Chamber Wall Effects on Polycrystalline-Si Reactive Ion Etching in Cl 2 : A Multiple Real- Time Sensors Study

Transcription

1 Chamber Wall Effects on Polycrystalline-Si Reactive Ion Etching in Cl : A Multiple Real- Time Sensors Study Fred L. Terry, Jr. Dept of EECS University of Michigan (fax) fredty@umich.edu 1

2 Acknowledgements Dr. Pete Klimecky (now with Intel) Dr. Craig Garvin (now with Inficon) Prof. Jessy Grizzle (UofM) Dr. Jay Jefferies (Stanford)

3 Outline Multi-sensor Study of Cl Etching of Poly-Si in Lam 9400 TCP / Variations with F-cleans OES/Actinometry for Cl Broadband RF for Plasma Density RTSE for Poly Si Etch Rate Wall Recombination Affects Both Neutral Species and Ion Concentrations Ion Density Measurement Control of Cl etch of Si Interpretation of Actinometry Results Requires Careful Consideration of Gas Dilution Effects on Actinometer Concentration HBr-Cl Mixtures 3

4 Motivation Chamber wall state as source of transient variations Loss rates at walls dependent on wall buildup Wall condition dynamically alters chemical and plasma densities Solutions for process drift: PMs, additional clean steps, test wafers gas inlet wall buildup losses TCP coil upper electrode Plasma generation (energetic e - ) ion transport & losses lower electrode losses Control of plasma density will improve process tolerance limits & OEE! 4

5 Previous Wall State Work Sawin: 1 st reported Etch Rate changes in Cl due to O ( ) & CF 4 ( ) chamber exposure. (JECS 199) Donnelly: Increasing Cl neutral conc. with time in a quartz tube helical resonator. (JVSTA 1996) Aydil: Atomic Cl drifts due to SiO wall conditioning & SF 6 wall cleans. (JVSTA 00) 5

6 This Work 1 st experimental evidence of Cl plasma density variation with F-cleans/wall prep. 1 st direct correlation of real-time plasma density & real-time etch rate variations 1 st direct real-time feedback control of plasma density to stabilize poly-si etch rate in Cl Improved Understanding of Wall Effects and Actinometry Results 6

7 Time Stamped Sensor System Actuator Signals Tool Signals Digital I/O Plasma Tool Outputs EMACS Real-Time Data Acq. & Control Computer Digital I/O Outputs Relay/Routing Box EMACS I/O Hardware Inputs Inputs EMACS Real-Time System GPIB TCP/IP Lock-ins Broad band Detector OEM Tool Controller LAM 9400SE TCP External Sensors (OES, lasers, etc.) 1.79µ m diode laser w/ alignment laser CCD RTSE for real time ER laser Sensor Computers Real-Time Monitors a) RTSE wafer state b) BroadBand RF plasma state c) FTIR exhaust chem; SiCl 4, SiF 4 d) Diode Laser Absorption chem state e) OES [F], [Cl] intensity in chamber 7

8 RTSE Real-Time Spectroscopic Ellipsometer (RTSE) Can optically model film etch depth, CD, sidewall slope Use for real-time etch rate monitoring & transients 8

9 BroadBand RF Antenna Quartz Tube Wafer 4 " Process Chamber Wall Network Analyzer Remarks High frequency (GHz), low power (mw) sweep of plasma Plasma impedance spectroscopy Must analyze broad spectrum of data (Broadband RF Probe) Yields plasma density metric 9

10 BroadBand RF Circuit Analogy Antenna TCP Coil Quartz Tube 4 " Wafer Network Analyzer Process Chamber Wall BB probe Z A Z p C coil ε(ω) Z p3 Z p1 C w Loss paths give many resonance peaks in Γ for single ω p Model peaks as RLC circuit resonances w/ ω 1 LC ni = L p (ω) C chuck Z p Elec. Stat. Chuck R p (ω) Wall C 10

11 BroadBand Signature Reflect Coeff. Condition 1 Condition Electron Density Frequency Signal sensitive to several important plasma outputs Plasma density Delivered plasma power Chamber wall state Wafer surface chemistry 11

12 BB Peak Shifts & Density 10 0 Seasoning etch rate recovery: BB response run1 run run3 run4 run5 LOG Γ 10 1 ω n ω n frequency in GHz Two prominent resonance modes, ω n1 & ω n, for these chamber conditions Peak frequencies shift right for increasing density 1

13 FTIR Effluent Measurements Moving Mirror Computer Beamsplitter Sample Stationary Mirror Detector Interferometer Source Fourier Transform InfraRed (FTIR) spectroscopy measures volatile etch products in foreline exhaust Yields dynamic chemical state changes in SiCl 4 & SiF 4 Used commercial INDUCT tm FTIR from On-line Tech. 13

14 Etch Conditions Lam 9400 TCP SE 10 mtorr 100 sccm Cl flow 100 sccm total etch gas flow for Cl /HBr experiments 5 sccm flow 50 W TCP Power Varied for Plasma Density Control (Closed Loop) Runs 100 W Bias Power Bias Voltage Measurement Not Available Unpatterned 150 mm Poly-Si/30nm SiO /Si Test Wafers 14

15 Experimental Definition 1 First project; 3 experiments Compensate for ion density losses due to F- cleaning of chamber walls 1) Nominal Etch: Run plasma chamber at steady state chlorine condition to establish real-time etch rate, BB peak position, and SiCl 4 effluent level ) Open loop recovery: Prep chamber walls using C F 6 clean to strip Silicon Oxychloride buildup, then run identical Cl recipe. 3) Closed loop compensation: Run identically as uncontrolled open loop etch, only now use TCP power to maintain BroadBand setpoint. 15

16 (OL) Open Loop Drift Recovery rate (nm/s) Cl etch from F prep chamber walls nominal open loop ω n (GHz) time (s) Nominal etch rate flat, OL rate increasing (upper plot) Nominal BroadBand ω n flat, OL ω n increasing (lower) OL signals do not recover in 60sec 16

17 (CL) Closed Loop Recovery rate (nm/s) Cl etch from F prep chamber walls nominal closed loop ω n (GHz) time (s) Both nominal & CL etch rate flat (upper plot) Both nominal & CL BroadBand ω n flat (lower plot) CL signals recover in ~5sec 17

18 SiCl 4 Effluent from FTIR 5000 SiCl 4 etch product from FTIR 4000 SiCl 4 (unscaled) nominal closed loop open loop time (s) Nominal SiCl 4 is flat with no disturbance (black) OL SiCl 4 effluent is suppressed = lower ER (green) CL SiCl 4 is mostly compensated by controller (blue) 18

19 TCP Power OL vs. CL Cl etch from F prep chamber walls TCP Pwr (W) open loop closed loop time (s) TCP power compensation in CL is very high at the start to make up for lost Cl + ions to the walls 19

20 Experimental Definition Second project; experiments, OL vs. CL 1 st wafer effect elimination with plasma density compensation Prep chamber walls using C F 6 clean Follow with 3 open loop etches for 30s each in Cl and measure etch depth Prep chamber with C F 6 clean again Follow with 3 closed loop etches for 30s each and compare etch depth variation with that in OL case 0

21 1 st Wafer Effect Reduction Three 30s Cl etches after single F-prep of chamber 1st wafer effect, open loop SP max-min: 156. Å, RTSE max-min: Å 1st wafer effect reduced, closed loop SP max-min: 49. Å, RTSE max-min: 56.7 Å etched material (Å) SP RTSE etched material (Å) SP RTSE run # run # Open loop etch depth Etch rate increases, both in situ (RTSE) & ex situ (Reflectometer) Etch depth variation ~150Å Closed loop etch depth with density correction Etch depth variation reduced to ~50Å 1

22 TCP Compensation RR sequential closed loop control action TCP power (W) (nominal) run1 run run time (s) Closed loop TCP power compensation reduces with each successive run as chamber begins to season

23 Summary 1 st evidence of real-time Poly-Si etch rate variation in Cl due to F-exposure. 1 st demonstration of ion density control in Cl to compensate for Poly-Si real-time etch rate transients. Effluent SiCl 4 chemistry verifies both real-time performance drifts and feedback correction. Significant 1 st wafer effect reduction after chamber cleans with density feedback control. Question: How Do We Explain the Results of Earlier Researchers? Actinometry Results & Interpretations Key Point Is That Even For Qualitative Conclusions, Actinometry/OES Results Must Be Carefully Analyzed Considering All Gasses Present In Chamber 3

24 Intensity Ratio I Cl /I λ : 750.4nm λ Cl : 8.nm I Cl /I Nominal I /I Nominal Cl I /I OL Cl I /I CL Cl After F-disturbance, both controlled & uncontrolled cases show similar Clneutral suppression and recovery. Simple Conclusion is that Ions (not neutrals) control etch rate for this process. time(s) 4

25 Cl Intensity I Cl Nominal I Cl Nominal I Cl Open Loop I Cl Closed Loop Cl Intensity is Flat in Nominal/Seasoned-wall case & varies in Open Loop and Closed Loop Cases time(s) 5

26 Intensity 5 I Nominal I Nominal I Open Loop I Closed Loop Intensity of Being Nearly Flat Was Previously Taken By Some Researchers To Show that the Plasma Density Was Constant This led to the conclusion that neutral Cl loss was responsible for Si etch rate variations We have shown that neither of these conclusions can be correct time(s) 6

27 OES Setup Equations d = Cl dissociation fraction f = mole fraction of in feed gas (5%) Mass balance: Cl dcl + (1-d)Cl Raw optical intensity signals: (n e ω n ) I = K ( T ) ω n e n I = K ( T ) ω n Cl Cl e n Cl coupled simply by d, f Intensity ratio: I Cl K Cl 1 f = d I K f ncl f = d ' αcl f if d 7

28 Detailed Look at Dissociation Diluation Effect on Cl dcl 1 d Cl Chlorine Dissociation + ( ) input gas mixture gascompositionin plasma Now including the actinometer concentration ( ) + ( 1 ) ( ) + ( 1 ) + ( 1 )( 1 ) f f Cl f d f Cl d f Cl input gas mixture gascompositionin plasma The concentration of is diluted by Cl dissociation So in the plasma, assuming all molecules, atoms, ions at the same temperature: f f n = ntot = n f + d( 1 f ) + ( 1 d)( 1 f ) 1+ d( 1 f ) d( 1 f ) d( 1 f ) n = n = n f + d( 1 f ) + ( 1 d)( 1 f ) 1+ d( 1 f ) Thus n d( 1 f ) ( 1 f Cl ) = = d n f f Cl tot tot tot 8

29 OES Fits Clean Chamber / High Recombination Case Yields Actinometry Data with Enough Structure to Extract α Cl & K by Nonlinear Regression Dissociation Fractions for Other Runs Estimated by Assuming α Cl is the same as the Clean Chamber Result Possible T e variations Errors Possible Window Variations 9

30 f I = K ( T ) ω n = Kω n e n 1 n tot 1+ d( 1 f ) Fitting of OES Data Fitting constants allows quantitative extraction of d from OES data ( f ) ( f ) d 1 I = K ( T ) ω n = K ω n Cl Cl e n Cl n tot 1+ d 1 I Cl K Cl 1 f 1 K f I Cl = d d = I K f K 1 f I Cl meas α cl I = Kntotω f f f n = Kntotω n = K ω n 1 f I Cl 1 I Cl 1 I Cl 1+ αcl ( 1 f ) 1 αcl f 1 αcl f 1 f I + + I I meas meas meas I = K ( T ) ω n = K n ω Cl e n Cl Cl tot n I I I α f α f f Cl Cl Cl cl cl I I meas I meas meas = K Cl ω n = K ω n 1 I Cl 1 I Cl 1 I Cl 1+ αcl f 1 αcl f 1 αcl f I + I + I meas meas meas 30

31 OES Signal & Fit: SiCl 4 Ignored 3.5 WNpoly11c_ol.txt Fitted Signal K' = / α' Cl = / (95.4% confidence limits) 3.5 Intensity (arbitrary units) Measured I Fitted I I -I,fit Time (s) 31

32 Cl OES Signal & Fit: SiCl 4 Ignored 4.5 WNpoly11c_ol.txt Fitted Cl Signal K' = / α' Cl = / (95.4% confidence limits) Measured I Cl Fitted I Cl I Cl -I Cl,f it Cl Intensity (arbitrary units) Time (s) 3

33 Cl Net Dissociation: SiCl 4 Ignored 0.75 WNpoly11c_ol.txt Dissociation Fraction d and n Cl /n g K' = / α' Cl = / (95.4% confidence limits) Fraction out of dissociation fraction (d) n cl /n g Time (s) 33

34 Fraction: SiCl 4 Ignored Precentage 3 n /n g Scaled ω (ne BB ) Minimum Possible Precentage Time (s) I (t) ~const. due to opposing effects of dilution ( ) & ne ( ) 34

35 Dissociation Fractions: SiCl 4 Ignored Dissociation Fractions WNpoly11c_ol.txt WNpoly10c_CL.txt WNpolygold_nom.txt Dissociation Fraction time (s) 35

36 Intensity Ratio I Cl /I λ : 750.4nm λ Cl : 8.nm I Cl /I Nominal I /I Nominal Cl I /I OL Cl I /I CL Cl Why is feedback controlled I Cl /I still low? Our Next AVS Paper : GENERATION of Cl Is Increased but COMSUMPTION by Si Etching & Dilution by SiCl 4 Offset Generation time(s) 36

37 Key Reactions Cl Cl Dissociation Cl Cl Recombination (wall & bulk gas phase) + Cl + e Cl + e Ionization & Bulk Deionization Si + 4Cl SiCl Etch 4 SiCl4 + SiO SiOxCly + Cl + Cl Deposition Reactions (unbalanced) SiCl4 + AlO3 AlvSizOxCly + Cl + Cl 37

38 molecules in Simplified Reaction Set Assuming Cl ionization and Si-species deposition Reactions have small effects on gas species concentrations, the other remaining reactions yield: FCl Cl + F + FSiSi xcl+ ycl + FSiSiCl4 + f ( ) Cl { Si } chamber gas phase molecules 1 x+ y+ FSi = FCl for Cl mass balance y = 1 d F where d = Net Dissociation Fraction of Cl F Si So = x= dfcl 4F atoms/s consumed by etching known from measured etch rate & flows Si 38

39 Result of Simplified Reaction Set n x+ y+ F + F g Si ( ) ( ) n df 4F + 1 d F + F + F g Cl Si Cl Si n 1+ d F + F 3F g Cl Si x df 4F Cl Si Cl g g n = n = n x+ y+ FSi + F ( 1+ d) FCl + F 3 FSi F F n = n = n x+ y+ F + F ( 1+ d) F + F 3F g g Si Cl Si I = K n n I = K Cl Cl Cl e Measured Actinometry Ratio: I 1 Cl 4 Cl KClSClnCln df F e Si Am = = ' I KSnne αcl F m n n e Cl Actinometry Signal Suppressed by Etch/Loading 39

40 PV = n RT n = g g g PV RT g Assume Tg is constant & n e=cωbbwhere C is a portionality constant fixed during the etch run. I Cl = Pω BB ' ' KClαCl Am F 1 F + 1+ α A F F ' Cl Cl m Si K ' & α are the only unknowns ' Cl They can be extracted if there is sufficient variation in I ( t) & I ( t) Cl = Pω BB ' K Am F 1 F + 1+ α A F F ' Cl Cl m Si I = Pω BB ' KF 1 F + 1+ α A F F ' Cl Cl m Si 40

41 OES Intensity & Fit: SiCl 4 Included from RTSE 3.5 WNpoly11c_ol.txt Fitted Signal K' = / α' Cl = / (95.4% confidence limits) 3.5 Intensity (arbitrary units) Measured I Fitted I I -I,fit Time (s) 41

42 Cl OES Intensity & Fit: SiCl 4 Included from RTSE 4.5 WNpoly11c_ol.txt Fitted Cl Signal K' = / α' Cl = / (95.4% confidence limits) Measured I Cl Fitted I Cl 3 I Cl -I Cl,fit Cl Intensity (arbitrary units) Time (s) 4

43 Fraction : SiCl 4 Included from RTSE Precentage 3 n /n g Scaled ω BB (ne ) Minimum Possible Precentage Time (s) 43

44 Dissociation Fraction : SiCl 4 Included from RTSE 0.9 WNpoly11c_ol.txt Dissociation Fraction d and n Cl /n g K' = / α' Cl = / (95.4% confidence limits) dissociation fraction (d) n cl /n g Fraction out of Time (s) 44

45 Dissociation Fractions: SiCl 4 Included from RTSE Dissociation Fraction Dissociation Fractions WNpoly11c_ol.txt WNpoly10c_CL.txt WNpolygold_nom.txt time (s) Net Dissociation Fraction (d) Is Increased by Higher TCP Power in Closed Loop Run Net d is higher than estimated from procedure ignoring SiCl 4 Wall Recombination Still Suppresses Cl, d 45

46 T e (EEDF) Issue 6 With some assumptions which we believe are justified: b Units Intensity Cl WNpoly 10c Closed Loop ω 4 /1.57 WNpoly10c Closed Loop BB Intensity Cl WNpoly11c ω 4 /1.388 WNpoly11c Open Loop BB Time 4 ω n = I Cl ftonly ( ) T e for open loop case appears ~constant T e is increased initially for closed loop case (constant α Cl assumption may not be accurate) e 46

47 Wall-State Effects Model n Cl reduced due to recombination on F-cleaned walls. n Cl+ reduced due to lower availability of n Cl precursor. ER decreases due to lower ion bombardment. Real-time feedback control corrects for n e n Cl+ losses by increasing T e, but does not fully recover n Cl. Model supports ion dominated etch of Si w/ Cl ; n Cl+ ER n Cl. High n Cl keeps surface Cl-saturated. ion bombardment is rate limiting step. Extracted d varies significantly, causing constant I. 47

48 HBr/Cl Etches HCl Is Formed In Mixing Manifold By HBr/Cl Reaction Collaboration With Stanford Group Shows Similar Plasma/Gas Chemistry Trends To Cl Only Cases HCl absolute concentration was measured by laser diode absorption HCl Dissociation follows BB-RF/plasma density trends Chamber cleaning suppresses dissociation of HCl & increases plasma density variation Open Loop Etch Rates Become More Constant With Increasing HBr & Show Less Sensitivity to Chamber Wall Condition Closed Loop Plasma Density Control Causes More Time Variation In Etch Rate for High HBr Concentration Cases HBr/Cl Etch Rates e Not Directly Ion Limited & Future Work is Needed Wafer Surface Temperature? 48

49 HBr/Cl Etch (80/0) Open Loop Closed Loop ER (nm/s) ER (nm/s) 5 wnpoly1c 5 wnpoly19c 4 4 ER (nm/s) 3 ER (nm/s) time (s) time (s) 49

50 Future Work Modeling of BB Signals to extract more from the shape of the data collision parameters Possible T e /EEDF Information Improved antenna designs for BB System Lower-cost electronics for BB reflectometry Apply density control to topography & profile variations. Expand to other ion-dominated etches besides Cl etching of Poly-Si. Larger scale, multi-wafer tests to verify control improvements. Ion density control most effective when etch is ion dominated. Chemically dominated etches do not show same effects. Combine ion density control with ion energy control. 50

51 Funding Acknowledgements Initial Work Funded by SRC Center of Excellence for Automated Semiconductor Manufacturing Project Ended August, 1999 Research funded in part by: AFOSR/DARPA MURI Center for Intelligent Electronics Manufacturing (AFOSR F ) Projected Ended August, 001 Research funded in part by: NIST Intelligent Control of the Semiconductor Patterning Process (70NANB8H4067) Project Ended June, 00 51