Upload
asif-basha
View
216
Download
0
Embed Size (px)
Citation preview
8/6/2019 ThakurMWSCAS06
1/5
Effects of Process and Enviromental Variations
on Adder Architectures
Arun Thakur, Dinesh Chilamakuri, and Dimitrios VelenisDept. of Electrical and Computer Engineering
Illinois Institute of Technology
Chicago, Illinois 60616
Email: [email protected], [email protected], and [email protected]
Abstract Scaling of the on-chip feature size and power supplyvoltage have significantly reduced the noise margins of anintegrated circuit and have aggravated the effects of processand enviromental variations. These effects can introduce delayvariations on the signals within a circuit, possibly causing aviolation of the timing constraints in a clocked register thatcan lead to system malfunctioning. The effects of parametervariations on the timing characteristics of adder structures areinvestigated in this paper. The sensitivity of the critical delay of
sum and carry signals under variations in power supply voltage,temperature, and gate oxide thickness is demonstrated for fourdifferent adder architectures.
I. INTRODUCTION
The dominant characteristic in the evolution of integrated
circuits has been the scaling of the on-chip feature size. This
trend has produced phenomenal improvements in circuit func-
tionality and system performance by increasing the number of
transistors on-chip as well as the transistor switching speed.
However, with shrinking feature sizes the sensitivity of the
circuit elements to imperfections in the manufacturing process
is aggravated [1]. Furthermore, as the power supply voltage is
reduced to decrease the power dissipation, the driving strengthof transistors is weakened [2]. Therefore the propagation delay
of a signal can be significantly affected by variations in
enviromental parameters and on-chip noise [3], [4]. These
effects can introduce variations in signal delay and cause a
violation of the tight timing constraints at a register, especially
at the most critical data paths within a system [5].
One of the most critical functional blocks within a syn-
chronous integrated circuit is the Arithmetic Logic Unit
(ALU). The basic component of an ALU is the adder circuit
which in many applications determines the overall speed of
a system. Therefore, the effects of process and enviromental
parameter variations on signals propagating along an adder
structure can cause a violation of the timing constraints,resulting in a system malfunction. The delay of a critical
path within an adder and the amount of delay variations is
determined by the adder architecture utilized in a particular
implementation.
In this paper the effects of process and environmental pa-
rameter variations on the delay of different adder architectures
are investigated. The sensitivity of the sum and carry signal
delays is evaluated for four different adder architectures. The
adder architectures considered in this paper are introduced in
Section II. The effects of variations in power supply voltage,
temperature, and gate oxide thickness upon the different adder
implementations are discussed in Section III. Finally some
conclusions are presented in Section IV.
I I . ADDER ARCHITECTURES
Addition of binary numbers is implemented in a bitwise
approach. At each bit position the sum value can be determinedbased upon the corresponding bit values of the operands and
the incoming carry value from the previous position. Therefore
the delay of an addition operation is determined by the total
number of bits in the operands since the incoming carry value
should be propagated from the least significant bit position to
the highest one. In order to decrease the total addition delay
different carry propagation techniques have been proposed at
both the logical and circuit level.
The four adder architectures considered in this paper are
introduced in this section. The operand length for each adder
is 16 bits and the different architectures are implemented
in
technology with 1.8 Volts nominal power supply
voltage. The delay of the sum and carry values at the mostsignificant bits are determined using Hspice simulations and
are listed in Table I.
TABLE I
SUM AND CARRY DELAYS AT NOMINAL PARAMETER VALUES
Adder Sum propagation Carry propagationArchitecture delay delay
Ripple Carry Adder 2234 ps 2158 psCarry Select Adder 1745 ps 1731 psSquare Root Adder 1438 ps 1264 ps
Carry Lookahead Adder 571 ps 135 ps
The adder designs listed in Table I are the Ripple Carry
Adder, the Carry Select Adder, the Square Root Adder, and
the Carry Lookahead Adder. These designs are presented in
the following sections.
A. Ripple Carry Adder
The Ripple Carry Adder (RCA) [6] is the adder architecture
that is implemented directly from of the addition logic. The
sum at each bit position is determined by the corresponding
bit values of the operands and the incoming bit value. The
addition is completed once the carry value propagates along
8/6/2019 ThakurMWSCAS06
2/5
8/6/2019 ThakurMWSCAS06
3/5
positions can be calculated using the carry propagate and carry
generate signals in the carry lookahead mirror logic structure
shown in Figure 4(b).
A 4-bit implementation of a Carry Lookahead Adder is
illustrated in Figure 5. Four mirror structures of different sizes
are utilized to determine the incoming carry values at each bit
position. All the carry values are calculated in parallel after the
carry propagate and carry generate signals become available.
Therefore the ripple effect of the carry is eliminated. Once the
carry values are determined, the sum values at each position
are calculated within the corresponding PFA structures. In
addition the group-generate and group-propagate signals are
produced by the carry lookahead logic as shown in Figure 5.
PFA
A0 B0
P0G0
PFA
A1 B1
P1G1
PFA
A2 B2
P2G2
PFA
A3 B3
P3G3
CIN
C1 C2 C3
COUT
GroupGenerate
C0
Sum0 Sum1 Sum2 Sum3
Carry Lookahead Logic
GroupPropagate
Fig. 5. 4-Bit Carry Lookahead Adder
The group-generate and group-propagate output signals are
utilized to implement a 16-bit Carry Lookahead adder by
adding one more level of hierarchy as illustrated in Figure6. The carry lookahead logic shown in Figure 6 is utilized to
produce the incoming carry values at bit positions 4, 8, and
12. The critical paths of the 16-bit Carry Lookahead Adder are
illustrated in Figure 6 with the grey arrows. The performance
enhancement achieved by the implementation of the Carry
4bit
CLA
A0-3 B0-3
P0G0
CIN
C0
S0-3
4bit
CLA
A4-7 B4-7
S4-7
4bit
CLA
A8-11 B8-11
S8-11
4bit
CLA
A12-15 B12-15
S12-15
P1G1 P2G2 P3G3
C3 C7 C11
GroupGenerate
GroupPropagate
Carry Lookahead Logic
C15
Fig. 6. 16-Bit Carry Lookahead Adder
Lookahead Adder is listed in Table I. Furthermore, notice
in Figure 6 that the final carry value at the most significant
bit position is determined at the highest hierarchical level.
Therefore it is produced much faster compared to the final
sum value at the same position, as listed in Table I. For the
calculation of the final sum value, the incoming carry value
at position 12 is required within a lower level of the adder
hierarchy, therefore increasing the delay of the sum signal.
III. PARAMETER VARIATION EFFECTS
The effects of process and enviromental parameter varia-
tions on the adder architectures presented in Section II are in-
vestigated in this section. In particular, the effects of variations
in power supply voltage ( ! " " ), temperature (% &
(
), and gate
oxide thickness (01 3
) on the sum and carry signals at the most
significant bit positions of each adder are demonstrated. The
effects of ! " " variations are discussed in section III-A. The
effects of variations in temperature are presented in section
III-B and the variations in gate oxide thickness are discussed
in section III-C.
A. Variations in power supply voltage
In high performance synchronous integrated circuits, effects
such as the 45
voltage drop and 67 9
7@
noise can affect the
voltage level at the power supply [10]. Furthermore, power
saving mechanisms such as clock gating and system standby
that control the switching activity of large circuit blocks within
an IC may also affect the power supply voltage level. The
effects of ! " " variations on the delay of the different adder
architectures are demonstrated in this section.
The effect in adder delay is evaluated for a drop in the power
supply voltage of 33% below of the nominal value of 1.8 V.
The result of the voltage drop on the sum output delay at the
most significant bit location for the four adder architectures is
illustrated in Figure 7. It is shown that the sum delay of the
Carry Select Adder is the most sensitive to ! " " variations.
Alternatively, the Square Root Adder is the most robust design
under a power supply drop.
0%
10%
20%
30%
40%
50%
60%
70%
Nominal
VDD6% 11% 17% 22% 28% 33%
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Sum
DelayVariation(%)
Power Supply Drop
Fig. 7. Percent increase in sum delay due to C D D drop
8/6/2019 ThakurMWSCAS06
4/5
In addition the effects of variations in the power supply
voltage on the delay of the carry signal are evaluated and the
results are illustrated in Figure 8. It is shown in Figure 8 that
the increase in carry delay is similar to the increase in the
sum signal delay for the RCA, CSA, and SRA architectures.
However, for the Carry Lookahead Adder, the increase in
the delay of the carry signal is larger than the increase in
sum delay. This is due to the long transistor stacks at the
carry lookahead mirror logic blocks utilized in the CLA. As
the power supply voltage drops, the driving strength of the
stacked transistors is farther reduced compared with the rest
of the transistors in the adder structure. Therefore, the percent
increase in the carry signal is higher than the corresponding
increase in the sum delay. However, the carry signal is still
faster than the sum signal. For the lowest level of ! " " drop,
at 33% below the nominal value, the delay of the carry signal
is 256 ps (90% increase) and the delay of the sum signal is
902 ps (58% increase).
CarryDelayVariation(%)
Power Supply Drop
0%
10%
20%
30%
40%
50%
60%
70%
80%90%
100%
Nominal
VDD6% 11% 17% 22% 28% 33%
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Fig. 8. Percent increase in carry delay due to C D D drop
B. Temperature Variations
With increasing die area, circuit density, and on-chip power
dissipation the variation of temperature across an integrated
circuit becomes significant. The operating temperature of a
circuit element depends upon the proximity to a hot spot
within a die. Therefore, the effects of temperature on circuit
performance are non-uniform across a die. The effects of
temperature variations on the delay of the sum and carry
signals along the different adder designs are investigated in
this section.
Variations in the operating temperature within the range ofF
G
1
C to
1
C are considered for the four adder architectures.
The effect of temperature variations on the delay of the sum
signal at the most siginificant bit is illustrated in Figure 9.
As shown in Figure 9, variations in temperature have larger
effect on the delay of the Ripple Carry Adder. Furthermore,
by comparing Figures 9 and 7, it is shown that the percent
increase in adder delay caused by power supply variations is
significantly larger than the percent increase on the adder delay
due to the effect of temperature variations.
Sum
DelayVariation(%)
Temperature (oC)
0%
2%
4%
6%
8%
10%
12%
27 40 60 80 100
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Fig. 9. Percent increase in sum delay due to temperature variations
In addition, the effect of temperature variations in the delay
of the carry signal are investigated and the results are shown
in Figure 10. It is shown in Figure 10 that for the RCA,
CSA, and SRA architectures the delay of the carry signal is
increased at the same percent as the sum signal. Alternatively,in the Carry Lookahead Adder the percent increase in the
delay of the carry signal is higher than the corresponding
increase in the delay of the sum signal. Therefore, the effect
of temperature variations is larger upon the stacked transistor
structures utilized in the carry lookahead mirror logic blocks
than the rest of the transistors in the adder circuit.
CarryDelayVariation(%)
Temperature (oC)
0%
2%
4%
6%
8%
10%
12%
14%
16%
18%
27 40 60 80 100
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Fig. 10. Percent increase in carry delay due to temperature variations
C. Gate Oxide Thickness Variations
As the on-chip feature size is decreased the effects ofvariations in the manufacturing process are aggravated. One of
the device parameters that is susceptible to imperfections in the
manufacturing process is the gate oxide thickness ( 0 13
) with a
nominal value below 20 angstroms for the current technology
nodes [11]. The slightest variation in the gate oxide deposition
process can create a significant variation in 0 13
. In this section
the effects of 01
3
variation within 20% of the nominal 01
3
value on the delay of the different adder architectures are
demonstrated.
8/6/2019 ThakurMWSCAS06
5/5
The effect of 01 3
variations on the delay of the sum signal
for theQ four adder designs is illustrated in Figure 11. It is
shown that 01 3
variations have significantly lower effect on
adder delay compared with variations in power supply voltage
and temperature. A RF
T
variation in gate oxide thickness
causes a maximum variation in the sum delay within the RU
T
range. As shown in Figure 11 the most sensitive design to0 1 3
variations is the Square Root Adder.
Sum
DelayVariation(%)
Gate Oxide Thickness Variation (%)
-6%
-4%
-2%
0
2%
4%
6%
8%
-20 -15 -10 -5 0 5 10 15 20
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Fig. 11. Percent variation in sum delay due to W X ` variations
Furthermore the effect of variations in the gate oxide
thickness on the carry delay of the adders is evaluated and
the results are illustrated in Figure 12. It is shown that the
percent variation in the carry delay in the Carry Lookahead
adder is higher compared with the variation in the delay of
the sum signal, due to the stacked transistor structures in the
carry lookahead mirror logic block.
CarryDelayVariation(%
)
Gate Oxide Thickness Variation (%)
-6%
-4%
-2%
0
2%
4%
6%
8%10%
-20 -15 -10 -5 0 5 10 15 20
Ripple Carry Adder
Carry Select Adder
Square Root Adder
Carry Look Ahead Adder
Fig. 12. Percent variation in carry delay due toW
X`
variations
IV. CONCLUSIONS
The effects of variations in power supply voltage, temper-
ature, and gate oxide thickness on the critical delays of the
sum and carry signals within four different adder architec-
tures are demonstrated in this paper. It is shown that power
supply variations have the largest effects on adder delay. The
variations demonstrated in the delay of the sum and carry
signals are similar for the Ripple Carry Adder, Carry Select
Adder, and Square Root Adder under the same parameter
variation effects. Alternatively in the Carry Lookahead Adder,
parameter variations have a greater effect on the delay of
the carry signal compared with the variations in the sum
signal delay due to the long stacked transistor structures
utilized within the carry lookahead mirror logic. Therefore, the
Carry Lookahead Adder provides an enhancement in addition
performance albeit an increase in sensitivity on the effects of
process and enviromental parameter variations.
REFERENCES
[1] S. Natarajan, M. A. Breuer, and S. K. Gupta, Process variations andtheir impact on circuit operation, Proceedings of the IEEE InternationalSymposium on Defect and Fault Tolerance in VLSI Systems, pp. 7381,November 1998.
[2] R. Sitte, S. Dimitrijev, and H. B. Harrison, Device parameter changescaused by manufacturing fluctuations of deep submicron mosfets, IEEETransactions on Electron Devices, vol. 41, no. 11, pp. 22102215,November 1994.
[3] S. Sauter, D. Schmitt-Landsiedel, R. Thewes, and W. Weber, Effectof parameter variations at chip and wafer level on clock skews, IEEETransactions on Semiconductor Manufacturing, vol. 13, no. 4, pp. 395
400, November 2000.[4] K. T. Tang and E. G. Friedman, Delay and noise estimation of cmoslogic gates driving coupled rc interconnections, Integration, the VLSI
Journal, vol. 29, no. 2, pp. 131165, September 2000.[5] D. Velenis, M. C. Papaefthymiou, and E. G. Friedman, Reduced delay
uncertainty in high performance clock distribution networks, Proceed-ings of the IEEE Design Automation and Test in Europe Conference,pp. 6873, March 2003.
[6] N. H. Weste and D. Harris, CMOS VLSI Design: A Circuits and SystemsPerspective, 3rd ed. Boston: Addison-Wesley, May 2004.
[7] O. J. Bedrij, Carry-select adder, IRE Transactions on ElectronicComputers, vol. EC-9, pp. 226231, 1960.
[8] J. M. Rabaey, Digital Integrated Circuits: A Design Perspective. Pren-tice Hall Inc., 1995.
[9] N. Quach and M. J. Flynn, High-speed addition in cmos, IEEETransactions on Computers, vol. 41, no. 12, pp. 16121615, December1992.
[10] A. V. Mezhiba and E. G. Friedman, Power Distribution Networks in HighSpeed Integrated Circuits. Norwell Massachusetts: Kluwer AcademicPublishers, 2004.
[11] S. I. A., The national technology roadmap for semiconductors, Semi-conductor Industry Association, Tech. Rep., 2004.