Clock Strategy

VLSI System Design

Clocked Systems Latch and Flip-flops System timing Clock skew High speed latch design Phase locked loop Dynamic logic Multiple phase clock Clock distribution
NCKUEE-KJLEE

Clocked Systems
Most VLSI systems are a combination of
pipelines and finite state machines(FSM) Pipelined systems
Input D Q Logic D Q ... Logic D Q output

CLK

CLK

Finite state machine

Comb. Logic

QD CLK or CLKS
VLSI System Design NCKUEE-KJLEE

Single-phase clock timing waveforms

Cycle Time (Tc) Clock Setup Time (Ts) CLK

Tc: clock cycle time (period)

D Q

data

Hold Time (Th)

Q Clock-to-Q Delay (Tq)

Ts: setup time -- the time before the clock edge during which the data input (D) has to be stable Th: hold time -- the time after the clock edge during which the data input (D) has to remain stable Tq: clock-to-Q delay -- The delay from the clock edge to the Q output
NCKUEE-KJLEE

VLSI System Design

Latches and flip-flops

D 0 Q 1 s CLK

clk D Q

(a) Negative Latch (Level sensitive)

clk
0 Q D 1 s CLK

D Q

(b) Positive Latch (Level sensitive)

VLSI System Design

NCKUEE-KJLEE

Latches and flip-flops

clk

0 0 1 s CLK master QM Q 1 s CLK

D QM Q

slave

(c) Positive edge-triggered register(single-phase clock)

clk=0

clk=1 master slave

(d) Pass transistor/inverter implementation

VLSI System Design

NCKUEE-KJLEE

System timing
(A) Positive-edge triggered
Tc

clock

Register A

Tq

Combinational Logic Td

Ts

Register B
Tq Td Ts

Tc >Tq + Td + Ts

(B) Alternatively, one may use latches as storage elements to save area.

clock

Latch A

Tq

Combinational Logic Td

Ts

Latch B

VLSI System Design

NCKUEE-KJLEE

System timing (cont)

(C)
A Latch A Tq Combinational Logic Tda Ts Latch B B Combinational Logic Tdb Latch c

clock

Tc1

Tco

Tc1>Tqa+Tda+Tsb Tsb=Tsc
Tqa
Tda Tsb Tqb Tdb Tsc

Tco>Tqb+Tdb+Tsc

If Tc=Tc1+Tco and Tc1=Tco, Tqa=Tqb, => The limit is Tc = Tda + Tdb + 2(Tq+Ts)

VLSI System Design

NCKUEE-KJLEE

Racing due to clock skew

1 REG 1-bit d q Logic Td 2 0 REG 1-bit d q

1. If Tc2 > Tc1 + tq1 + td2

M2 may sample a wrong data (current data) Transparency problem

M1 clk delay Tc1 delay Tc2

M2

2. If Tc1 + tq1 + td2 >Tc (cycle time)

clk Tc1 Td 2 Old Data Tc2 New Data

M2 cannot sample the previous data

VLSI System Design

NCKUEE-KJLEE

Single-phase clock using D-FF

C1 L1 C2 L2

1) C2=C1

C2

C1

C1 C2 Wrong data in L2 2) C1=C2 C1 C2 Correct data

C1 = C2
C2 C1

or

C2 = C1
C1 C2

L1

L2

Comb
Wrong only if Comb.

CLK

Logic

FSM --> "feedback" or Pipeline --> "feedthrough"

VLSI System Design

NCKUEE-KJLEE

To eliminate (reduce) time skew (clock skew)

1.Balanced delay clock driver
clk-in clk

2.Use buffers where necessary

clk clk
Large Load

clk

clk

clk

clk-in

clk

3.Very careful simulation(HSPICE) 4.Very small rise and fall time on the clock-- large buffer for large load 5.Multiple clocking strategies

clk

usually slightly smaller than the inverter

VLSI System Design

NCKUEE-KJLEE

Some Implementations of clocked latches

Use a weak trickle inverter
-ck D ck
small inverter (low-gain, smaller W or large L)

D clk clk Q clk clk

eliminate a metal connection smaller area

or
D c lk c lk c lk

Transmission-gate latch
Q D

c lk

buffered input
VLSI System Design

compared with a tri-state buffer

NCKUEE-KJLEE

Typical symbolic layouts for latches

(a)
V

(b)
DD
V

(c)
DD
V

DD

Q D

clk

-clk

SS

clk

-clk

clk

-clk

SS

SS

-clk
D clk clk clk Q clk

-clk

clk D -clk

clk Q

clk D -clk

clk Q

VLSI System Design

NCKUEE-KJLEE

logic gate based latches

(a) Level sensitive
D Q
clk

-Q clk
D

(b) Edge triggered

D Q

-Q clk

VLSI System Design

NCKUEE-KJLEE

Asynchronously settable and resettable F/Fs

-clk -reset Q clk D -clk -clk clk -reset
Q clk D -clk
VLSI System Design

(a)

clk

-clk

clk

-clk

(b)

clk

-clk

clk

clk

-set -clk
NCKUEE-KJLEE

Dynamic single clock latches

The feedback inverter and transmission gate are eliminated. The latched value is stored on the capacitance of the input to
the inverter (mainly gate capacitance) Clock-to-Q (Tq) is very small need to be very careful to prevent transparency problem. Internal inversion of the clock is often necessary. Dynamic nodes should be always refreshed or clamped to a known state when in stand-by or low-power mode.
clk D -clk -Q
D -clk clk

(a)

(b)

VLSI System Design

NCKUEE-KJLEE

Dynamic single clock latches (cont)

clk D
D

clk

-clk Q

clk D -clk

-clk Q clk

-clk
-clk clk

(c)Tristate inverter

(d)master-slave F/F

(e)

VLSI System Design

NCKUEE-KJLEE

Refreshing for Dynamic latches

Dynamic storage nodes are
usually a gate capacitance. Assume the leakage current= 1 nA and the storage capacitance = 0.02PF
C V 5 = 0.02 10 12 9 = 100s i 10

Even if the storage of the

correct state is unimportant, the leakage may cause the storage node to assume a level that causes the inverter to draw significant current.
large current 5V -> 2.5V

must refresh every 100s

VLSI System Design

NCKUEE-KJLEE

Phase locked loop (PLL) clock techniques

(1) To synchronize internal and external clocks. (2) To synchronize data transfers between chips. (3) To operate the internal clock at a higher rate.
(1)
clock clock pad clock route dclk output pad dclk+dpad clock dclk data out chip PLL clock clock pad clock route dclk output pad dclk+dpad clock dclk data out

(a) Without PLL

(a) With PLL

VLSI System Design

NCKUEE-KJLEE

PLL techniques
(2)
bus high speed tristate bus

(3)
chip PLL /4

clock clock pad clock route dclk output pad dclk+dpad clock d clk

clock PLL PLL

clock

system clock

To ensure the output of chips are synchronized with each other.

Clock rate at d clk 4 = clock d clk = 4 clock

NCKUEE-KJLEE

VLSI System Design

Block diagram of a PLL circuit

n Ffb Phase Detector reference clock(F ) in
Phase detector: detect the difference between Ffb and Fin.
If Ffb > Fin =>D pulse If Ffb > Fin =>U pulse

U D Charge pump Filter VCO n*Fin

Charge pump: charge or Discharge a capacitor according to D and U. Filter: filter the capacitor output (smoother). VCO: Change the oscillation frequency depending on the control voltage. (Voltage Control Oscillator)

VLSI System Design

NCKUEE-KJLEE

Phase Detector
F1
F1 16/8 16/8 16/8 16/8 16/8 UP 16/8

F2
16/8 16/8

If F1 falls before F2 => UP=1

16/8

If F2 falls before F1 => DN=1

VLSI System Design

F2 16/8

16/8 16/8

16/8 16/8 DN

NCKUEE-KJLEE

Charge Pump
40/2 P1 P-REF CHGUP N2 40/2 CHGDN N4 N-REF P-REF P1 40/2

SW0

2/5 2/5

N1

P-SWITCHSW0 SW1 CAP P-SWITCH SW0 SW1 N-REF N5 40/2

N3 10/2

IN

8/1 16/1 10/2 10/2 SW1

Bias circuit

CAP charges when CHGUP=1 discharge when CHGDN=1

VLSI System Design

NCKUEE-KJLEE

Filter
VCO
in 2/6 4/6
32/1
13stages

out

32/1

32/1

32/1 32/1 32/1 16xFsc 16/1

2400/6

2400/6

control voltage 16/1

16/1 16/1
VCO

16/1 16/1

in

out

VLSI System Design

NCKUEE-KJLEE

Metastability Problem
D clk clk Q Q
clock 4ns delay 2ns Q -Q No Problem data

delay=2.2ns

If the setup or hold time is not satisfied, I.e., D changes at the activation edge of the clock, then the output Q will have a state depending on the timing relation between D and CLK

delay=2.3ns

Q metastable Long delay point -Q

Q metastable point -Q delay=2.4ns

Output error

VLSI System Design

NCKUEE-KJLEE

Metastable state in a pair of inverters

Inv1 A B
To Solve the metastability problem:
Setup time is shorter than the clock-to-Q delays in a synchronization system. For asynchronous input : need a special circuit called synchronizer.

VB

Inv2

Inv1 metastable point Inv2 VA 5V

0V
VLSI System Design

2.5V

NCKUEE-KJLEE

Single-phase N-P CMOS dynamic logic

Combine N-P section of domino logic with clocked CMOS (C2MOS) latch as the output stage.
to n-logic blocks to p-logic blocks clk -clk -clk Inputs from -clk stages clk to -clk section to p-logic blocks to n-logic blocks

clk

-clk -clk clk to -clk section

(b)

n-logic block n-p CMOS -clk logic stage

p-logic block

C2MOS latch

n-logic block (a)

VLSI System Design

p-logic C2MOS latch block From n or buffered p-logic n-p CMOS clk logic stage

-clk logic

clk logic

-clk logic evaluation Precharge

evaluation Precharge Precharge evaluation (c)

clk -clk 0 1 1 0

NCKUEE-KJLEE

Design rules for N-P CMOS dynamic logic

Two problems to be solved
1.Each section must be internally race free. 2.When different section are cascaded to from pipelined system, clock skew should not cause a problem.
clk clk

R1: During precharge, logicblocks must be switched off. R2: During evaluation, internal inputs can make only one transition. When a static logic is used in a N-P CMOS dynamic logic, it should be placed after dynamic logic (I.e., one should keep the static logic up to the C2MOS latch. Reason: static logic after creates a glitch at its output.

VLSI System Design

NCKUEE-KJLEE

R3: There exists in each logic block Reason: at least one dynamic gate that is separated from 1 1 or the previous C2MOS 2 output stage by an even number C MOS of inventions. clk or clk C2MOS R4: The total number of inversions or Domino between two consecutive C2MOS stage is even.
-clk clk clk -clk

clk

clk

The same evaluation phase in a section

even number of inversions C2MOS latch output stage (-clk section)

at least one dynamic stage

C2MOS latch (clk section)

OR

even umber of inversions

VLSI System Design NCKUEE-KJLEE

Two phase clocking

phi1 phi 2 -phi1 D phi (a) 1

phi1 phi2
-phi2 phi2 DFF1 Q

phi1 logic phi2 logic overlap

skewed clokcs

(c) small dealy

phi1=1 phi2=0 phi =0 1 phi =1 2 (b) C1 C2

slow rise time overlap phi1=1 phi2=0 (d)

C1

C2

C2

VLSI System Design

NCKUEE-KJLEE

Two phase clock generation

Two-phase clock generator 1. Globally distribute two clocks with or without their 1 clk complements. 3. A single global clock and locally generated two-phase clocks

Delay for non-overlap period

VLSI System Design

NCKUEE-KJLEE

Two phase registers

-phi1 D DEF1
-phi 2
high level =V

-phi 2
D

DD

-V

tn Q DEF1A

Q
(a) phi1 phi 2

phi1
-phi1 D phi1

phi2
-phi 1

p leakers

-phi

2 n

D phi

n 1 phi2

D (b) phi D 1 phi 2 DEF1B

phi2

DEF2 Both of these dynamic registers have to drive a local storage gate.
VLSI System Design

DEF3

clk (c)

NCKUEE-KJLEE

Two phase logic

1. Static logic with two phase registers
Logic 1 Logic 2

2. Dynamic logic
-phi1 from phi2 stage -phi1 phi1 n-logic -phi2 -phi2 phi1 n-logic to phi 1 phi2 stage phi2

phi1 phi1

phi1

evaluate phi1 logic precharge phi1 logic latch phi data 2 evaluate phi2 logic precharge phi2 logic latch phi1 data latch phi2 data

precharge phi1 logic

phi2

evaluate phi2 logic

VLSI System Design

NCKUEE-KJLEE

Four-Phase clock
11 12 nonoverlapping 23 24
clk 1 clk 2 D clk 1 clk 3 n 1 clk 4 clk 3 in v2 clk 3 clk 34 clk 3 clk 1 Q clk 2 clk 3 clk 4
NCKUEE-KJLEE

Four-Phase logic clocking method

Slave Latch

Logic

Master Latch

Logic

clk 1 Q clk 2 clk 3 clk 4

clk1

clk2

clk3

clk4

(a) in v1 clk 1 clk 12 D clk 1 (b )

VLSI System Design

Clock distribution
n-bit datapath

1. A single large buffer 2. A distributed-clock-tree approach

n-bit datapath n-bit datapath n-bit datapath n-bit datapath n-bit datapath clock n-bit datapath delays have to match between stages n-bit datapath n-bit datapath n-bit datapath n-bit datapath n-bit datapath

VLSI System Design

NCKUEE-KJLEE

Trends in clock strategy

For first-time designer, use static logic, single-phase
static registers. For standard cell and gate-array design, single-phase may be the only choice. Two phase clocking make timing design of RAMs, ROMs and PLAs easier. In modern process and circuits, cycle time is the main concern => single phased Processes are extremely dense => single phase

VLSI System Design

NCKUEE-KJLEE