Estimation of etch rate and uniformity with plasma impedance monitoring Daniel Tsunami IC Design and Test Laboratory Electrical & Computer Engineering Portland State University dtsunami@lisl.com 1
Introduction Plasma Processing in semiconductor manufacturing Plasma Impedance Monitoring (PIM) Advanced Process Control (APC) 2
Introduction Plasma Impedance Data Upper and lower frequency ~ 2 MHz and 27 MHz Lower voltage, current and phase 5 harmonics Upper voltage, current and phase 3 harmonics 4 tools, 2-4 chambers per tool Up to 12 recipes for each tool/chamber combination Collected ~ 150k files full data set probably 250k files Only a few will be used for regression 3
Background Plasma Impedance Data 4
Introduction Etch Rate Data Etch rate measured no less frequently than every seven days using test wafers and procedures for each tool/chamber. Etch rate measured at 17 sites on test wafer Range and mean are used as parameters plotted on statistical process control (SPC) charts 5
ER measurements : wafermaps 6
ER measurements : wafermaps 7
ER measurements : wafermaps 8
Etch Rate Variability Etch Rate Data 9
Etch Rate Distributions Estimated pdf 4.5 x 10 3 4 3.5 3 2.5 2 1.5 1 0.5 pm1 pm2 pm3 pm4 0 6300 6400 6500 6600 6700 6800 6900 7000 Etch Rate 10 10 m/min 10
Big Idea Plasma impedance measurements contain information about the etch rate of the plasma tool. Regression analysis is a standard tool for discovering relationships in data. Create regression model to characterize relationship between PIM signals and etch rate measurements 11
Idea - Details Model Selection/Construction Stepwise regression Model Validation R 2 PRESS Statistics 12
Initial Literature Search Garvin and Grizzle : Demonstration of broadband radio frequency sensing: Empirical polysilicon etch rate estimation in a Lam 9400 etch tool Estimate etch rate using regression model and broadband plasma impedance measurements specifically the reflection coefficient phase and magnitude at several different frequencies. Only 20 experiments Good results R 2 = 0.997 and methodology Active as opposed to passive plasma impedance sensing No model validation to prove that results are not due to overfitting 13
Variable Selection 14
Variable Selection 1 R a 2 Adjusted R 2 0.9 0.8 0.7 0.6 0.5 2 1.5 1 0.5 0 0.5 1 1.5 2 10 8 PRESS 10 6 10 4 10 2 2 1.5 1 0.5 0 0.5 1 1.5 2 Data interval from etch rate measurement 15
ER Predictions 16
ER Estimates 17
SPC Chart Mean Etch Rate 10 10 m/min Etch Rate Range 10 10 m/min 7000 6800 6600 6400 6200 0 50 100 150 200 250 5000 4000 3000 2000 1000 0 0 50 100 150 200 250 Time (days) 18
Initial Yield Correlation 0.9 21609 pm1 lig12low E(yld ER) 0.85 0.8 Yield 0.75 0.7 0.65 6550 6600 6650 6700 6750 6800 6850 6900 6950 7000 7050 Etch Rate 10 10 m/min 19
Yield Nonparametric Regression 21609 PM1 LIG12LOW 0.9 Data Points Kernel Regression 95% Confidence Intervals 0.85 0.8 Yield 0.75 0.7 0.65 0.6 6600 6650 6700 6750 Estimated Etch Rate 20
DD Nonparametric Regression 21609 PM1 LIG12LOW 0.45 0.4 Data Points Kernel Regression 95% Confidence Intervals 0.35 Defect Density 0.3 0.25 0.2 0.15 6640 6660 6680 6700 6720 6740 6760 Estimated Etch Rate 21
APC Controller Etch Tool Production Data PIM Test Wafer Data Etch Time Test Wafer Etch Rate Measurements Measured Etch Rates Predictors APC Controller Estimated Etch Rates Adaptive Linear Model 22
PM1 Predictions 7800 7600 7400 Measurements Predictions Etch rate mean 7200 7000 6800 6600 6400 6200 50 100 150 200 250 2500 2000 Etch rate range 1500 1000 500 0 500 1000 50 100 150 200 250 Days since Jan 1 2004 23
PM2 Predictions 7000 6500 Etch rate mean 6000 5500 5000 4500 4000 Measurements Predictions 50 100 150 200 250 1500 Etch rate range 1000 500 0 50 100 150 200 250 Days since Jan 1 2004 24
PM4 Predictions 7400 7200 Measurements Predictions Etch rate mean 7000 6800 6600 6400 6200 0 50 100 150 200 250 2000 Etch rate range 1500 1000 500 0 50 100 150 200 Days since Jan 1 2004 25