Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Module 3 PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

VLSI Design 15EC63

Module 3
Scaling of MOS Circuits
 High density chips in VLSI technology requires packing density of MOSFETS used
is high and also the size of the transistors to be as small as possible. This reduction
of size i.e. the dimension of MOSFET is referred as 'scaling'.
 It is expected that characteristics of the transistors will change with scaling and
physical limitations will restrict the extent of scaling.
 Microelectronic technology is characterized in terms of indicators or figures of
merit. The common figure of merits are
o Minimum feature size
o Number of gates on one chip
o Power dissipation
o Maximum operational frequency
o Die size
o Production cost
 The figure of merits can be improved by shrinking the dimensions of transistors,
interconnections and the separation between features, adjusting doping levels and
supply voltages.
 There are 2 types of size reduction/scaling strategies/ scaling models
1. Full scaling or constant-field scaling.
2. Constant-voltage scaling.
 Recently combined voltage and dimension model is presented. It is also called as
‘Lateral scaling’
 Two scaling factors 1/α and 1/β are used.
 1/β is chosen as the scaling factor for supply voltage VDD and gate oxide thickness D
 1/α is used as scaling factor for other linear dimensions
 Scaling theory indicates that the characteristics of MOS devices can be maintained
and the basic operational characteristics have to be preserved if the parameters of a
device are scaled in accordance to a given criteria.

 Constant field scaling: scaled device is obtained by applying


dimension less factor to
 All dimensions
 Device voltages
 Concentration densities

Scaling factors for device parameters:


The device dimensions are shown in Fig. 3.1

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Fig. 3.1 scaled nMOS transistor

1. Gate Area Ag

Ag = L W
L and W defines channel length and width respectively. They are scaled by factor
1/α.
Thus Agis scaled by 1/α2.

2. Gate Capacitance per Unit Area Co or Cox


Ԑox
Co =
𝐷

Ԑoxis the permittivity of gate oxide and D is the thickness of gate oxide (thinox).
1
Thus Cois scaled by =𝛽
1/β

3. Gate Capacitance Cg
Gate capacitance is given by, Cg = CoL W
Thus Cgis scaled by β*1/α2= β/α2

4. Parasitic Capacitance Cx
Cxis proportional to Ax/d
Where d depletion width around source or drain, it is scaled by 1/α. Ax is area of
depletion region around source and drain, it is scaled by 1/α2
Thus Cx is scaled by (1/α2)/(1/α) = 1/α

5. Carrier Density in channel Qon


Qonis the average charge per unit area in the channel in the ‘on’ state.
Qon= CoVgs [Co is scaled by β and Vgsis scaled by 1/β]
Thus Qonis scaled by β* 1/β= 1

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

6. Channel Resistance Ron


𝐿
Ron = *1/Qonμ
𝑊
L is scaled by 1/α, W is scaled by 1/α, Qonis scaled as 1, μ is carrier mobility in the
channel and is a constant.
Thus Ron is scaled by (1/α)/( 1/α)*1/1=1

7. Gate Delay Td
Tdis proportional to Ron. Cg
Ron is scaled to 1 and Cg is scaled by β/α2
Thus Td is scaled to 1*β/α2= β/α2

8. Maximum Operating Frequency fo


fo = W/L *(μCoVDD)/Cg
orfois inversely proportional to Td
Thus fois scaled as 1/(β/α2) = α2/β
9. Saturation Current Idss

Co is scaled by β, μ and 2 are constants, W and L is scaled by 1/α, Vgs and Vtboth
voltages are scaled by 1/β.
Thus Idssis scaled by β*(1/β)2= 1/β

10. Current Density J


𝐼𝑑𝑠𝑠
J=
𝐴
A is the cross sectional area of channel in ‘on’ state and is scaled by 1/α 2 and Idss is
scaled by 1/β.
Thus J is scaled by (1/β)/(1/α2) = α2/β

11. Switching Energy per gate Eg

Cg is scaled by β/α2 and VDD voltage is scaled by 1/β.


ThusEg is scaled by (β/α2)*(1/β)2= 1/α2β

12. Power Dissipation Per Gate Pg


Pg comprises 2 components, i.e., Pg= Pgs + Pgd
Pgsis the static component, given by Pgs= (VDD)2/Ron
Pgdis the dynamic component given by Pgd= Eg*fo
VDDis scaled by 1/β, Ronis scaled by 1
Eg is scaled by 1/α2β, fois scaled by α2/β
Thus Pgsand Pgd both are scaled by (1/β)2

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Thus Pg is scaled by 1/β2

13. Power Dissipation Per Unit Area Pa


Pais defined as scaled by

14. Power-Speed Product PT


PT=Pg* Td
Pg is scaled by 1/β2 and Td is scaled by β/α2
Thus PTis scaled by (1/β2)*(β/α2) = 1/α2β

Summary of Scaling effects of all the 3 models is given in the table below

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Subsystem Design Processes


Here the chapter deals with design of a digital system using a top down approach.
The system considered is a 4 bit microprocessor.
The microprocessor includes ALU, control unit, I/O unit and memory. Here only
ALU or data path is considered. The data path by itself is further divided into subsystem and
in this ‘shifter’ unit is considered.
General Considerations:
 The considerations provide ways of handling problems, provides way of designing
and realizing systems which are too complex
 Also help in understanding and appreciating technologies. The considerations are as
follows:
 Lower unit cost : with different approaches available for same requirement
lower unit cost is appreciable
 Higher reliability: high levels of system integrations greatly reduces
interconnections and this in turn provides good reliability.
 Lower power dissipation, lower weight and lower volume: in comparison
with other approaches.
 Better performance: particularly in terms of speed power product.
 Enhanced repeatability: if there are fewer process to be controlled in the
whole system or very large part to be realized on a single chip, this reduces
the repeatability which is appreciable
 Possibility of reduced design/development periods: for more complex
systems reduced development time is appreciable.

 Some Problems related with VLSI design are


1. How to design complex systems in a reasonable time and reasonable effort.
2. The nature of architectures best suited to take full advantage of VLSI and the
technology
3. The testability of large/complex systems once implemented on silicon
For the problem seen the solution is as follows:
Problem 1 & 3 are greatly reduced if two aspects are followed.
 a) Top-down design approach with adequate CAD tools
 b) Partitioning the system sensibly
 c) Aiming for simple interconnections
 d) High regularity within subsystem
 e) Generate and then verify each section of the design
 Devote significant portion of total chip area to test and diagnostic facility
Problem 2 can be solved by
 Select architectures that allow design objectives and high regularity in realization

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Illustration of Design Processes:


 Structured design begins with the concept of hierarchy. This involves dividing any
complex function into less complex sub-functions. This can be done until bottom
level referred to leaf cells is reached.
 This process is known as top-down design
 As a systems complexity increases, its organization changes as different factors
become relevant to its creation
 Coupling can be used as a measure the sub-modules interconnection. Clever system
aims at minimum sub-module interconnection resulting in independent design.
 Concurrency should be exploited so that all gates on the chip do useful work most of
the time
 As technology is changing fast, the adaptation to a new process must occur in a short
time.

General Arrangement of a 4-bit Arithmetic Processor:


The basic architecture of microprocessor is shown below.

 It has a unit which processes data when applied at one port and gives output at
second port.
 It is also possible that both data ports can be combined to form a single bidirectional
port if storage facility is available in the data path.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 The data path can be decomposed into having main subunits as it will be helpful in
deciding the possible floor plan.

 The sub units can be linked with different bus architecture. It can be either one-bus,
two-bus or three-bus architecture.
One bus architecture:

The sequence is:


1. First operand is moved from register to ALU and stored there.
2. Second operand is moved from register to ALU where operands are added
(subtraction or any other arithmetic operation) and result is stored in ALU
3. The result is then passed through shifter and stored in register.
4. The process takes 3 clock cycles and this be fastened by using two bus architecture.

Two bus architecture:

The sequence is:


1. Both operands (A & B) are sent from register(s) to ALU & are operated upon, result
S in ALU.
2. Result is passed through the shifter & stored in registers.
Dept. of ECE, SVIT 2017-18
VLSI Design 15EC63

3. This requires 2 clock cycle.

Three bus architecture:

1. Both operands (A & B) are sent from registers, operated in the ALU and result
which is shifted is returned to another register. All these happen in same clock
period.

 During the design process, care must be taken for allocating layers to various data
path and guidelines. The guidelines are as follows:
 Metal can cross poly or diffusion
 Poly crossing diffusion form a transistor.
 Whenever lines touch on the same level an interconnection is formed
 Simple contacts can be used to join diffusion or poly to metal.
 Buried contacts or a butting contacts can be used to join diffusion and poly
 If 2nd metal layer is available, this can cross over any layer and can be used
for power rails.
 Two metal layers may be joined using a via
 Each layer has particular electrical properties which must be taken into
account
 For CMOS layouts, p-and n-diffusion wires must not directly join each other
nor may they cross either a p-well or an n-well boundary
Design of 4 bit shifter
 Ageneral purpose n-bit shifter should be capable of shifting n incoming data up to n-
1 place either in a right or leftdirection.
 Shifting should take place in ‘end-around’ basis i.e., any bit shifted out at one end of
a data word will be shifted in at the other end of the word. Thus the problem of right
shift or left shift can be easily eased.
 It can be analyzed that for a 4-bit word, that a 1-bit shift right is equivalent to a 3-bit
shift left and a 2-bit shift right is equivalent to a 2-bit left etc. Hence, the design of
either shift right or left can be done. Here the design is of shift right by 0, 1, 2, or 3
places.
 The shifter must have:
• input from a four line parallel data bus
• Four output lines for the shifted data
• The input data can be transferred to output lines using shift from 0, 1, 2 to 3 bits
 While designing the strategy should be decided. The chosen strategy here is data
flow direction is horizontal and control signal flow direction is vertical.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 To meet the criteria a 4 × 4 crossbar switch is used. The MOS switch


implementation of 4 × 4 crossbar switch is shown in the Fig below.

 To drive each cross bar switch - 16 control signals SC00 to SC15 are provided to each
transistor switch.
 Arrangement is general and can be expanded to accommodate n-bit inputs/outputs.
 In this arrangement any input can be connected to any or all the outputs.
 If all switches are closed all inputs are connected to all outputs and it forms a short
circuit.
 As it needs 16 control signal it increases the complexity and to reduce the
complexity, the switch gates are coupled in groups (in this case it is grouped into 4)
 Here 4 groups of 4 is formed which corresponds to shift 0, shift 1, shift 2 and shift 3
bits.This arrangement is called ‘barrel shifter’. (In this only 4 control signals is
needed)

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 The interbus switches have their gate inputs connected in a staircase fashion in
groups of four and there are now four shift control inputs which must be mutually
exclusive in the active state.
 CMOS transmission gates may be used in place of the simple pass transistor
switches if appropriate. Barrel shifter connects the input lines representing a word to
a group of output lines with the required shift determined by its control inputs (sh0,
sh1, sh2, sh3). Control inputs also determine the direction of the shift. If input word
has n – bits and shifts from 0 to n-1 bit positions are to be implemented.

SH0 SH1 SH2 SH3 SH4


O0 O1 O2 O3 O0
O1 O2 O3 O2 O1
O2 O3 O0 O1 O2
O3 O0 O1 O0 O3

Block diagram of barrel shifter and shifting of data is shown in the table.
Illustration of the Design Process
Regularity: It is an essentiality of any design. It reduces design efforts required for a
system. Regularity of any particular design can be gauged as
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑎𝑛𝑠𝑖𝑠𝑡𝑜𝑟𝑠 𝑜𝑛 𝑡ℎ𝑒 𝑐ℎ𝑖𝑝
𝑅𝑒𝑔𝑢𝑙𝑎𝑟𝑖𝑡𝑦 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑎𝑛𝑠𝑖𝑠𝑡𝑜𝑟 𝑐𝑖𝑟𝑐𝑢𝑖𝑡𝑠 𝑡ℎ𝑎𝑡 𝑚𝑢𝑠𝑡 𝑏𝑒 𝑑𝑒𝑠𝑖𝑔𝑛𝑒𝑑 𝑖𝑛 𝑑𝑒𝑡𝑎𝑖𝑙
 For 4×4 bit barrel shifter, regularity factor is given by
16
𝑅𝑒𝑔𝑢𝑙𝑎𝑟𝑖𝑡𝑦 =
1
In case of barrel shifter it needs only one transistor designing and the same
can be applied to all the 16 transistors.
 Higher the regularity lesser will be the design effort. Good system design achieve
regularity factor of 50 or 100.

Design of ALU subsystem


 Heart of the ALU is the adder circuit
 4 bit adder will perform the sum of 2 4-bit number and here it is assumed that
parallel operation form is done.
 The input to the adder needs two 4-bit buses, a single 4 bit is needed to move data
from adder to shifter and other another 4-bit bus for shifting output back to register
array. It also provide ‘carry out’ and possible ‘carry in’ signal
 The block diagram of 4 bit carry adder is shown below

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 Considering an example of binary arithmetic operation.

 In any column we observe, there are 3 input. This number represents 3 inputs i.e.,
the corresponding input bits and previous carry/carry in
 If we consider kth column, it includes Ak, Bk representing input bits. Ck-1
represents previous carry
 There are 2 outputs, sum and new carry. i.e., in Kth column Sk and Ck
respectively.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 The sum carry equation we get is as follows:


 Sum Sk= HkC’k-1+ Hk’ Ck-1

New carry Ck=Ak Bk + Hk Ck-1


Where Hk is the half sum given by Hk= A’k Bk+ AkB’k
 The above sum and new carry equation can be implemented using ‘AND-OR’ or
using ‘EX-OR’ gates. But in VLSI this implementation becomes complicated and
alternate method is to find a standard element through which the above equations
can be implemented.
 Analyzing the table of full adder again, we can see that
For Sk, if Ak= Bk then Sk= Ck-1
elseSk= C’k-1
For Ck if Ak= Bk then Ck= Ak= Bk
else Ck= Ck-1
 A standard adder element (1 bit) can be implemented as shown below

 To form n-bit adder, n adder elements must be cascaded with carry-out of 1 element
with carry-in of next most significant bit.
 The adder element can be implemented using multiplexer either in nMOS or CMOS
technology. This method of implementation is easy to follow and results obtained are
in the required forms.
 Implementation of adder using MUX (pass transistor ) is shown in the last page.
Below figure shows the stick diagram of the same.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Implementing ALU function with adder element:


 An ALU must be able to add and subtract two binary numbers, perform logical
operations such as AND, OR and Equality (EX-OR) functions.
 Subtraction can be easily implemented using the adder element. Suppose we need
find A – B, then
 Compliment B and add 1 to it to get B’
 Add this to A i.e., A+B’
 This gives subtraction using adder
 Here compliment for subtraction can be served from ‘logical compliment’ (negate)
 In order to keep architecture as simple as possible it would be better if all the logical
operations can be performed on the adder.
 Considering the possibility
We have the sum and new carry equations as
Sum Sk= HkC’k-1+ Hk’ Ck-1
New carry Ck=Ak Bk + Hk Ck-1
Where Hk =A’k Bk+ AkB’k

 If in Sk, if Ck-1 is held at logic 0 then the expression would be


 Sk= Hk .1+ Hk’ 0
 Sk= Hk = A’k Bk+ AkB’k
 This represents the EX-OR function/operation
 If in Sk, if Ck-1 is held at logic 1 then the expression would be
 Sk= Hk .0+ Hk’. 1
 Sk= H’k = A’kB’k+ Ak Bk
 This represents the EX-NOR function/operation
 If in Ck, if Ck-1 is held at logic 0 then the expression would be
 Ck=Ak Bk + Hk0
 Ck=Ak Bk
 This represents the AND function/operation
 If in Ck, if Ck-1 is held at logic 1 then the expression would be
 Ck=Ak Bk + Hk. 1=Ak Bk + Hk
 Ck=Ak Bk + A’k Bk+ AkB’k
 Ck= Bk+ AkB’k
 Ck= Bk+ Ak
 This represents the OR function/operation
 Thus with suitable switching of carry line between adder element will give ALU
logical functions.
 One such arrangement for both arithmetic and logical function is shown in figure..

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Further considerations of Adders:


 Simple and direct implementation of adder circuit was observed now alternate form
of adder equations will be observed.
 In the adder carry will be propagated when either Akor Bk is high. Thus this equation
can be written as
Pk= A’k Bk+ AkB’k= AkXOR Bk
 Also carry will be generated when both Ak and Bk are high. This equation can be
written as
Gk= Ak Bk
 With this the expression for sum Skand carry Ck the expression can be written as
Sk= A’k BkC’k-1+ AkB’kC’k-1+A’kB’kCk-1+ Ak BkCk-1
Sk= AkXORBk XOR Ck-1

And Ck=Ak Bk + A’k BkCk-1+ AkB’k Ck-1


Ck= Gk+ PkCk-1

 Thus adder based on pass/generate carry concept is shown in the Fig


 Using this circuit carry can either be propagated to next stage or generated.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 Here propagate signal from ex-or gate is used to drive the transmission gate which
acts like a controlling signal to propagate the carry signal. Only when P signal is
high TG will be enabled and propagates Ck-1
 When P signal is 0 hen depending on the input A and B carry will be generated.
 The nMOS and pMOS pass transistor will generate the carry signal. (When A=B=1,
carry is generated and when A=B=0 carry is not generated.)
 This is a direct realization which leads to the concept of carry chain and leads to
popular arrangement known as Manchester carry chain.

Manchester Carry Chain:

 It is a fast adder circuit, it is a carry propagate adder in which the delay taken by the
carry to reach the last stage of output is reduced.
 As seen in the previous concept of passing the carry the transmission gate, the path
can be precharged by the clock
 Then the path can be gated by the n-type pass transistor.
 When clock is 0, output will be charged to logic high due to pMOS. When clock is 1
pMOS will be off. If Pkis high, then the carry will be propagate.
 If Pk is 0, then Ck-1 will not be propagated.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 Depending on inputs at Gk,(Ak= Bk =1) carry will be generated or (Ak= Bk =0) no


carry generation.
 Even though Manchester carry cells are faster while cascading delay is observed.
Cascading is done by connecting pass transistor in series.
 As n pass transistor is cascaded the delay also increases as square of n. thus in order
to reduce delay buffers are included at after every 4 chain as shown in the block
diagram

Fig. Cascading of Manchester carry adder

Adder Enhancement Techniques:


 In any adder element the previous carry bit is necessary to compute its own sum and
carry out bits.
 For smaller adder (n<8) the delay observed for carry to propagate in ripple carry
adder is small but at the n increases the delay is more and hence the time to find sum
also increases.
 Thus some special techniques are used to improve the time taken for addition.
 This includes 3 methods
o Carry Select Adder
o Carry Skip Adder
o Carry Look Ahead Adder
Note: All the above 3 techniques is a modification of ripple carry adder.
Carry Select Adder:
 In carry select adder the adder will be divided into group.
 It requires two identical parallel adders in each group except the least significant
group as seen in the block diagram.
 Each group generates a group carry. Here, two sums are generated simultaneously.
One sum assumes that the carry in is equal to one as the other assumes that the carry
in is equal to zero.
 A mux is used to select valid result.
 It can be observed that the group carries logic increases rapidly when more high-
order groups are added to the total adder length.
 This complexity can be decreased, with a subsequent increase in the delay, by
partitioning a long adder into sections, with four groups per section, similar to the
CLA adder.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Fig. Block diagram of Carry Skip Adder


Optimization of Carry Skip Adder:
 For n bit ripple carry adder, the computational time T is given by
T = K1n , where K1is the delay through 1 adder cell
 As carry select adder is a modification of carry skip which is having 2 adder in each
block i.e., each block has 2 parallel paths. Thus computation time T becomes
T = K1n/2 + K2 where K2 is the time needed by the mux to select the actual carry
output.
 If there are many multiplexer then ripple through effect of carry is observed in mux
rather than in the carry chain. Thus optimum value of mux should be selected for the
size of each block.
 Suppose if there is an n-bit adder divided into M-blocks and each block contains P
adder cells (n = M.P). The computation time for overall carry has 2 components
o Propagation delay through the first block
o Propagation delay through mux
Thus, T = PK1 + (M – 1) K2
Minimum value for T is seen when M=√𝑛𝐾1/𝐾2

Note: except the first group all other groups have mux. If M is the number of groups the
number of mux for n bit adder will be (M-1) and delay observed is (M -1)K2.
Carry Skip Adder:
 In a Ripple Carry Adder, if the input bits Ai and Bi are different for all position i,
then the carry signal is propagated at all positions (thus never generated), and the
addition is completed when the carry signal has propagated through the whole adder.
 In this case, the Ripple Carry Adder is as slow as it is large. Actually, Ripple Carry
Adders are fast only for some configurations of the input words, where carry signals
are generated at some positions
 Carry Skip Adders take advantage both of the generation or the propagation of the
carry signal.
 They are divided into blocks, where a special circuit detects quickly if all the bits to
be added are different. This signal is called ‘block propagation signal’

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 If in the block, if A & B bits differ then the output i.e., block propagation signal = 1.
If it is 1 then carry entering the block can bypass and transmit the carry to the block
through multiplexer.
 Fig below shows 24 bit adder with 4 blocks and each block has 6 adder elements.

Fig. Block diagram of 24 bit carry skip adder

Optimization of carry skip adder:

 Let K1 represents the time needed by the carry signal to propagate through an adder
cell, and K2 the time it needs to skip over one block. Suppose the N-bit Carry Skip
Adder is divided into M blocks, and each block contains P adder cells. The actual
addition time of a Ripple Carry Adder depends on the configuration of the input
words. The completion time may be small but it also may reach the worst case, when
all adder cells propagate the carry signal.
 In the same way, we must evaluate the worst carry propagation time for the Carry
Skip Adder. One of the worst case of carry propagation is depicted in Figure.

Fig. Worst case carry propagation for Carry Skip adder

 The configuration of the input words (for the worst case) is such that a carry signal is
generated at the beginning of the first block. Then this carry signal is propagated by
all the succeeding adder cells but the last which generates another carry signal.
 In the first and the last block the block propagation signal is equal to 0, so the
entering carry signal is not transmitted to the next block.
 Consequently, in the first block, the last adder cells must wait for the carry signal,
which comes from the first cell of the first block. When going out of the firstblock,
the carry signal is distributed to the 2nd, 3rd and last block, where it propagates.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

 In these blocks, the carry signals propagate almost simultaneously (we must account
for the multiplexer delays).
 For the above condition the time to compute addition is given by
T = 2 (P – 1) K1 + (M – 2) K2

Note: in the first block except the first bit all other bits have for carry propagation hence it
(P – 1) and as the same is observed for last block it is 2(P-1) and delay is K1. Also the input
carry signal does not skip over the first and last block hence K2 is multiplied with (M-2).
Carry Look Ahead Adder:
 This is another method to improve the throughput time of adder.
 Here the prediction of carry is done and based on this designing is done.
 With the carry generate Gk= AkBk and carry propagate Pk= (AkXORBk)Cinthe carry
propagation can be avoided and carry outputs can be calculated.
 If n (number of adder) is large, the carry output bits increases and expression larger
and complex. Thus ‘carry look ahead adder’ and ‘ripple carry adder’ are clubbed
together. This can be seen in the block diagram.

C11 4 bit adder C7 4 bit adder C3 4 bit adder Cin


Cout 4 bit adder
with CLA with CLA with CLA with CLA

S15-S12 S11-S8 S7-S4


S3-S0
Fig. Block diagram of 16 bit carry look adder

 Considering the first block, carry out C3 can be calculated as follows


Cout= Gk+ PkCin
Thus for the first bit in first adder block is given by
C0 = G0 + P0Cin
Similarly for 2nd bit it is given by
C1 = G1 +P1C0 using C0 in the equation for C1
C1 = G1+ P1G0 +P1P0Cin
rd
For 3 bit it is given by
C2 = G2 +P2 C1 using C1 equation for C2
C2 = G2+ P2G1 +P2P1G0 + P2P1P0Cin
For the 4th bit it is given by

C3 = G3 +P3 C2 using C2 equation for C3


C3 = G3 +P3G2+ P3P2G1 +P3P2P1G0 +P3P2P1P0Cin

 From the above equation we see that carry out C3 can be calculated in prior without
the need of need of carry being propagated through the adder cells.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

B inputs A inputs

Cin
Carry Look Ahead Adder
Cout

Logic for Logic for Logic for Logic for


Sum Sum Sum Sum

Sn+3 Sn+2 Sn+1 Sn

 Each adder block has a 4 bit CLA(Carry Look Ahead) unit and is shown below.

G3 P3 G2 P2 G1 P1 G0 P0

C3 Cin
CLA UNIT

C2 C1 C0

 From the CLA unit carry outputs C0, C1, C2 and C3 can be simultaneously determined.
In this C0, C1, C2 are required to determine the sum and will not ripple through.
However C3 will propagate to the next block.
 By observing the expression for C0, C1, C2 and C3 the last term in the equation become
zero when Cin = 0.
 Thus to increase the speed, the expression for C3 can be written as
C3 = γ + π
Where π = P3P2P1P0Cin and
γ = G3 +P3G2+ P3P2G1 +P3P2P1G0
 Here the term γ is independent of Cin and π becomes 0 when Cin = 0. Hence the speed
of adder can be minimized by using these 2 defined functions γ and π.
 The implementation is shown in the block diagram. The carry outputs C3, C7 and
C11 are propagated sequentially. This propagation delay can be reduced by using γ
and π outputs, where C3, C7 and C11 are computed seperately.

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Dept. of ECE, SVIT 2017-18


VLSI Design 15EC63

Implementation of Adder using MUX.

Dept. of ECE, SVIT 2017-18

You might also like