ECE 3060 VLSI and Advanced Digital Design Lecture Datapath2: - - PowerPoint PPT Presentation

ece 3060 vlsi and advanced digital design
SMART_READER_LITE
LIVE PREVIEW

ECE 3060 VLSI and Advanced Digital Design Lecture Datapath2: - - PowerPoint PPT Presentation

ECE 3060 VLSI and Advanced Digital Design Lecture Datapath2: Registers,Comparators, Shifters, Decoders D Latch basic latch in out in out loading storing transparent latch in out load / store circuit schematic ECE 3060 Lecture


slide-1
SLIDE 1

ECE 3060 VLSI and Advanced Digital Design

Lecture Datapath2: Registers,Comparators, Shifters, Decoders

slide-2
SLIDE 2

ECE 3060 Lecture 16–2

D Latch

basic latch transparent latch loading storing

  • ut
  • ut

in in circuit schematic

  • ut

in

load / store

slide-3
SLIDE 3

ECE 3060 Lecture 16–3

Single Phase FF (Register)

  • PROBLEM: transparent latch can cause “oscillations”

when used in a sequential circuit

  • SOLUTION: a master-slave flip-flop (register cell)

transparent latch transparent latch in

  • ut

clock master slave clock master slave loading storing loading storing storing loading storing loading

slide-4
SLIDE 4

ECE 3060 Lecture 16–4

Delay and Clock Overlap

master slave transparent latch transparent latch in

  • ut

clk clk clk clk

ideal

clk clk

with delay clock

  • verlap
slide-5
SLIDE 5

ECE 3060 Lecture 16–5

Non-overlapping Two Phase Clocks

  • Advantages over Single-Phase Clocking
  • No Potential for Latches to be Transparent Simultaneously
  • Resistant to Delays in Circuit

transparent latch transparent latch in

  • ut

master slave

φ1 φ2 φ1 φ2

non-overlap periods

slide-6
SLIDE 6

ECE 3060 Lecture 16–6

Register Layout

  • Data and control flows left to right
  • Register is formed by tiling in the vertical direction
  • Attempt is to minimize vertical pitch

LD LD Vdd Gnd LD LD

D

Q Q

slide-7
SLIDE 7

ECE 3060 Lecture 16–7

Alternate Layout

  • Data still flows left to right, but control is vertical
  • Register is tiled vertically
  • Horizontal pitch is minimized

Vdd Vdd Gnd

D

LD LD Q Q

slide-8
SLIDE 8

ECE 3060 Lecture 16–8

Points to Note

  • Vdd and Gnd must be sized to carry necessary current

(1–2 mA/micron)

  • Cells may be flipped vertically to share Vdd and Gnd
  • Well contacts not shown
  • Vdd and Gnd inter-digitated

Vdd Gnd

slide-9
SLIDE 9

ECE 3060 Lecture 16–9

Comparator Circuits

  • Comparator may be implemented using adder/subtrac-

tor, or:

  • Pass gate implementation:
  • Note the use of nfets to pass ‘1’! This saves significant

area, at a slight penalty in speed and sensitivity to noise.

A0 B0 A1 B1

Vdd Vdd

A B =

slide-10
SLIDE 10

ECE 3060 Lecture 16–10

Comparator (cont.)

  • Transmission gates could also be used.
  • Signal must be restored every 3-4 stages.
  • Note the use of feedback at the output to sharpen the
  • utput transition, and bring it all the way to the upper

rail (Vdd)

slide-11
SLIDE 11

ECE 3060 Lecture 16–11

Multi-bit Shifters

  • There are several types of multibit (barrel) shifters
  • An arithmetic right shift by

is multiplication by

  • An arithmetic left shift by

is multiplication by

  • A logical shift by

ignores sign

  • Rotate by
  • There are several implementation
  • Transmission gate (

needs decoder)

  • Multiplexor (

no decoder needed)

k 2 k

k 2k k A O n2 ( ) = A n n log =

slide-12
SLIDE 12

ECE 3060 Lecture 16–12

T-gate Shifter

O0 O1 O2 O3 I6 I5 I4 I3:0 S3 S2 S1 S0

slide-13
SLIDE 13

ECE 3060 Lecture 16–14

Cell Layout

  • Cell layout requires consideration of signal flow

Overlap Needed No Overlap Needed 2x2 Tiling

slide-14
SLIDE 14

ECE 3060 Lecture 16–13

Shifter Operation

  • By multiplexing the inputs to the shifter, one can spec-

ify the shifter operation

  • Logical right shift
  • Arithmetic right shift
  • Logical left shift (complement

)

  • Rotate left (complement

)

  • Rotate right

000D3:0 D3D3D3D3:0 D3:0000 k D3:0D3:1 k D2:0D3:0

slide-15
SLIDE 15

ECE 3060 Lecture 16–15

Multiplexor Shifter

  • Basic idea: rows of 2x1 muxes which shift by
  • Does not need decoder, but wiring is expensive
  • May use inverting mux to simplify implementation

2k Shift by 1 Shift by 2 Shift by 4

I7 I6 I5 I4 I3 I2 I1 I0

slide-16
SLIDE 16

ECE 3060 Lecture 16–16

Predecoder Scheme

  • Cascaded stages of 2-4

input NAND/NOR gates

  • Layout is PLA like tiles
  • Vertical pitch is deter-

mined by rightmost column

  • Branching effort is reduced

for some inputs