1 3 rd Annual SFR Workshop November 8

1 3 rd Annual SFR Workshop, November 8, 2000 8: 30 – 9: 00 – 9: 50 – 10: 10 – 11: 00 – 12: 00 – 1: 50 – 2: 40 – 3: 30 – 4: 30 – 9: 00 9: 50 10: 10 11: 00 11: 50 1: 00 1: 50 2: 40 4: 30 5: 30 11/8/2000 Research and Educational Objectives / Spanos Plasma, Diffusion / Graves, Lieberman, Cheung, Haller break Lithography / Spanos, Neureuther, Bokor Sensors & Metrology / Aydil, Poolla, Smith, Dunn lunch CMP / Dornfeld, Talbot, Spanos Integration and Control / Poolla, Spanos Poster Session and Discussion, 411, 651 Soda Steering Committee Meeting in room 373 Soda Feedback Session

2 Recipe Generation, Process Optimization and Control SFR Workshop November 8, 2000 Costas Spanos, Kameshwar Poolla, Berkeley, CA 11/8/2000

3 Outline • Historical review and definitions – Traditional open loop production – Recent trends • Current work in progress – – Process Diagnosis Process Control In-line and in-situ sensing Recipe generation 11/8/2000

4 The early production method processing • Terrible Yield and Reliability • Terrible Equipment Utilization • Very inefficient learning 11/8/2000 testing Good wafers (for now) Discarded wafers

5 The Pilot Wafer Method / Trial and Error wafers waiting to be processed testing equipment processing equipment “pilots” Trial and error, using “pilot” wafers • Labor Intensive • Terrible Equipment Utilization 11/8/2000

6 Problems? • • Dramatic Yield Loss Very slow yield “learning” rates Lots of wasted equipment time. Lots of wasted labor. 11/8/2000

7 A large arsenal of tools IS available • Equipment modeling • Factory modeling • Process modeling – Model building – Model tuning • Expert systems and process diagnostics • Data mining • Factory models 11/8/2000

8 Model-based Control wafers waiting to be processed processing equipment Process/Equipment Model and Controller 11/8/2000 testing equipment

9 Requirements for Model-based Control • Models (both empirical and first-principle) for Lithography, CMP and Plasma. – Litho simulation studies – Global CMP model • Sensors and metrology – – Smart Wafers Plasma sensors Scatterometry Existing chamber sensors 11/8/2000

10 Model characterization is far from trivial Pad Roughness Pad Hardness Chemical Reaction Model (RR 0)chem Wafer Hardness Fluid Model Slurry Concentration, Abrasive Shape, Density, Size and Distribution Down Pressure Slurry Chemicals Relative Velocity Model of Active Abrasive Number N Model of Material Removal VOL by a Single Abrasive Physical Mechanism; MRR: N´VOL Preston’s Coefficient Ke Dishing & Erosion 11/8/2000 Surface Damage MRR Wafer, Pattern, Pad and Polishing Head Geometry and Material Contact Pressure Model Pressure and Velocity Distribution Model (FEA and Dynamics) (RR 0 )mech WIDNU WIWNU

11 Modeling Strategy • Detailed process models are very complex • Many parameters need estimation from experiments – High estimation variance – Limited data • Strategy – First principles analysis to determine critical subsystems – Sensitivity analysis to compute parameter identifiability – Numerical trials using experiments to gauge parameter variances – Experiment-based cross validation 11/8/2000

12 Model-based Diagnosis Real-time equipment model Diagnostic Engine Real time monitoring of key equipment variables processing equipment 11/8/2000

13 Furnace System A five-zone batch furnace: Insulator Thermocouple 11/8/2000 Heater Wafer Boat Door

14 Approximate Furnace Model --- Electrical Analogy P 5 T 5 R 45 T 4 R 34 T 3 R 23 T 2 R 12 C 5 R 5 P 4 C 4 R 4 P 3 C 3 R 3 P 2 C 2 R 2 P 1 Electrical Equivalents of Thermal Parameters: 11/8/2000 T 1 C 1 R 1

15 Model is tuned to match equipment behavior Comparison of Actual and Simulated system model 11/8/2000

16 Overall Diagnostic System Simulated Plant Malfunctions: Noise T(k+1) = AT(k)+BP(k) Power Controller + Temperature Actual System Temperature Setting 11/8/2000 Diagnosis Power Temperature Sensor Heater Controller Heat insulation Power supply …

17 Diagnostic Goals Temperature Settings Temperature Sensor Readings Is there any fault? Diagnostics Power Delivery What type of fault? Which sensor/which zone? How serious it is? (parameterization) Accurate System Model Ultimate Goal: Different combinations of faults can be distinguished and the corresponding fault parameters can be estimated with high accuracy. 11/8/2000

18 Diagnose Results by Least Square Approach * Results are based on experimental data from furnace tylan 17 in Berkeley Microlab 11/8/2000

19 Towards a Complete Plasma Diagnostic System • Plasma signals are difficult to characterize, and are subject to preventive maintenance, machine aging, chamber memory effect, etc. • Need to describe signals both qualitatively and qualitatively. • Need to discover meaningful “features” hidden within a large amount of data. • Need to establish “rules” that can be used to routinely characterize signals. 11/8/2000

20 Available Data from LAM 9400 • Optical emission spectroscopy (200 – 1100 nm), 1 nm, 1 hz. • Z-scan signals of current voltage, impedance, five harmonics of 13. 56 Mhz, 1 hz. • Regular machine signals & settings, power, pressure, temperature, gas flow rate, 1 hz. • Monitor the chamber continuously to build a large database suitable for data mining. OES 11/8/2000 machine settings signals

21 Experimental Setup on Lam 9400 TCP Etcher Workstation SECII OES sensor 11/8/2000 Lam 9400 Z-scan sensor

22 Labview Interface 11/8/2000

23 Archival System Data Gathering Process Data Gathering Process 11/8/2000 Database Registry N E T W O R K Database Store Archive User Interface (GUI)

24 Some Features Suitable for Exploration Wafer cycle Signal Intensity Characteristic pattern time 11/8/2000

25 Data Mining • • • Need an effective tool to explore large database. Explore maintenance, aging, memory effect. Explore meaningful attributes of the OES spectrum. Explore the database population for machine settings. Explore trends of the signals subject to variation in machine settings. 11/8/2000

26 OES Peak Exploration Must determine the peaks that change due to certain faulty condition, and establish threshold criteria. baseline faulty 11/8/2000

27 Exploring the Database Population • For this example, RF top power values of 280~310 W should be used as operating points for exploration purpose. 11/8/2000

28 Exploring Trends • E. g. , choose database entries that vary pressure and “fix” all other machine settings. 11/8/2000

29 Syntactic Analysis raw data Encoder Preprocessing Classifier Pre-processing • Extract OES peak values. • Select RF harmonics and parameters. • Compute average value for machine signals. 11/8/2000 fault category

30 Encoder • Scan the historical database to establish a baseline reference. • Assign codes to OES peaks, z-scan harmonics, and machine signals: Ø Ø Ø Large increase: 2 Moderate increase: 1 Unchanged: 0 Moderate decrease: -1 Large decrease: -2 Put the code in a stream: {OES peak codes}{Z-scan harmonic codes}{machine signal codes}, e. g. , {0 0 0 1 0 2 0…. 0 – 1}{0 0 – 2 – 1 … 0 1 0}{0 2 1 0 … 0 – 1 0} 11/8/2000

31 Classifier: Regular Expression • • • x*: zero or more x, i. e. , <empty>, x, xx, or xxxxxxx. x+: one or more x, i. e. , x, xx, or xxxxxxx. x? : zero or one x, i. e. , <empty>, x. x|y: either x or y. {0 0 1 2 0|1|2 … }{0 0 – 2|– 1 …}{…} 11/8/2000

32 Monitor Numerical Values • Monitor signals on a within-wafer and wafer-to-wafer basis. • For example, if pressure consistently drifts away from the setting point, we should generate an alarm. Actual pressure Target pressure 11/8/2000

33 Early Sensitivity Analysis OES and RF Sensors “complement” each other HBr Cl 2 Z-scan no weak yes yes OES yes yes strong no 11/8/2000 pressure RF top RF bottom

34 Future Directions in Diagnosis • Deploy automated fault detection system using high sampling rate RF fingerprinting. • Study automated generation of syntactic analysis rules for RF fingerprinting. • Study systems of real-time instability detection and plasma stabilization control; • Perform field studies of automated OES classification for fault diagnosis. 11/8/2000

35 In-line Metrology… processing equipment Built-in Testing! Far more efficient use of clean room space and labor. Ability to measure any fraction (0 -100%) of product flow. But, can in-line sensors be as good as stand-alone metrology? 11/8/2000

36 In-line ellipsometry/reflectometry/scatterometry • Noise analysis of commercial ellipsometers is carried out to determine detectability of EM response variations. • We are focusing on the capability of compact, in-line implementations of the required optics. Rounding Slope Angle Height PR Footing ARC Poly-Si Si 11/8/2000 d(ISource) d(IDetector) CD d(q. Analyzer) d(q. Polarizer) d(Beam Divergence) Sample

37 Smart wafers as in-line sensors processing equipment Smart Wafer data feedback process control wafers waiting to be processed 11/8/2000 base station

38 Process Monitoring using in-line Metrology on ASML/SVG workcell Process monitoring & control based on tuned simulator n&k thickness resist thickness ellipsometer reflectometer Spin coat & softbake Exposure Smart wafer monitors 11/8/2000 DITL resist profile reflectometer scatterometer PEB Develop

39 Process Window Engineering • Change the shape of the process window by adjusting the operating point settings and material parameters, in order to maximize the overlap between the joint p. d. f of inputs and the process window, • In this way, the projected lithography yield is maximized. 11/8/2000

40 Milestones, to September 30 th, 2001 • Install automated OES on LAM 9400 reactor and build large statistical database of process fingerprinting data. • Install the Z-scan sensor and explore the spectral RF signature of plasma instabilities. • Publish ellipsometric detection specifications for fullprofile, 100 nm metrology. • Demonstrate simulator tuning for full profile matching over a range of focus and exposure conditions. • Develop performance metrics for CMP and lithography. Assess input sensitivity and controllability. 11/8/2000

41 Milestones, to September 30 th, 2002 • Automated fault detection using RF fingerprinting. Automated generation of syntactic rules for RF fingerprinting. • Feasibility of building 100 nm capable profile extraction using small footprint, in-line spectroscopic ellipsometry. • Demonstrate simulator tuning (using Prolith) for full statistical profile matching over a range of conditions. • Design optimal recipes for unit processes, evaluate robustness using simulators and experiments. 11/8/2000

42 Milestones, to September 30 th, 2003 • Study systems of real-time instability detection and plasma stabilization control. • Perform field studies of automated OES classification for fault diagnosis. • Implement lithography controller that merges full profile in-line information with available metrology. • Model-based RTR control schemes, assess theoretical, simulated and experimental performance on SFR variance. 11/8/2000

43 3 rd Annual SFR Workshop, November 8, 2000 8: 30 – 9: 00 – 9: 50 – 10: 10 – 11: 00 – 12: 00 – 1: 50 – 2: 40 – 3: 30 – 4: 30 – 9: 00 9: 50 10: 10 11: 00 11: 50 1: 00 1: 50 2: 40 4: 30 5: 30 11/8/2000 Research and Educational Objectives / Spanos Plasma, Diffusion / Graves, Lieberman, Cheung, Haller break Lithography / Spanos, Neureuther, Bokor Sensors & Metrology / Aydil, Poolla, Smith, Dunn lunch CMP / Dornfeld, Talbot, Spanos Integration and Control / Poolla, Spanos Poster Session and Discussion, 411, 651 Soda Steering Committee Meeting in room 373 Soda Feedback Session