Digital Logic &
Verilog Design

$f = x_1 \cdot x_2 + \overline{x_1} \cdot x_3$

$f = (a+b) \cdot (c+d)$

$\therefore \boxed{\overline{f} = \bar{a}\bar{b} + \bar{c}\bar{d}}$

$f(x_1, x_2, x_3) = \sum m(1,2,5,6,7)$

Canonical SOP (sum of minterms) — convert each minterm number to binary, complement 0s:

Canonical POS (product of maxterms, missing minterms = 0,3,4) — convert each missing minterm to binary, then complement each bit to form an OR term:

K-map simplification:

       x₂x₃
x₁  00 01 11 10
 0:  0  1  0  1
 1:  0  1  1  1

Groups:

• $\{1,5\}$ (column 01, both rows) → $\overline{x_2}\,x_3$ — only $x_1$ changes, eliminated
• $\{2,6\}$ (column 10, both rows) → $x_2\,\overline{x_3}$ — only $x_1$ changes, eliminated
• $\{5,7\}$ (row 1, columns 01 & 11) → $x_1\,x_3$ — only $x_2$ changes, eliminated

Minimized SOP: $\boxed{f = \overline{x_2}\,x_3 + x_2\,\overline{x_3} + x_1\,x_3}$

(Canonical SOP had 5 terms / 15 literals → K-map reduced to 3 terms / 7 literals)

$f = \sum m(0,1,2,5,6,7)$. K-map:

       x₂x₃
x₁  00 01 11 10
 0:  1  1  0  1
 1:  0  1  1  1

Groups:

• $\{0,2\}$ → $\overline{x_1}\,\overline{x_3}$
• $\{1,5\}$ → $\overline{x_2}\,x_3$
• $\{6,7\}$ → $x_1\,x_2$

Minimized SOP: $\boxed{f = \overline{x_1}\,\overline{x_3} + \overline{x_2}\,x_3 + x_1\,x_2}$

$f = \sum m(0,2,5,7,8,10,13,15)$. 4-variable K-map:

        x₃x₄
x₁x₂  00 01 11 10
  00:   1  0  0  1
  01:   0  1  1  0
  11:   0  1  1  0
  10:   1  0  0  1

Groups:

• $\{0,2,8,10\}$ → $\overline{x_2}\,\overline{x_4}$
• $\{5,7,13,15\}$ → $x_2\,x_4$

Minimized SOP: $\boxed{f = \overline{x_2}\,\overline{x_4} + x_2\,x_4 = x_2 \odot x_4 \;\text{(XNOR)}}$

$f = \sum m(1,3,4,6,9,11,12,14) + D(0,5)$. 4-var K-map with don't cares at 0, 5:

        x₃x₄
x₁x₂  00 01 11 10
  00:   d  1  1  0
  01:   1  d  0  1
  11:   1  0  0  1
  10:   0  1  1  0

Groups:

• $\{1, 3, 9, 11\}$ (rows 00 & 10, cols 01 & 11) → $\overline{x_2}\,x_4$ — $x_2=0$, $x_4=1$ constant
• $\{4, 6, 12, 14\}$ (rows 01 & 11, cols 00 & 10) → $x_2\,\overline{x_4}$ — $x_2=1$, $x_4=0$ constant

These two quads cover all 8 minterms without needing don't-cares.

Minimized SOP: $\boxed{f = \overline{x_2}\,x_4 + x_2\,\overline{x_4} = x_2 \oplus x_4 \;\text{(XOR)}}$

Functional completeness requires implementing NOT and AND (or NOT and OR).

NOT from NAND:

AND from NAND:

OR from NAND:

Since $\{\text{NOT}, \text{AND}\}$ is functionally complete and both are realizable with NAND only, NAND is functionally complete. $\blacksquare$

3-way light: output changes when any switch toggles → XOR of three inputs.

$f = \sum m(1,2,4,7)$. Min SOP: $\boxed{f = x_1 \oplus x_2 \oplus x_3}$

(No simpler SOP exists for XOR.)

module mux2to1 (
    input  wire s, d0, d1,
    output reg  y
);
    always @(*) begin
        if (s)
            y = d1;
        else
            y = d0;
    end
endmodule

Also valid: assign f = ~s & w0 | s & w1; — the Boolean mux equation directly as a continuous assignment.

Majority of 3 inputs: output 1 when 2 or more inputs are 1.

$f = \sum m(3,5,6,7)$. K-map gives: $\boxed{f = ab + ac + bc}$

Circuit: three AND gates for $ab$, $ac$, $bc$; one 3-input OR gate.

• Minterm: A product term in which every variable appears exactly once (complemented or uncomplemented); evaluates to 1 for exactly one input combination. e.g. $m_5 = x_1\overline{x_2}\,x_3$ equals 1 only at $(1,0,1)$.
• Maxterm: A sum term in which every variable appears exactly once; evaluates to 0 for exactly one input combination. e.g. $M_0 = x_1 + x_2 + x_3$ equals 0 only at $(0,0,0)$.
• Implicant: A product term that covers only rows where f=1 (never a 0-row). e.g. $x_1 x_3$ covers $\{m_5, m_7\}$, both $f=1$.
• Prime implicant (PI): An implicant that cannot be combined with another implicant to eliminate a variable (cannot be enlarged). e.g. $x_3$ covers $\{m_1,m_3,m_5,m_7\}$; $x_1 x_3$ is not prime since it is contained in $x_3$.
• Essential PI: A prime implicant that is the only one covering at least one minterm; must be in every minimum cover. e.g. $\overline{x_1}\,x_2$ is essential because $m_2$ is covered only by it.

中文版：

• 最小項 (Minterm)：乘積項，每個變數恰好出現一次（原變數或補數）；僅在一組輸入組合下為 1。
• 最大項 (Maxterm)：和項，每個變數恰好出現一次；僅在一組輸入組合下為 0。
• 蘊涵項 (Implicant)：一個乘積項，只涵蓋 f=1 的列（不涵蓋任何 f=0 的列）。
• 質蘊涵項 (Prime Implicant)：無法再與其他蘊涵項合併以消去變數的蘊涵項（無法再擴大）。
• 必要質蘊涵項 (Essential PI)：唯一涵蓋某個最小項的質蘊涵項；在任何最小覆蓋中都必須包含。

Example 範例： $f(x_1,x_2,x_3) = \sum m(1,2,3,5,7)$

• Minterm 最小項： $m_5 = x_1\overline{x_2}\,x_3$ — only equals 1 at row $(1,0,1)$.
  僅在 $(1,0,1)$ 時為 1。
• Maxterm 最大項： $M_0 = x_1 + x_2 + x_3$ — only equals 0 at row $(0,0,0)$.
  僅在 $(0,0,0)$ 時為 0。
• Implicant 蘊涵項： $x_1 x_3$ covers $\{m_5, m_7\}$, both have $f=1$. A single minterm like $m_3$ is also an implicant.
  $x_1 x_3$ 涵蓋 $\{m_5, m_7\}$，兩者皆 $f=1$。單一最小項如 $m_3$ 也是蘊涵項。
• Prime Implicant 質蘊涵項： $x_3$ covers $\{m_1,m_3,m_5,m_7\}$ — cannot be enlarged further. $\overline{x_1}\,x_2$ covers $\{m_2,m_3\}$ — cannot be enlarged. But $x_1 x_3$ is not prime because it is contained within the larger PI $x_3$.
  $x_3$ 涵蓋 $\{m_1,m_3,m_5,m_7\}$，無法再擴大。$\overline{x_1}\,x_2$ 涵蓋 $\{m_2,m_3\}$，無法再擴大。而 $x_1 x_3$ 不是質蘊涵項，因為它被更大的 $x_3$ 包含。
• Essential PI 必要質蘊涵項： $\overline{x_1}\,x_2$ is essential — $m_2$ is covered only by this PI. $x_3$ is essential — $m_1, m_5, m_7$ are covered only by it.
  $\overline{x_1}\,x_2$ 是必要的，因為 $m_2$ 只被它涵蓋。$x_3$ 是必要的，因為 $m_1, m_5, m_7$ 只被它涵蓋。

Minimum SOP 最簡積項和：$\boxed{f = x_3 + \overline{x_1}\,x_2}$

$f_1 = \sum m(0,2,3,4,5,7)$, $f_2 = \sum m(1,2,3,5,6,7)$.

K-map for $f_1$:

       x₂x₃
x₁  00 01 11 10
 0:  1  0  1  1
 1:  1  1  1  0

No quads (only 6 ones, no 4-in-a-row/2×2 block of all 1s). Three pairs cover all six 1s:

• $\{m_0, m_4\}$ (column 00) → $\overline{x_2}\,\overline{x_3}$
• $\{m_2, m_3\}$ (top row, cols 10–11) → $\overline{x_1}\,x_2$
• $\{m_5, m_7\}$ (bottom row, cols 01–11) → $x_1 x_3$

$\boxed{f_1 = \overline{x_2}\,\overline{x_3} + \overline{x_1}\,x_2 + x_1 x_3}$

K-map for $f_2$:

       x₂x₃
x₁  00 01 11 10
 0:  0  1  1  1
 1:  0  1  1  1

Two essential quads cover all six 1s:

• $\{m_1, m_3, m_5, m_7\}$ (cols 01 ∪ 11, all $x_3{=}1$) → $x_3$
• $\{m_2, m_3, m_6, m_7\}$ (cols 11 ∪ 10, all $x_2{=}1$) → $x_2$

$\boxed{f_2 = x_2 + x_3}$

Shared-gate implementation: Note that $\overline{x_2}\,\overline{x_3} = \overline{(x_2 + x_3)} = \overline{f_2}$. So one OR gate computes $f_2$, and its inverted output supplies the $\overline{x_2}\,\overline{x_3}$ term of $f_1$:

1. $T = x_2 + x_3$   (OR2)  — this is $f_2$
2. $\overline{T}$   (NOT)
3. $A = \overline{x_1}\,x_2$   (AND2)
4. $B = x_1 x_3$   (AND2)
5. $f_1 = \overline{T} + A + B$   (OR3)

Cost (gates + total inputs, complemented literals free):

• Independent: $f_1$ = 4 gates + 9 inputs = 13; $f_2$ = 1 + 2 = 3 → 16
• Shared: 5 gates + 10 inputs = 15  (saving 1 by reusing the $x_2{+}x_3$ OR-gate)

Prove $x + x \cdot y = x$:

This is the absorption law.

$f = a \cdot b + c$. Convert to NAND-NAND using double inversion:

Implementation:

• Gate 1: $\text{NAND}(a, b) = \overline{ab}$
• Gate 2: $\text{NAND}(c, c) = \overline{c}$ (acts as NOT)
• Gate 3: $\text{NAND}(\overline{ab},\; \overline{c}) = \overline{\overline{ab} \cdot \overline{c}} = ab + c = f$

Three NAND gates total.

f = (a AND b) OR (NOT a AND c)

module structural_f (
    input  wire a, b, c,
    output wire f
);
    wire not_a, and1, and2;
    not  g0 (not_a, a);
    and  g1 (and1, a, b);
    and  g2 (and2, not_a, c);
    or   g3 (f, and1, and2);
endmodule

Distributive law: $x \cdot (y + z) = x \cdot y + x \cdot z$

Draw three overlapping circles for $x, y, z$ inside a universe rectangle.

LHS: $x \cdot (y+z)$ = region in $x$ AND (in $y$ OR in $z$) $= (x \cap y) \cup (x \cap z)$.

RHS: $x \cdot y + x \cdot z = (x \cap y) \cup (x \cap z)$.

Both expressions represent the same shaded area in the Venn diagram. $\blacksquare$

$f = \prod M(0,1,4,5,9,11,15)$. Zeros at minterms 0,1,4,5,9,11,15.

4-var K-map for $f=0$ positions, group the 0s for POS:

• $\{0,1,4,5\}$ → $(x_1 + x_2)$
• $\{9,11\}$ → $(\overline{x_1} + x_2 + \overline{x_4})$
• $\{15\}$ → $(\overline{x_1} + \overline{x_2} + \overline{x_3} + \overline{x_4})$

Min POS: $\boxed{f = (x_1+x_2)(\overline{x_1}+\overline{x_2}+x_4)(\overline{x_1}+\overline{x_2}+\overline{x_3})}$

$f = x_1 x_2 + \overline{x_1}\,x_3 + x_2 x_3$

Literal count: $x_1 x_2$ (2) + $\overline{x_1}\,x_3$ (2) + $x_2 x_3$ (2) = 6 literals.

Gate count (2-level): 3 AND gates (2-input each) + 1 OR gate (3-input) = 4 gates.

Cost (gates + inputs): $3 \times 2 + 1 \times 3 = 9$ gate inputs + 4 gates = cost of 9.

Note: $x_2 x_3$ is a redundant consensus term; minimized: $f = x_1 x_2 + \overline{x_1}\,x_3$ (cost = 5).

Circuit: $g = \overline{a}$; $h = a \cdot b$; $f = g + h$

$\boxed{f = \overline{a} + b}$ (equivalent to $a \rightarrow b$, implication).

Structural XOR:

module xor_struct (input a, b, output f);
    wire na, nb, t1, t2;
    not g0(na, a);
    not g1(nb, b);
    and g2(t1, a, nb);
    and g3(t2, na, b);
    or  g4(f, t1, t2);
endmodule

Behavioral XOR:

module xor_behav (input a, b, output f);
    assign f = a ^ b;
endmodule

Structural explicitly instantiates gates; behavioral describes function mathematically and lets synthesis choose implementation.

From 2.6: $f = \overline{x_2}\,\overline{x_4} + x_2\,x_4$

module f_2_6 (
    input  wire x1, x2, x3, x4,
    output wire f
);
    assign f = (~x2 & ~x4) | (x2 & x4);
endmodule

module tb;
    reg  x1,x2,x3,x4;
    wire f;
    f_2_6 uut(.x1(x1),.x2(x2),.x3(x3),.x4(x4),.f(f));
    integer i;
    initial begin
        $display("x1 x2 x3 x4 | f");
        for (i=0; i<16; i=i+1) begin
            {x1,x2,x3,x4} = i[3:0];
            #10;
            $display("%b  %b  %b  %b  | %b", x1,x2,x3,x4,f);
        end
        $finish;
    end
endmodule

Alarm = 1 when 2 or more of 4 sensors active. $f = \sum m(3,5,6,7,9,10,11,12,13,14,15)$.

4-var K-map groups:

• $\{3,7,11,15\}$ → $x_3 x_4$
• $\{5,7,13,15\}$ → $x_2 x_4$
• $\{6,7,14,15\}$ → $x_2 x_3$
• $\{9,11,13,15\}$ → $x_1 x_4$
• $\{10,11,14,15\}$ → $x_1 x_3$
• $\{12,13,14,15\}$ → $x_1 x_2$

All 6 prime implicants are essential. Minimal SOP:

$f = \sum m(0,1,2,3,4,6,8,9,12)$. 4-var K-map, convert to NAND-NAND.

K-map groups:

• $\{0,1,2,3\}$ → $\overline{x_1}\,\overline{x_2}$
• $\{0,1,8,9\}$ → $\overline{x_2}\,\overline{x_3}$
• $\{0,4,8,12\}$ → $\overline{x_3}\,\overline{x_4}$
• $\{4,6\}$ → $\overline{x_1}\,x_2\,\overline{x_4}$

Minimal SOP: $\boxed{f = \overline{x_1}\,\overline{x_2} + \overline{x_2}\,\overline{x_3} + \overline{x_3}\,\overline{x_4} + \overline{x_1}\,x_2\,\overline{x_4}}$

To prove or disprove $f = g$ for 5-variable functions, build truth tables (32 rows each) or use algebraic manipulation / BDD construction.

Method: Compute $h = f \oplus g$. If $h = 0$ for all 32 input combinations, $f = g$; otherwise find a counterexample row where $h = 1$.

In practice: represent both as BDDs with the same variable order. If their ROBDD roots point to the same node (canonical form), they are equivalent; otherwise inequivalent.

The truth table is

Expand the LHS using distributivity: $(x_1+\overline{x_3})(\overline{x_1}+x_3) = x_1\overline{x_1} + x_1 x_3 + \overline{x_1}\,\overline{x_3} + \overline{x_3} x_3 = 0 + x_1 x_3 + \overline{x_1}\,\overline{x_3} + 0 = x_1 x_3 + \overline{x_1}\,\overline{x_3}$, which equals the RHS. (This is the XNOR identity.)

Group the LHS as $(\overline{x_1}+x_1) x_3 \cdot \text{(impossible)}$… better: combine pairs by the variable $x_3$. $\overline{x_1}\,\overline{x_3} + x_1 x_3$ is one XNOR; $\overline{x_2}x_3 + x_2\overline{x_3}$ is one XOR. These two cancel some terms when expanded with $x_2$/$x_1$, and applying $x+\overline{x}=1$ and absorption gives the simplest expression $f = x_1 \oplus x_2 \cdot 0 + \dots = \overline{x_1\oplus x_2} + x_1\overline{x_2}$, equivalently the RHS $x_1 x_2 + \overline{x_1}\,\overline{x_2} + x_1\overline{x_2}$. By absorption $x_1\overline{x_2} + x_1 x_2 = x_1$, so the simplest form is $f = x_1 + \overline{x_1}\,\overline{x_2} = x_1 + \overline{x_2}$.

Draw three overlapping circles for $x,y,z$. The set $y \cdot z$ is the lens-shaped intersection of $y$ and $z$; $x + y\cdot z$ shades all of $x$ plus that lens. On the right side, $x+y$ shades $x \cup y$ and $x+z$ shades $x \cup z$; their intersection $(x+y)(x+z)$ contains every point that lies in $x$ OR in both $y$ and $z$, which is the same shaded region. Hence the equality holds.

In a Venn diagram with two circles $x$ and $y$, the region $x\cdot y$ is their intersection. $\overline{x\cdot y}$ is everything outside that intersection. $\overline{x}$ is everything outside circle $x$ and $\overline{y}$ is everything outside circle $y$; their union $\overline{x}+\overline{y}$ covers exactly the same region (everything not in both circles). Therefore $\overline{x\cdot y} = \overline{x}+\overline{y}$.

Specification: $f = s_3 + s_1 s_2$ (reject if too large, OR both too light and too small). Truth table has $f=1$ for $(s_1 s_2 s_3) \in \{001,011,100,101,110,111\}$. Canonical SOP: $f = \sum m(1,3,4,5,6,7)$. Algebraic minimization: $f = \overline{s_1}\overline{s_2}s_3 + \overline{s_1}s_2 s_3 + s_1\overline{s_2}\overline{s_3} + s_1\overline{s_2}s_3 + s_1 s_2\overline{s_3} + s_1 s_2 s_3$. Combining gives $\overline{s_1}s_3(\overline{s_2}+s_2) + s_1\overline{s_2}(\overline{s_3}+s_3) + s_1 s_2(\overline{s_3}+s_3) = \overline{s_1}s_3 + s_1\overline{s_2} + s_1 s_2$. Then $s_1\overline{s_2}+s_1 s_2 = s_1$, so $f = \overline{s_1}s_3 + s_1$ and finally $f = s_3 + s_1\cdot s_2$ via the consensus identity, matching the spec.

Start from the canonical SOP and replicate $s_1 s_2\overline{s_3}$ (allowed since $a+a=a$): $f = \overline{s_1}\overline{s_2}s_3 + \overline{s_1}s_2 s_3 + s_1\overline{s_2}\overline{s_3} + s_1\overline{s_2}s_3 + s_1 s_2\overline{s_3} + s_1 s_2 s_3 + s_1 s_2\overline{s_3}$. Group $(\overline{s_1}\overline{s_2}s_3 + \overline{s_1}s_2 s_3) + (s_1\overline{s_2}\overline{s_3}+s_1 s_2\overline{s_3}) + (s_1\overline{s_2}s_3 + s_1 s_2 s_3) + s_1 s_2\overline{s_3} = \overline{s_1}s_3 + s_1\overline{s_3} + s_1 s_3 + s_1 s_2\overline{s_3}$. Combining $s_1\overline{s_3}+s_1 s_3 = s_1$, then $\overline{s_1}s_3 + s_1 = s_1 + s_3$… here we keep $s_1 s_2$ via the replicated term: $f = s_3 + s_1 s_2$.

Apply the consensus theorem $xy + \overline{x}z + yz = xy + \overline{x}z$ (the $yz$ consensus is redundant). Here from the full SOP, take $x = s_3$. Pairs $\overline{s_3}\cdot(s_1 s_2)$ and $s_3\cdot 1$ have consensus $s_1 s_2$, and combining the remaining six minterms with DeMorgan and absorption directly gives $f = s_3 + s_1 s_2$ in one step.

$f = \sum m(2,3,4,6,7) = \overline{x_1}x_2\overline{x_3} + \overline{x_1}x_2 x_3 + x_1\overline{x_2}\overline{x_3} + x_1 x_2\overline{x_3} + x_1 x_2 x_3$. Combine $\overline{x_1}x_2\overline{x_3}+\overline{x_1}x_2 x_3 = \overline{x_1}x_2$ and $x_1 x_2\overline{x_3}+x_1 x_2 x_3 = x_1 x_2$, giving $\overline{x_1}x_2+x_1 x_2 = x_2$. Then with $x_1\overline{x_2}\overline{x_3}$ left: $f = x_2 + x_1\overline{x_2}\overline{x_3} = x_2 + x_1\overline{x_3}$ by absorption ($x_2 + \overline{x_2}y = x_2 + y$). Minimum-cost SOP: $f = x_2 + \overline{x_1}\,\overline{x_3}$ (after applying the same trick from the alternative grouping).

The 7 minterms of $f(x_1,x_2,x_3,x_4) = \sum m(3,7,9,12,13,14,15)$ group on a K-map as: $x_3 x_4$ covers $m_3, m_7, m_{15}$; $x_1 x_2$ covers $m_{12}, m_{13}, m_{14}, m_{15}$; $x_1\overline{x_2}x_3 x_4$ ($m_9$ alone needs $x_1\overline{x_2}x_4$). Combining $m_9$ with $m_{13}$ gives $x_1 x_4\overline{x_2}+x_1 x_4 x_2 = x_1 x_4$. So minimum SOP is $f = x_1 x_2 + x_3 x_4 + x_1 x_4$.

$\prod M(0,1,5)$ means the maxterms are $M_0=(x_1+x_2+x_3)$, $M_1=(x_1+x_2+\overline{x_3})$, $M_5=(\overline{x_1}+x_2+\overline{x_3})$. Combine $M_0$ and $M_1$ on $x_3$: $(x_1+x_2)$. Combine $M_1$ and $M_5$ on $x_1$: $(x_2+\overline{x_3})$. Minimum POS: $f = (x_1+x_2)(x_2+\overline{x_3})$. (Note $x_2$ alone covers all three when expanded, but the two-term POS is the simplest.)

Take $f = (\overline{x_1}+x_2)(x_2+\overline{x_3})$. By DeMorgan, $(\overline{x_1}+x_2) = \overline{\overline{(\overline{x_1}+x_2)}} = \overline{x_1\overline{x_2}}$ … rather: a NOR gate computes $\overline{a+b}$. Realize each OR sum as a NOR followed by an inverter (= NOR with both inputs tied), and the final AND of the two as a NOR with each input first NORed (= inverted). Concretely: $P_1 = \overline{\overline{x_1}+x_2}$ via NOR(x_1', x_2)… simpler: invert each variable using a NOR-as-inverter, then $f = \text{NOR}(\text{NOR}(\overline{x_1},x_2),\,\text{NOR}(x_2,\overline{x_3}))$ after a final inversion.

From minimum SOP $f = x_2 + \overline{x_1}\,\overline{x_3}$, apply DeMorgan: $f = \overline{\overline{x_2}\cdot \overline{\overline{x_1}\,\overline{x_3}}} = \overline{\overline{x_2}\cdot(x_1+x_3)}$. Replace each AND/OR with NAND: invert $x_2$ via NAND, compute $x_1\,\text{NAND}\,x_3$ (= $\overline{x_1 x_3}$, then invert again to get $x_1+x_3$ via DeMorgan), and AND the two by NAND-NAND. Total: 4 NAND gates.

Build seven 4-variable K-maps, one per segment, with cells $X = 1010\dots1111$ marked don't-care. Minimum SOPs (standard result):

a = x3 + x1 + x2·x0 + x2'·x0'
b = x2' + x1·x0 + x1'·x0'
c = x2 + x1' + x0
d = x2'·x0' + x1·x0' + x2·x1'·x0 + x2'·x1 + x3
e = x2'·x0' + x1·x0'
f = x3 + x2·x1' + x2·x0' + x1'·x0'
g = x3 + x2'·x1 + x2·x1' + x1·x0'

Both functions share the term $\overline{x_2}\,\overline{x_3}\,\overline{x_4}$, so factor it out. $f_1 = \overline{x_1}\,\overline{x_3} + x_1 x_3 + \overline{x_2}\,\overline{x_3}\,\overline{x_4}$, $f_2 = \overline{x_1} x_3 + x_1\overline{x_3} + \overline{x_2}\,\overline{x_3}\,\overline{x_4}$. Note $f_1$'s first two terms form XNOR($x_1, x_3$) and $f_2$'s first two form XOR($x_1, x_3$). Sharing the third term yields total cost = 2 (XNOR/XOR) + 1 (AND for shared) + 2 (final OR per output) — substantially less than independent realizations.

Suppose $f_3 = \overline{x_1}\,\overline{x_2}\,\overline{x_4} + x_1 x_2 x_3$ and $f_4 = \overline{x_1}\,\overline{x_3}\,\overline{x_4} + x_2 x_3 x_4$ (Figure 2.65 pair). Their separate primes share none. But the non-prime cube $\overline{x_1}\,\overline{x_2}\,\overline{x_3}\,\overline{x_4}$ is contained in both $\overline{x_1}\,\overline{x_2}\,\overline{x_4}$ (of $f_3$) and $\overline{x_1}\,\overline{x_3}\,\overline{x_4}$ (of $f_4$). Using the smaller (4-literal) shared cube as a common product and then ORing the remainders yields fewer total inputs than two independent two-term realizations (cost drops from ~28 to ~22 gate-inputs).

Convert each function of Example 2.16 to its minimum POS via maxterms. $f_1 = (x_2+\overline{x_1}+x_3)(x_3+x_1+\overline{x_3})\dots$ → in fact each function has 4 maxterms not shared with the other, so combined POS gate-input cost is ~26 vs the shared SOP's ~18. Hence the SOP-based shared realization is cheaper.

The POS forms of $f_3$ and $f_4$ share the maxterm $(x_1+x_2+x_3+x_4)$ (covering $m_0$ in both). Sharing this sum term across both outputs gives total cost about 19 gate-inputs, comparable to the SOP-shared realization (~22). Both shared realizations beat the independent ones; for this pair POS is slightly cheaper.

Expand each side to canonical SOP. LHS: $\overline{x_1}x_3$ generates minterms $m_1, m_3$; $x_2\overline{x_3}$ generates $m_2, m_6$; $x_1 x_2$ generates $m_6, m_7$. So LHS = $\sum m(1,2,3,6,7)$. RHS: $x_1\overline{x_2}$ → $m_4, m_5$; $\overline{x_1}x_2$ → $m_2, m_3$; $\overline{x_2}\,\overline{x_3}$ → $m_0, m_4$. So RHS = $\sum m(0,2,3,4,5)$. Different minterm sets → equation is not valid.

$f = \sum m(0,2,4,5,6,7,8,10,12,14,15)$ has zeros at $m(1,3,9,11,13)$, i.e. maxterms $M_1, M_3, M_9, M_{11}, M_{13}$. Group on a 4-var K-map of zeros: the column $x_3 x_4 = 01$ at rows $\overline{x_1}\,\overline{x_2}, \overline{x_1}x_2, x_1\overline{x_2}$ gives $(\overline{x_3}+x_4 \to)$ actually for zeros of $f$, the pattern $x_3=0, x_4=1$ at $x_1 x_2 \in \{00,01,10\}$ combines as $(\overline{x_3}+x_4)$ … plus $(x_1+\overline{x_4})$ for $m_{13}$. Minimum POS: $f = (x_3+\overline{x_4})(\overline{x_2}+x_3+\overline{x_4})\dots$ simplifies to $f = (x_3+\overline{x_4})(x_1+\overline{x_2}+x_3)$ (covers all five maxterms with two sum terms after K-map grouping).

By definition $f = AB+AC+BC$ (two-out-of-three of the conditions). Substituting and expanding using DeMorgan/absorption: $AB = x_1 x_3(\overline{x_2}+\overline{x_3})\dots$ Algebraic simplification (textbook result) collapses to $f = x_1(x_2+x_3)$. Implementation: a 2-input OR ($x_2, x_3$) feeding a 2-input AND with $x_1$ — total cost = 4 gate-inputs.

Draw 3 circles for $A,B,C$ derived from inputs $x_1, x_2, x_3$. Marking the regions where each $A,B,C$ is true and shading where at least two overlap, then projecting back to the $(x_1, x_2, x_3)$ space, the only points where two-of-three are true are those satisfying $x_1 \wedge (x_2 \vee x_3)$. So $f = x_1(x_2+x_3)$, confirming the algebraic result.

Add the consensus of $\overline{x_2}x_3\overline{x_4}$ and $x_1 x_2\overline{x_4}$ which is $x_1 x_3\overline{x_4}$, then absorb. $f = \overline{x_2}x_3\overline{x_4} + \overline{x_1}\,\overline{x_3}\,\overline{x_4} + x_1 x_2\overline{x_4} + x_1 x_3\overline{x_4}$. Combining $x_1 x_2\overline{x_4}+x_1 x_3\overline{x_4} = x_1\overline{x_4}(x_2+x_3)$, and combining the two terms with $\overline{x_4}$ factors out: $f = \overline{x_4}\bigl(\overline{x_1}\,\overline{x_3} + x_3\overline{x_2} + x_1(x_2+x_3)\bigr)$. Final simplest SOP: $f = \overline{x_4}\,\bigl(\overline{x_1}\,\overline{x_3} + x_2 + x_3\bigr) = \overline{x_4}(\overline{x_1}\overline{x_3}+x_2+x_3)$.

Apply the dual consensus. Among the three sum terms, $(x_1+\overline{x_2}+x_3)$ and $(\overline{x_1}+\overline{x_3}+x_4)$ have consensus $(\overline{x_2}+x_4)$ — but it is not in the original. Use the identity $(A+B)(A+C) = A+BC$ on pairs sharing a literal: $(x_1+\overline{x_2}+x_3)(x_1+x_2+\overline{x_4}) = x_1 + (\overline{x_2}+x_3)(x_2+\overline{x_4}) = x_1 + x_3 x_2 + \overline{x_2}\,\overline{x_4} + x_3\overline{x_4}$. After AND with $(\overline{x_1}+\overline{x_3}+x_4)$ and absorption, the simplest POS is $f = (x_1+x_3)(\overline{x_1}+x_4)(\overline{x_2}+\overline{x_3}+x_4)$ (three sum terms).

On a 4-var K-map with $\sum m(4,6,8,10,11,12,15) + D(3,5,7,9)$: Minimum SOP using don't-cares: $f = x_2\overline{x_3} + x_1\overline{x_3} + x_3 x_4$ (three product terms — uses $D=5,7$ to extend cubes). Minimum POS using the same don't-cares: $f = (x_1+x_2+x_3)(\overline{x_3}+x_4)$. The two forms encode different functions on the don't-care cells (e.g. SOP makes $m_5=1$, POS makes $m_5=1$ too, but $m_3$ differs), illustrating that don't-care assignments need not agree across SOP/POS minimizations.

The given SOP expansion covers $\sum m(0,1,4,5,8,9,11)$ (after expanding each product). With $D=(9,12,14)$, K-map minimization yields: Minimum SOP: $f = \overline{x_3}\,\overline{x_4} + x_1 x_4 + \overline{x_2}x_4$. Minimum POS: $f = (\overline{x_2}+\overline{x_3})(\overline{x_3}+x_4)(x_1+\overline{x_3}+\overline{x_4})$. Cost (literals + gates+1): SOP ≈ 11, POS ≈ 13, so SOP is preferred.

Distribute: $f = s_3 s_1 + s_3 s_2 + \overline{s_1}\,\overline{s_2}$. On a 3-var K-map this is $\sum m(0,4,5,6,7)$. Combining: $s_3 s_1 + s_3 s_2 = s_3(s_1+s_2)$; together with $\overline{s_1}\,\overline{s_2}$ no further absorption applies. Minimum SOP: $f = s_1 s_3 + s_2 s_3 + \overline{s_1}\,\overline{s_2}$ (3 product terms).

module ex2_e29(input x, y, z, output f, g);
  wire t;
  assign t = y ^ z;
  assign f = (~t) & z | t & x;
  assign g = t ^ x;
endmodule

module mux2(input a, b, s, output f); assign f = s ? b : a; endmodule
module add1(input x, y, output s, c); assign {c,s} = x + y; endmodule

module shared_adder(input a,b,c,d,m, output s, cout);
  wire x, y;
  mux2 u1(.a(a), .b(c), .s(m), .f(x));
  mux2 u2(.a(b), .b(d), .s(m), .f(y));
  add1 u3(.x(x), .y(y), .s(s), .c(cout));
endmodule

Each product term has at most 3 distinct variables, so each fits in one 3-LUT directly:

L1 = x1·x2·x4'
L2 = x2'·x3·x4
L3 = x1'·x2'·x3'
L4 = L1 + L2 + L3   (3-LUT OR)

$3\text{F7A}_{16}$: convert each hex digit to 4 bits.

$3=0011,\; F=1111,\; 7=0111,\; A=1010$

Binary: 0011 1111 0111 1010

Decimal: $3\times16^3 + 15\times16^2 + 7\times16 + 10 = 12288 + 3840 + 112 + 10 = \mathbf{16250}$

73 (positive, 12-bit): $73 = 64+8+1 = 001001001_2$ → pad to 12: 000001001001 (all three formats same)

$-95$: $|95| = 001011111_2$ → 12-bit = $000001011111$

• Sign-magnitude: 100001011111 (MSB=1)
• 1's complement: 111110100000
• 2's complement: 111110100001

$-1630$: $|1630| = 11001011110_2$ → 12-bit = $011001011110$

• Sign-magnitude: 111001011110
• 1's complement: 100110100001
• 2's complement: 100110100010

Half-adder: adds two bits A, B → Sum S, Carry C.

Equations: $S = A \oplus B$, $C = A \cdot B$

Circuit: one XOR gate for S, one AND gate for C.

8-bit 2's complement range: −128 to +127. Overflow occurs when carry into MSB $\neq$ carry out of MSB ($c_{n-1} \oplus c_n = 1$).

(a) 00110110 + 01000101 = 54 + 69 = 123. Sum = 01111011; $c_{\text{in MSB}}=0$, $c_{\text{out}}=0$ → no overflow. Result = 01111011 = 123 ✓

(b) 01110101 + 11011110 = 117 + (−34) = 83. Sum = 1 01010011; $c_{\text{in MSB}}=1$, $c_{\text{out}}=1$ → no overflow. Discard carry → 01010011 = 83 ✓

(c) 11011111 + 10111000 = −33 + (−72) = −105. Sum = 1 10010111; $c_{\text{in MSB}}=1$, $c_{\text{out}}=1$ → no overflow. Discard carry → 10010111 = −105 ✓

   c:  1 1 1 1 1 0 0 0      <- c[i] = carry INTO bit i
       ─────────────────
   A:  0 1 1 0 1 1 0 0   (108)
   B:  0 0 1 1 0 1 0 1   (53)
       ─────────────────
   S:  1 0 1 0 0 0 0 1   (161 / −95)   Cout = 0
       ↑
       bit 7 (MSB)  →  c_in MSB = c[7] = 1 (leftmost of c row)

For $n$-bit 2's complement addition, $c_n$ is carry out of MSB, $c_{n-1}$ is carry into MSB.

Proof by cases:

• pos + pos → result must be positive. If result has MSB=1, $c_{n-1}=1$, $c_n=0$ → overflow=1 $\checkmark$
• neg + neg → result must be negative. If result MSB=0, $c_{n-1}=0$, $c_n=1$ → overflow=1 $\checkmark$
• pos + neg → no overflow possible; $c_n = c_{n-1}$ always → overflow=0 $\checkmark$

4-bit ripple-carry adder: chain four full adders.

A[3:0] ──┐           B[3:0] ──┐
          │                    │
  A0,B0 →[FA0]→ S0; C0→[FA1]→ S1; C1→[FA2]→ S2; C2→[FA3]→ S3,C3
  Cin=0    cₒᵤₜ₀       cₒᵤₜ₁         cₒᵤₜ₂         Cout

Each Full Adder (FA): inputs A_i, B_i, C_in; outputs Sum_i, C_out.

Carry ripples from FA0 → FA1 → FA2 → FA3 (critical path = 4 FA delays).

Use a control signal Sub. When Sub=0: add; when Sub=1: subtract $(A - B = A + \overline{B} + 1)$.

XOR each B bit with Sub: $B'_i = B_i \oplus \text{Sub}$. Feed Sub as $C_{in}$ to the adder.

module add_sub4 (
    input  [3:0] A, B,
    input        Sub,
    output [3:0] S,
    output       Cout, Overflow
);
    wire [3:0] B_in;
    wire c4, c3;
    assign B_in = B ^ {4{Sub}};
    assign {c4, S} = A + B_in + Sub; // sub is for 2的補數+1
    // carry into MSB is c3 (internal), overflow = c4 ^ c3
    assign Cout = c4;
    assign Overflow = c4 ^ (A[3] ^ B_in[3] ^ S[3]); // simplified
endmodule

Multiply $1101$ (13) $\times$ $1011$ (11):

        1101   (multiplicand)
      × 1011   (multiplier)
      ------
        1101   (1101 × 1)
       1101    (1101 × 1, shift 1)
      0000     (1101 × 0, shift 2)
     1101      (1101 × 1, shift 3)
     --------
    10001111   = 143 ✓

$-6.5$ in IEEE 754 single precision (32-bit):

Sign = 1 (negative). $|6.5| = 110.1_2 = 1.101 \times 2^2$.

Exponent $= 2 + 127\;\text{(bias)} = 129 = 10000001_2$.

Mantissa (fraction) = 101 followed by 20 zeros = 10100000000000000000000.

Result: 1 10000001 10100000000000000000000

Hex: C0D00000

module full_adder (
    input  a, b, cin,
    output sum, cout
);
    assign sum  = a ^ b ^ cin;
    assign cout = (a & b) | (a & cin) | (b & cin);
endmodule

module ripple_adder #(parameter N=8) (
    input  [N-1:0] A, B,
    input          Cin,
    output [N-1:0] Sum,
    output         Cout
);
    wire [N:0] c;
    assign c[0] = Cin;
    genvar i;
    generate
        for (i = 0; i < N; i = i+1) begin : fa_loop
            full_adder fa_inst (
                .a(A[i]), .b(B[i]), .cin(c[i]),
                .sum(Sum[i]), .cout(c[i+1])
            );
        end
    endgenerate
    assign Cout = c[N];
endmodule

The scan-from-right method: copy rightmost bits up to and including the first 1, then invert all remaining bits.

Why it works: Let $N = \ldots 1\underbrace{0\ldots0}_{k}$ (last 1 at position $k$, zeros below). 2's comp $= \overline{N} + 1$. Adding 1 to $\overline{N}$ propagates carries through all the 1s at positions $0 \ldots k-1$ (which were 0s in $N$, inverted to 1s in $\overline{N}$), zeroing them and carrying into position $k$. Position $k$ was 1 in $N$, inverted to 0 in $\overline{N}$, $+1$ carry makes it 1. Higher bits: $\overline{N}$ bits (inverted from $N$). Net result: bits $0 \ldots k-1$ are 0 (same as $N$), bit $k$ is 1 (same as $N$), bits above $k$ are inverted. $\blacksquare$

2's complement has one extra negative number (−128) because 10000000 has no positive counterpart; sign-magnitude and 1's complement have two representations of zero.

Define: $g_i = A_i B_i$ (generate), $p_i = A_i \oplus B_i$ (propagate).

All carries computed in parallel in 2 logic levels (AND-OR), vs. 8 levels for ripple-carry.

2's complement: invert all bits, add 1.

01010101: invert → 10101010, +1 → 10101011 (= −85)

11110000: invert → 00001111, +1 → 00010000 (= +16; confirms −16 → +16)

10000000: invert → 01111111, +1 → 10000000 (special case: −128, no positive counterpart in 8-bit)

0.625: 0.625×2=1.25→1; 0.25×2=0.5→0; 0.5×2=1.0→1; exact.

$0.625 = 0.101_2$ (exact in 3 bits)

0.3: 0.3×2=0.6→0; 0.6×2=1.2→1; 0.2×2=0.4→0; 0.4×2=0.8→0; 0.8×2=1.6→1; 0.6→1; 0.2→0; 0.4→0 (repeats)

$0.3 \approx 0.01001100_2$ (8 bits, repeating)

Prove $(A \oplus B) \oplus C = A \oplus (B \oplus C)$:

Note $A \oplus B = A\overline{B} + \overline{A}B$. Expand $(A \oplus B) \oplus C$:

Expanding $A \oplus (B \oplus C)$ yields the same 4 minterms $(1,2,4,7)$. $\blacksquare$

BCD encodes each decimal digit in 4 bits.

947: 9 = 1001, 4 = 0100, 7 = 0111

BCD(947) = 1001 0100 0111

BCD addition 67 + 45:

  0110 0111  (67 BCD)
+ 0100 0101  (45 BCD)
-----------
  1010 1100  (binary sum)

Lower nibble 1100 = 12 > 9 → add 0110 correction, generate carry:

  1100 + 0110 = 0010 carry 1
Upper nibble 1010 + 1 (carry) = 1011 > 9 → add 0110:
  1011 + 0110 = 0001 carry 1

Result: 0001 0001 0010 BCD = 112 ✓ (67+45=112)

module adder8 (
    input  [7:0] A, B,
    input        Cin,
    output [8:0] Result
);
    assign Result = {1'b0, A} + {1'b0, B} + Cin;
endmodule

The 9-bit result Result[8] captures the carry out. Using {1'b0, A} zero-extends to 9 bits before addition, giving a correct 9-bit sum.

01110101 (117) − 11010110. Subtraction = add 2's complement of subtrahend.

2's complement of 11010110: invert→00101001, +1→00101010 (= −42 represented, but +42 in unsigned).

  01110101  (117)
+ 00101010  (+42... wait, -(-42)=+42)
----------

Actually: 01110101 − 11010110 = 01110101 + 2'scomp(11010110).

11010110 inverted = 00101001, +1 = 00101010.

01110101 + 00101010 = 10011111. Carry out = 0.

Result $10011111$ in 2's complement $= -97$. Check: $117 - (-42) = 159$, but in 8-bit signed, $11010110 = -42$, so $117-(-42)=159$ overflows. Alternately if $11010110$ unsigned $= 214$: $117-214 = -97$. Result $= 10011111_2 = -97$.

$(752)_8$ to binary: $7=111,\; 5=101,\; 2=010$ → $111101010_2$

To hex: group binary in 4s from right: $0001\;1110\;1010$ → $1\text{EA}_{16}$

$(\text{B3D})_{16}$ to binary: $B=1011,\; 3=0011,\; D=1101$ → $101100111101_2$

To octal: group binary in 3s from right: $101\;100\;111\;101$ → $5475_8$

For an $n$-bit CLA adder:

• Generate/Propagate: $n$ XOR gates ($p$) + $n$ AND gates ($g$) = $2n$ gates.
• Carry logic ($c_1 \ldots c_n$): each $c_i$ is AND-OR: $c_i$ has $i+1$ AND gates + 1 OR gate. Total $\approx n(n+1)/2$ AND + $n$ OR.
• Sum bits: $n$ XOR gates.

Total gate count $\approx O(n^2)$ gates but $O(1)$ depth (constant 4 levels for standard 4-bit CLA block). For $n$-bit using hierarchical CLA: $O(n \log n)$ gates with $O(\log n)$ depth.

Ripple-carry adder ($n$-bit): Critical path $= n \times t_{FA} \approx n \times 2\;\text{gate levels} = O(n)$ delay.

For 4-bit at 1 ns per gate: $\sim 8$ ns (through 4 FAs, each 2 levels).

Carry-lookahead adder (CLA): Carries computed in 2 logic levels; sum then 1 more XOR level = 3 levels total, $O(\log n)$ for hierarchical CLA.

For 4-bit CLA: $\sim 3$ gate delays regardless of $n$ (within one block).

Tradeoff: CLA uses more gates (fan-in grows) but is significantly faster for wide adders.

Hierarchical CLA using two 2-bit CLA blocks (each handles bits 0-1 and bits 2-3):

Block 0 (bits 0,1): computes $c_1, c_2$, and group generate $G_0 = g_1 + p_1 g_0$, group propagate $P_0 = p_1 p_0$.

Block 1 (bits 2,3): uses $c_2$ from Block 0; computes $c_3, c_4, G_1, P_1$.

Top-level carry:

This achieves the same $O(1)$ carry depth within 4 bits, and extends to 16-bit with another hierarchical level.

module array_mult4 (
    input  [3:0] A, B,
    output [7:0] P
);
    wire [3:0] pp [3:0];
    wire [7:0] sum;
    genvar i;
    generate
        for (i = 0; i < 4; i = i+1) begin : partial
            assign pp[i] = A & {4{B[i]}};
        end
    endgenerate
    // Sum partial products with shifts
    assign P = pp[0]
             + (pp[1] << 1)
             + (pp[2] << 2)
             + (pp[3] << 3);
endmodule

Each row pp[i] is the AND of A with a replicated B[i]. Shifted addition produces the 8-bit product.

Add $10^n$ and subtract: $A - B = A + (10^n - B) - 10^n$. $10^n - B$ is the 10's complement of $B$. After the addition $A + (10^n - B)$ we obtain a value whose leading position carries $1$ (representing $10^n$) when $A \ge B$; discarding that carry-out subtracts the $10^n$ for free. When $A < B$, no carry occurs and the result is $10^n - (B-A)$, which is the 10's complement of the negative answer.

$045 - 027$: 10's-comp of $027$ is $1000 - 027 = 973$. $045 + 973 = 1018$; discard the leading 1 → $018 = +18$. $027 - 045$: 10's-comp of $045$ is $955$. $027 + 955 = 982$; no carry-out, so the result is $982$, which is the 10's complement of $-18$ (since $1000 - 982 = 18$).

$+5 = 0101$, $+2 = 0010$. 2's-comp of $0010$ is $1110$. $0101 + 1110 = 1\,0011$. Discard the carry-out from bit 4 (= $2^4 = 16$) → $0011 = +3$. The discarded bit corresponds to the $2^n$ term, exactly as predicted.

$+2 = 0010$, $+5 = 0101$. 2's-comp of $0101$ is $1011$. $0010 + 1011 = 1101$ with no carry-out. $1101$ in 4-bit 2's complement is $-(0010+1) = -3$, ✓.

$+5 = 0101$, $-2 = 1110$. 2's-comp of $-2$ is $0010$. $0101 + 0010 = 0111 = +7$. ✓ (No carry to discard in this case because the addend is positive.)

Repeated division by 16:

Multiply 0.8254 by 2 repeatedly, recording the integer part: $1.6508, 1.3016, 0.6032, 1.2064, 0.4128, 0.8256, 1.6512, 1.3024$ → bits 1,1,0,1,0,1,1,1. So $0.8254_{10} \approx \mathbf{0.11010111_2}$ (with truncation error).

Integer part: $214_{10} = 11010110_2$ (by repeated /2). Fractional part $0.45$: ×2 yields $0.9, 1.8, 1.6, 1.2, 0.4, 0.8, 1.6, \dots$ → bits $0,1,1,1,0,0,1,\dots$. Result: $214.45_{10} \approx \mathbf{11010110.0111001\ldots_2}$.

With 2's-complement subtraction $S = X - Y$ and signed flags:

• $X = Y$: $Z = 1$
• $X < Y$: $N \oplus V = 1$ (i.e., result negative without overflow OR overflow without negative)
• $X \le Y$: $Z + (N\oplus V) = 1$
• $X > Y$: $\overline{Z}\,\overline{(N\oplus V)} = 1$
• $X \ge Y$: $\overline{N\oplus V} = 1$

Structural version instantiates four full-adder modules with $Y$ inverted and $c_0=1$ to form $X-Y$, then derives $Z = !|S$, $N = S[3]$, $V = c_3 \oplus c_2$.

module comp4(input [3:0] X, Y, output Z, N, V, AeqB, AltB, AleB, AgtB, AgeB);
  wire [3:0] S; wire [3:0] c;
  full_add fa0(X[0], ~Y[0], 1'b1, S[0], c[0]);
  full_add fa1(X[1], ~Y[1], c[0], S[1], c[1]);
  full_add fa2(X[2], ~Y[2], c[1], S[2], c[2]);
  full_add fa3(X[3], ~Y[3], c[2], S[3], c[3]);
  assign Z = ~|S, N = S[3], V = c[3]^c[2];
  assign AeqB = Z, AltB = N^V, AleB = Z|(N^V), AgtB = ~(Z|(N^V)), AgeB = ~(N^V);
endmodule

Ripple-carry array (Figure 3.35): each row is an $n$-bit ripple adder of delay $\approx n\cdot t_{FA}$, and there are $n-1$ rows, so total delay $\approx (n-1)n\, t_{FA} + t_{AND}$. For $n=4$: $\approx 12 t_{FA} + t_{AND}$. Carry-save array (Figure 3.48): the partial-product reduction uses $n-1$ rows of carry-save adders (each $\approx t_{FA}$) plus one final ripple adder of $n$ FAs. Total $\approx (n-1) t_{FA} + n\, t_{FA} + t_{AND} = (2n-1) t_{FA} + t_{AND}$. For $n=4$: $\approx 7 t_{FA} + t_{AND}$. Carry-save is roughly $n/2$× faster.

$f = \sum m(0,2,3,4,5,7)$. A 3-to-8 decoder asserts exactly one output for each minterm.

Connect decoder outputs $Y_0, Y_2, Y_3, Y_4, Y_5, Y_7$ to an OR gate:

The decoder has inputs $x_1 x_2 x_3$ and active-high outputs. No minimization needed; the OR gate directly implements the canonical SOP.

$f = \sum m(0,2,3,6)$. Shannon expansion on $x_1$:

$f|_{x_1=0}$: minterms $\{0,2,3\}$ → $f_0(x_2,x_3) = \overline{x_3} + x_2$

$f|_{x_1=1}$: minterm $\{6\}$ → $(x_2 x_3) = 10$ → $f_1 = x_2\overline{x_3}$

2-to-1 mux: select $= x_1$; $D_0 = \overline{x_3} + x_2$, $D_1 = x_2\overline{x_3}$.

4-to-1 mux with select $S_1 S_0$, inputs $D_0 \ldots D_3$, output $Y$:

Circuit:
- 2 NOT gates: S̄₁, S̄₀
- 4 AND gates (3-input each): one per term
- 1 OR gate (4-input)

Total: 2 NOT + 4 AND + 1 OR = 7 gates.

4-to-2 priority encoder: highest-priority active input determines output; V=1 if any input active.

$Y_1 = I_3 + I_2$; $\;Y_0 = I_3 + I_1\overline{I_2}$; $\;V = I_0 + I_1 + I_2 + I_3$

2-to-4 decoder with enable E (active-high):

module dec2to4 (input E, A1, A0, output reg [3:0] Y);
    always @(*) begin
        if (!E)
            Y = 4'b0000;
        else
            case ({A1, A0})
                2'b00:   Y = 4'b0001;
                2'b01:   Y = 4'b0010;
                2'b10:   Y = 4'b0100;
                2'b11:   Y = 4'b1000;
                default: Y = 4'b0000;
            endcase
    end
endmodule

BCD to 7-segment (common cathode, segments a-g). Digits 0-9; inputs 10-15 are don't-cares.

K-map minimization with don't-cares for inputs 10–15 (BCD = wxyz, MSB = w):

2-bit magnitude comparator. Inputs $A = A_1 A_0$, $B = B_1 B_0$.

$A > B$ when $A_1 > B_1$, or $A_1 = B_1$ and $A_0 > B_0$.

Note: exactly one of AgtB, AeqB, AltB is 1 for any input.

Two 2-to-4 decoders (each with enable) build a 3-to-8 decoder:

• Decoder A handles A₂=0: enable = Ā₂; inputs A₁,A₀ → outputs Y₀..Y₃
• Decoder B handles A₂=1: enable = A₂; inputs A₁,A₀ → outputs Y₄..Y₇

When A₂=0, decoder A is enabled and selects among Y₀..Y₃. When A₂=1, decoder B is enabled and selects among Y₄..Y₇. Exactly one of the 8 outputs is active at any time.

module priority_enc8 (
    input  [7:0] in,
    output reg [2:0] out,
    output reg valid
);
    always @(*) begin
        valid = 1'b1;
        if      (in[7]) out = 3'd7;
        else if (in[6]) out = 3'd6;
        else if (in[5]) out = 3'd5;
        else if (in[4]) out = 3'd4;
        else if (in[3]) out = 3'd3;
        else if (in[2]) out = 3'd2;
        else if (in[1]) out = 3'd1;
        else if (in[0]) out = 3'd0;
        else begin out = 3'd0; valid = 1'b0; end
    end
endmodule

An incomplete case statement without a default in a combinational always block infers latches for any output not assigned in all branches.

The synthesizer holds the previous value for unspecified cases, which requires storage elements (latches). This is usually unintended and leads to unexpected behavior and synthesis warnings.

Fix: always include a default branch, or use always @(*) with full assignments.

Gray code: consecutive codes differ in exactly 1 bit. For 4-bit: $G_3 G_2 G_1 G_0$ from binary $B_3 B_2 B_1 B_0$:

module bin2gray (input [3:0] B, output [3:0] G);
    assign G = B ^ (B >> 1);
endmodule

Shannon's expansion:

Proof: Every input combination has $x_i = 0$ or $x_i = 1$.

Case $x_i = 0$: RHS $= 1 \cdot f(\ldots,0,\ldots) + 0 \cdot f(\ldots,1,\ldots) = f(\ldots,0,\ldots)$ = LHS. $\checkmark$

Case $x_i = 1$: RHS $= 0 \cdot f(\ldots,0,\ldots) + 1 \cdot f(\ldots,1,\ldots) = f(\ldots,1,\ldots)$ = LHS. $\checkmark$

Since both cases hold for all input combinations, the expansion is valid. $\blacksquare$

Build 8-to-1 mux from 2-to-1 muxes in a tree:

Level 1 (4 muxes): pair up D₀/D₁, D₂/D₃, D₄/D₅, D₆/D₇, selected by S₀.

Level 2 (2 muxes): pair level-1 outputs, selected by S₁.

Level 3 (1 mux): pair level-2 outputs, selected by S₂.

Total: 4+2+1 = 7 two-to-one muxes. S₂ is the MSB of select.

module mux4to1 (
    input  [1:0] sel,
    input  [3:0] d,
    output reg   y
);
    always @(*)
        y = d[sel];
    // alternatively using conditional operator:
    // assign y = (sel==2'b00) ? d[0] :
    //            (sel==2'b01) ? d[1] :
    //            (sel==2'b10) ? d[2] : d[3];
endmodule

1-to-4 demultiplexer: routes single input I to one of four outputs based on S₁S₀.

$Y_i = I \cdot \text{(decoder output for } i\text{)} = I \cdot m_i(S_1, S_0)$

module comparator4 (
    input  [3:0] A, B,
    output       AgtB, AeqB, AltB
);
    assign AgtB = (A > B);
    assign AeqB = (A == B);
    assign AltB = (A < B);
endmodule

Synthesis tools expand the relational operators into optimized combinational logic.

module bcd_to_7seg (
    input  [3:0] bcd,
    output reg [6:0] seg  // seg[6]=a, seg[5]=b,..., seg[0]=g
);
    function [6:0] decode;
        input [3:0] d;
        case (d)
            4'd0: decode = 7'b1111110;
            4'd1: decode = 7'b0110000;
            4'd2: decode = 7'b1101101;
            4'd3: decode = 7'b1111001;
            4'd4: decode = 7'b0110011;
            4'd5: decode = 7'b1011011;
            4'd6: decode = 7'b1011111;
            4'd7: decode = 7'b1110000;
            4'd8: decode = 7'b1111111;
            4'd9: decode = 7'b1111011;
            default: decode = 7'b0000000;
        endcase
    endfunction
    always @(*) seg = decode(bcd);
endmodule

module popcount8 (
    input  [7:0] data,
    output reg [3:0] count
);
    integer i;
    always @(*) begin
        count = 0;
        for (i = 0; i < 8; i = i+1)
            count = count + data[i];
    end
endmodule

The for loop is unrolled by synthesis into a tree of adders. Result is the number of 1-bits (0 to 8, needs 4 bits).

Apply Shannon's expansion recursively on all variables:

Each cofactor $f(a,b,x_3)$ is further expanded on $x_3$:

$f(a,b,c) \in \{0,1\}$ for each specific minterm. This recursively produces the canonical sum of minterms: $f = \sum_i m_i \cdot f(m_i)$, where the sum is over all input combinations where $f = 1$. $\blacksquare$

casez: Treats z (high-impedance) and ? as don't-care bits in case item comparisons. Used for priority encoders and partial matches.

casex: Treats both x (unknown) and z as don't-care. More aggressive masking — also masks unknown values in the case expression itself, which can hide simulation mismatches.

Best practice: Prefer casez (or casez with ?) for synthesis; avoid casex as it masks Xs in simulation, potentially hiding bugs.

module sort3 (
    input  [7:0] a, b, c,
    output reg [7:0] x, y, z  // x≤y≤z
);
    task swap;
        inout [7:0] p, q;
        reg   [7:0] t;
        begin
            if (p > q) begin t = p; p = q; q = t; end
        end
    endtask
    always @(*) begin
        x = a; y = b; z = c;
        swap(x, y);   // ensure x≤y
        swap(y, z);   // bubble largest to z
        swap(x, y);   // final sort x≤y
    end
endmodule

CLA carry: $c_{i+1} = g_i + p_i c_i$. This is a 2-to-1 mux structure:

$c_{i+1} = \text{mux}(p_i,\; g_i,\; c_i)$: if $p_i = 1$, output $= c_i$ (propagate); if $p_i = 0$, output $= g_i$ (generate or kill).

This is a 2-to-1 mux with select $= p_i$, $D_1 = c_i$, $D_0 = g_i$.

This "carry select" mux structure underlies fast carry-select adders.

$f = w_1 w_2 + w_2 w_3 + \overline{w_1}\,\overline{w_2}\,w_3$. Expand by Shannon on $w_1$:

Use 2-to-1 mux with select $= w_1$: $D_0 = w_3$, $D_1 = w_2$.

module decoder_n #(parameter N=3) (
    input  [N-1:0]   addr,
    input            en,
    output [(1<

Cascadable 4-bit comparator has additional inputs: GIN (greater-in), EIN (equal-in), LIN (less-in) from the less-significant stage.

GOUT = AgtB_internal OR (AeqB_internal AND GIN)

EOUT = AeqB_internal AND EIN

LOUT = AltB_internal OR (AeqB_internal AND LIN)

Initialize the LSB stage with GIN=0, EIN=1, LIN=0. Chain stages from LSB to MSB; the MSB stage outputs the final result.

Two 2-to-1 muxes both controlled by $s$: $y_1 = \overline{s}\,x_1 + s\,x_2$, $y_2 = \overline{s}\,x_2 + s\,x_1$. When $s=0$ each output forwards its same-index input; when $s=1$ they swap.

Maj($w_1, w_2, w_3$) = $w_1 w_2 + w_1 w_3 + w_2 w_3$. With $w_1, w_2$ as the 4-to-1 mux selects:

Recursively decompose $w_1\oplus w_2\oplus w_3 = w_1 \oplus (w_2\oplus w_3)$. Use one mux selected by $w_2$ to compute $w_2\oplus w_3$ (data inputs $w_3$ and $\overline{w_3}$). Then a second mux selected by $w_1$ takes that XOR result and its complement as data, producing $f$. Total: 2 muxes plus inverters for the complemented data lines.

$f = \overline{w_1}\,\overline{w_3} + w_2 w_3$. On $w_1$: $f_{w_1}=w_2 w_3$, $f_{\overline{w_1}} = \overline{w_3}+w_2 w_3 = \overline{w_3}+w_2$. → 6 inputs. On $w_2$: $f_{w_2} = \overline{w_1}\,\overline{w_3}+w_3$, $f_{\overline{w_2}} = \overline{w_1}\,\overline{w_3}$. → 5 inputs. On $w_3$: $f_{w_3} = w_2$, $f_{\overline{w_3}} = \overline{w_1}$. So $f = \overline{w_3}\,\overline{w_1} + w_3 w_2$ — only 4 gate-inputs. Expansion on $w_3$ wins.

Shannon on $w_1$: $f_{\overline{w_1}} = \overline{w_3}$, $f_{w_1} = w_2 + w_3$. (a) 2-to-1 mux selected by $w_1$, data inputs $\overline{w_3}$ and $(w_2+w_3)$ — needs an inverter and an OR. (b) Shannon on $\{w_1, w_3\}$: cofactors at $(w_1, w_3) \in \{00,01,10,11\}$ are $\{1, 0, 1, w_2\to 1\}\to\{1,0,1,1\}$. So 4-to-1 mux selected by $\{w_1, w_3\}$ has data inputs $1, 0, w_2, 1$.

Maj $= w_1 w_2 + w_1 w_3 + w_2 w_3$. Shannon on $w_1$: $f_{\overline{w_1}} = w_2 w_3$, $f_{w_1} = w_2 + w_3$. Shannon each cofactor on $w_2$: $f_{\overline{w_1}\overline{w_2}}=0$, $f_{\overline{w_1}w_2} = w_3$; $f_{w_1\overline{w_2}} = w_3$, $f_{w_1 w_2} = 1$. Build the tree as two muxes selected by $w_2$ (data $0,w_3$ and $w_3,1$) feeding one mux selected by $w_1$. Total: 3 2-to-1 muxes.

A 2-to-4 decoder with En=1 outputs the four minterms of its select inputs $s_0, s_1$. AND each minterm $m_i$ with the corresponding data $w_i$ and OR all four results: $f = m_0 w_0 + m_1 w_1 + m_2 w_2 + m_3 w_3$ — exactly the function of a 4-to-1 mux.

module mux2to1(input w0, w1, s, output f);
  assign f = s ? w1 : w0;
endmodule

module mux2to1(input w0, w1, s, output reg f);
  always @(*) begin
    if (s) f = w1;
    else   f = w0;
  end
endmodule

module mux16to1(input [15:0] W, input [3:0] S, output f);
  wire [3:0] m;
  mux4to1 g0(W[3:0],   S[1:0], m[0]);
  mux4to1 g1(W[7:4],   S[1:0], m[1]);
  mux4to1 g2(W[11:8],  S[1:0], m[2]);
  mux4to1 g3(W[15:12], S[1:0], m[3]);
  mux4to1 g4(m, S[3:2], f);
endmodule

module mux4to1(input [3:0] W, input [1:0] S, output reg f);
  always @(*) case (S)
    2'b00: f = W[0];
    2'b01: f = W[1];
    2'b10: f = W[2];
    2'b11: f = W[3];
  endcase
endmodule

module dec2to4(input [1:0] W, input En, output reg [3:0] Y);
  always @(*) case ({En, W})
    3'b100: Y = 4'b0001;
    3'b101: Y = 4'b0010;
    3'b110: Y = 4'b0100;
    3'b111: Y = 4'b1000;
    default: Y = 4'b0000;
  endcase
endmodule

module dec2to4(input [1:0] W, input En, output reg [3:0] Y);
  always @(*) begin
    if (En == 0) Y = 4'b0000;
    else case (W)
      2'b00: Y = 4'b0001;
      2'b01: Y = 4'b0010;
      2'b10: Y = 4'b0100;
      2'b11: Y = 4'b1000;
    endcase
  end
endmodule

module dec4to16(input [3:0] W, input En, output [15:0] Y);
  wire [3:0] M;
  dec2to4 d0(W[3:2], En, M);
  dec2to4 d1(W[1:0], M[0], Y[3:0]);
  dec2to4 d2(W[1:0], M[1], Y[7:4]);
  dec2to4 d3(W[1:0], M[2], Y[11:8]);
  dec2to4 d4(W[1:0], M[3], Y[15:12]);
endmodule

module seg7(input [3:0] hex, output reg [6:0] leds);
  always @(*) case (hex)  // active-low; bits a..g
    4'h0: leds = 7'b0000001;
    4'h1: leds = 7'b1001111;
    4'h2: leds = 7'b0010010;
    4'h3: leds = 7'b0000110;
    4'h4: leds = 7'b1001100;
    4'h5: leds = 7'b0100100;
    4'h6: leds = 7'b0100000;
    4'h7: leds = 7'b0001111;
    4'h8: leds = 7'b0000000;
    4'h9: leds = 7'b0000100;
    4'hA: leds = 7'b0001000;
    4'hB: leds = 7'b1100000;
    4'hC: leds = 7'b0110001;
    4'hD: leds = 7'b1000010;
    4'hE: leds = 7'b0110000;
    4'hF: leds = 7'b0111000;
  endcase
endmodule

module alu74381(input [2:0] S, input [3:0] A, B, output reg [3:0] F);
  always @(*) case (S)
    3'b000: F = 4'b0000;       // clear
    3'b001: F = B - A;
    3'b010: F = A - B;
    3'b011: F = A + B;
    3'b100: F = A ^ B;
    3'b101: F = A | B;
    3'b110: F = A & B;
    3'b111: F = 4'b1111;       // preset
  endcase
endmodule

module prio4(input [3:0] W, output reg [1:0] Y, output reg z);
  always @(*) begin
    z = |W;
    casex (W)
      4'b1xxx: Y = 2'd3;
      4'b01xx: Y = 2'd2;
      4'b001x: Y = 2'd1;
      4'b0001: Y = 2'd0;
      default: Y = 2'd0;
    endcase
  end
endmodule

module dec2to4(input [1:0] W, input En, output reg [3:0] Y);
  integer k;
  always @(*) begin
    Y = 4'b0;
    for (k = 0; k < 4; k = k+1)
      if (En && (W == k)) Y[k] = 1'b1;
  end
endmodule

module prio4(input [3:0] W, output reg [1:0] Y, output reg z);
  integer k;
  always @(*) begin
    Y = 2'b0; z = 1'b0;
    for (k = 0; k < 4; k = k+1)
      if (W[k]) begin Y = k[1:0]; z = 1'b1; end
  end
endmodule

module cmp4(input [3:0] A, B, output reg AeqB, AgtB, AltB);
  always @(*) begin
    AeqB = 0; AgtB = 0; AltB = 0;
    if      (A == B) AeqB = 1;
    else if (A >  B) AgtB = 1;
    else             AltB = 1;
  end
endmodule

module rca #(parameter n=4)(input [n-1:0] A, B, input cin,
                              output [n-1:0] S, output cout);
  wire [n:0] c; assign c[0] = cin; assign cout = c[n];
  genvar i;
  generate for (i=0; i

module mux16to1(input [15:0] W, input [3:0] S, output reg f);
  reg [3:0] M;
  task mux4; input [3:0] X; input [1:0] s; output y;
    case (s) 0:y=X[0]; 1:y=X[1]; 2:y=X[2]; 3:y=X[3]; endcase
  endtask
  always @(*) begin
    mux4(W[3:0],   S[1:0], M[0]);
    mux4(W[7:4],   S[1:0], M[1]);
    mux4(W[11:8],  S[1:0], M[2]);
    mux4(W[15:12], S[1:0], M[3]);
    mux4(M, S[3:2], f);
  end
endmodule

module mux16to1(input [15:0] W, input [3:0] S, output f);
  function mux4; input [3:0] X; input [1:0] s;
    case (s) 0:mux4=X[0]; 1:mux4=X[1]; 2:mux4=X[2]; 3:mux4=X[3]; endcase
  endfunction
  wire [3:0] M;
  assign M[0] = mux4(W[3:0],   S[1:0]);
  assign M[1] = mux4(W[7:4],   S[1:0]);
  assign M[2] = mux4(W[11:8],  S[1:0]);
  assign M[3] = mux4(W[15:12], S[1:0]);
  assign f    = mux4(M,        S[3:2]);
endmodule

Use a 3-to-8 decoder to produce minterms $m_0\dots m_7$, then OR the outputs corresponding to $m_0, m_1, m_3, m_4, m_6, m_7$: $f = m_0+m_1+m_3+m_4+m_6+m_7$. One 6-input OR gate suffices.

One-hot inputs $w_0\dots w_7$. Truth table maps $w_i=1$ to $y_2 y_1 y_0 = i$. $y_0 = w_1+w_3+w_5+w_7$, $y_1 = w_2+w_3+w_6+w_7$, $y_2 = w_4+w_5+w_6+w_7$ (three 4-input OR gates).

Shannon on $(w_1, w_4)$ as the mux selects. Cofactors are computed from $f$: $f|_{00}=0$, $f|_{01}=w_5(\overline{w_2}+w_3)$, $f|_{10}=\overline{w_2}+w_3$, $f|_{11}=1$. So a 4-to-1 mux selected by $\{w_1, w_4\}$ with data inputs $0,\,w_5(\overline{w_2}+w_3),\,\overline{w_2}+w_3,\,1$ realizes $f$. Extra logic: one OR ($\overline{w_2}+w_3$) feeding both an AND with $w_5$ and the third data input — assuming $\overline{w_2}$ comes from an inverter.

Shannon: $f = \overline{w_1} f_{\overline{w_1}} + w_1 f_{w_1}$. (a) If $f_{\overline{w_1}} = 0$: $f = w_1\, f_{w_1}$ — the mux degenerates to a single AND gate. (b) If $f_{\overline{w_1}} = 1$: $f = \overline{w_1} + w_1 f_{w_1} = \overline{w_1} + f_{w_1}$ — the mux degenerates to a single OR gate.

A 4-to-1 mux has 6 inputs ($w_0..w_3, s_0, s_1$). With 4-LUTs, two LUTs suffice: LUT1 takes $(w_0, w_1, s_0, s_1)$ and produces $a = \overline{s_1}\overline{s_0}w_0 + \overline{s_1}s_0 w_1$; LUT2 takes $(a, w_2, w_3, s_0, s_1)$ — too many. Better: LUT1 takes $(w_0, w_1, s_0)$ → $\overline{s_0}w_0+s_0 w_1$; LUT2 takes $(w_2, w_3, s_0)$ → $\overline{s_0}w_2+s_0 w_3$; LUT3 takes those two outputs and $s_1$ → final mux. That is 3 LUTs of 3 inputs (or 4-LUTs with one input unused). The optimized 2-LUT mapping: LUT1$(w_0,w_1,s_0,s_1) = \overline{s_1}(\overline{s_0}w_0+s_0 w_1)$ and LUT2$(\text{LUT1}, w_2, w_3, s_0)$ ORs in the upper half — 2 4-LUTs.

Five 2-to-1 muxes selected by Shift; data lines (Shift=0, Shift=1): $y_3:(w_3, 0)$, $y_2:(w_2, w_3)$, $y_1:(w_1, w_2)$, $y_0:(w_0, w_1)$, $k:(0, w_0)$. Shift=0 passes $W$ unchanged with $k=0$; Shift=1 right-shifts (LSB into $k$, MSB filled with 0).

Each output $y_i$ takes the rotated bit $w_{(i+S)\bmod 4}$. Use four 4-to-1 muxes, all sharing select $S$: $y_0$ data $\{w_0,w_1,w_2,w_3\}$, $y_1$ data $\{w_1,w_2,w_3,w_0\}$, $y_2$ data $\{w_2,w_3,w_0,w_1\}$, $y_3$ data $\{w_3,w_0,w_1,w_2\}$.

module mux4to1(input [3:0] W, input [1:0] S, output f);
  wire [3:0] D;
  dec2to4 g(.W(S), .En(1'b1), .Y(D));
  assign f = | (D & W);
endmodule

// Style 1: explicit if-else
module shr1(input [3:0] W, input Shift, output reg [3:0] Y, output reg k);
  always @(*) if (Shift) {Y, k} = {1'b0, W};
              else       {Y, k} = {W, 1'b0};
endmodule

// Style 2: shift operator
module shr2(input [3:0] W, input Shift, output [3:0] Y, output k);
  assign {Y, k} = Shift ? {1'b0, W} : {W, 1'b0};
endmodule

module barrel4(input [3:0] W, input [1:0] S, output [3:0] Y);
  wire [7:0] T;
  assign T = {W, W};
  assign Y = T >> S;
endmodule

module evenparity(input [6:0] b_in, output [7:0] b_out);
  assign b_out[6:0] = b_in;
  assign b_out[7]   = ^b_in;        // XOR-reduction = even parity bit
endmodule

SR latch using two cross-coupled NOR gates:

Q = NOR(R, Q̄); Q̄ = NOR(S, Q)

Latch: Level-sensitive; transparent when clock/enable is high — output follows input while enabled. Can have timing issues (multiple transitions per clock cycle).

Flip-flop: Edge-triggered; samples input only at clock edge (rising or falling). Output changes only once per cycle.

Why edge-triggered preferred:

• Predictable: data captured once per cycle at known time.
• Simplifies timing analysis (setup/hold constraints well-defined).
• Prevents feedback loops that can cause oscillation in combinational logic.
• Enables pipelined design with well-defined stage boundaries.

D input sequence (sampled at rising edge): 0,1,1,0,1,0,0,1

Clk:  ↑     ↑     ↑     ↑     ↑     ↑     ↑     ↑
D:    0     1     1     0     1     0     0     1
Q:    0     1     1     0     1     0     0     1

Q follows D with one clock cycle delay. Q captures D at each rising edge.

T flip-flop from D flip-flop: D = T⊕Q (XOR T with current output).

• T=0: D=Q → holds state
• T=1: D=Q̄ → toggles state

Characteristic equation: Q(t+1) = T⊕Q(t)

3-bit synchronous up counter using T flip-flops:

T₀ = 1 (always toggles — LSB)

T₁ = Q₀ (toggle when Q₀=1)

T₂ = Q₁·Q₀ (toggle when Q₁Q₀=11)

State sequence: 000→001→010→011→100→101→110→111→000

All T FFs clocked simultaneously — synchronous design.

Assume each FF has t_FF = 5 ns and each gate = 1 ns.

Ripple counter: Each FF clocked by previous FF output. Critical path through all 4 FFs: 4 × 5 ns = 20 ns. Max freq = 1/20 ns = 50 MHz.

Synchronous counter: All FFs share clock. Critical path = t_FF + t_AND (for T₃ logic) ≈ 5 + 1 = 6 ns. Max freq = 1/6 ns ≈ 167 MHz.

Synchronous counter is ~3× faster for 4-bit; advantage grows with bit width.

module counter_updown (
    input        clk, rst, up,
    output reg [3:0] count
);
    always @(posedge clk or posedge rst) begin
        if (rst)
            count <= 4'd0;
        else if (up)
            count <= count + 1;
        else
            count <= count - 1;
    end
endmodule

Blocking (=): Executes sequentially within always block; each assignment takes effect immediately for the next statement.

Non-blocking (<=): All RHS evaluated first, then all LHS updated simultaneously at end of time step. Models register behavior correctly.

Wrong type example — swap with blocking:

// WRONG: blocking swap in sequential logic
always @(posedge clk) begin
    a = b;   // a gets b's new value
    b = a;   // b gets a's NEW value (a already changed)
end
// CORRECT: non-blocking
always @(posedge clk) begin
    a <= b;  // RHS evaluated simultaneously
    b <= a;  // b gets original a
end

module dff_srst_en (
    input  clk, srst, en, d,
    output reg q
);
    always @(posedge clk) begin
        if (srst)
            q <= 1'b0;
        else if (en)
            q <= d;
        // else q holds
    end
endmodule

Synchronous reset: reset takes effect only on rising clock edge. Enable gates the data input.

Setup time (t_su): Minimum time the data input D must be stable before the active clock edge for reliable capture.

Hold time (t_h): Minimum time D must remain stable after the active clock edge.

Clock-to-Q delay (t_cQ): Time from active clock edge until the output Q reaches its new stable value.

Violations: setup violation → metastability (indeterminate output); hold violation → data captured incorrectly regardless of clock period.

Master-slave D FF: Two latches in series (master then slave), clocked by CLK and CLK̄ respectively.

• When CLK=0: master is transparent (captures D), slave is held.
• When CLK=1: master is held (locks captured value), slave is transparent (propagates to Q).

Data is captured at the rising clock edge (master closes, slave opens) — effectively edge-triggered.

Why edge-triggered: The double-latch structure means input can only affect output during a brief transition window at the clock edge, providing full edge-triggered behavior.

Metastability: When a flip-flop's setup or hold time is violated, the output may settle to an indeterminate voltage level between 0 and 1 for an unpredictable time. This can propagate errors through the circuit.

Synchronizer: A chain of two or more flip-flops that allows a metastable signal time to resolve before use. The second FF sees a stable input in most cases. A single FF is insufficient because its output could still be metastable when sampled by subsequent logic.

MTBF (mean time between failures) improves exponentially with each additional synchronizer stage.

module counter_load (
    input        clk, arst, load, en,
    input  [3:0] d,
    output reg [3:0] q
);
    always @(posedge clk or posedge arst) begin
        if (arst)
            q <= 4'd0;
        else if (load)
            q <= d;
        else if (en)
            q <= q + 1;
    end
endmodule

Async reset clears on arst edge regardless of clock. Load takes parallel data on clock edge. Enable controls counting.

Characteristic equation: Q(t+1) = J·Q̄ + K̄·Q

4-bit SISO (Serial-In Serial-Out) shift register: 4 D flip-flops in series. Input 1,0,1,1 (MSB first).

Clk  In   Q₀  Q₁  Q₂  Q₃(out)
 0   -     0   0   0   0
 1   1     1   0   0   0
 2   0     0   1   0   0
 3   1     1   0   1   0
 4   1     1   1   0   1  ← first bit out

After 4 clocks, the input sequence 1,0,1,1 appears at Q₃ in order.

4-bit ring counter, init = 1000. Each clock, the single 1 bit rotates right (or left).

Cycle:  Q₃Q₂Q₁Q₀
  0:    1  0  0  0
  1:    0  1  0  0
  2:    0  0  1  0
  3:    0  0  0  1
  4:    1  0  0  0  (repeats)

Period = 4. Only 4 of 16 states are used.

4-bit Johnson counter: complement of Q₃ fed back to D₀ (or Q₃ fed to rightmost). Init = 0000.

Cycle:  Q₃Q₂Q₁Q₀
  0:    0  0  0  0
  1:    1  0  0  0
  2:    1  1  0  0
  3:    1  1  1  0
  4:    1  1  1  1
  5:    0  1  1  1
  6:    0  0  1  1
  7:    0  0  0  1
  8:    0  0  0  0  (repeats)

Johnson counter has 2n = 8 valid states (for n=4 FFs); the other 8 states are unused/invalid.

module shift_reg8 (
    input        clk, s0, s1,  // mode: s1s0
    input        si_l, si_r,   // serial in left/right
    input  [7:0] d,
    output reg [7:0] q
);
    // 00=hold, 01=shift right, 10=shift left, 11=parallel load
    always @(posedge clk) begin
        case ({s1,s0})
            2'b00: q <= q;
            2'b01: q <= {si_r, q[7:1]};  // shift right
            2'b10: q <= {q[6:0], si_l};  // shift left
            2'b11: q <= d;
        endcase
    end
endmodule

module bcd_counter (
    input  clk, rst,
    output reg [3:0] count
);
    always @(posedge clk) begin
        if (rst || count == 4'd9)
            count <= 4'd0;
        else
            count <= count + 1;
    end
endmodule

Counts 0-9; synchronous reset to 0 when reaching 9 or when rst is asserted.

module counter_async_rst (
    input        clk, arst,
    output reg [3:0] q
);
    always @(posedge clk or posedge arst) begin
        if (arst)
            q <= 4'd0;
        else
            q <= q + 1;
    end
endmodule

Async reset: the posedge arst in sensitivity list causes immediate reset independent of clock.

module shift_detect (
    input  clk, rst, si,
    output detected
);
    reg [3:0] sr;
    always @(posedge clk or posedge rst) begin
        if (rst) sr <= 4'b0;
        else     sr <= {sr[2:0], si};
    end
    assign detected = (sr == 4'b1011);
endmodule

Shift register captures last 4 bits. Output asserts when pattern 1011 is in the register.

Modulo-6 counter sequences: 000→001→010→011→100→101→000.

State 110 and 111 unused. Using D FFs, derive next-state equations via K-maps:

D₂: K-map over present states (with don't cares at 6,7): D₂ = Q₂Q̄₀ + Q₁Q₀

D₁: D₁ = Q₁⊕Q₀ (standard counter rule for bit 1)

D₀: D₀ = Q̄₂Q̄₁Q̄₀ + Q₂Q̄₁Q̄₀... simplified: D₀ = Q̄₀·(Q̄₂+Q̄₁)

Output can be count value or a terminal count = Q₂Q̄₁Q₀ ... Verify by tracing all 6 states.

Given: t_su=2 ns, t_h=1 ns, t_cQ=3 ns, t_logic=5 ns, t_skew=0.5 ns.

Timing constraint: T_clk ≥ t_cQ + t_logic + t_su + t_skew

T_clk ≥ 3 + 5 + 2 + 0.5 = 10.5 ns

Max clock frequency = 1 / 10.5 ns ≈ 95.2 MHz

Hold constraint: t_cQ + t_logic_min ≥ t_h + t_skew → must also be verified separately.

LFSR with polynomial x³+x+1 (taps at positions 3 and 1). Feedback = Q₃⊕Q₁. Starting from 001:

State (Q₃Q₂Q₁): feedback = Q₃⊕Q₁
001 → fb=0⊕1=1 → 100...
Wait, standard LFSR: Q₁←Q₂, Q₂←Q₃, Q₃←fb
001: fb=Q₁⊕Q₃=1⊕0=1 → next: Q₃=1,Q₂=0,Q₁=0 → 100
100: fb=0⊕1=1 → 110
110: fb=0⊕1=1 → 111
111: fb=1⊕1=0 → 011... wait recheck taps x³+x+1: feedback into position 0 = Q₃⊕Q₁?
For right-shift LFSR: new_bit = Q[3]⊕Q[1]:
001→100→110→111→011→101→010→001 (7 states)

Period = 7 = 2³−1. Yes, maximal-length LFSR (2ⁿ−1 = 7 for n=3).

Divide-by-6 with 50% duty cycle: count 0-5, output high for 0-2, low for 3-5 (but this gives 3/6 = 50%).

module div6 (
    input  clk, rst,
    output reg out
);
    reg [2:0] cnt;
    always @(posedge clk or posedge rst) begin
        if (rst) begin cnt <= 0; out <= 0; end
        else if (cnt == 3'd5) begin cnt <= 0; out <= ~out; end
        else if (cnt == 3'd2) begin cnt <= cnt+1; out <= ~out; end
        else cnt <= cnt + 1;
    end
endmodule

Alternatively use two edges: toggle on count 0 and count 3 to achieve 50% duty cycle at f_clk/6.

module D_latch(input D, Clk, output reg Q);
  always @(D, Clk)
    if (Clk) Q = D;
endmodule

module flipflop(input D, Clock, output reg Q);
  always @(posedge Clock)
    Q <= D;
endmodule

With blocking assignments inside always @(posedge Clock), the statements execute in order, so $Q1$ is updated first to $D$, then $Q2$ is set to the new $Q1$ — i.e. both registers receive $D$ on the same edge. The synthesizer infers a single flip-flop driven by $D$ whose output fans out to both $Q1$ and $Q2$ (or two flip-flops with their D inputs tied together).

Non-blocking assignments evaluate all RHS at the edge, then update LHS. So $Q2$ samples the old $Q1$ (the value before this edge), then $Q1$ receives $D$. This is exactly the behavior of two cascaded flip-flops: $D \to Q1 \to Q2$ — a 2-stage shift register.

Blocking assignments execute sequentially, so when the synthesizer sees $f = x_1\,\&\, x_2$ followed by $g = f\,|\, x_3$, the variable $f$ is treated as a wire (its value is the AND result on this cycle). The result: $f$ is registered, but $g$'s flip-flop is fed by $(x_1 x_2)\,|\, x_3$ — i.e. the AND gate output goes through an OR with $x_3$ before the $g$ flip-flop, with no register between them on this path.

With non-blocking, $g \le f \,|\, x_3$ uses the previous value of $f$ (the flip-flop output). So the OR gate that feeds the $g$ flip-flop is driven by the registered $f$, not the live $x_1 x_2$ AND output. Effectively $g$ is one cycle delayed relative to the blocking version, and the AND output no longer feeds OR directly.

module dff_async_reset(input D, Clock, Resetn, output reg Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= 1'b0;
    else         Q <= D;
endmodule

module dff_sync_reset(input D, Clock, Resetn, output reg Q);
  always @(posedge Clock)
    if (!Resetn) Q <= 1'b0;
    else         Q <= D;
endmodule

module regn #(parameter n=8)(input [n-1:0] R, input Clock, Resetn,
                              output reg [n-1:0] Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= {n{1'b0}};
    else         Q <= R;
endmodule

module muxdff(input D0, D1, Sel, Clock, output reg Q);
  always @(posedge Clock) Q <= Sel ? D1 : D0;
endmodule

module shift4(input [3:0] R, input L, w, Clock, output [3:0] Q);
  muxdff s3(Q[3], R[3], L, Clock, Q[3]);  // bit 3: serial-in is w? -> see below
  // Correct wiring (left-shift serial-in = w):
  // bit3 D0 = Q[2] (shift), D1 = R[3] (load)
  muxdff u3(Q[2], R[3], L, Clock, Q[3]);
  muxdff u2(Q[1], R[2], L, Clock, Q[2]);
  muxdff u1(Q[0], R[1], L, Clock, Q[1]);
  muxdff u0(w,    R[0], L, Clock, Q[0]);
endmodule

module shift4(input [3:0] R, input L, w, Clock, output reg [3:0] Q);
  always @(posedge Clock) begin
    if (L) Q <= R;
    else   Q <= {Q[2:0], w};   // shift left, w in at LSB
  end
endmodule

module shiftn #(parameter n=8)(input [n-1:0] R, input L, w, Clock,
                                 output reg [n-1:0] Q);
  integer k;
  always @(posedge Clock) begin
    if (L) Q <= R;
    else begin
      for (k = n-1; k > 0; k = k-1) Q[k] <= Q[k-1];
      Q[0] <= w;
    end
  end
endmodule

module upcnt4(input Clock, Resetn, E, output reg [3:0] Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= 4'b0;
    else if (E)  Q <= Q + 1;
endmodule

module upcnt4_load(input [3:0] R, input Clock, Resetn, L, E,
                    output reg [3:0] Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= 4'b0;
    else if (L)  Q <= R;
    else if (E)  Q <= Q + 1;
endmodule

module dncnt #(parameter n=4)(input [n-1:0] R, input Clock, Resetn, L, E,
                                output reg [n-1:0] Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= {n{1'b0}};
    else if (L)  Q <= R;
    else if (E)  Q <= Q - 1;
endmodule

module updncnt #(parameter n=4)(input [n-1:0] R, input Clock, Resetn, L, E,
                                  input up_down, output reg [n-1:0] Q);
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) Q <= {n{1'b0}};
    else if (L)  Q <= R;
    else if (E)  Q <= up_down ? Q + 1 : Q - 1;
endmodule

Let $t_{skew} = \delta_2 - \delta_1$. The data launch time at $Q_1$ is $\delta_1 + t_{cQ}$; data must reach $Q_2$ before $\delta_2 + T - t_{su}$. (a) Setup: $\delta_1 + t_{cQ} + t_L + t_{su} \le \delta_2 + T$, so $T_{min} = t_{cQ} + t_L + t_{su} - t_{skew}$. (b) Hold: data from the same edge must not reach $Q_2$ before its hold time after the edge: $\delta_1 + t_{cQ} + t_l \ge \delta_2 + t_h$, i.e. $t_{cQ} + t_l \ge t_h + t_{skew}$. Negative skew increases the hold-time pressure.

With each gate adding delay $\Delta$, internal node $A$ (after one gate) lags $C$ by $\Delta$; $B$ (after two gates) lags by $2\Delta$. Edges that occur within a window of width $\Delta$ on $C$ produce glitches (hazards) on $B$ when an AND/OR re-converges paths of unequal delay. The waveforms show $A$ and $B$ as $C$ with phase shifts of $\Delta$ and $2\Delta$, and any glitch widths equal to the path-delay difference.

Build the excitation table for the JK pair given inputs $J_0=w$, $K_0=w$, $J_1=w Q_0$, $K_1=w$ (typical Figure 5.71 wiring). Starting at $Q_1 Q_0 = 00$: pulse → $01$; pulse → $10$; pulse → $00$; … sequence $00 \to 01 \to 10 \to 00 \dots$ — a modulo-3 (divide-by-3) counter. Output $Q_1 Q_0$ takes 3 distinct states.

Datapath: a 6-bit register $S$ accumulates the running total. On each Coin pulse, encode $\{N,D,Q\}$ to a 5-bit value $V \in \{5,10,25\}$: $V = N\cdot 5 + D\cdot 10 + Q\cdot 25$. A 6-bit adder forms $S + V$ and clocks the result back into $S$ on the negative edge of Coin. Reach-30 detection: $Z = 1$ when $S \ge 30$. Since $30 = 011110_2$, $Z = s_5 + s_4 s_3 s_2 s_1$ covers all values $\ge 30$ in the achievable range. Resetn asynchronously zeros the register.

module vending(input N, D, Q, Coin, Resetn, output Z);
  reg [5:0] S;
  wire [4:0] inc = N ? 5'd5 : D ? 5'd10 : Q ? 5'd25 : 5'd0;
  always @(negedge Resetn, negedge Coin)
    if (!Resetn) S <= 6'd0;
    else         S <= S + {1'b0, inc};
  assign Z = S[5] | (S[4] & S[3] & S[2] & S[1]);
endmodule

The straight design has cascaded ANDs with delay $(n-1)t_{AND}$ between $Q_0$ and the highest XOR. Refactor by using an XOR for each toggle bit and an AND tree of fan-in 2 driven directly from $Q_0\dots Q_{i-1}$, so the longest path is one AND followed by one XOR. Then $T_{min} = t_{cQ} + t_{AND} + t_{XOR} + t_{su} = 1.0 + 1.4 + 1.2 + 0.6 = 4.2$ ns, so $F_{max} = 1/4.2\,\text{ns} \approx 238$ MHz.

Worst-case path through the FSM: $T_{min}(i\to j) = t_{cQ} + t_{path} + t_{su} - (\delta_j - \delta_i)$. Hold check: $t_{cQ} + t_{min\,path} \ge t_h + (\delta_j - \delta_i)$. (i) zero skew: typical $T_{min}\approx 1.0+t_{path}+0.6$. (ii) skew on $\delta_2$: paths into FF2 gain $0.7$ ns slack, paths out of FF2 lose $0.7$ ns. Compute worst path; one of the data paths probably becomes the bottleneck, producing a slightly slower $F_{max}$. (iii) $\delta_1=1, \delta_2=0, \delta_3=0.5$: paths $1\to 2$ have $t_{skew}=-1$ ns (penalty), paths $2\to 3$ get +0.5 ns slack. The worst-case path tightens the clock; check hold: $0.8 + t_{l} \ge 0.4 + (\delta_j-\delta_i)$ — for $1\to 2$ with skew $-1$, hold is automatic. Compute each $F_{max}$ from the dominant path.

Moore FSM detecting 101 (overlapping). States: S0(initial), S1(got 1), S2(got 10), S3(got 101, output=1).

S0 --1--> S1 --0--> S2 --1--> S3(out=1)
S0 --0--> S0
S1 --1--> S1
S2 --0--> S0
S3 --1--> S1 (overlap: 101, last 1 starts new)
S3 --0--> S2 (overlap: 1010...)

Output z=1 only in S3 (Moore: output on state, not transition).

State encoding: S0=00, S1=01, S2=10, S3=11. Using D FFs:

D₁ = Q₁Q̄₀w + Q̄₁Q₀w̄ + Q₁Q₀w̄; D₀ = Q̄₁w + Q₀; z = Q₁Q₀

Moore FSM: output 1 after three consecutive 1s. States represent count of consecutive 1s: A(0), B(1), C(2), D(3+, output=1).

A --0--> A, --1--> B
B --0--> A, --1--> C
C --0--> A, --1--> D
D --0--> A, --1--> D

Output: z=0 for A,B,C; z=1 for D.

4 states needed. D FF next-state equations derived from state table.

Mealy FSM for three consecutive 1s: output z=1 on the transition that completes the third 1.

States: A(0 ones), B(1 one), C(2 ones). Output on A→B→C→C (z=1 on C→C and all further 1s from C).

A --0/0--> A, --1/0--> B
B --0/0--> A, --1/0--> C
C --0/0--> A, --1/1--> C

State count comparison: Mealy needs 3 states; Moore needs 4 states. Mealy FSMs typically need fewer states because the output depends on both state and input.

Moore FSM: Output depends only on current state. Output is stable for entire clock period; simpler to analyze.

Mealy FSM: Output depends on current state AND current input. Output can change mid-cycle if input changes; typically needs fewer states; output arrives one cycle earlier than Moore for same sequence.

Moore advantages: Glitch-free outputs (registered), easier timing analysis.

Mealy advantages: Fewer states (more compact), faster response (output same cycle as input).

One-hot encoding: One flip-flop per state; exactly one FF is 1 at any time. Simple next-state logic (each state is directly a FF output).

5-state comparison:

• Binary encoding: ⌈log₂5⌉ = 3 flip-flops.
• One-hot encoding: 5 flip-flops (one per state).

One-hot uses more FFs but reduces combinational logic complexity, which is preferable on FPGAs where FFs are plentiful and LUTs are the scarce resource.

module seq_detect_101 (
    input  clk, rst, w,
    output z
);
    parameter S0=2'd0, S1=2'd1, S2=2'd2, S3=2'd3;
    reg [1:0] state, next;
    always @(posedge clk or posedge rst)
        state <= rst ? S0 : next;
    always @(*) begin
        case (state)
            S0: next = w ? S1 : S0;
            S1: next = w ? S1 : S2;
            S2: next = w ? S3 : S0;
            S3: next = w ? S1 : S2;
            default: next = S0;
        endcase
    end
    assign z = (state == S3);
endmodule

module mealy_01detect (
    input  clk, rst, w,
    output z
);
    // Output 1 when last two inputs were 01
    parameter A=1'b0, B=1'b1;
    reg state;
    always @(posedge clk or posedge rst)
        state <= rst ? A : (w ? A : B);
    // state B = last input was 0
    // Mealy output: z=1 when in state B and w=1
    assign z = (state == B) & w;
endmodule

State A: last input was 1 (or initial). State B: last input was 0. Output z=1 when transitioning B→A on w=1 (sequence 0 then 1 = "01").

Given a state table with states, inputs, next states, and outputs, derive equations by:

1. Assign binary codes to states.
2. Build next-state table (D₁, D₀ for 2 FFs).
3. Minimize each D function via K-map.
4. Build output equation from output column.

For a specific table (example with 3 states, 1 input): the equations emerge directly from K-map minimization of each column (D₁, D₀, z) treating state bits and input as variables.

State minimization via partitioning (Myhill-Nerode / implication chart):

1. Initial partition: group states by output values.
2. Refine: within each group, if two states have transitions to different groups under same input, split them.
3. Repeat until no more splits.
4. States remaining in same partition class are equivalent — merge them.

For states A–F: build implication chart, mark incompatible pairs (different outputs), propagate implications. Remaining unmarked pairs are equivalent. Merge to minimal state machine.

Vending machine ASM: accepts nickels (N) and dimes (D); item costs 15¢; dispenses change.

States: S0(0¢), S5(5¢), S10(10¢), S15(dispense)
ASM blocks:
S0: if N→S5; if D→S10
S5: if N→S10; if D→S15(no change)
S10: if N→S15(no change); if D→S15+change(5¢)
S15: dispense item; if change needed→output nickel; →S0

Decision boxes for N, D inputs; conditional outputs for item and change; state boxes for each amount. ASM chart uses rectangles (states), diamonds (decisions), ovals (conditional outputs).

module traffic_light (
    input  clk, rst,
    output reg [1:0] light  // 00=red,01=green,10=yellow
);
    reg [4:0] cnt;
    // Green 20 cycles, Yellow 5, Red 20
    always @(posedge clk or posedge rst) begin
        if (rst) begin cnt<=0; light<=2'b01; end
        else begin
            cnt <= cnt + 1;
            case (light)
                2'b01: if(cnt==19) begin light<=2'b10; cnt<=0; end
                2'b10: if(cnt==4)  begin light<=2'b00; cnt<=0; end
                2'b00: if(cnt==19) begin light<=2'b01; cnt<=0; end
            endcase
        end
    end
endmodule

Detect 1001 or 1111 (overlapping). Build a state machine tracking the longest suffix of the input that is a prefix of either target:

States: S0, S1(1), S2(10), S3(100), S4(11), S5(111)
S0: 0→S0, 1→S1
S1: 0→S2, 1→S4
S2: 0→S0, 1→S3
S3: 0→S0, 1→S4; on completing 1001 at S3+input 1: output=1, goto S1
S4: 0→S2, 1→S5; completing 1001: S3→w=1→output+goto S1
S5: 0→S2, 1→S1; completing 1111: S5+1→output=1, goto S4

Output 1 when completing 1001 (state S3, input 1) or 1111 (state S5+, input 1). Moore output can be added with an accepting state.

A finite state machine is a 5-tuple M = (Q, Σ, δ, q₀, F) where:

• Q — finite set of states
• Σ — finite input alphabet
• δ: Q × Σ → Q — transition function (next-state function)
• q₀ ∈ Q — initial state
• F ⊆ Q — set of accepting/output states (for Moore: output function λ: Q → Λ; for Mealy: λ: Q × Σ → Λ)

Given: D₁ = x⊕Q₂, D₂ = Q₁ (using Q₁,Q₂ as state, x as input).

Next state: Q₁⁺=x⊕Q₂, Q₂⁺=Q₁.

State diagram: 4 states, each with two transitions. No output specified → count states/cycles for analysis or add output function.

Arbiter FSM for r₁, r₂: grants access one at a time.

States: IDLE, GRANT1, GRANT2
IDLE: if r₁&r₂→GRANT1 (r₁ priority); if r₁ only→GRANT1; if r₂ only→GRANT2; else→IDLE
GRANT1: g₁=1; if ~r₁→IDLE (r₁ releases); else hold
GRANT2: g₂=1; if ~r₂→IDLE; else hold

Outputs: g₁=1 in GRANT1, g₂=1 in GRANT2. The arbiter ensures mutual exclusion: only one grant active at a time. Priority given to r₁ when both assert simultaneously.

Advantages of ASM charts over state diagrams:

• Clearly show data operations (register transfers) alongside control flow.
• Conditional outputs (on arcs/diamonds) are explicit without cluttering state bubbles.
• More readable for hardware designers familiar with flowcharts.
• Directly model datapath+control interactions; easier to derive RTL from ASM block structure.
• Timing is implicit: each ASM block executes in one clock cycle.

1. State diagram / state table: Capture desired behavior with states, inputs, transitions, outputs.
2. State minimization: Merge equivalent states to reduce hardware.
3. State assignment: Assign binary codes to states (binary, Gray, one-hot).
4. Flip-flop selection: Choose FF type (D, JK, T); derive excitation equations.
5. Minimization: Minimize next-state and output equations (K-maps or Quine-McCluskey).
6. Circuit implementation: Realize equations with gates and FFs; verify with simulation.

// Binary encoding (3 FFs for 8 states)
parameter [2:0] S0=3'd0, S1=3'd1, ..., S7=3'd7;

// Gray encoding (3 FFs, adjacent states differ by 1 bit - reduces glitches)
parameter [2:0] S0=3'b000, S1=3'b001, S2=3'b011, S3=3'b010,
                S4=3'b110, S5=3'b111, S6=3'b101, S7=3'b100;

// One-hot (8 FFs, simplest next-state logic - preferred for FPGAs)
parameter [7:0] S0=8'b00000001, S1=8'b00000010, S2=8'b00000100,
                S3=8'b00001000, S4=8'b00010000, S5=8'b00100000,
                S6=8'b01000000, S7=8'b10000000;

One-hot: 8 FFs but next-state = OR of incoming state FFs (no decoding needed). Best for FPGAs. Binary: 3 FFs but complex decode. Gray: reduces output glitches for counters.

Modulo-8 up counter via sequential design. States 000–111, next state = present state + 1 (mod 8). Using D FFs:

State table same as binary counter. K-maps for D₂,D₁,D₀:

D₀ = Q̄₀ (toggles every cycle)

D₁ = Q₁⊕Q₀

D₂ = Q₂⊕(Q₁Q₀)

Output = state value (Q₂Q₁Q₀). Implementation: D₀=NOT Q₀; D₁=XOR(Q₁,Q₀); D₂=XOR(Q₂,AND(Q₁,Q₀)).

Mod-8 counter with JK FFs. JK excitation table: J=0,K=x (hold 0); J=x,K=0 (hold 1); J=1,K=x (set); J=x,K=1 (reset).

J₀=1, K₀=1 (always toggle for LSB).

J₁=Q₀, K₁=Q₀ (toggle when Q₀=1).

J₂=Q₁Q₀, K₂=Q₁Q₀ (toggle when Q₁Q₀=1).

Gate count comparison: JK implementation uses same XOR structure but expressed as J/K. D FF version needs explicit XOR gates; JK version has simpler excitation equations (J₀=K₀=1 is trivially wired). Total gates similar; JK can save gates when J≠K not needed.

Output 1 when w₁=w₂ for four consecutive cycles. States track count of consecutive equal cycles: 0,1,2,3,4(output).

5 states, 2 inputs. Let eq = (w₁ XNOR w₂).

S0 --eq=1--> S1, --eq=0--> S0
S1 --eq=1--> S2, --eq=0--> S0
S2 --eq=1--> S3, --eq=0--> S0
S3 --eq=1--> S4, --eq=0--> S0
S4 --eq=1--> S4, --eq=0--> S0

Moore output: z=1 in S4. Need 3 FFs (⌈log₂5⌉=3). Reduce by noting the counter simply counts eq=1 consecutive cycles.

Serial adder FSM: Adds two serial bit streams A, B. State = carry. Inputs: aᵢ, bᵢ. Output: sᵢ = aᵢ⊕bᵢ⊕carry.

Moore version: 2 states (carry=0, carry=1). Output sᵢ must depend on input too, so pure Moore would need 4 states. Mealy is more natural.

Mealy version (2 states):

Converting Mealy FSM to Moore FSM:

1. For each Mealy state-input pair (s, i) that produces output z, create a new Moore state s_z.
2. Any transition leading to s under input i now leads to s_z; s_z has output z.
3. Transitions from s_z for all inputs are the same as from s in the Mealy machine.

Result: Moore FSM typically has more states (one per unique (state, output) combination). If the Mealy FSM has n states and 2 outputs, the Moore FSM can have up to 2n states.

Detect 110 or 101 (overlapping). States track suffix matching either pattern:

S0: initial
S1: saw 1
S2: saw 11
S3: saw 10
S0 --1--> S1, --0--> S0
S1 --1--> S2, --0--> S3
S2 --1--> S2(output), --0--> S3(output for 110)
S3 --1--> S1(output for 101), --0--> S0

Moore output: z=1 in states reached when pattern completed. Minimal: 4 states. Output state added or Moore output on S2(w=0) and S3(w=1).

For Mealy: output on transitions S2→(0) and S3→(1).

Four states $A, B, C, D$. State $A$ idles; on $w=1$ go to $B$, then $C$, then $D$ unconditionally, then back to $A$. Outputs (Moore): $B$: $R2_{out}=R3_{in}=1$ (copy R2→R3); $C$: $R1_{out}=R2_{in}=1$ (R1→R2); $D$: $R3_{out}=R1_{in}=Done=1$ (R3→R1). State assignment $A=00, B=01, C=10, D=11$. With $y_2 y_1$ as state bits: $Y_1 = w\overline{y_2}\overline{y_1} + y_2$, $Y_2 = y_1 + y_2\overline{y_1}$ (and so on). The four-cycle sequence performs the swap via R3 as scratch.

Swapping $C$ and $D$ assigns $C=11, D=10$. Recompute next-state K-maps: $Y_1$ now requires $w\overline{y_2}\overline{y_1}$ for $A\to B$ and additional terms for $B\to D$ and $C\to A$ that no longer combine as nicely. Output expressions also become more complex. The total gate-input cost rises by ~3 inputs versus the straightforward assignment, illustrating that state-assignment choice matters.

One-hot: $y_1=A, y_2=B, y_3=C, y_4=D$. Next-state: $Y_1 = (y_1\overline{w}) + y_4$, $Y_2 = w y_1$, $Y_3 = y_2$, $Y_4 = y_3$. Outputs come directly: e.g. $R1_{out} = y_3$, Done = $y_4$. Each transition is one product term. Trade-off: more flip-flops (4 vs 2), but simpler combinational logic and faster speed.

Mealy version uses 3 states $A, B, C$. In $A$, on $w=1$, output $R2_{out}=R3_{in}=1$ and go to $B$. In $B$, output $R1_{out}=R2_{in}=1$ and go to $C$. In $C$, output $R3_{out}=R1_{in}=Done=1$ and go to $A$. The output is generated combinationally with the transition (one clock cycle earlier than Moore), saving one cycle. Total: 3 cycles instead of 4. Care must be taken so outputs do not glitch on combinational input changes.

module swap_fsm(input Clock, Resetn, w, output R1in,R1out,R2in,R2out,R3in,R3out,Done);
  reg [1:0] y, Y;
  parameter A=2'b00, B=2'b01, C=2'b10, D=2'b11;
  always @(*) case (y)
    A: Y = w ? B : A;
    B: Y = C;
    C: Y = D;
    D: Y = A;
  endcase
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) y <= A; else y <= Y;
  assign R2out = (y==B), R3in = (y==B);
  assign R1out = (y==C), R2in = (y==C);
  assign R3out = (y==D), R1in = (y==D), Done = (y==D);
endmodule

Partition refinement (output classes first). $P_1$: split states by output → e.g. $\{S_1, S_3, S_5\}$ (output 0) and $\{S_2, S_4, S_6, S_7\}$ (output 1). $P_2$: refine by where each state's transitions go (which $P_1$ block). After several iterations $P_3 = P_4$ and the equivalent classes form 4 states. The reduced FSM has 4 states with the same I/O behavior.

States track cumulative deposit modulo 25 (with credit): $S_0=0$¢, $S_5, S_{10}, S_{15}, S_{20}, S_{25}$. Inputs: nickel ($n$), dime ($d$). Transitions: $S_0$: n→$S_5$, d→$S_{10}$; $S_5$: n→$S_{10}$, d→$S_{15}$; $S_{10}$: n→$S_{15}$, d→$S_{20}$; $S_{15}$ (dispense)→$S_0$; $S_{20}$ (dispense + 5¢ credit) goes to $S_5$. Apply partitioning to merge $S_{15}$ and $S_{20}$ if their behaviors match — they don't (different next states), so 5 states minimum after dispense logic added.

Four unspecified entries can be set to 0 or 1 to maximize state merging. With both = 0: partitioning yields a coarser refinement and merges 7 states down to (e.g.) 4. With both = 1: a different merging produces (e.g.) 5 states. The exact counts depend on the table; the lesson is that completing don't-cares strategically minimizes states.

Four states $00, 01, 10, 11$. Compute $Y_1 Y_2$ for each $(y_1, y_2, w)$: Substitute and tabulate. The state transitions trace: $00 \xrightarrow{w=1} 10 \xrightarrow{w=1} 11 \xrightarrow{w=1} 11$ (with $z=1$). On $w=0$ from any state, return to $00$ or $01$. Sequence detection: three consecutive 1s drive the FSM to state $11$ where $z=1$. Hence the FSM detects the pattern 111.

Excitation $J_1=w$, $K_1=\overline{w}+y_2$, $J_2=wy_1$, $K_2=\overline{w}$. Compute next-state from JK rules ($Q^+=J\overline{Q}+\overline{K}Q$): $y_1^+ = w\overline{y_1} + (w\overline{y_2}) y_1$, $y_2^+ = wy_1\overline{y_2} + wy_2$. Tabulate four states. The same input sequence $w=111\dots$ produces $00\to 10\to 11\to 11$ with $z=1$ in state $11$ — the same three-1s detector as Example 6.9.

$Y_1^+ = D_1 = w(y_1+y_2)$. $Y_2^+ = y_2 \oplus T_2$ for the T flip-flop, where $T_2 = \overline{w}y_2 + wy_1\overline{y_2}$. Build the excitation table: starting at 00, with $w=1\to 00, 10, 11, 11$… The output $z = y_1 y_2$ asserts in state 11. Same FSM behavior as 6.9 and 6.10 — three consecutive 1s detector.

States: $A$ (idle), $B$ (just saw 0), $C$ (saw 00, output $z=1$), $D$ (just saw 1), $E$ (saw 11, output $z=1$). Transitions: $A$: 0→$B$, 1→$D$. $B$: 0→$C$, 1→$D$. $C$: 0→$C$, 1→$D$. $D$: 0→$B$, 1→$E$. $E$: 0→$B$, 1→$E$. Output $z=1$ in $C$ and $E$. State assignment for 5 states needs 3 flip-flops. Derive $Y_3 Y_2 Y_1$ next-state from K-maps.

FSM_1 (detect 11): states $A_1$ (idle), $B_1$ (saw 1), $C_1$ (saw 11, $z_1=1$). FSM_2 (detect 00): mirror. Output $z = z_1 + z_2$. Each FSM has 3 states, total 6 states, but they run independently and combine more cleanly than the monolithic 5-state version. The cost trade-off depends on shared inputs.

Mealy: 3 states $A$ (idle), $B$ (just saw 0), $C$ (just saw 1). $A$: 0/0→$B$, 1/0→$C$. $B$: 0/1→$B$, 1/0→$C$. $C$: 0/0→$B$, 1/1→$C$. Output asserts on the input edge that completes the pair. With 2 flip-flops $y_1 y_2$: $Y_1 = w$, $Y_2 = \overline{w}y_2 + wy_1\dots$ (assignment-dependent), $z = wy_1 + \overline{w}y_2$ (output high when current input matches the previous one).

For each present-state/next-state transition, JK excitation is: $00\to 0$: J=0, K=X; $0\to 1$: J=1, K=X; $1\to 0$: J=X, K=1; $1\to 1$: J=X, K=0. Build K-maps for $J_1, K_1, J_2, K_2, J_3, K_3$ from the assigned table. Resulting expressions typically simpler than D-FF case because of the X cells. The book's solution shows a ~30% input-count reduction over D realization.

module seq_moore(input Clock, Resetn, w, output reg z);
  reg [2:0] y, Y;
  parameter A=0, B=1, C=2, D=3, E=4;
  always @(*) case (y)
    A: Y = w ? D : B;
    B: Y = w ? D : C;
    C: Y = w ? D : C;
    D: Y = w ? E : B;
    E: Y = w ? E : B;
    default: Y = A;
  endcase
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) y <= A; else y <= Y;
  always @(*) z = (y == C) || (y == E);
endmodule

module seq_mealy(input Clock, Resetn, w, output reg z);
  reg [1:0] y, Y;
  parameter A=0, B=1, C=2;
  always @(*) begin
    z = 0;
    case (y)
      A: if (w) Y = C; else Y = B;
      B: if (w) Y = C; else begin Y = B; z = 1; end
      C: if (w) begin Y = C; z = 1; end else Y = B;
      default: Y = A;
    endcase
  end
  always @(negedge Resetn, posedge Clock)
    if (!Resetn) y <= A; else y <= Y;
endmodule

Frame: 1 start bit (0) + 7 data bits + 1 stop bit (1) = 9 bits. State machine with 11 states: idle, start, $b_0\dots b_6$, stop, done. A counter cycles through bit positions; on each clock the output line presents the next bit of the shift register holding the byte. Done asserts after the stop bit. Verilog uses a parallel-load shift register and a state-FSM controller.

Tri-state bus: Multiple drivers share a wire; only one enabled at a time via OE (output-enable). Bus floats when all disabled. Simple wiring but requires careful control; floating bus can pick up noise.

Mux-based bus: A multiplexer selects one of several sources; no tri-state buffers needed. Cleaner logic, easier to synthesize for FPGAs (which have no tri-state internal routing), but requires a physical mux and select signals.

FPGAs prefer mux-based; external PCB buses often use tri-state.

Simple processor datapath with 4 registers R0–R3, ALU, and bus:

Registers R0-R3 ──┐
                  ├──► MUX_A ──► ALU_A
                  └──► MUX_B ──► ALU_B
ALU: ADD/SUB/AND/OR with opcode input
ALU result ──► Result register ──► Write-back MUX ──► Rn
Control signals: RegSel_A[2], RegSel_B[2], ALUop[2],
                 WE (write enable), RegDest[2]

The controller FSM sequences: fetch operands → execute ALU op → write back to destination register.

Arithmetic mean of four 8-bit numbers: sum A+B+C+D (10-bit result), then divide by 4 (right-shift 2).

ASM chart:
IDLE ──start──► LOAD: Sum:=0, cnt:=4, ptr:=0
ACCUM: Sum:=Sum+Data[ptr]; ptr:=ptr+1; cnt:=cnt-1
  ──(cnt≠0)──► ACCUM
  ──(cnt=0)──► COMPUTE: Mean:=Sum>>2
COMPUTE ──► DONE: output Mean; done:=1 ──► IDLE

Datapath: 10-bit accumulator, 4-element memory or 4 input ports, 2-bit right shifter, done flag.

Why debouncing is needed: Mechanical switches bounce for 1–20 ms, generating multiple transitions. Without debouncing, a single press registers as multiple pulses.

Hardware approach: SR latch with two SPDT switch positions; once set/reset, further bounces have no effect. Alternatively: RC low-pass filter + Schmitt trigger.

Software approach: Sample the switch; if value is stable for N consecutive samples (e.g., 20 ms at 1 ms intervals), accept it as a valid press. Implemented with a counter in software or HDL.

Clock skew: Difference in arrival time of the clock signal at different flip-flops. Caused by wire delays, buffer delays.

Setup violation due to skew: If the destination FF clock arrives earlier than the source FF clock, the effective cycle time is reduced. Constraint: T_clk + t_skew_neg ≥ t_cQ + t_logic + t_su.

Hold violation due to skew: If destination FF clock arrives later, data may change before being captured. Constraint: t_cQ + t_logic_min ≥ t_h + t_skew_pos. Hold violations cannot be fixed by slowing the clock — they require adding buffers on the data path.

Microinstructions for a simple processor (horizontal microcode):

Each microinstruction specifies: source register(s), ALU operation, destination register, memory operation, and next microaddress.

Register Transfer Level (RTL) design describes a digital system as registers storing data and combinational logic transferring data between registers in each clock cycle.

Key concepts: registers hold state; data operations described as register transfers (e.g., A ← B + C); controller FSM determines which transfers occur each cycle; datapath implements the hardware to perform those operations.

RTL abstraction sits between algorithmic level and gate level — high enough for synthesis tools to optimize, detailed enough to specify hardware behavior precisely.

GCD using Euclidean algorithm: GCD(A,B) = GCD(B, A mod B). In hardware use repeated subtraction: GCD(A,B): while A≠B, if A>B then A←A−B else B←B−A.

ASM chart:
IDLE ──start──► LOAD: A:=a, B:=b
TEST: if A==B → DONE: gcd:=A
      if A>B → SUBA: A:=A-B → TEST
      else → SUBB: B:=B-A → TEST

Datapath: two registers A, B; subtractor; comparator. Cycles for worst case: O(max(a,b)).

module counter_param #(parameter N=8) (
    input              clk, rst, en, load,
    input  [N-1:0]     d,
    output reg [N-1:0] q
);
    always @(posedge clk or posedge rst) begin
        if (rst)       q <= {N{1'b0}};
        else if (load) q <= d;
        else if (en)   q <= q + 1'b1;
    end
endmodule

An ASM block consists of three elements:

1. State box (rectangle): Represents one state of the FSM. Contains the state name and any Moore (unconditional) outputs. Executes in one clock cycle.
2. Decision box (diamond): Tests a condition (input or status signal). Has two exit paths (Yes/No or 1/0). Does not consume clock cycles — purely combinational.
3. Conditional output box (oval/rounded rectangle): Contains Mealy-type outputs that are asserted only when the associated decision condition is true. Also combinational (no additional clock cycle).

module bit_counter (
    input        clk, rst, start,
    input  [7:0] data,
    output reg [3:0] count,
    output reg done
);
    reg [7:0] sr;
    reg [2:0] cnt;
    reg [1:0] state; // 0=IDLE,1=LOAD,2=COUNT,3=DONE
    always @(posedge clk or posedge rst) begin
        if (rst) begin state<=0; done<=0; count<=0; end
        else case (state)
            0: if (start) begin sr<=data; cnt<=0; count<=0; state<=2; end
            2: begin
                   count <= count + sr[0];
                   sr <= sr >> 1;
                   cnt <= cnt + 1;
                   if (cnt==7) begin done<=1; state<=3; end
               end
            3: begin done<=0; state<=0; end
        endcase
    end
endmodule

Shift-and-add multiplier datapath:

• Register A (8-bit): accumulator, initialized to 0.
• Register Q (4-bit): multiplier, loaded with B.
• Register M (4-bit): multiplicand A.
• Shift register: A:Q shifted right 1 bit each cycle.
• Adder: A + M when Q[0]=1.

ASM chart: INIT state → LOAD A:=0, Q:=B, cnt:=4; LOOP: if Q[0]=1 then A:=A+M; shift A:Q right; cnt:=cnt−1; if cnt=0 → DONE else → LOOP.

Product = A:Q after 4 iterations.

System: read eight 4-bit values, store them, compute sum and average.

Datapath: 8×4-bit memory (or shift reg), 7-bit accumulator, 3-bit counter
Controller ASM:
IDLE ──start──► INIT: acc:=0, cnt:=0
READ: store data[cnt]; acc:=acc+data[cnt]; cnt:=cnt+1
  ──(cnt<8)──► READ
  ──(cnt=8)──► COMPUTE: sum:=acc; avg:=acc>>3 (divide by 8)
DONE: output sum, avg; done:=1 ──► IDLE

Average = sum/8 implemented as 3-bit right shift (exact for multiples of 8, truncated otherwise).

module tristate_bus (
    input      [7:0] data_in,
    input            oe,       // output enable
    output     [7:0] bus
);
    assign bus = oe ? data_in : 8'bz;
endmodule

When oe=1, drives bus with data_in. When oe=0, bus is high-impedance (Z). Multiple such modules can share the same bus wire provided only one has oe=1 at a time.

Sorting network for three 4-bit numbers A, B, C into sorted order X≤Y≤Z:

Step 1: Compare-swap (A,B): if A>B, swap → ensures min(A,B) in A
Step 2: Compare-swap (B,C): if B>C, swap → ensures max in C (= Z)
Step 3: Compare-swap (A,B): if A>B, swap → ensures A≤B (= X≤Y)

3 compare-swap elements total. Each comparator is a 4-bit magnitude comparator with a 4-bit 2-to-1 mux for each output. Fully combinational, depth = 3 comparator levels.

Moore in ASM: Outputs are listed inside the state box (rectangle). Output active for the entire clock cycle regardless of input. Example: state S2 box contains "z=1".

Mealy in ASM: Outputs are listed in oval conditional output boxes on arcs, adjacent to the decision diamond. Output is conditional on both state and input. Example: on the arc from S1 when w=1, an oval box contains "z=1".

Both representations preserve the one-clock-per-ASM-block timing relationship.

Synchronizer circuit: Two cascaded D flip-flops clocked by the destination domain. The first FF captures the asynchronous input (may go metastable); the second FF samples the first FF's output after a full clock period — by which time the first FF has almost certainly resolved.

Why a single FF is insufficient: The first FF's output may still be metastable (not fully resolved to a valid 0 or 1) when the next logic stage samples it, propagating an invalid value. Adding a second FF gives the first FF a full clock period to resolve its metastable state, reducing the probability of failure exponentially.

Repeated-subtraction division: quotient Q=0; while dividend ≥ divisor: dividend−=divisor, Q++.

Pseudo-code:
  Q := 0
  R := dividend  // 255
  D := divisor   // 7
  while R >= D:
      R := R - D
      Q := Q + 1
  // result: Q=36, R=3 (255 = 36×7 + 3)

Cycles for 255÷7: quotient = 36, so loop executes 36 times plus setup = 37 clock cycles (36 subtractions + final comparison). In hardware, each subtraction takes 1 clock cycle.

Shift-and-add multiplier cycles = number of bits in multiplier:

4-bit multiplier: 4 iterations + 1 cycle for initialization + 1 cycle for done = 4 clock cycles (excluding load/done).

8-bit multiplier: 8 clock cycles for the shift-and-add loop.

Generally: n-bit × n-bit multiplier requires n clock cycles for the sequential shift-and-add algorithm. A combinational (array) multiplier requires 0 clock cycles but uses O(n²) gates.

Debouncing circuit using SR latch with SPDT switch and pull-up resistors:

Vcc
 |
R1 (pull-up)
 |──────── S input of SR latch
 |   ← switch common
R2 (pull-up)
 |──────── R input of SR latch
GND

When switch moves to S position: S line is grounded → SR latch Set. Bounces on S side re-apply Set — no effect (already set). When switch moves to R: R line grounded → Reset. Result: Q output changes cleanly once per switch throw, regardless of bouncing contacts.

Clock distribution challenges: Long interconnects introduce varying delays to different FFs (clock skew). Skew causes setup/hold violations.

Techniques to minimize skew:

• Clock tree synthesis (CTS): CAD tool balances the clock tree so all leaf FF clock pins see equal delay. H-tree or balanced binary tree topology.
• Clock buffers: Symmetric placement of equal-strength buffers at each level.
• Low-skew clock networks: FPGAs have dedicated global clock routing with matched delays to all CLBs.
• Clock domains: Limit the size of each synchronous domain; use synchronizers at domain crossings.
• Minimize clock loading: Avoid excessive fan-out on clock net; use clock enable instead of gated clocks.

Restoring divider: 8-bit ÷ 4-bit.

Datapath: 8-bit partial remainder register P, 4-bit divisor D.

Each step: P_shifted = P left-shifted by 1 (bring in next dividend bit); attempt subtraction P_trial = P_shifted − D; if P_trial ≥ 0: quotient bit = 1, P = P_trial; else: quotient bit = 0, P restored (P remains).

ASM chart: INIT(load dividend into P[7:0], D into divisor reg, cnt=4) → SHIFT(P=P<<1) → SUBTRACT(P_trial=P−D) → decision(P_trial[MSB]=0 → quotient bit=1, P=P_trial else bit=0) → cnt−1 → if cnt≠0 repeat → DONE.

module multiplier (
    input        clk, rst, start,
    input  [3:0] A, B,
    output reg [7:0] product,
    output reg done
);
    reg [7:0] acc; reg [3:0] mplier; reg [2:0] cnt;
    always @(posedge clk or posedge rst) begin
        if (rst) begin done<=0; acc<=0; end
        else if (start) begin
            acc<=0; mplier<=B; cnt<=4; done<=0;
        end else if (cnt>0) begin
            if (mplier[0]) acc <= acc + {4'b0,A};
            mplier <= mplier>>1; acc[7:1]<=acc[6:0]; // shift
            // actually: {acc,mplier} >>= 1 with add to upper half
            cnt <= cnt-1;
            if (cnt==1) begin product<=acc+(mplier[0]?{4'b0,A}:0); done<=1; end
        end
    end
endmodule
// Testbench: 13*11=143
// A=4'd13, B=4'd11; expect product=8'd143

For signed (2's complement) multiplication:

• The Baugh-Wooley or Booth algorithm handles signed multiplication directly.
• Simple modification: treat the MSB of both operands as sign bits with weight −2^(n−1).
• Booth encoding: recode the multiplier to reduce the number of partial products; handles sign automatically.
• For shift-and-add: use arithmetic (sign-extending) right shift instead of logical shift; negate the last partial product if the MSB of the multiplier is 1.

UART transmitter: sends 8-bit data with start (0) and stop (1) bits.

ASM: IDLE(tx=1) ──send──► START: tx:=0; cnt:=8
BIT: tx:=data[0]; data:=data>>1; cnt:=cnt-1
  ──(cnt≠0)──► BIT
  ──(cnt=0)──► STOP: tx:=1 ──► IDLE

module uart_tx (
    input       clk, rst, send,
    input [7:0] data,
    output reg  tx, busy
);
    reg [7:0] sr; reg [3:0] cnt;
    reg [1:0] state;
    always @(posedge clk or posedge rst) begin
        if (rst) begin tx<=1; state<=0; busy<=0; end
        else case (state)
            0: if(send) begin sr<=data;cnt<=8;tx<=0;busy<=1;state<=1; end
            1: begin tx<=sr[0];sr<=sr>>1;cnt<=cnt-1;
                     if(cnt==1) state<=2; end
            2: begin tx<=1; busy<=0; state<=0; end
        endcase
    end
endmodule

Datapath: $n$-bit shift register holding the input (MSB shifts out), $\lceil\log_2 n\rceil$-bit counter accumulating ones, and an $n$-bit-cycle counter. Control signals: load, shift, incr, done. ASM chart: state $S_1$ loads input on Start. $S_2$ loops $n$ times: shift the register; if the shifted-out bit is 1, increment the count; decrement the cycle counter. When the cycle counter reaches 0, go to $S_3$ which asserts done.

module bitcount #(parameter n=8)(input Clk, Rst, Start, input [n-1:0] D,
                                  output reg [3:0] cnt, output reg done);
  reg [n-1:0] sr; reg [3:0] k; reg [1:0] s;
  always @(posedge Clk, posedge Rst)
    if (Rst) begin s<=0; cnt<=0; done<=0; end
    else case (s)
      0: if (Start) begin sr<=D; cnt<=0; k<=n; s<=1; done<=0; end
      1: if (k==0) begin done<=1; s<=2; end
         else begin if (sr[n-1]) cnt<=cnt+1; sr<=sr<<1; k<=k-1; end
      2: if (!Start) s<=0;
    endcase
endmodule

Datapath: $n$-bit multiplicand register $M$, $n$-bit multiplier register (LSB shifted out each cycle), $2n$-bit accumulator $A$ initially 0. ASM: in state $S_1$ load operands. In $S_2$ (looping $n$ times): if multiplier LSB = 1, add $M$ to upper half of $A$; right-shift $A$ by one (capturing the next product bit); shift multiplier right. After $n$ iterations the product sits in $A$. State $S_3$ asserts Done.

module shift_add #(parameter n=4)(input Clk, Rst, Start, input [n-1:0] M, Q,
                                    output reg [2*n-1:0] P, output reg done);
  reg [2*n-1:0] A; reg [n-1:0] q; reg [3:0] k; reg [1:0] s;
  always @(posedge Clk, posedge Rst)
    if (Rst) begin s<=0; done<=0; end
    else case (s)
      0: if (Start) begin A<=0; q<=Q; k<=n; s<=1; done<=0; end
      1: begin
          if (q[0]) A[2*n-1:n-1] <= A[2*n-1:n-1] + {1'b0,M};
          A <= A>>1; q <= q>>1; k<=k-1;
          if (k==1) begin s<=2; done<=1; P<=A; end
        end
    endcase
endmodule

Restoring divider: $n$-bit divisor $B$, $2n$-bit register $\{R, Q\}$ where dividend loaded into $Q$ and $R$ initially 0. ASM: each cycle, shift $\{R, Q\}$ left by 1 (MSB into $R$); subtract $B$ from $R$; if result $\ge 0$, place 1 into $Q$'s LSB; else restore (add $B$ back) and place 0 in $Q$'s LSB. After $n$ cycles, $R$ holds the remainder and $Q$ holds the quotient. Verilog: case-based FSM with shift, subtract-or-restore conditional, and a cycle counter.

Datapath: an accumulator (width $n+\lceil\log_2 k\rceil$), a counter for the number of inputs received, and a divider (or shift-by-$\log_2 k$ if $k$ is a power of 2). ASM: in state $S_1$ wait for Start. $S_2$: on each input strobe, add input to accumulator and increment counter; loop until counter = $k$. $S_3$: divide accumulator by $k$ (use the divider FSM if $k$ is not a power of 2, or right-shift if it is). $S_4$: assert Done with the mean as output.

Datapath: a register file of $k$ entries × $n$ bits, two read ports, one write port; a comparator and a swap mechanism. Selection-sort ASM: outer loop $i = 0\dots k-2$; inner loop finds index of minimum in $RF[i\dots k-1]$ and swaps with $RF[i]$. Each iteration uses two registers (current min index, current scan index) and a comparator. Verilog uses a 2-counter nested-loop FSM with comparator-driven conditional swaps. Total time: $O(k^2)$ clock cycles.

f = x₁x₃ + x₁x₄ + x₂x₃ + x₂x₄ = x₃(x₁+x₂) + x₄(x₁+x₂) = (x₁+x₂)(x₃+x₄)

SOP cost: 4 AND gates (2-input) + 1 OR gate (4-input) = 8 gate inputs + 5 gates.

Factored form cost: 2 OR gates (2-input) + 1 AND gate (2-input) = 3 gates, 6 gate inputs.

Saving: 2 gates, multilevel is cheaper and faster for this function.

f = ac+ad+bc+bd+e = (a+b)(c+d)+e. Convert to NAND-NAND:

Use double inversion: f = (a+b)(c+d) · e

NAND of NAND (applying DeMorgan):

t1 = NAND(a,c); t2=NAND(a,d); t3=NAND(b,c); t4=NAND(b,d)
t5 = NAND(t1,t2,t3,t4)  // = ac+ad+bc+bd
ne = NAND(e,e)            // = ē  (not needed if e uncomplemented)
f  = NAND(t5, NAND(e,e)̄)  // adjust...

More directly: f = NAND(NAND(a,c), NAND(a,d), NAND(b,c), NAND(b,d), NAND(e,e)̄) — using e with self-NAND for inversion, then a 5-input NAND for the final OR layer.

f = Σm(0,1,2,5,6,7,8,9,14). Quine-McCluskey grouping by number of 1s:

Group 0: 0000(0)
Group 1: 0001(1), 0010(2), 1000(8)
Group 2: 0101(5), 0110(6), 1001(9)
Group 3: 0111(7), 1110(14)
Combine:
(0,1)=000-, (0,2)=00-0, (0,8)=-000
(1,5)=0-01, (1,9)=-001, (2,6)=0-10
(8,9)=100-
(5,7)=01-1, (6,7)=011-
(9,...)  check (1,9)=-001 ✓
Second pass:
(0,1,8,9)=-00-, (0,2,1,...)...

Prime implicants: -00- (0,1,8,9); 0-0- (0,2)? ; 0-1- (5,7)=0-1-? ; 011- (6,7); 0-10 (2,6); -001 (1,9); 1110 (14). Full QM table needed for exact set.

f = x₁x₂ + x̄₁x₃, order x₁ < x₂ < x₃.

         x₁
        /   \
       0     1
      x₃    x₂
     / \    / \
    0   1  0   1
    0   1  0   1

Build Shannon expansion tree. After reduction:

• Root: x₁. Left (x₁=0): f = x₃. Right (x₁=1): f = x₂.
• Left child: x₃ node, 0→0, 1→1.
• Right child: x₂ node, 0→0, 1→1.

After ROBDD reduction (x₃ and x₂ nodes both map to identity, but different variables — cannot merge). 4 nodes: root x₁, left x₃, right x₂, terminal 0 and 1.

For n variables, the K-map has 2ⁿ cells. Practical issues with 5+ variables:

• A 5-variable map has 32 cells — displayed as two 4-variable maps; adjacency rules extend across the maps, making visual grouping error-prone.
• 6 variables: 64 cells, four 4-variable maps — essentially unmanageable by hand.

Alternatives:

• Quine-McCluskey: Tabular, systematic, scalable to any number of variables.
• BDDs: Canonical representation for verification and optimization.
• Espresso: Heuristic minimization algorithm used in CAD tools for large functions.

f = Σm(0,3,4,7,9,10,13,14). K-map grouping with fan-in ≤ 2 constraint means only 2-input gates.

Standard K-map: {0,3,4,7}=x̄₂x̄₃·? No: 0=0000,3=0011,4=0100,7=0111 → x̄₁(x̄₂x̄₃x̄₄+x̄₂x₃x₄+x₂x̄₃x̄₄+x₂x₃x₄)... group {0,3,4,7}= x̄₁(x₃⊙x₄)... {0,3,4,7}=x̄₁·(x₂⊕x₃x₄)?

Groups: {0,4,9,13}? No. Best K-map cover: {0,3,4,7}=x̄₁(x̄₂x̄₃+x₂x₃)... ={x̄₁(x₂⊙x₃)}... Actually {0,3,4,7}: bits differ in x₂,x₃,x₄ — these are x̄₁x̄₄(x̄₂x̄₃ impossible since 3=0011,7=0111). With fan-in≤2: decompose using a tree of 2-input gates.

Two-level logic advantages: Simple structure (SOP/POS); well-understood minimization (K-map, QM); predictable delay (exactly 2 levels); easy to implement with PLAs.

Two-level disadvantages: Can have large fan-in for complex functions; cannot share common sub-expressions; gate count can be exponential in the number of variables.

Multilevel advantages: Shares sub-expressions → fewer gates/literals; supports any fan-in constraint; more area-efficient for complex functions.

Multilevel disadvantages: More logic levels → longer critical path; harder to optimize systematically; less predictable timing.

f = abd + abe + acd + ace. Factor using algebraic division:

= a(bd + be + cd + ce) = a·(b+c)(d+e)

Extract t₁ = (b+c), t₂ = (d+e):

f = a · t₁ · t₂ = a·(b+c)·(d+e)

Original: 4 product terms × 3 literals = 12 literals + 1 OR gate.

Factored: 2 OR gates + 3-input AND gate (or 2 two-input ANDs) = 3 gates, 6 literals. Significant saving.

ROBDD (Reduced Ordered Binary Decision Diagram): A directed acyclic graph where each internal node is labeled with a variable, has two children (low=0, high=1), and the variable ordering is consistent along all root-to-leaf paths.

Two reduction rules:

1. Merge rule: If two nodes have the same variable, same low-child, and same high-child → merge into one node (share identical sub-graphs).
2. Elimination rule: If a node's low-child and high-child are identical → remove the node and redirect its parents to the child (variable is irrelevant for this sub-function).

Canonical form importance: For a fixed variable ordering, the ROBDD is unique for each Boolean function. This enables O(1) equality checking (same pointer = same function) and efficient BDD operations.

Technology mapping transforms a technology-independent (generic gate) netlist into one using cells from a specific cell library (standard cells for ASIC, or LUTs for FPGA).

Steps:

1. Represent the circuit as a tree or DAG of simple gates (INV, AND, OR, NAND, NOR).
2. Pattern-match sub-trees to available library cells using dynamic programming or recursive partitioning.
3. Select the best-matching cells to minimize area or delay.
4. Handle fan-out reconvergence by partitioning into trees.

The result is a mapped netlist with specific cell instances rather than generic gates.

Mux f = s·d₁ + s̄·d₀. Variables: s, d₀, d₁.

Ordering 1: s < d₀ < d₁

Root: s
  s=0: d₀ (output = d₀)
  s=1: d₁ (output = d₁)
BDD has 3 nodes: s, d₀, d₁ plus terminals 0,1.

Ordering 2: d₀ < d₁ < s

Root: d₀; must branch on d₁ then s → more nodes needed.
Effectively 7 nodes in worst case.

Ordering 1 (s first) gives the smaller BDD because s is the "controlling" variable for the mux function.

Functional decomposition: express f(x₁..x₅) = h(g(x₁,x₂,x₃), x₄, x₅) where g is a simpler sub-function.

Identify a "bound set" (e.g., {x₁,x₂,x₃}) and "free set" ({x₄,x₅}).

For each combination of x₄,x₅, list the required function of x₁,x₂,x₃. If multiple (x₄,x₅) combinations require the same sub-function of x₁,x₂,x₃, they share the same g output column.

A decomposition exists if the number of distinct columns ≤ 2ᵏ for k bits. Extract g (3-input function) and h (3-input: g-output + x₄ + x₅).

f = Σm(0,2,3,5,7,8,10,11,14,15,24,26,27,29,31) — 5 variable function.

QM: Group by number of 1s in binary representation, then combine pairs (differ by 1 bit), then quads, etc. With 5 variables (00000–11111), systematically combine until no further reduction possible.

Pattern: minterms (0,2,8,10,24,26)=−−0−0; (3,7,11,15,27,31)=−−111; (5,7,29,31)=−10−1; etc. Full enumeration yields prime implicants from which minimum cover is selected via PI chart.

f = (a+b)(c+d)+e using NOR gates:

NOR-NOR implements POS. f' = f = (a+b)(c+d)+e.

f' = NOR(NOR(a,b), NOR(c,d), e) — using De Morgan repeatedly.

Actually: complement the function and implement complement of complement with NOR-NOR:

NOR(a,b) → nor_ab = ā·b̄
NOR(c,d) → nor_cd = c̄·d̄
NOR(nor_ab, nor_cd, ē) → f

This requires complemented inputs (ā,b̄,c̄,d̄,ē) or NOR inverters for each.

f = x̄₁x₂ + x₁x̄₃. Cubical representation — each cube is a product term represented as a ternary vector (0=complemented, 1=uncomplemented, −=don't care):

The cover (set of cubes) = {(0,1,−), (1,−,0)}. Each cube represents a set of minterms: cube (0,1,−) = minterms {010, 011}={2,3}; cube (1,−,0)={100,110}={4,6}.

For a complex 5-variable function, multilevel synthesis extracts common sub-expressions:

1. Express f in SOP form.
2. Identify kernel expressions (common factors) using algebraic division.
3. Extract kernels as intermediate signals t₁, t₂, etc.
4. Rewrite f using extracted signals.

Example: if f = abcd + abce + fgh, extract t₁=ab: f = t₁(cd+ce)+fgh = t₁·c·(d+e)+fgh. This reduces the literal count from 9 to 7 (one ab → one t₁ used twice, saving literals).

f = Σm(4,7,8,11) + D(12,15). K-map (4-var):

       x₃x₄
x₁x₂  00 01 11 10
  00:   0  0  0  0
  01:   1  0  1  0
  11:   d  0  d  0
  10:   1  0  1  0

Groups: {4,8,12d}=x̄₃x̄₄? No, 4=0100,8=1000,12=1100 → differ in x₁,x₂. {4,12d}=x̄₃x̄₄·x₂? {4,8}=x̄₂x̄₄? No..

Minterms: 4=0100, 7=0111, 8=1000, 11=1011, DC 12=1100, 15=1111.

Groups: {4,12}=?x100; {7,15d}=?111; {8,12d}=1?00; {11,15d}=1?11.

Min SOP: f = x̄₃x̄₄ + x₃x₄ (covers 4,8,12dc,0? — check). Recheck: {4,8,12,0}: x₃=0,x₄=0 → covers 0,4,8,12 — but 0 not in f. Use {4,12}: x₂=1,x₃=0,x₄=0→x₂x̄₃x̄₄; {8}: x₁=1,x₂=0,x₃=0,x₄=0→x₁x̄₂x̄₃x̄₄; {7,15d}: x₂x₃x₄; {11,15d}: x₁x₃x₄.

Min SOP: x₂x̄₃x̄₄ + x₁x̄₂x̄₃x̄₄ + x₁x₃x₄. Factored: (x₂+x₁x̄₂)x̄₃x̄₄ + x₁x₃x₄ = (x₁+x₂)x̄₃x̄₄ + x₁x₃x₄.

Cubical representation definitions:

• Literal: A single variable or its complement; represented as a ternary value (0, 1, or −) in one position of a cube.
• Cube: A product term represented as a ternary vector; a cube with k "−" positions covers 2ᵏ minterms.
• Cover: A set of cubes whose union covers all the on-set minterms (positions where f=1).
• Prime implicant: A cube not contained in any other cube in the cover (cannot be enlarged while remaining valid).
• Essential PI: A prime implicant that is the only cube covering at least one minterm; must be in every minimum cover.

In Quine-McCluskey, a "check mark" (✓) is placed next to a minterm entry in a group when it has been successfully combined with another minterm. Any row without a check mark at the end is a prime implicant (cannot be combined further).

PI chart selection:

1. Build table: rows=PIs, columns=minterms.
2. Mark each PI's covered minterms.
3. Identify essential PIs (single mark in a column) — these must be selected.
4. Remove covered minterms; if minterms remain, use Petrick's method or heuristic (choose PI covering the most remaining minterms).
5. Repeat until all minterms covered.

f = x₁x₂ + x₃x₄ + x₅x₆.

Interleaved ordering (x₁,x₂,x₃,x₄,x₅,x₆): Each pair xᵢxᵢ₊₁ is adjacent; BDD can share the AND-structure for each pair compactly. BDD size = O(n) nodes.

Grouped ordering (x₁,x₃,x₅,x₂,x₄,x₆): Each product term's variables are separated; the BDD cannot share structure between the first halves — exponential blowup possible. BDD size = O(2^(n/2)).

General rule: For a function that is a sum of independent pairs, interleaving the pair members minimizes BDD size. Variable ordering is NP-hard to optimize optimally; dynamic variable reordering heuristics (sifting) are used in practice.

PI chart: columns = minterms, rows = prime implicants. Mark each minterm covered by each PI.

Find essential PIs: columns with only one mark → that PI is essential. Select all essential PIs, remove covered minterms, then use Petrick's method or greedy selection for remaining.

For f from 8.5: PI -00- covers {0,1,8,9}; if any of 0,1,8,9 is only covered by -00-, it's essential. Continue until all minterms covered. The minimum cover = essential PIs + minimum additional PIs for uncovered minterms.

Carry function for n-bit adder: cₙ = f(a₁,b₁,...,aₙ,bₙ,c₀).

Interleaved ordering (a₁,b₁,a₂,b₂,...aₙ,bₙ): BDD size is polynomial O(n) — each carry bit depends on its and lower bits' a,b pairs; BDD can share nodes efficiently.

Grouped ordering (a₁,...,aₙ,b₁,...,bₙ): BDD size grows exponentially O(2ⁿ) because cₙ depends on all aᵢ and bᵢ in a way that requires exponentially many nodes when variables are split into two separate groups.

Lesson: variable ordering dramatically affects BDD size; interleaved is optimal for adder carry.

f = Σm(0,4,8,13,14,15) with uncomplemented inputs only (no inverters).

Without complements, we can only form product terms using uncomplemented variables. Minterms requiring all-1 variables are easy; minterms with 0s require complements.

Minterm 0 = x̄₁x̄₂x̄₃x̄₄ requires all complements → not realizable with uncomplemented inputs only.

For this constraint: use NAND/NOR decomposition, or accept that some minterms (0,4,8) are not directly implementable; consider alternative: implement f̄ using uncomplemented inputs and invert, or use XOR-based form.

If only minterms 13,14,15 covered: f_partial = x₁x₂x₃x̄₄ + x₁x₂x̄₃x₄ + x₁x₂x₃x₄ = x₁x₂(x₃+x₄). Add minterms 0,4,8 by alternative decomposition.

f = ab+cd. Cell library example: {INV, NAND2, NOR2, AND2, OR2, AOI21 (AND-OR-Invert)}.

Minimize area: f = ab+cd = NAND(NAND(a,b), NAND(c,d)). Uses 3 NAND2 gates (area = 3 units). Alternatively: AND2+AND2+OR2 = 3 cells but likely larger area.

Minimize delay: f can be implemented as single AOI21 cell if available: ̄f = AOI(a,b,c,d) then INV. Or NAND-NAND: 2 levels × t_NAND delay — same as any 2-level implementation.

Circuit: t₁=a+b, t₂=c·d, t₃=t₁·e, f=t₂+t₃ = c·d + (a+b)·e

Analysis: f = cd + ae + be. Literal count = 5. Gate count = AND(c,d) + AND(a,b→OR) + 2 ANDs + OR = 4 gates.

Re-factor: f = cd + e(a+b). This is already factored. Alternative: no simpler form exists since cd and e(a+b) share no common factor.

Could also write f = c·d + (a+b)·e — same. Verify: expanding gives cd+ae+be which matches original enumeration.

Note $f_2 = (x_1+x_2)(x_3+x_4)$ already factored. Examine $f_1$: it equals $f_2$ minus the case $x_1=x_2=0$ where it should be $\overline{x_1}\overline{x_2}(x_3+x_4)$ — i.e. $f_1 = (x_1+x_2)(x_3 x_4) + \overline{x_1}\overline{x_2}(x_3+x_4)$. Both share the subexpression $g = x_3+x_4$ and $h = x_1+x_2$. Total cost: $g, h$ shared; then $f_2 = h\cdot g$, $f_1 = h\cdot x_3 x_4 + \overline{h}\cdot g$. Sharing $g$ and $h$ saves several gate-inputs versus building each output independently.

Group on the K-map columns: in columns 01 ($\overline{x_3}x_4$) and 10 ($x_3\overline{x_4}$) the function takes value $x_1 \oplus \overline{x_2}$; in columns 00 and 11 it is 0. So $g(x_3, x_4) = x_3 \oplus x_4$ (asserts in the two middle columns). Then $f = g \cdot (x_1 \oplus \overline{x_2})$ — a 2-level XOR product, two XORs feeding one AND. Cost: 3 gates with 2 inputs each, vs four 3-input ANDs in the SOP form.

By inspection of the K-map, when rows where $g(x_1,x_2,x_5)=0$ all have one column pattern and rows where $g=1$ have a different pattern, define $g$ as the row indicator. Similarly $k(x_3,x_4)$ as the column index. The book's example produces $f = g \oplus k$ or $f = g\overline{k} + \overline{g}k$. Cost: 3 small subfunctions + 1 XOR ≈ 8 inputs, versus minimum SOP ≈ 18 inputs — a substantial saving.

5-NAND XOR (standard):

g1 = NAND(x1,x2)
g2 = NAND(x1,g1)
g3 = NAND(g1,x2)
f  = NAND(g2,g3)

g  = NAND(x1, x2)
f1 = NAND(x1, g)
f2 = NAND(x2, g)
f  = NAND(f1, f2)

Bubble-pair insertion: AND-OR levels alternate, so place a bubble pair at every signal between adjacent levels. After absorbing each bubble into the gate it touches: every AND becomes NAND and every OR (with bubbles on inputs and output) becomes NAND too — yielding (a) all-NAND. For (b) all-NOR: insert bubble pairs differently — turn the AND gates into NORs (with bubbled inputs) and the ORs into NORs (with bubbled outputs absorbed). The structure flips between AND-OR-AND-OR and OR-AND-OR-AND topology.

Label intermediate nodes $P_1, P_2, P_3$ inside Figure 8.12. Trace from output back: $f = P_1 + P_2$; $P_1 = x_1 P_3$, $P_3 = x_2 + x_3$; $P_2 = \overline{x_1} x_4$. So $f = x_1(x_2+x_3) + \overline{x_1}x_4 = x_1 x_2 + x_1 x_3 + \overline{x_1}x_4$. SOP cost: 3 ANDs + 1 OR = 4 gates, 9 inputs.

Replace the XOR/XNOR gates of Example 8.3 by their SOP equivalents ($x \oplus y = x\overline{y}+\overline{x}y$). Label internal nodes and trace: each leg of the multilevel circuit collapses to one of the original product terms. The expression reconstructs to $f = $ the original specification — verifying functional equivalence.

Label $P_1=$ NAND($x_1, x_2$), $P_2=$ NAND($x_3, x_4$), $P_3=$ NAND($P_1, P_2, x_5$). Apply DeMorgan to each: $P_1 = \overline{x_1 x_2}$, $P_2 = \overline{x_3 x_4}$, but feeding into the next-level NAND with bubble absorption gives an OR-of-AND structure. The textbook reduction yields $f = x_1 x_2 x_4 + \overline{x_3} x_4 + x_5$ after simplification.

Label $P_1\dots P_4$ at gate outputs. NAND outputs equal $\overline{\prod}$; NOR outputs equal $\overline{\sum}$. Propagate bubbles by DeMorgan until all bubbles cancel. After algebraic simplification: $f = x_5(x_1 + x_2 + \overline{x_3} + x_4) = x_1 x_5 + x_2 x_5 + \overline{x_3} x_5 + x_4 x_5$. ✓

Build a small truth table over $(x_1, x_2)$ for each terminal of the original BDD, recording where edges from the swap-boundary go. Then re-create the BDD with $x_1$ above $x_2$. Many edges merge if the function is symmetric in those variables. Result is a BDD of equivalent or smaller size depending on the variable ordering.

Shannon on $x_1$: $f|_{x_1=0} = x_2 x_3 + x_2 x_4 + \overline{x_2}\overline{x_3}\overline{x_4}$, $f|_{x_1=1} = x_3 + x_4 + x_2 x_3 + x_2 x_4 = x_3 + x_4$. Shannon $f|_{x_1=1}$ on $x_2$: same $x_3+x_4$ either way (independent of $x_2$). For $f|_{x_1=0}$ on $x_2$: $x_2=1$ gives $x_3+x_4$; $x_2=0$ gives $\overline{x_3}\overline{x_4}$. BDD: root $x_1$, both children point to an $x_2$ node (or merge if equal). The $x_3+x_4$ subgraph appears multiple times — merge it. Multilevel SOP from BDD: $f = x_1(x_3+x_4) + \overline{x_1}[\,x_2(x_3+x_4) + \overline{x_2}\overline{x_3}\overline{x_4}\,]$.

Quine-McCluskey on $\sum m(0,2,5,6,7,8,9,13) + D(1,12,15)$: Group minterms by 1-count, combine adjacent ones across iterations until no more merges. Resulting prime implicants (after marking unmerged): typically $\{\overline{x_2}\overline{x_3}\overline{x_4}, x_2 x_4, x_2 x_3, x_1\overline{x_3}, x_1 x_4\}$ or similar. Build cover table over essential minterms, eliminate dominated rows/columns, and select. Minimum cover (one valid result): $f = x_2 x_4 + x_2 x_3 + \overline{x_2}\overline{x_3}\overline{x_4} + x_1\overline{x_3}$.

Apply tabular method: prime implicants include several 2-cubes and 1-cubes, no essential primes. Cover table has no row or column dominance. Use Petrick's method or branching: select an arbitrary prime, recurse, choose minimum-cost cover. Typical result: $f = $ 4 prime implicants of 2 literals each, gate-input cost $\approx 12$.

$\star$-product pairs cubes that differ in exactly one position, replacing that position with x. $C_0 = \{000, 001, 010, 011, 111\}$. $C_1$: $000\star 001 = 00x$, $000\star 010 = 0x0$, $001\star 011 = 0x1$, $010\star 011 = 01x$, $011\star 111 = x11$. $C_2$: $00x\star 01x = 0xx$. No further merges. After removing covered cubes, primes are $\{0xx, x11\}$. ✓

Apply $\star$-product to $C_0=\{0101,1101,1110,011x,x01x\}$. $0101\star 1101 = x101$. $011x \star x01x$: differ in two positions → no merge. $0101\star 011x = 01x1$ (differ in bit 1, x merge). $1101\star 1110$: two diff → no. $C_1 = \{x101, 01x1, 1110, 011x, x01x\}$. Continue: $x101 \star 01x1$: two diff, no. Final primes: $x101, 01x1, 1110, 011x, x01x$. Algebraic: $\{x_2\overline{x_3}x_4, \overline{x_1}x_2 x_4, x_1 x_2 x_3\overline{x_4}, \overline{x_1}x_2 x_3, \overline{x_2}x_3\}$.

The $\#$-operation: $A \# B$ returns minterms in $A$ not in $B$. For each prime $p$, compute $p \# (P \setminus \{p\})$ over the on-set. If non-empty, $p$ is essential (covers a minterm no other prime covers). For $\{x11, 0xx\}$ on the function of Figure 2.56: $x11 \# 0xx = 111$ (non-empty) → essential. $0xx \# x11 = \{000, 001, 010\}$ (non-empty) → essential. Both primes are essential.

For each $p \in P$, compute $p \# (P\setminus\{p\})$. $x01x \#$ rest: $\{0010, 0011\}$ → essential. $x101$: subset of $x101$ alone — empty after removing self → not essential. $01x1$: $\{0101\}$ — essential. $0x1x \#$ rest: $\{0010, 0011\}$ — already covered by $x01x$ → check: actually $0x1x \# x01x = \{0110, 0111\}$ — essential. $xx10$: most cells covered by others → not essential. Essentials: $x01x, 01x1, 0x1x$. (Exact set depends on minterm coverage in the function.)

Apply $\star$-product to find primes (many cubes after several iterations on the 5-var function). Use $\#$-op to identify essentials covering minterms unique to each. After essentials are chosen, branch on the remaining cover table to minimize cost. Final minimum SOP (illustrative): $f = \overline{x_1}\overline{x_2}\overline{x_3}\overline{x_4} + x_4 x_5(x_1\oplus x_2) + x_2 x_3 x_4 + \overline{x_3}\overline{x_5}$ (or equivalent ~4-term cover, depending on don't-care assignments).

Straightforward 3-LUT mapping: LUT1 = $x_1 x_2\overline{x_4}$, LUT2 = $\overline{x_2} x_3 x_4$, LUT3 = $x_1 x_2 x_3$, LUT4 = OR of these — total 4 LUTs. Decomposition: factor $x_1 x_2(\overline{x_4} + x_3) = x_1 x_2(x_3 + \overline{x_4})$ in one 3-LUT, $\overline{x_2}x_3 x_4$ in another, OR them in a third — total 3 LUTs.

Group: $f = (x_1+x_2)(x_3+x_4) + \overline{x_1}\overline{x_2}\overline{x_3}\overline{x_4}$. Use $\overline{x_1}\overline{x_2}\overline{x_3}\overline{x_4} = (\overline{x_1+x_2})(\overline{x_3+x_4}) = \overline{x_3}\overline{x_4}\cdot(\overline{x_1+x_2}) = \overline{x_3}\overline{x_4}$ when $x_1+x_2=0$. Net: $f = \overline{x_3}\overline{x_4} + (x_1+x_2)(x_3+x_4)$. ✓ (Identity verified by expansion: when $x_3+x_4=0$, only the first term remains; when $x_3+x_4=1$, the AND term covers it.)

Shannon on $x_1$: $f|_{x_1=0} = x_2 x_4$, $f|_{x_1=1} = x_3 + x_2 x_4$. Shannon $f|_{x_1=1}$ on $x_2$: $f|_{x_1=1, x_2=0} = x_3$, $f|_{x_1=1, x_2=1} = x_3 + x_4$. BDD: root $x_1$. Right branch ($x_1=1$) goes to $x_2$ node with two cofactors (one is $x_3$, one is $x_3+x_4$). Left branch ($x_1=0$) goes to a node testing $x_2$ — only $x_2 x_4$ branch. Continue with $x_3$ then $x_4$ levels, merging duplicate subgraphs.

Track the 4 entries of the truth table over $(x_2, x_3)$ at each subgraph crossing the swap. After swapping inner $x_2$ and $x_3$, build a new BDD with root $x_1$, then $x_3$, then $x_2$, then $x_4$. Some subgraphs merge differently. The new BDD usually has the same node count for symmetric functions, but variable ordering can change BDD size dramatically in general.

Minterms of $f$: $\sum m(0,1,2,4,5,8,9,13) + D(9,12,14)$ (after expanding, given the SOP form). Tabular method: group by 1-count, combine. Generated cubes: $\{00x0, 0x0x, 1x01, 10x1, 1101\}$ and primes after removal of covered ones. Cover table: essentials from rows with single mark. Minimum-cost SOP: $f = \overline{x_3}\overline{x_4} + x_1 x_4 + \overline{x_2} x_4$ (as in 2.38).

$C_0$ from minterms+don't-cares. $C_1$ via $\star$-product yields 2-cubes; $C_2$ yields 3-cubes. Convergence after ~3 iterations. Resulting primes (in cube notation, $x_1 x_2 x_3 x_4$): $\{xx00, x0x1, 1x01, 1xx1\}$ — algebraically $\{\overline{x_3}\overline{x_4}, \overline{x_2}x_4, x_1\overline{x_3}x_4, x_1 x_4\}$. The book's minimum-cost cover selects a subset matching 8.35.

(a) Minimum SOP $f = \overline{x_3}\overline{x_4} + x_1 x_4 + \overline{x_2} x_4$: 3 ANDs + 1 OR, 11 gate-inputs. (b) Minimum POS $f = (\overline{x_2}+\overline{x_3})(\overline{x_3}+x_4)(x_1+\overline{x_3}+\overline{x_4})$: 3 ORs + 1 AND, ~12 inputs. (c) Multilevel: factor $f = \overline{x_3}\overline{x_4} + x_4(x_1+\overline{x_2})$: 2 levels for the second term + final OR — about 8 inputs but 3 levels deep. Trade-off: SOP is fastest (2 levels) at 11 inputs; multilevel saves area (8 inputs) at the cost of 3 levels of delay.

$h[g(x_1,x_2,x_3,x_4), x_5, x_6, x_7]$ — first LUT computes any 4-var function $g$, second computes any 4-var function $h$ of $g$ and three other vars. $f_1 = x_1 x_2 x_3 x_4 x_5 x_6 x_7$: set $g = x_1 x_2 x_3 x_4$, $h = g \cdot x_5 x_6 x_7$ — works. $f_2 = x_1+\dots+x_7$: $g = x_1+x_2+x_3+x_4$, $h = g + x_5+x_6+x_7$ — works. But e.g. $f = x_1 x_5 + x_2 x_6 + x_3 x_7 + x_4$ requires LUT2 to know all of $x_1\dots x_4$ individually — only $g$ (one bit summarizing them) is available. So no 2-LUT realization exists for this $f$.

Fundamental mode: Asynchronous circuit operating assumption where inputs change only when the circuit is in a stable state — i.e., one input changes at a time, and the circuit is allowed to settle before the next input change.

Input constraints:

• Only one input variable changes at a time (no simultaneous changes).
• The circuit must reach a stable state before the next input transition.
• No clock; the circuit responds directly to input changes.

These constraints prevent races and ensure well-defined behavior.

Cross-coupled NOR gates (SR latch): Q = NOR(R,Q̄), Q̄ = NOR(S,Q). Flow table (S,R inputs; stable states circled):

Stable states (circled): Q=0 with SR=00,01; Q=1 with SR=00,10. SR=11 is the forbidden input (both outputs attempt to become 0). When SR=11→00, output is indeterminate (race).

Stable state: A state where the next state equals the present state for the current input combination — the circuit has finished transitioning and no further changes occur. Y = y (feedback variable equals its driving function).

Unstable state: A state where Y ≠ y — the circuit's next state differs from its current state. The circuit will continue transitioning (may pass through multiple unstable states) until reaching a stable state. The sequence of unstable-to-stable transitions is called a transition.

Critical race: When multiple state variables must change simultaneously, and the outcome depends on which changes first. Different orderings lead to different final states — non-deterministic behavior.

Example: State 00→11 transition. If y₁ changes first: 00→10→11 (correct). If y₂ changes first: 00→01→11 (may be correct). But if 10 or 01 are stable states for the current input, the circuit may get stuck there.

Fix: Use race-free state assignment (e.g., assign codes so any transition changes only one variable) or use one-hot encoding for asynchronous circuits.

f = x₁x₂ + x̄₂x₃. Static 1-hazard occurs when both x₁x₂ and x̄₂x₃ are 0→1 simultaneously — when x₂ transitions from 1→0 with x₁=1, x₃=1: transition from x₁x₂=1 to x̄₂x₃=1, momentarily neither term is 1.

Hazard condition: At x₁=1, x₃=1, as x₂ transitions. The two minterms 110 and 011 are not in the same K-map group.

Eliminate: Add the consensus term x₁x₃ (covers the adjacent minterms 110 and 011, bridging the gap):

f = x₁x₂ + x̄₂x₃ + x₁x₃ (hazard-free)

Static 1-hazard: Output should remain 1 but momentarily glitches to 0 during an input transition.

Static 0-hazard: Output should remain 0 but momentarily glitches to 1 during an input transition. Occurs in POS implementations; eliminated by adding extra sum terms.

Dynamic hazard: Output should change once (0→1 or 1→0) but transitions 3 or more times (e.g., 0→1→0→1). Caused by multiple paths of different delays when there are reconvergent paths. More complex to eliminate — requires removing the reconvergent topology or balancing delays.

Compatible states: Two states sᵢ and sⱼ are compatible if (1) they have the same output for all input combinations where both are defined, and (2) their successor state pairs are also compatible (or both undefined) for all inputs.

Merger diagram: A graph where nodes are states and undirected edges connect compatible state pairs. Groups of mutually compatible states can be merged into a single state.

Process: find all compatible pairs (from implication table), draw merger diagram, find maximum compatible groups (cliques), select minimum number of merged states that covers all original states.

f = Σm(1,2,3,5,7). K-map groups: {1,3,5,7}=x₃; {2,3}=x̄₁x₂; {3,7}=x₂x₃.

Min SOP = x₃ + x̄₁x₂. Check for static 1-hazard: adjacent 1s not in same group?

Minterms 2(010) and 3(011): covered by x̄₁x₂ and x₃ respectively. Transition m2→m3 (x₃: 0→1) with x₁=0,x₂=1: initially x̄₁x₂=1, finally x₃=1. During transition: momentarily neither? No — x̄₁x₂=1 throughout (x₁,x₂ constant). No hazard on this path.

Check m5(101)→m7(111): x₂:0→1 with x₁=1,x₃=1. Before: x₃=1(covers). After: x₃=1 and x₂x₃=1. No hazard.

Hazard exists between m2(010) and m3(011): x̄₁x₂ covers m2; x₃ covers m3. Adding x̄₁x₂x₃ as bridge or the extra PI x₂x₃: f = x₃ + x̄₁x₂ + x̄₁x₃ is hazard-free (extra term bridges).

Y₁ = x̄₁(x₂+y₁); Y₂ = x₁(x̄₂+y₂). Build flow table: states = (y₁y₂), inputs = (x₁x₂).

Stable states (Y=y): (00,x=00), (01,x=01), (10,x=10), (11,x=11). Transitions determined by evaluating Y₁,Y₂ for each (y₁y₂,x₁x₂) combination.

Essential hazard: An inherent hazard in an asynchronous sequential circuit that cannot be eliminated by adding or removing gates. It arises when a single input change causes the circuit to cycle through multiple unstable states, and the required ordering of state variable changes cannot be guaranteed by pure combinational logic delay matching.

Differs from static/dynamic hazards: Static/dynamic hazards occur in combinational logic and can be eliminated by adding redundant terms. Essential hazards are sequential — they result from the interaction between feedback loops and input changes.

Can gates fix it? Not by adding redundant terms. Instead, essential hazards require introducing deliberate delays (e.g., RC circuits or buffer chains) on specific feedback lines to ensure the proper sequencing of state variable updates.

Muller C-element: output C changes to match inputs only when all inputs agree; holds previous value otherwise.

For 2-input C-element: C(t+1) = AB + (A+B)·C = AB + AC + BC (majority function of A, B, C_current).

Implementation using async techniques:

Y = A·B + Y·(A + B)
  = A·B + Y·A + Y·B

Circuit: two AND gates for AB and YA and YB; one OR gate for Y. This is a hazard-free equation since the feedback Y ensures that C holds its value when A≠B.

The C-element is fundamental to handshaking asynchronous protocols (Micropipelines, NULL convention logic).

SR latch synthesis from primitive flow table:

Primitive flow table: 2 rows (Q=0, Q=1), 4 input columns (SR: 00,01,10,11). Assign y=Q.

Excitation Y (next state) from the flow table:

Y = S + R̄·y (standard SR latch equation, SR≠11 assumed).

Output Q = y (state variable directly).

Implementation: Y = NOR(R, NOR(S,Y)) — cross-coupled NOR gates realize this feedback equation naturally.

State reduction on an async flow table:

1. Build implication chart for states A–D.
2. Mark pairs with different outputs as incompatible.
3. For remaining pairs, check if their implied pairs are compatible.
4. Propagate incompatibilities until no more changes.
5. Compatible pairs can be merged.

Example: if A and C have same outputs and their successors under all inputs are also compatible pairs → A≡C, merge to a single state. Reduces the flow table rows, simplifying the implementation.

One-hot state encoding for 3-state async FSM: one SR latch (or FF) per state.

States S₁,S₂,S₃ → y₁,y₂,y₃ with only one yᵢ=1 at a time.

Transitions are controlled by setting the target state latch and resetting the current state latch. Since only one variable changes per transition (the one being set or reset), one-hot is inherently race-free for asynchronous circuits.

Advantage: no critical races. Disadvantage: uses more state variables (3 instead of 2 for 3 states).

Three challenges of async circuit design:

1. Races: When multiple state variables change simultaneously, the outcome is non-deterministic. Requires careful state assignment or one-hot encoding to ensure race-free transitions.
2. Hazards: Glitches in combinational logic (static 1/0, dynamic) can cause spurious state transitions since there is no clock to mask them. All hazards must be eliminated by adding redundant logic.
3. Fundamental mode violations: Design assumes one input changes at a time and circuit settles before next change. If violated, behavior is undefined. Testing and verifying these timing assumptions is difficult.

Primitive flow table: One row per stable state. Each input combination in that row has exactly one stable state (the current state) and shows the next state for each input combination. Rows are not merged — one unique stable state per row.

Merged flow table: Multiple compatible stable states combined into one row, reducing the number of state variables (fewer FFs needed).

Why merging is needed: A primitive flow table with N stable states requires ⌈log₂N⌉ state variables or N state variables (one-hot). Merging compatible states reduces the number of rows (and thus state variables), yielding a more compact implementation and simplifying the excitation equations.

Dynamic hazard: A signal that should make a single transition (0→1 or 1→0) instead makes three or more transitions (0→1→0→1 or 1→0→1→0) due to multiple paths with different delays in the circuit.

Conditions: Requires at least three distinct paths from input to output with different delays, and reconvergent fan-out where the paths merge.

Elimination: Dynamic hazards cannot be eliminated simply by adding terms (unlike static hazards). They require restructuring the circuit — either eliminating reconvergent paths, ensuring that no competing paths exist, or using hazard-free logic design techniques such as implementing in a single fan-in stage.

"Don't happen" input combinations in async design are input state combinations that cannot physically occur — either due to circuit constraints or by design convention.

They appear as blanks or "–" in the flow table cells. For example, if inputs are R and S of a latch, the combination R=1, S=1 may be defined as a "don't happen" condition (never applied simultaneously).

These are treated as don't-care conditions during excitation equation minimization — may be assigned 0 or 1 freely to minimize logic cost — provided the circuit designer guarantees they never actually occur.