1 / 45

Statistical Characterization of the Chemical-Mechanical Polishing Process

Statistical Characterization of the Chemical-Mechanical Polishing Process. A Presentation at the XVI Oklahoma State University Research Week . Prahalada K Rao School of Industrial Engineering and Management. Introduction: Why CMP.

Download Presentation

Statistical Characterization of the Chemical-Mechanical Polishing Process

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.


Presentation Transcript

  1. Statistical Characterization of the Chemical-Mechanical Polishing Process A Presentation at the XVI Oklahoma State University Research Week Prahalada K Rao School of Industrial Engineering and Management

  2. Introduction: Why CMP • Conventional semiconductor manufacturing processes (CVD) have flaws. • Notably planarity concerns. Notice the “bird beak” structure in the adjoining figure (Zantye, 2004). • CMP is the only known method to achieve both local and global planarity and thus facilitate miniaturization. • The basic principle: Use a chemical reaction to soften material and then mechanically polish off this layer. IEM-MAE OSU

  3. The Advantages and Challenges • Excellent Planarization. • Indifference to wafer surface. • Reduce defects. IEM-MAE OSU Zantye, 2004

  4. Resources LapMaster Lapping M/c Wafer Carrier Polishing Pad Strasborough CMP M/c Retaining Ring IEM-MAE OSU

  5. Challenges in CMP • Complexity of CMP Process IEM-MAE OSU

  6. Method of Analysis • The material in these slides is presented using the Theory of Constraints approach. • The data used for this research is primarily sourced from US patent 6564116 B2.The objective of the research can be briefly summarized as follows. • Part 1- Classification: To understand the behavior of key process input variables (KPIV’s). • Part 2 – Correlation : To illuminate the effect of the KPIV’s on key process output variables (KPOV’s) namely MRR and Within-wafer-non uniformity (WIWNU) . • Part 3 – Effect-Cause-effect : Contribute to the physical understanding of the process. IEM-MAE OSU

  7. Design of Experiment and ANOVA - US patent 6564116 B2. (Wang et al, 2001) IEM-MAE OSU

  8. Part 1: Classification. The main factors that affect the MRR and WIWNU are classified – What’s happening?

  9. Part 1: Classification for MRR • Platen Speed, Down Force • and Solid Content are the • KPIV’s that affect MRR most. • Back Pressure and Time are • relatively benign KPIV’s. IEM-MAE OSU

  10. Part 1: Classification for MRR – how the variables behave IEM-MAE OSU

  11. Part 1: Classification for WIWNU Most significant A: Solid Content B: Down Force C: Back Pressure D: RPM E: Time • Platen Speed and Back Pressure • are KPIV’s that affect WIWNU most. • Solid Content and Down Force are • relatively benign. • Interaction effects are more significant • than the KPIV’s (knobs) themselves. Least Significant IEM-MAE OSU

  12. Part 1: Classification for WIWNU – how the variables behave. A: Solid Content B: Down Force C: Back Pressure D: Platen speed E: Time Needs further investigation Non Linear Non Linear Non Linear Monotonically Increasing • MRR and WIWNU are governed by different set of factors (variables), some of them acting in opposition in each response. IEM-MAE OSU

  13. Part 2: Correlation Correlate the KPIV’s to the KPOV’s. How is it happening?

  14. Part 2: Correlation for MRRA: Solid Content (wt%), B: Down Force (psi), C: Back Pressure (psi), D: Platen Speed (RPM), E: Time (Sec) • The regression equation for mean MRR obtained as a results of the statistical analysis can be written as A: Solid Content B: Down Force C: Back Pressure D: Platen speed E: Time IEM-MAE OSU

  15. Model Validation for MRR IEM-MAE OSU

  16. Explaining Significant Interactions – MRR IEM-MAE OSU


  18. Part 2: Correlation for WIWNUA: Solid Content (wt%), B: Down Force (psi), C: Back Pressure (psi), D: Platen Speed (RPM), E: Time (Sec) • On the basis of a regression model connecting the mean WIWNU with the various factors can be written as… • Notice the three way interactions in the equation, these are not seen in the MRR equation. A: Solid Content B: Down Force C: Back Pressure D: Platen speed E: Time IEM-MAE OSU

  19. Model Validation for WIWNU IEM-MAE OSU

  20. Significant Interaction effects for WIWNU IEM-MAE OSU




  24. Logarithmic Regression • The data was subjected to a logarithmic regression procedure. • The regression equation is • The equation above is in close conjunction with the Preston’s equation. It is imperative to note that a time domain is considered. Preston’s Equation Form C: Solid Content. P: Down Force V:RPM t: Time γ:Back Pressure IEM-MAE OSU

  25. Part 3: Effect- Cause-Effect Leverage the understanding, optimize towards the goal and stretch the correlation.

  26. Part 3: Effect-Cause-Effect - MRR IEM-MAE OSU

  27. Effect-Cause-Effect - WIWNU IEM-MAE OSU

  28. Statistical and Dynamic Comparison to the Lapping Process Take a designed experiments approach to characterize the lapping process Extrapolate results to the CMP process.

  29. Motivation and Introduction • Can we provide an experimental basis for our conclusions regarding CMP? • A study of a similar process and analysis of the same would provide an excellent starting point. • Lapping is a super finishing process, utilized to achieve submicron surface roughness. Prominent researchers contend Lapping and CMP are closely related. • Material removal mechanisms have been shown to match. IEM-MAE OSU


  31. Lapping Machine Polishing Pads Sensor DAQ system 8728 A 500 Kistler Accelerometer IEM-MAE OSU

  32. How do we decide the KPIV’s and KPOV’s Customer Requirement IEM-MAE OSU

  33. Results from DOE (Taguchi L8) analysis. • Cannot just predict roughness given parameters alone. BAD! IEM-MAE OSU

  34. Sensor Based Modeling DOE doesn't work that well, so what do we do?

  35. Principal Component Analysis- PCA • The following features were extracted from the sensor based data. • Peak to Peak amplitude – P2P • Energy • Variance • Kurtosis. • A grand matrix consisting of these features was constructed. • This matrix, also called the features matrix is subjected to a PCA, to recognize the most telling features IEM-MAE OSU

  36. The Features Matrix IEM-MAE OSU

  37. Underlying Hypothesis and Results. • Implicit Conclusions • Cannot predict the process based on parameters alone • Process is essentially non-linear • Need sensor based modeling IEM-MAE OSU

  38. Summary of results from Hypothesis 3 Notice closeness With CMP IEM-MAE OSU

  39. System Development Based on: Dornfeld et al , ‘An investigation of material removal mechanisms in Lapping’ ,Journal of Manufacturing Science and Engineering, ASME , Vol 122 pp. 413- 2000 From Hanna-Tobias machine tool dynamics model IEM-MAE OSU

  40. Model development Measuring this Simulink Model Block Diagram IEM-MAE OSU

  41. Simulation parameters. • %------------------------------------------------- • %M-File for 'tuning' the model to observed results • %------------------------------------------------- • m =1 %mass of the system in kg • lambda =1.485e7 %hanna tobias stiffness fcn • HW =45 %VHN of aluminium • beta1 = 350000; %nonlinear constant coeff • beta2 = 350000; %nonlinear constant coeff • HP =91 %Hardness of Plate - VHN of Silicon Carbide • A = (pi*9/4)*2.56/100 %surface area exposed to lapping - 3" dia workpiece • Dm = 78e-3 %mean dia of abrasive in mm • Dmin = 25e-3 %min dia of abrasives in mm • Dmax = 125e-3 %max dia of abrasives in mm • Dsigma = (Dmax-Dmin)/2.99 %stdev of the grain size in mm • alpha = 0.95; %geometric coeffcient • T = alpha*HW/((1+(HW/HP)^2)^0.5) • P = 3.2e3 %density of Silicon Carbide Grit kg/m3 • Pd = 1000 %density of water kg/m3 • M = 0.16 %grain to fluid work ratio • K = (1+(P/(Pd*M))) • N = 6*A*Dmax/(pi*((Dm)^3)*K) • G = N*T • dint = 0.001; %initial displacement • h =1008050 %Hysteresis damping coefficent • w = 10 %Region where energy peaks are observed in the data IEM-MAE OSU


  43. Reality Theory IEM-MAE OSU

  44. My Research tool kit. • Designed Experiments • Response Surface methodology. • Regression modeling. • Theory of Constraints. • Process Simulation. • Process Modeling • Six Sigma Approaches. • Principal Component Analysis. • Sensor based monitoring and modeling. Future focus • Illuminate the reason for the interactions. • Provide a physical model for CMP. • Introduce sensor based control techniques. IEM-MAE OSU

  45. Acknowledgements • Dr R. Komanduri. http://www.okstate.edu/MAE/maerl/mpmrl.html http://www2.mae.okstate.edu/Faculty/koman/koman.html • Dr S.T.S. Bukkapatnam. http://www.okstate.edu/ceat/iem/iepeople/bukkapatnam/default.htm • We thank the National Science Foundation (Grant # 0427840) for their generous support of this research. • WenChen Lih – PhD Student, MAE. IEM-MAE OSU

More Related