measurements of eth rates (using spetrosopi ellipsometry), plasma eletron density (using broadband RF refletometry), and Cl neutral atinometry data during high-density Cl reative ion ething of poly-si in a Lam 9 SE TCP system. All of these quantities are found to vary over time after a seasoned hamber is leaned by exposure to a fluorine-ontaining plasma. Following hamber leaning/fluorination, variations in real-time poly-si eth rates orrelate losely with eletron density transients. Proportional-integral (PI) feedbak ontrol of the plasma density using the TCP power restores both the nominal density and poly-si real-time eth rate. Using orresponding hlorine atinometry measurements, we suggest that the ompensation mehanism is aomplished by inreasing the eletron temperature, Te, and therefore the ionization rate. Neutral Cl onentrations, n Cl, are found to drift as the hamber walls season in a Cl environment, and PI feedbak ontrol of density may ompensate for some of the n Cl losses. These results are fully onsistent with a signifiant body of experimental and theoretial work suggesting that high density plasma Cl ething of Si is ion-dominated and is sensitive to hamber wall state. However, they are the first to learly demonstrate these onnetions to real-time eth rate variations and provide the first demonstration mehanism for feedbak ontrol orretion to wall-state drift. Our in situ spetrosopi ellipsometry measurements allowed us to: () onnet the data of the plasma density and hemial speies onentrations to the real-time eth rates; and, () aurately estimate the onentration of eth by-produt (SiCl ) present in the hamber. With the estimated SiCl onentration, we were able to orret for gas dilution effets and improve the atinometri estimate of the neutral Cl onentration. The simultaneous use of multiple realtime proess and wafer state sensors allowed us to get an improved overall qualitative and quantitative understanding of this eth proess. Experimental All experiments all used unpatterned 5 mm Si () orientation wafers oated with ~5 nm of LPCVD polyrystalline Si on.7 nm of thermally grown SiO. Reative ion ethes were performed in a Lam 9 SE TCP high density, indutively oupled eth system. The indutive soure and bias supplies operated at.56 MHz. During poly-si eth, the hamber pressure was mtorr and the gas flows onsisted of sm Cl and 5 sm Ar. The soure power was nominally 5 W (but was automatially varied during the PI ontrol experiments) and the bias supply power was W. 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
physia p s s P. I. Klimeky and F. L. Terry, Jr.: Cl ething of polyrystalline Si The eth system was equipped with a high speed rotating polarizer spetrosopi ellipsometer with a prism spetrometer/ccd detetion system (Sopra RTSE). SE data were ollet at a sample/.8 s rate. It was also equipped with a broadband RF refletometry system (BBRF) for plasma density measurents [], and a two wavelength optial emission spetrosopy (OES) system. The OES system used two SPEX 5M grating monohromators with ooled photomultiplier detetors monitored using a hopper/lok-in amplifier onfiguration. This approah allowed simultaneous high speed, high signal to noise monitoring of two spetral lines (the 75. nm Ar line and the 8. nm atomi Cl line) for performing Cl neutral atinometry. The BBRF and OES data were ollet at sample/.5 s rate. All data was olleted using a LABVIEW -based olletion system with areful attention to maintaining a synhronized, aurate time base for all data. We also olleted data on eth byproduts from an Online Tehnologies Fourier transform infrared spetrometer (FTIR) in the foreline of the vauum system. The sampling of this system was slower (typially sample/ s depending on desired SNR). This data was used for orroboration of our results rather than for quantitative real-time analysis. Results and disussion In our earlier work of [], we showed that when the eth hamber had been seasoned following ~ minutes of Si ething (several separate wafer eth runs), the poly Si eth rate (measured by RTSE) was onstant as was the plasma density (BBRF). If the hamber was leaned using CF to remove SiO x Cl y eth byproduts from the hamber walls, the poly-si eth rate dereased and beame time-dependent. The plasma density also showed similar time dependene. Using a PI ontroller whih adjusted the TCP soure power to maintain a onstant target plasma density, the poly-si eth rate was stabilized. The results of one of these sets of experiments are shown in Figs. and. The atinometry ratio (Cl OES emission intensity I Cl /Ar OES emission intensity I Ar ) for these experiments is plotted in Fig.. This ratio is often taken to be proportional to the neutral atomi Cl onentration (n Cl ). This point will be addressed further in this paper. Sine the poly-si eth rates follow the trends of the BBRF plasma density measurements and the losed loop results do not follow the atinometry ratio, we onlude that the poly-si eth rate is dominated by the ion density rather than neutral Cl onentration. In quasi-neutral plasmas, the ion and eletron onentrations are nearly equal, so the BBRF signal (whih is proportional to the eletron density) provides a indiator for both. We onlude that in all tested ases, there is suffiient neutral Cl to hemially saturate the Si surfae, and that the eth rate is ontrolled by the Cl + ion onentration. While we feel these onlusions well supported by this data, there are important details that were unresolved. The most notable are: () the Ar OES signal I Ar is nearly onstant with time in the open loop experiments even though the plasma density is varying; and () the atinometry ratio is the same for the open loop and losed loop experiments (why does the extra TCP power in the losed loop ase not raise the apparent neutral Cl onentration?). Careful simultaneous analysis of the real-time sensor data shows that both of these effets are explained by timevarying hamber gas omposition. While the input gas flows are held virtually onstant by mass flow ontrollers, the time varying dissoation rate of Cl into Cl (due to plasma density hanges), the time varying reombination rate of Cl into Cl (due to hamber wall seasoning ), and the time-varying evolution of SiCl eth byprodut (due to Poly-Si Eth Rate (nm/s) 6 5 Eth Rate (nm/s) Nominal Eth Rate (nm/s) Open Loop Eth Rate (nm/s) Closed Loop 5 6 Figure Real-time poly-si eth rates measured with RTSE: (solid line) nominal/seasoned, (dashed/dots) losed loop/lean, (dashed/squares) open loop/leaned. Broad Band Peak Freq (GHz).8.7.6.5.. Broad Band (GHz) Nominal Broad Band (GHz) Open Loop Broad Band (GHz) Closed Loop. 5 6 Figure BBRF refletometry data ( to plasma density) for the experiments of Fig.. (solid line) nominal/seasoned, (dashed/ dots) losed loop/lean, (dashed/squares) open loop/leaned. 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.pss-.om
Contributed Artile phys. stat. sol. () 5, No. 5 (8) Atinometry Ratio (IC l /I Ar ).8.6...8 ICl/IAr Nominal ICl/IAr Open Loop ICl/IAr Closed Loop.6 5 6 time(s) Figure Cl/Ar atinometry ratio for the representative eth runs. (solid line) nominal/seasoned, (dashed/dots) losed loop/ lean, (dashed/squares) open loop/leaned. the varying poly-si eth rate) all ontribute strongly to the atual relative onentration of Ar present in the hamber and thus to the quantitative interpretation of the OES/ atinometry results. To model the OES data, we assume the following simplified eth reation: F Cl + F Ar+ F SiÆ Cl Ar Si moleules in xcl + ycl + F SiCl + F Ar Si Ar hamber gas phase moleules where FCl and F Ar are the flow rates of Cl and Ar into the hamber (measured by mass flow meters), and F Si is the flow rate of Si into the eth hamber gas environment (as SiCl ). This quantity is derived from the poly-si eth rate and the area of the wafer. We assume that the density of the poly-si is that of single rystal signal (this introdues a negligible systemati error). With an eth rate of nm/s, there is an effetive flow rate of ~7.9 sm of Si byprodut (SiCl x ) into the eth hamber. Mass balane requires that: x+ y+ FSi = FCl () and y = ( - d) FCl () where d is the net fration of Cl whih is dissoiated into Cl. This is affeted by both dissoiation in the plasma, bulk reombination, and wall reombination. These two mass balane equations yield: x = df - F () ÈÎ Cl Si. () The OES signal an be modelled as: I = C n n = K ω n (5) Ar Ar e Ar Ar n Ar I = C n n = K ω n (6) Cl Cl e l Cl n Cl sine the eletron density (n e ) is proportional to the BBRF frequeny squared ( ω n ). The onstants C and K are sensitive to the eletron energy distribution funtion (sometimes simplified to an eletron temperature, T e ). For atinometry, we assume that the ratio KCl KAr is approximately onstant with T e variations. The following set of mass balane equations, with n g representing the total gas partile onentration in the hamber, allows us to reate a time varying model for the OES data: ng µ x+ y+ FSi + FAr µ df - F + ( - d ) F + F + F µ ( + d) F + F -F Cl Si Cl Si Ar Cl Ar Si È x ncl = Í ng x y FSi F = Î + + + Ar È dfcl - F Si Í ng Î( + d) FCl + F Ar - FSi È FAr nar = Í ng x y FSi F = Î + + + Ar È F Ar Í ng Î( + d) FCl + F Ar - FSi Thus the measured atinometry ratio is: ÈI Cl Cl KClnClne KCl È df - F Si ÍI = = Ar KArnArne K Í Ar F Î Î Ar m (7) (8) (9) () For low dissoiation fration (d) and low eth rates (thus low F Si ), the atinometry ratio given in the last equation will be proportional to the neutral Cl onentration n Cl. However, in high density plasma eth systems, both of these assumptions are violated and more detailed analysis is required. Assuming a onstant hamber gas temperature yields a onstant n g. Thus we now have a model with only adjustable time-onstant parameters (K Cl and K Ar ). The timevarying Cl dissoiation fration d(t) is found from the last equation using the measured atinometry ratio. Regression analysis of the open loop data allows estimation of both unknown onstants. These fits are shown in Figs. and 5. www.pss-.om 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
physia p s s P. I. Klimeky and F. L. Terry, Jr.: Cl ething of polyrystalline Si.5 Ar Intensity (arbitrary units).5.5.5 Measured I Ar Fitted I Ar I Ar -I Ar,fit Ar Preentage.5 n /n Ar g Saled ω (n ) BB e Minimum Possible Ar Preentage -.5 5 6 Figure Measured (solid) Ar OES line intensity, fit (dashed) and error (dotted). The resulting relative Ar onentration in the hamber is shown in Fig. 6 along with a saled plot of the BBRF signal. This plot shows that the nearly onstant I Ar signal (Fig. ) observed in the open loop ase is the result of a dereasing Ar fration in the gas (due to dilution effets) and an inreasing plasma density (with the resulting produt being nearly onstant). This anellation affet had lead prior researhers to onlude that the plasma density was onstant [, ]. With a time-varying eth rate (and thus SiCl evolution rate) and a time-varying Cl reombination rate, the Ar onentration annot be onstant and thus using the nearly onstant behaviour of I Ar as a plasma density indiator is inorret. Cl Intensity (arbitrary units) 5 Measured I Cl Fitted I Cl I Cl -I Cl,fit - 5 6 Figure 5 Measured (solid) Cl OES line intensity, fit (dashed), and error (dotted). Assuming that these onstants hold for the onditions of the nominal and losed loop runs, we an then estimate the net dissoiation fration d. This is illustrated in Fig. 7. Based on our estimated orretions, we again assert that the poly-si eth rate is ontrolled by the Cl + ion onentration (diretly related to plasma density). Neutral atomi Cl reombination on the hamber walls redues this speies onentration in the open loop ase, and the lak of atomi neutrals redues the ionization rate, leading to a lower plasma density. In the losed loop ase, the BBRF PI.5 5 6 Figure 6 Ar fration (solid), saled BBRF signal (dashed), and minimum possible Ar fration (dotted) showing the opposing ations of Ar dilution and plasma density inrease. ontrolled inrease in the TCP power yields a onstant plasma density approximately the same as that of the open loop ase. This leads to only a small observable inrease in the net Cl dissoiation fration (d). Thus, the stabilization of the plasma density (and thus eth rate) must be the result of an inrease in the eletron temperature T e with TCP soure power (this is ertainly physially reasonable). Thus the Cl ionization and Cl dissoiation rates are both inreased, but the net dissoiation fration is still suppressed by the wall reombination. Also, most of the newly generated atomi neutral Cl is onsumed by the inrease in the poly-si eth rate, leading to another interesting balane of effets resulting in no apparent differene in the atinometry results of two very different ases (the open and lose loop experiments). Dissoiation Fration.8.6.. Open Loop Close Loop Nominal 5 6 Figure 7 Net Cl dissoiation fration d for the open loop (solid), losed loop (dashed) and nominal seasoned hamber (dotted) ases. Conlusions We have self-onsistently analyzed the results of simultaneous real-time measurements of plasma, neutral hemial, and wafer state data to develop a more omplete piture for the fators that lead to timevarying poly-si eth rates in Cl plasmas. Use of in situ, real-time spetrosopi ellipsometry was ritial to both link the plasma and neutral hemistry data to the eth rates, 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.pss-.om
Contributed Artile phys. stat. sol. () 5, No. 5 (8) 5 as well as to provide a ritial quantitative estimate for the amount of eth byprodut gas (SiCl ) being released into the hamber. We also developed a test for validating the onstant T e assumption. Due to spae onstraints, this annot be disussed in detail in this report. We will state without proof that examining the BBRF to Cl OES signal ration ( ω n ICl ) provides an indiator of variation in T e. For the open loop and nominal ases, we find this ratio to be approximately onstant. For the losed-loop ase, it varies signifiantly with time. Our work ould be further improved if an independent measure of T e (suh as by Langmuir probe or multiple speies rare gas OES methods) were also inluded. Referenes [] P. I. Klimeky, J. W. Grizzle, and F. L. Terry, Jr., J. Va. Si. Tehnol. A, 76 (). [] S. J. Ullal, A. R. Godfrey, E. Edelberg, L. Braly, V. Vahedi, and E. S. Aydil, J. Va. Si. Tehnol. A, () [] V.M. Donnelly, J. Va. Si. Tehnol. A, 76 (996). www.pss-.om 8 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim