Title: A Study of Delay and Power Consumption of Adders Built with Different Logic Styles
1A Study of Delay and Power Consumption of Adders
Built with Different Logic Styles
2Objectives
- Implementing 1-bit full adder module with
different logic styles, such as static CMOS,
TG-CMOS, Differential Cascade Voltage Switch
Logic(DCVSL), and Domino logic. - Measuring the delay and power consumption of the
32-bit ripple-carry adders. - Measuring the delay and power consumption of the
32-bit carry-lookahead adders. - Discussing the simulation results.
31-bit Full Adder
- The Logics of the Sum and Carry-out
- s x ? y ? C in
- Cout x?y x ? cin y ? cin
- x, y two 1-bit operands
- cin carry-in
- s sum
- cout carry-out
41-bit Full Adder (Static CMOS)
51-bit Full Adder(TG-CMOS)
62-input XOR Gate(TG-CMOS)
7The Sum Logic of 1-bit Full Adder(DCVSL)
from Dr. Reeses ece8273 lecture notes
8The Carryout Logic of 1-bit Full Adder(DCVSL)
from Dr. Reeses ece8273 lecture notes
91-bit Full Adder (Domino)
102-input Domino OR Gate
11Transistor Counts of 1-bit Full Adders
Table 1. The Transistor Counts of 1-bit Full
Adders
Static CMOS TG-CMOS DCVSL Domino
Transistor Count 34 36 22 41
1232-bit Ripple-carry Adder
1332-bit Carry Lookahead Adder
16-bit carry lookahead adder
16-bit carry lookahead adder
C0
C16
P12
G12
G02
P02
2-bit carry lookahead generator
P23
G23
1416-bit Carry Lookahead Adder Module
From Koren Text (Chapter 5)
15Circuit Structure Used for Simulations
X
Sum
32-bit adder
Y
Cout
Cin
CLK
16Spectre Simulation Results
Table 2. Power Consumption and Delay of 32-bit
Ripple-carry Adders (50MHz )
Adder Delay (ns) Power (mW) PowerDelay
Static CMOS 6.679 0.965 6.445
TG-CMOS 19.669 1.004 19.748
DCVSL 10.230 1.220 12.481
Domino 6.124 2.479 15.181
17Spectre Simulation Results
Table 3. Power Consumption and Delay of 32-bit
Carry Lookahead Adders (50MHz )
Adder Delay (ns) Power (mW) PowerDelay
Static CMOS 1.392 1.293 1.800
TG-CMOS 2.163 1.129 2.442
DCVSL 1.842 1.917 3.531
Domino 1.125 6.988 7.861
18Spectre Simulation Results
Table 4. The Delay of 32-bit Carry Lookahead
Adders for Different Input Switching Frequencies
unit (ns)
Adder 50 MHz 10 MHz 0.1 MHz
Static CMOS 1.392 1.407 1.409
TG-CMOS 2.163 2.191 2.176
DCVSL 1.842 1.858 1.845
Domino 1.125 1.138 1.142
19Discussions and Conclusions
- For any of the logic style, the delay of 32-bit
carry lookahead adder is much shorter than that
of 32-bit ripple-carry adder. - The rankings of the delay and power consumption
are not necessarily related to the transistor
counts. - TG-CMOS implementation has the longest delay
since its 1-bit full adder module cannot provide
enough driving capability. The switching
activities inside the adder take more time to
accomplish. - The delay of the domino logic implementation is
the shortest due to its single transistor pull up
network(less drain/source, parasitic capacitance
at the output node).
20Discussions and Conclusions(Cont.)
- The DCVSL implementation has the second longest
delay although it has the lowest transistor
count. The more complex pull down network
contributes this conclusion( more drain/source
and parasitic capacitances need to be charge or
discharged during the switching procedure). - The 32-bit ripple-carry adder consumes less power
than the 32-bit carry lookahead adder for any of
the logic style because of the carry lookahead
generator circuit. - TG-CMOS implementation has the least power
consumption since the switching activity occurs
less frequently than the other implementations.
21Discussions and Conclusions(Cont.)
- The domino logic consumes more power than the
other three logic styles. The large amount of
power is consumed during the precharge period.
(2.72mW for evaluation period and 4.27mW for
precharge period). - For a given logic style, the delay and power
consumption are determined by the input pattern. - For any of the logic style, the 1-bit full adder
module can be optimized for either speed or power
consumption. - The product of the delay and power consumption is
the smallest for the static logic style. The
domino logic is the best choice when the faster
circuit is required.
22References
- Ahmed M. Shams, Tarek K. Darwish, and Magdy A.
Bayoumi, Performance Analysis of Low-Power
1-bit CMOS Full Adder Cells, IEEE Trans. on Very
Large Scale Integration Systems, Vol.10, No.1,
pp20-29, February 2002. - Jan. M. Rabaey, Digital Integrated Circuits A
Design Perspective , Prentice-Hall Inc., 1996.