VLSI CAD: Lecture 2.

1
Logic to Layout Computational Boolean
Algebra: Basics

Rob A. Rutenbar
University of Illinois
Computational Boolean Algebra…?
•  Background…
•  You’ve done Boolean algebra, hand manipulations, Karnaugh maps to simplify…
•  But this is not sufficient for real designs

•  Example: simplify Boolean function of 40 variables via Kmap

•  It has 1,099,511,627,776 squares
•  You could fit this on an American
style football field...
•  …but each Kmap square would
be just 60 x 60 microns!
•  There must be a better way…

Slide 2 © 2013, R.A. Rutenbar

Need a Computational Approach
•  Need algorithmic, computational strategies for Boolean stuff
•  Need to be able to think of Boolean objects as data structures + operators
•  What will we study?
•  Decomposition strategies
•  Ways of taking apart complex functions into simpler pieces
•  A set of advanced concepts, terms you need to be able to do this
•  Computational strategies
•  Ways to think about Boolean functions that let them be manipulated by programs
•  Interesting applications
•  When you have new tools, there are some useful new things to do

Slide 3 © 2013, R.A. Rutenbar

Advanced Boolean Algebra
•  Useful analogy to calculus...
•  You can represent complex functions like exp(x) using simpler functions
•  If you only get to use 1,x,x2,x3,x4,... as the pieces...
•  ...turns out exp(x) = 1 + x + x2/2! + x3/3! + …
•  In Calculus, we tell you the general formula, the Taylor series expansion

•  f(x) = f(0) + f’(0)/1! x + f’’(0)/2! x2 + f’’’(0)/3! x3 + …

•  If you take more math, you might find out several other ways:
•  If it’s a periodic function, can use a Fourier series

•  Question: Anything like this for Boolean functions?

Slide 4 © 2013, R.A. Rutenbar
Boolean Decompositions
•  Yes. Called the Shannon Expansion

Slide 5 © 2013, R.A. Rutenbar

Shannon Expansion
•  Suppose we have a function F(x1,x2, ..., xn)
•  Define a new function if we set one of the xi=constant
•  Example: F(x1, x2, ..., xi=1, ..., xn)
•  Example: F(x1, x2, ..., xi=0, ..., xn)

•  Easy to do one by hand

F(x,y,z) = xy + xz’ +y(x’z +z’)

F(x=1,y,z) = F(x,y=0,z) =

•  Note: this is a new function, that no longer depends on this

variable (var)
Slide 6 © 2013, R.A. Rutenbar
Shannon Expansion: Cofactors
•  Turns out to be an incredibly useful idea
•  Several alternative names and notations
•  Shannon Cofactor with respect to xi
•  Write F(x1, x2, ..., xi=1, ...xn) as:

•  Write F(x1, x2, ..., xi=0, ...xn) as:

•  Often write this as just F(xi=1) F(xi=0) which is easier to type

•  Why are these useful functions to get from F?

Slide 7 © 2013, R.A. Rutenbar

Shannon Expansion Theorem
•  Why we care: Shannon Expansion Theorem
•  Given any Boolean function F(x1, x2, ..., xn) and pick any xi in F( )’s inputs
F( ) can be represented as
F(x1, x2, …, xi, …..., xn) = xi • F(xi=1) + xi’ • F(xi=0)

•  Pretty easy to prove...

Slide 8 © 2013, R.A. Rutenbar

Shannon Expansion: Another View
a
b
• function
•
F F
•
z
same as

function
0 x F mux
0
1 sel
function
1 x F x

Slide 9 © 2013, R.A. Rutenbar

Shannon Expansion: Multiple Variables
•  Can do it on more than one variable, too
•  Just keep on applying the theorem
•  Example F(x,y,z,w) = x • F(x=1) + x ’ • F(x=0) expanded around x

Expand each F(x=1) = y • F(x=1, y=1) + y ’ • F(x=1,y=0)

cofactor
around y F(x=0) = y • F(x=1, y=1) + y ’ • F(x=1,y=0)
F(x,y,z,w) = xy • F(x=1,y=1) + x’y • F(x=0,y=1)
+ xy’ • F(x=1,y=0) + x’y’ • F(x=0,y=0)

= expanded around variables x and y

Slide 10 © 2013, R.A. Rutenbar
Shannon Cofactors: Multiple Variables
•  BTW, there is notation for these as well
•  Shannon Cofactor with respect to xi and xj
•  Write F(x1, x2, ..., xi=1, ..., xj=0, ..., xn) as Fxi xj’ or Fxi xj
•  Ditto for any number of variables xi, xj, xk, ...

•  Notice also that order does not matter: (Fx)y = (Fy)x = Fxy
•  For our example

F(x,y,z,w) = xy • Fxy + x’y • Fx’y + xy’ • Fxy’ + x’y’ • Fx’y’

•  Again, remember: each of the cofactors is a function, not a number

Fxy = F(x=1, y=1, z, w) = a Boolean function of z and w

Slide 11 © 2013, R.A. Rutenbar
VLSI CAD: Lecture 2.2
Logic to Layout Computational Boolean
Algebra: Boolean Difference

Rob A. Rutenbar
University of Illinois
Next Question: Properties of Cofactors

•  What else can you do with cofactors?
•  Suppose you have 2 functions F(X) and G(X), where X=(x1,x2,...xn)
•  Suppose you make a new function H, from F and G, say…
•  H = F
•  H = (F • G) ie, H(X) = F(X) • G(X)
•  H = (F + G) ie, H(X) = F(X) + G(X)

•  H = (F ⊕ G) ie, H(X) = F(X)⊕G(X)

•  Interesting question
•  Can you tell anything about H’s cofactors from those of F, G…?
(F•G)x = what? (F ’)x = what? etc.
Slide 13 © 2013, R.A. Rutenbar
Nice Properties of Cofactors
•  Cofactors of F and G tell you everything you need to know
•  Complements
•  (F’ )x =
•  In English: cofactor of complement is complement of cofactor

•  Binary boolean operators

•  (F • G)x = Fx • Gx cofactor of AND is AND of cofactors
•  (F + G)x = Fx + Gx cofactor of OR is OR of cofactors

•  (F ⊕ G)x = Fx ⊕ Gx cofactor of EXOR is EXOR of cofactor

•  Very useful: can often help in getting cofactors of complex formulas

Slide 14 © 2013, R.A. Rutenbar
Combinations of Cofactors
•  OK, now consider operations on cofactors themselves
•  Suppose we have F(X), and get Fx and Fx’
•  Fx ⊕ Fx = ?
’
•  Fx • Fx’ = ?
•  Fx + Fx’ =?
•  Turns out these are all useful new functions
•  Indeed – they even have names!

•  Next: let’s go look at these interesting, useful new things

•  First up: the EXOR of the cofactors

Slide 15 © 2013, R.A. Rutenbar

Calculus Revisited: Derivatives
•  Remember way back to how you defined derivatives?
•  Suppose you have y = f(x)
y=f(x)
defined as slope of curve as
a function of point x

Considered f(x + Δ) - f(x) f(x+Δ)

Δ
f(x)

x x+Δ

Let Δ go to 0 in the limit and you got df(x)
Slide 16 dx © 2013, R.A. Rutenbar
Boolean Derivatives
•  So, do Boolean functions have “derivatives”…?
•  Actually, yes. Trick is how to define them...

•  Basic idea
•  For real-valued f(x), df/dx tell how f changes when x changes
•  For 0,1-valued Boolean function, we cannot change x by small delta
•  Can only change 0 ßà 1, but can still ask how f changes with x ...

Boolean f(x): Compares value of f( ) when

x=0 against when x=1 ;
∂f/∂x = ==1 just if these are different

Slide 17 © 2013, R.A. Rutenbar

It’s Got a Name: Boolean Difference
•  Hey, we have seen these pieces before!
•  ∂f/∂x = exor of the Shannon cofactors with respect to x
•  But… for Boolean variables, it’s usually written with the “∂” symbol
•  It also behaves sort of like regular derivatives...
•  Order of variables (vars) does not matter
•  ∂f / ∂x∂y = ∂f / ∂y∂x
•  Derivative of exor is exor of derivatives

•  ∂( f ⊕ g ) / ∂x = ∂f/∂x ⊕ ∂g/∂x
•  If function f is actually constant (f=1 or f=0, always, for all inputs)
•  ∂f/∂x = 0 for any x

Slide 18 © 2013, R.A. Rutenbar

Boolean Difference
•  But some things are just more complex, though...
•  Derivatives of (f •g) and (f + g) do not work the same...

∂ ⎡ ∂g ⎤ ⎡ ∂ f ⎤ ⎡ ∂ f ∂g ⎤
( f • g) = ⎢ f • ⎥ ⊕ ⎢g • ⎥ ⊕ ⎢ • ⎥
∂x ⎣ ∂x ⎦ ⎣ ∂x ⎦ ⎣ ∂x ∂x ⎦
∂ ⎡ ∂g ⎤ ⎡ ∂ f ⎤ ⎡ ∂ f ∂g ⎤
( f + g) = ⎢ f • ⎥ ⊕ ⎢g • ⎥ ⊕ ⎢ • ⎥
∂x ⎣ ∂x ⎦ ⎣ ∂x ⎦ ⎣ ∂x ∂x ⎦
•  Why?
•  Because AND and OR on Boolean values do not always behave like
ADDITION and MULTIPLICATION on real numbers

Slide 19 © 2013, R.A. Rutenbar

Boolean Difference: Gate-level View
•  Try the obvious “simple” examples for ∂f/∂x

x x
f f
y

x x
f y f
y

Meaning: when ∂f/∂x = 1, then f changes if x changes(!)

Slide 20 © 2013, R.A. Rutenbar
Interpreting the Boolean Difference
a
b
c
• A BIG F(a,b,c,....,w, x)
• BLOB OF
•
LOGIC
w
x
When ∂F / ∂x (a,b,c, ..., w) = 1, it means that ...

If you apply a pattern of other inputs (not X) that makes ∂F / ∂x = 1,

Any change in X will force a change in output F( )

Slide 21 © 2013, R.A. Rutenbar

Boolean Difference: Example
a 1 bit s sum s = a ⊕ b ⊕ cin
b
Cin full add Cout Cout = ab + (a + b)Cin

∂Cout / ∂Cin = ?

Slide 22 © 2013, R.A. Rutenbar

Boolean Difference
•  Things to remember about Boolean Difference
•  Not like the physical interpretation of the ordinary calculus derivative
(ie, no “slope of the curve” sort of stuff)…
•  …but it explains how an input-change can cause output-change for a Boolean F( )

•  ∂f / ∂x is another Boolean function, but it does not depend on x

•  It cannot; it is made out of the cofactors with respect to (“wrt”) x
•  …and they eliminate all the x and x’ terms by setting them to constants

•  Surprisingly useful (we will see more, later…)

Slide 23 © 2013, R.A. Rutenbar
VLSI CAD: Lecture 2.3
Logic to Layout Computational Boolean
Algebra: Quantification
Operators
Rob A. Rutenbar
University of Illinois
Computational Boolean Algebra, Cont…
•  What you know
•  Shannon expansion lets you decompose a Boolean function
•  Combinations of cofactors do interesting things, e.g., the Boolean difference

•  What you don’t know

•  Other combinations of cofactors that do useful things
•  The big ones: Quantification operators (this lecture)
•  Applications: Being able to do something impressive (next lecture)

Slide 25 © 2013, R.A. Rutenbar

AND: Fx • Fx’ is Universal Quantification
•  Have F(x1, x2, …, xi, … xn)
•  AND cofactors: Fxi • Fxi’
•  Name: Universal Quantification of
function F with respect to (wrt) x F(x=0)
variable xi 0
AND
(∀xi F) [x1,x2, …, xi-1, xi+1, … xn]
1 x F(x=1)

•  “(∀xi F)” is a new function

•  Yes, the “∀” sign is the “for all”
symbol from logic (predicate calculus)
•  And, it does not depend on xi…
Slide 26 © 2013, R.A. Rutenbar
OR: Fx + Fx’ is Existential Quantification
•  Have F(x1, x2, …, xi, … xn)
•  OR the cofactors: Fxi +Fxi’
•  Name: Existential Quantification of
function F wrt variable xi 0 x F(x=0)
•  (∃xi F) [x1, x2, …, xi-1, xi+1, … xn] OR

1 x F(x=1)
•  “(∃xi F)” is a new function
•  “∃” sign is “there exists” from logic;
and function also does not depend on xi

Note, like anything involving cofactors, both these new functions do not depend on xi
So: (∀xi F) and (∃xi F) both omit xi
Slide 27 © 2013, R.A. Rutenbar
Quantification Notation Makes Sense…
(∀x F) ( all original vars but x) ==1 when?
0 x F(x=0)

AND

1 x F(x=1)

(∃x F) ( all original vars but x) ==1 when?

0 x F(x=0)

OR

1 x F(x=1)

Slide 28 © 2013, R.A. Rutenbar

Extends to More Variables in Obvious Way
•  Additional properties
•  Like Boolean difference, can do with respect to more than 1 var
•  Suppose we have F(x,y,z,w)
•  Example: (∀xy F)[z,w] = (∀x (∀y F)) = Fxy • Fx’y • Fxy’ • Fx’y’
•  Example: (∃xy F)[z,w] = (∃x (∃y F)) = Fxy + Fx’y + Fxy’ + Fx’y’

•  Remember!
•  (∀x F), (∃x F), and ∂F / ∂x are all functions...
•  ..but they are functions of all the vars except x
•  We got rid of variable x and made 3 new functions
Slide 29 © 2013, R.A. Rutenbar
Quantification Example
•  Consider this circuit, it adds X=0 or X=1 to a 2-bit number A1A0
•  It’s just a 2-bit adder, but instead of B1B0 for the second operand, it is just 0X
•  It produces a carry out called C
and also has a carry in called D A1 0 A0 X

C a1 b1 a0 b0 D
adder
•  What is (∀A1,A0 C)[X,D]…?
•  A function of only X,D. Makes a 1 for values of X,D that make carry C=1 for
all values of operand input A1A0, i.e., makes a carry C=1 for all values of A1A0
•  What is (∃A1,A0 C)[X,D]…?
•  A function of just X,D. Makes a 1 for values of X,D that make carry C=1, for
some value of A1A0, i.e., there exists some A1A0 that, for this X,D makes C=1
Slide 30 © 2013, R.A. Rutenbar
 Quantification Example
A1 0 A0 X
Do the math: C = A1A0X + A1(A0+X)D
C a1 b1 a0 b0
adder
D Need all 4 CA1A0 CA1’A0 CA1A0’ CA1’A0’
cofactors:

•  Compute (∀A1,A0 C)[X,D] •  Compute (∃A1,A0 C)[X,D]

•  CA1A0 • CA1’A0 • CA1A0’ • CA1’A0’ •  CA1A0 + CA1’A0 + CA1A0’ + CA1’A0’ BUG fix.
C=1 for
some
values
of A1,A0
•  In words: No values of X,D that •  In words: Yes, if at least one of
make C=1 independent of A1,A0 X,D =1 à C=1 for some A1,A0
Slide 31 © 2013, R.A. Rutenbar
VLSI CAD: Lecture 2.4
Logic to Layout Computational Boolean
Algebra: Application to
Logic Network Repair
Rob A. Rutenbar
University of Illinois
Quantification App: Network Repair
•  Suppose ...
•  I specified a logic block for you to implement... f(a,b) = ab + b’
•  …but you implemented it wrong: in particular, you got ONE gate wrong

Specified Got
b
ab + b’ f(a,b) b ?? g(a,b)
a a
•  Goal
•  Can we deduce how precisely to change this gate to restore correct function?
•  Lets go with this very trivial test case to see how mechanics work...
Slide 33 © 2013, R.A. Rutenbar
 Network Repair
•  Clever trick
•  Replace our suspect gate by a
4:1 mux with 4 arbitrary new vars ?? g(a,b)
b
•  By cleverly assigning values to a
d0 d1 d2 d3, we can fake any gate
•  Question is: what are the right values
of d’s so g is repaired (==f) Replace with

b
a s1 s0
d0 00
d1
d2
01 mux G(a,b,d0,d1,d2,d3)
10
d3 11

Slide 34 © 2013, R.A. Rutenbar

Aside: Faking a Gate with a MUX
•  Remember...
•  You can do any function of 2 vars with one 4 input multiplexor (MUX)

y
x s1 s0
0 00
0 01
mux
0 10
1 11

y
x s1 s0
0 00
1 01
mux
1 10
0 11
Slide 35 © 2013, R.A. Rutenbar
Network Repair: Using Quantification
•  Next trick
•  Make new function Z(a,b,d0,d1,d2,d3) that =1 just when G == f

b
a s1 s0
d0 00
d1 01 G(a,b,d0,d1,d2,d3)
d2 mux
10
d3 11
exnor
(==) Z(a,b,d0,d1,d2,d3)

b f(a,b)
a
known good version of logic
Slide 36 © 2013, R.A. Rutenbar
Using Quantification
•  What now?
•  Think hard about exactly what we want:

•  Want to know values for (d0 d1 d2 d3) that,

independent of whatever a, b are, force Z(a,b,d0,d1,d2,d3) = 1

•  But this is something we have seen!

•  Universal quantification of function Z wrt variables a,b!
•  Any pattern of (d0 d1 d2 d3) that makes (∀ab Z)(d0,d1,d2,d3)==1 will do it!
•  (Aside: do you know where a, b went??)
Slide 37 © 2013, R.A. Rutenbar
 Network Repair via Quantification: Try It…
d0s1’s0’ + d1s1’s0 + d2s1s0’ + d3s1s0
b
a s1 s0
d0 00

d1 01
mux G(a,b,d0,d1,d2,d3)
d2 10
d3 11

exnor Z(a,b,d0,d1,d2,d3)
(==)

b f(a,b)
a

ab+b’

Slide 38 © 2013, R.A. Rutenbar

 Network Repair via Quantification: Continued
G f Reminder:
Q exnor 0 = Q’
Z = [d0a’b + d1b’ + d2ab] ⊕ [ab +b’] = G exnor f
Q exnor 1 = Q
Use nice property: cofactor of exnor is exnor of cofactors!

Za’b’ = Ga’b’ ⊕ fa’b’ à set a=0,b=0 à

Za’b = Ga’b ⊕ fa’b à set a=0,b=1 à
Zab’ = Gab’ ⊕ fab’ à set a=1,b=0 à
Zab = Gab ⊕ fab à set a=1,b=1 à

Slide 39 © 2013, R.A. Rutenbar

Repair via Quantification: Continued
•  So, we got this: (∀ab Z)[d0,d1,d2,d3] = d0’ •d1•d2
•  And know this: if can solve (∀ab Z)[d0,d1,d2,d3]==1 à repaired!

•  Hey – this one is not very hard…

Slide 40 © 2013, R.A. Rutenbar

Network Repair
•  Does it work? What do these d’s represent?

b
d0= 00
a
d0 00
s1 s0
d1= 01 mux
d1
d2
01
10
mux G(a,b,d0,d1,d2,d3)
d2= 10
d3 11
d3= 11
exnor Z(a,b,d0,d1,d2,d3)
(==)

b f(a,b)
a

•  This example is tiny...

•  But in a real example, you have a big network-- 100 inputs, 50,000 gates
•  When it doesn’t work, it’s a major hassle to go thru in detail
•  This is a mechanical procedure to answer: Can we change 1 gate to repair?
Slide 41 © 2013, R.A. Rutenbar
Computational Boolean Alg Strategies
•  What haven’t we seen yet? Computational strategies
•  Example: find inputs to make (∀ab Z)(d0,d1,d2,d3)==1 for gate debug
•  This computation is called Boolean Satisfiability (also called SAT)

•  Ability to do Boolean SAT efficiently is a big goal for us

•  We will see how to do this in later lectures…

Slide 42 © 2013, R.A. Rutenbar

VLSI CAD: Lecture 2.5
Logic to Layout Computational Boolean
Algebra: Recursive
Tautology
Rob A. Rutenbar
University of Illinois
Important Ex Computation: Tautology
•  Let’s build a real computational strategy for a real problem
•  Tautology:
•  I will give you a representation – a data structure -- for a Boolean function f( )
•  You build an algorithm to tell– yes or no – if this function f( ) == 1 for every input

•  You might be thinking: Hey, how hard can that be…??

•  Very, very hard.
•  What happens if I give you a function with 50 variables…?
•  Turns out this is a great example: illustrates all the stuff we need to know...

Slide 44 © 2013, R.A. Rutenbar

Start with: Representation
•  We use a simple, early representation scheme for functions
•  Represent a function as a set of OR’ed product terms (i.e., a sum of products)
•  Simple visual: use a 3-var Boolean cube, with solid circles where f() = 1
•  So: each product term (circle in a Kmap) called a “cube” == 2k corners circled

111
011

001 101

c
b 010 110

a 000 100
Cube is just a “graph”
of a Boolean function,
but only vertices have
Slide 45 values for the function © 2013, R.A. Rutenbar
 Positional Cube Notation (PCN)
•  So, we say ‘cube’ and mean ‘product term’
•  So, how to represent each cube? PCN: one slot per variable, 2 bits per slot
•  Write each cube by just noting which variables are true, complemented, or absent
•  In slot for var x: put 01 if product term has ...x... in it
•  In slot for var x: put 10 if product term has ...x’... in it
•  In slot for var x: put 11 if product terms has no x or x’ in it

111
011

001 101

c 010 110
b
a 000 100
Slide 46 © 2013, R.A. Rutenbar
PCN Cube List = Our Representation
•  So, we represent a function as a cover of cubes (circle its 1’s)
•  This is a list of cubes (this is the ‘sum’) in positional cube notation (of products)

•  Ex: f(a,b,c)=a’ +bc +ab => [10 11 11], [11 01 01], [01 01 11]

BUG fix.
[10 11 11] 111
011
First slot is
for a’ 001 101

Other are c
don’t care b 010 110

a 000 100

Slide 47 © 2013, R.A. Rutenbar

Tautology Checking
•  How do we approach tautology as a computation?
•  Input = cube-list representing products in an SOP cover of f
•  Output = yes/no, f ==1 always or not

•  Cofactors to the rescue

•  Great result: f is a tautology if and only if fx and fx’ are both tautologies
•  This makes sense:
•  If function f() = 1 à then cofactors both obviously = 1
•  If both cofactors = 1 à x•F(x=1) + x’•F(x=0) = x•1 + x’•1 = x+x’ = 1

Slide 48 © 2013, R.A. Rutenbar

Recursive Tautology Checking
•  Suggests a recursive computation strategy:
•  If you cannot tell immediately that f==1 ...go try to see if each cofactor == 1 !

f=1?? split on x
(fx =1 AND fx’ =1)??
y/n y/n
fx=1?? fx’=1??
•  What else do we need here?
•  Selection rules: which x is good to pick to split on?
•  Termination rules: how do we know when to quit splitting, so we can
answer ==1 or !=1 for function at this node of tree?
•  Mechanics: how hard is it to actually represent the cofactors?

Slide 49 © 2013, R.A. Rutenbar

VLSI CAD: Lecture 2.6
Logic to Layout Computational Boolean Algebra:
Recursive Tautology—
URP Implementation
Rob A. Rutenbar
University of Illinois
Recursive Tautology Checking
•  So, we think a recursive computation strategy will do it:
•  If you cannot tell immediately that f==1 ...go try to see if each cofactor == 1 !

f=1?? split on x
(fx =1 AND fx’ =1)??
y/n y/n
fx=1?? fx’=1??
•  We need these:
•  Selection rules: which x is good to pick to split on?
•  Termination rules: how do we know when to quit splitting, so we can
answer ==1 or !=1 for function at this node of tree?
•  Mechanics: how hard is it to actually represent the cofactors?

Slide 51 © 2013, R.A. Rutenbar

Recursive Cofactoring
•  Do mechanics first (easy!). For each cube in your list:
•  If you want cofactor wrt var x=1, look at x slot in each cube:
•  [... 10 ...] => just remove this cube from list, since it’s a term with an x’
•  [... 01 ...] => just make this slot 11 == don’t care, strike the x from product term
•  [... 11 ...] => just leave this alone, this term doesn’t have any x in it
•  If you want cofactor wrt var x=0, look at x slot in each cube:
•  [...01 ...] => just remove this cube from list, since it’s a term with an x
•  [...10 ...] => just make this slot 11 == don’t care, strike the x’ from product term
•  [...11 ...] => just leave this alone, this term doesn’t have any x in it

f = abd+bc’ fa fc

[01 01 11 01]
[11 01 10 11]
Slide 52 © 2013, R.A. Rutenbar
Unate Functions
•  Selection / termination, another trick: Unate functions
•  Special class of Boolean functions
•  f is unate if a SOP representation only has each literal in exactly one polarity,
either all true, or all complemented
•  Ex: ab + ac’d + c’de’ is …
•  Ex: xy + x’y + xyz’ + z’ is ...

•  Terminology
•  f is positive unate in var x -- if changing x 0à1 keeps f constant or makes f: 0à1
•  f is negative unate in var x -- if changing x 0à1 keeps f constant or makes f: 1à0
•  Function that is not unate is called binate
Slide 53 © 2013, R.A. Rutenbar
Unate Functions
•  Analogous to monotone continuous functions

y
This function is monotone non-decreasing,
whenever x2 ≥ x1, then f(x2) ≥ f(x1)
x
•  Example: for a Boolean function f positive unate in x

Boolean
function f( ... x ...)
variable x
f

Slide 54 © 2013, R.A. Rutenbar

Can Exploit Unate Func’s For Computation
•  Suppose you have a cube-list for f
•  A cube-list is unate if each var in each cube only appears in one polarity, not both
•  Ex: f(a,b,c)=a +bc +ac à [01 11 11], [11 01 01], [01 11 01] is unate
•  Ex: f(a,b,c)=a +b’c +bc à [01 11 11], [11 10 01], [11 01 01] is not
•  Easier to see if draw vertically

a+b’c+ac UNATE a+b’c+bc NOT

[01 11 11] [01 11 11]

[11 10 01] [11 10 01]
[01 11 01] [11 01 01]
a
b b , b’
Slide 55 c © 2013, R.A. Rutenbar
Using Unate Functions in Tautology Checking
•  Beautiful result
•  It is very easy to check a unate cube-list for tautology
•  Unate cube-list for f is tautology iff it contains all don’t care cube = [11 ... 11]

•  Reminder: what exactly is [11 11 11 ... 11] as a product term?

[01 01 01] = abc [01 01 11] = ab [01 11 11] = a [11 11 11] =
cd
ab 00 01 11 10
•  This result actually makes sense... 00
•  Cannot make a “1” with only product terms 01
where all literals are in just one polarity. (Try it!) 11
10
Slide 56 © 2013, R.A. Rutenbar
So, Unateness Gives Us Termination Rules
•  We can look for tautology directly, if we have a unate cube-list
•  If match rule, know immediately if ==1, or not
•  Rule 1: ==1 if cube-list has all don’t care cube [11 11 ... 11]
Why: function at this leaf is (stuff + 1 + stuff) == 1
•  Rule 2: !=1 if cube-list unate and all don’t care cube missing
Why: unate ==1 if and only if has [11 11 ... 11] cube

f==1? f==1?

split split
fx=1?? fx=1??
... ==1 YES 01 11 10 ==1 NO
11 11 ... 11 Rule 1 11 01 10 Rule 2 - It’s unate but
... 01 01 10 all don’t care
cube is not here
Slide 57 © 2013, R.A. Rutenbar
Recursive Tautology Checking
•  Lots more possible rules...
•  Rule 3: ==1 if cube list has single var cube that appears in both polarities
Why: function at this leaf is (stuff + x + x’ + stuff) == 1

•  You get the idea...

f==1?

fx=1?? split

01 11 11 ==1 YES
10 11 11 Rule 3 -
11 11 10 = ... + x + x’ + ...

Slide 58 © 2013, R.A. Rutenbar

Recursive Tautology Checking
•  But can’t use easy termination rules unless unate cubelist
•  Selection rule...? Pick splitting var to make unate cofactors!
•  Strategy: pick “most not-unate” (binate) var as split var
•  Pick binate var with most product terms dependent on it (Why? Simplify more cubes)
•  If a tie, pick var with minimum | true var - complement var | (L-R subtree balance)
x y z w
A 01 01 01 01 (If no binate variables? Pick a unate
B 10 11 01 01 variable that appears in most cubes)
C 10 11 11 10
D 01 01 11 01
binate, in 4 cubes, | true - compl | = 3-1 = 2
unate
unate
binate, in 4 cubes, | true - compl | = 2-2 = 0
Slide 59 © 2013, R.A. Rutenbar
Recursive Tautology Checking: Done!
•  Algorithm tautology(f represented as cubelist) {
/* check if we can terminate recursion */
if (f is unate) {
apply unate tautology termination rules directly
if (==1) return (1)
else return (0)
}
else if (any other termination rules, like rule 3, work?) {
return the appropriate value if ==1 or ==0
}
else { /* can’t tell from this -- find splitting variable */
x = most-not-unate variable in f
return ( tautology( fx ) && tautology( fx’ ) )
}
Slide 60 } © 2013, R.A. Rutenbar
 Recursive Tautology Checking: Example
•  Tautology example: f = ab + ac + ab’c’ + a’
Dear Courserians: a b c
ab 01 01 11 All vars are binate, var a
Small bug fix to “c’ ” ac 01 11 01 affects most implicants, so
label on the side ab’c’ 01 10 10 split on a...
a’ 10 11 11
fa fa’

Same bug fix to “c’ ” 0b [ ]

label for this cube 0c [ ]
b 11 01 11 0b’c’ [ ]
c 11 11 01 0’=1 [11 11 11]
b’c’ 11 10 10
= tautology YES
Same bug fix:
it’s “c’ ” not “c “
Slide 61 © 2013, R.A. Rutenbar
Recursive Tautology Checking: Example
b 11 01 11
No unate vars yet, no simplifications.
c 11 11 01
Let’s split again, now on b.
b’c’ 11 10 10
fab fab’

‘1’ 11 11 11 c 11 11 01
c 11 11 01 c’ 11 11 10

f
•  So we are done:
fa fa’ ==1
•  Our tree has tautologies at all leaves!
•  Note -- if any leaf !=1, then f != 1 too,
this is how tautology fails fab ==1 fab’ ==1
Slide 62 © 2013, R.A. Rutenbar
Computational Boolean Algebra
•  Computational philosophy revisited
•  Strategy is so general and useful it has a name: Unate Recursive Paradigm
•  Abbreviated usually as “URP”

•  Summary
•  Cofactors and functions of cofactors are important and useful
•  Boolean difference, quantifications; real applications like network repair
•  Representations (data structures) for Boolean functions are critical
•  Truth tables, Kmaps, equations cannot be manipulated by software

•  Saw one real representation: Cube-list, positional cube notation

Slide 63 © 2013, R.A. Rutenbar