You are on page 1of 5

CH5020 - Assignment 4

Suraj Y [CH11B063]
Saturday 15 November 2014

Background
In semiconductor processes, Integrated Circuits (ICs) are built on silicon wafers using a
multi-step photolithography process. The wafers are processed in groups called lots. Film
thickness, critical width dimensions, and electrical properties are measured on each wafer,
at different sites. The success of each step is directly related to the uniformity of these
measurements.
In the paper, nine measurements were made on each of the 2 wafers from 20 lots.
The variable names have not been revealed for proprietary reasons.

Objectives of the Experiment


The objectives of the experiment are
To quantify the total amount of process variation
To quantify the contribution of various factors - lots, wafers and sites (disturbances) to the total variability

Type of experimental design


A randomized fully nested and balanced designed is used here:
The effect due to the wafer type is nested under the lot, and the effect due to the site
is nested under the wafer.
Each lot has the same number of wafers and the same number of sites are measured on
each wafer.
All the factors - lots, wafer,sites- are considered random.

Statistical Model
Let the number of lots be l, the number of wafers in each lot be w, and the number of sites
measured on each wafer be s.
The statistical model for this design is given by
Yijk = + i + j(i) + k(ij)

(1)

i = 1, . . . , l

j = 1, . . . , w

k = 1, . . . , s
where
Yijk is the measurement from the k th site on the j th wafer in the lth lot.
is the overall mean
i is the random effect of the ith lot.
j(i) is the nested random effect of the j th wafer in lot i
k(ij) is the nested random effect of the k th site on wafer j in lot i.
j(i) indexes the wafer in lot i and k(ij) indexes the replicates/sites

Analysis of Variance
Theory
The contribution due to the total variation can be split as follows:
SStotal = SSlot + SSwaf er + SSE
l X
w X
s
X

(Yijk Y... )2 = ws

i=1 j=1 k=1

l
X

l X
w
X

i=1

i=1 j=1

(Yi.. Y... )2 + s

(Yij. Yi.. )2 +

l X
w X
s
X

(Yijk Yij. )2

i=1 j=1 k=1

(2)
The degrees of freedom are as follows
l.w.s 1 = (l 1) + l.(w 1) + l.w.(s 1)
THe expected values of the mean square terms are as follows:
E[M Slot ] = s2 + sw2 + wsl2
E[M Swaf er ] = s2 + sw2
E[M SE] = s2
2

(3)

Results
The calculations have been performed in R. The code for the same is presented later in the
document. The split-up of the total variation is listed in table 1.

Source
Lot
Wafer
Residuals
Total

Table 1: Analysis of Variance Table


Degrees of
Sum Squares Mean Squares F value
Freedom
19
20
320
359

14050.9
1959.6
6089.6
22100.1

739.52
97.98
19.03

p-value

38.8608 2.14E-071
5.1488 3.71E-011

From the p-values, we can see that the variation contributions from the lot and the
wafer are significant.
Variance Estimates
The variance estimates are calculated as follows;
s 2 = M SE
= 19.030
M SW M SE
w 2 =
= 8.772
s
M SL M SW
= 35.641
l 2 =
ws
About 56% of the total variation is due to lot. About 30% is due to measurement error (site
variation). Wafer variation is the least contributor, about 14% of the total.

Comparison with the published results


The author of the paper uses an identical experimental design. Hence I have obtained identical
results as presented in the paper. In addition, I have statistically determined that the
contributions of the lot and the wafer to the total variation are significant.

R-code
The ANOVA steps have been performed in R. Before analyzing, the data had to be tidied
into a suitable format for processing in R.

# Reading in the Variables


dat<- read.csv("DataSetNo18to20/data.txt",sep=" ",header=T)
dat <- transform(dat,Lot=as.factor(Lot),Wafer=as.factor(Wafer))

Manual Code
#
l
w
s

Parameters
<- max(dat$Lot)
<- max(dat$Wafer)
<- max(dat$Site)

# Loading the reshape2 library


library(reshape2)
# Melt and Cast Operations
mold <- melt(dat,id.vars = c("Site","Wafer","Lot"),measure.vars =
c("Measurement"))
dat_lot <- dcast(mold,Lot~variable,fun.aggregate = sum)
dat_wafer <- dcast(mold,Lot+Wafer~variable,fun.aggregate = sum)
# Sum-square calculations
ss_total <- sum(dat$Measurement^2) - sum(dat_lot$Measurement)^2/l/w/s
ss_lot <- sum(dat_lot$Measurement^2)/s/w - sum(dat_lot$Measurement)^2/l/w/s
ss_wafer <- sum(dat_wafer$Measurement^2)/s - sum(dat_lot$Measurement^2)/w/s
ss_site <- ss_e <- ss_total - ss_wafer - ss_lot
# Mean-square calculations
ms_lot <- ss_lot/(l-1)
ms_wafer <- ss_wafer/l/(w-1)
ms_site <- ss_site/l/w/(s-1)

1
2
3
4

Source Df SumSquares MeanSquares Fvalue


pValue
Lot 19
14050.9
739.52 38.8608 2.14e-71
Wafer 20
1959.6
97.98 5.1488 3.71e-11
Site 320
6089.6
19.03
Total 359
22100.1

Using R routines

fit <-lm(Measurement~Lot+Lot/Wafer,data=dat)
anova(fit)
Analysis of Variance Table
Response: Measurement
Df Sum Sq Mean Sq F value
Pr(>F)
Lot
19 14050.9 739.52 38.8608 < 2.2e-16 ***
Lot:Wafer 20 1959.6
97.98 5.1488 3.708e-11 ***
Residuals 320 6089.6
19.03
--Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Variance Estimates
# Variance Estimates
var_s <- ms_site
var_w <- (ms_wafer-ms_site)/s
var_l <- (ms_lot - ms_wafer)/w/s
round(c(Lot=var_l,Wafer=var_w,Residual=var_s),3)
Lot
35.641

Wafer Residual
8.772
19.030

References
1. Charles R. Jensen (2002), Variance Component Calculations: Common Methods and Misapplications in the Semiconductor Industry, Quality Engineering,
14:4, 647-657, DOI: 10.1081/QEN-120003564
2. Douglas C. Montgomery and George C. Runner (2011), Applied Statistics and
Probability for Engineers (5th ed.), John Wiley & Sons (Asia) Pre Ltd.
3. Nested Treatment Design, STAT 502 - Analysis of Variance and Design of Experiments, Penn. State Statistics Online Resources. Retrived Novmember 14, 2014 from
https://onlinecourses.science.psu.edu/stat502/node/158
4. Special Case: Fully Nested Random Effects Design, STAT 502 - Analysis of
Variance and Design of Experiments, Penn. State Statistics Online Resources. Retrived
Novmember 14, 2014 from https://onlinecourses.science.psu.edu/stat502/node/166
5. Nested Designs in R, Retrived Novmember 15, 2014 from http://www.personal.psu.
edu/mar36/stat_461/nested/nested_designs.html
5

You might also like