Professional Documents
Culture Documents
Bob Pease Lab Notes Part 2
Bob Pease Lab Notes Part 2
Bob Pease Lab Notes Part 2
F
uzzy logic, introduced in 1965 with each element x of U a membership in ables as control signals and the signals for
[ I ] . was originally developed as a the interval [O,l]; and (2) the support o f t h e measuring the system performance; (2) de-
method for implementing deci- fuzzy set, which is a set of element .Y. The riving inference rules: (3) designing fuzzy
sion-making functions with $- value of the membership function PA(.\-) rep- membership functions for each variable; (4)
rlien rules. Now, it is becoming more widely resents the grade of x in A. A larger mem- performing fuzzy inference based on the
recognized as a practical tool for dealing b e r s h i p v a l u e of x indicates a closer inference rules; (5) obtaining a control sig-
with the topic of uncertainty and systems belonging. For example, let A be a set of nal, performing defuzzification to derive
that are poorly understood from the mathe- young engineers, where 1-1 = 25 and 12 = 35 an actual control action.
matical standpoint. During the past several are the ages of two such engineers, and
years, the interests in applying fuzzy logic where the membership function ~ A ( x I ) =
to control, signal processing, and manufac- F u u y Reasoning
0.8, p ~ ( s 2=) 0.6 indicate that the first engi-
turing has been experiencing rapid growth. neer fits much closer into the group of young Fuzzy reasoning is based on the composi-
Examples include aircraft flight control 121. engineers. tional rules of inference. There is more than
accessing equipment reliability 13J , control- A fuzzy set can be represented as fol- one way of generating fuzzy reasoning
ling chemical reactors and processes [4], lows: rules. Generally, their design requires sig-
tracking 2D motion [5], and controlling nificant insight and understanding of the
typically nonlinear systems, such as backing problem. Very often, the rules can be de-
up a truck and trailer [ 6 ] . rived heuristically rather than from closed-
Working with the semiconductor indus- form mathematical formulae. Hence, it is
try, we have recently developed techniques difficult to automate the design process.
based on fuzzy logic for automated visual where the integral stands for the union. Nevertheless, the rules should be derived ( I )
inspection and control applications. In this Some most widely adopted basic opera- to reflect a human experts experience and
article, we will briefly introduce the concept tions on fuzzy sets include the min-max knowledge; ( 2 ) to have sound analytical
of fuzzy logic, then describe how fuzzy logic model [ I ] : background; and (3) to exhibit some leam-
can be used to detect dimple defects on ing and generalization capability. Consider
polished wafer surfaces. Finally, we will the following example:
discuss our in-house built demonstration Rule 1: I f x i is Ail andx2 isA12 theny is
system. Bi
Rule 2: If xi is A21 and x2 is A22 then y is
Fuzzy Set Theory A UB = Jma(pA(x), Pdx))/ x B2
U (3) The truth values of the premises are
Basic Coiic,ept.c where both A and B are fuzzy sets. given by W I = min ( p ~ i(d,
i ~ ~ (x2)), 1 wi2
A fuzzy set. A, of a universe, U, consists of The process of designing fuzzy logic = min ( ~ ~ (xi),
2 1 ~ ~ (x2)).
2 2
two critical components: ( I ) a membership application usually involves five steps: ( I ) The actual control signal as the inference
functions : U + [O,l]. which associates formulating the problem and selecting vari- result is calculated by
A
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Y 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0
0 0 1 1 1 1 1 0 0 0 0 0 0 0 0
0 1 1 1 1 1 1 1 0 0 0 0 0 0 0
0 1 1 1 1 1 1 1 0 0 0 0 0 0 0
0 1 1 1 1 1 1 1 0 0 0 0 0 0 0
0 0 1 l 1 1 1 0 0 0 0 0 0 0 0
0 0 0 1 1 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1. Fuzzy inference anda defuzzification proc- 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
ess. 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
,
boundaries. Figure 2c shows a dimple defect Distance Measurement for Circles Model: Y-[A/(X+B)]+C
taken from actual experimental data. 25 ,Euclidean distance (1E-4) 00, ,Euclidean distance (Yaxis)
Using a fuzzy membership function. we
20t :
can describe the patterns closeness to a
circle. Our algorithm involves three stages.
First, we reject all badly-asymmetric pat-
terns. Then, among the remaining patterns,
006
0 005
0 004
tI
we use fuzzy membership to define the ac-
0 003
ceptance of distorted circular-shaped dim-
ples. Finally, decision-making based on the 0 002
membership of acceptance is performed
0 001
to identify dimple defects.
n
0 20 40 60 80
0 05 1 15 2 25 3 35 4 45 5
SelectinCqSymmetric und Ci~~cular~ Pattwns Area of pattern (X-axis)
Radius of circle
Discriminating badly-asymmetric patterns
from symmetric or slightly-asymmetric pat-
- Circles +Squares -a+ Average
K es I I Its
E.Y/XVi/?/('I 1trr I
Thc algorithms were implemented on a 386
machine. Test pattern images werc used to
verify the algorithms. Then an %bit wafer
image with 256 x 240 resolution was used
as input data. Resolution reduction of the
input image was first performed before the
preprocessing. Fuzzy membership func-
tions wcrc evaluated according to the algo-
rithm given in the previou\ secrion. Fuzzy
decision-making was performed by using a
properly choscn threshold Ta. The experi-
mental results confirm our design. Figure 5
shows a processed wafer image. As one can
see, the dimples are detected.
Ream-Balancing System
In the \emiconductor manuiacturing indus-
try. many systems are highly dynamic. non-
l i n e a r , a n d c o n s t a n t l y under r a n d o m
disturbances. These includes gas distribu-
tion sy\tems, furnace\. and so on. Closed--
loop control, adaptive and model-based
control techniques have been used quite suc-
cessfully for controlling certain processes,
but not that effective for dealing with the
processes having high nonlinearity and ran-
dom disturbances. We recently developed a
fuzzy control demonstration system, called
a beam-balancing system I IO], to demon-
strate fuzzy control. 6. In-house beam balancing system and its block diagram
Circuits .8 Devices 33
-- - ~- ~
I - -_______- ~~ .
If angle I S +MEDIUM
+SMALL LARGE
+LARGE +MEDIUM -
,,+SMALL -~ SMALL-- -____ MEDIUM -
i-
LARGE-
+MEDIUM - +SMALL LARGE , LARGE -
+%MALL -SMALL L
-LARGE -
-MEDIUM L
-MEDIUM
-
-LARGE 4
~
-LARGE A
-LARGE -LARGE J
Then fur7y membership functions for A'i for fuzzy inference rules. we perform each control action for each inference rule is
each linguistic variable are defined. Mem- x U x / t o cover :ill the possible colnbina- dcterniined by the mi,/operation, as p c + =~
bership functions ~ D + I . .p o + anti
~ ~ D + Silrc tions of the inputs. Using the compositional iiiiii (,UD+L. po+L. pI+L f o r rule I . etc
,iloun tor .D+L, D+M.and D+s. respec-
F ~7 ~ rule of inference [ 1 1 1 and a connective Computing the control signal is done
tively. The membership functions for the "AND" operation, we have derived 216 in- through linguistic variables for each set of
negative part are the mirror images of these ference rules. The list of some rules are input signal from U x U x 1. Then defuzzifi-
functions. given in Table I . The membership value for cation is perlbrmed by the following center
of gravity algorithm:
1
SPEED Membership Functions I
A2 -82 ~ D D1
' 82 A2
E.vp,c~r.in/riir.s
ANGLE Membership Function
The prototyping system was built in-house.
It consist\ of ii host microprocessor Intel
80486. a joystick that allows human inter-
vention so that thc comparison between h u -
man control and fuzzy control can be made.
and a fuzzy logic control program stored in
7. Membership funcbons for D+L.D+M,and D+s, respectively the host computer. The algorithm for fuzzi-