Journal Pre-Proof: Microprocessors and Microsystems
Journal Pre-Proof: Microprocessors and Microsystems
B. Sakthivel , A. Padma
PII: S0141-9331(19)30528-9
DOI: https://doi.org/10.1016/j.micpro.2019.102958
Reference: MICPRO 102958
Please cite this article as: B. Sakthivel , A. Padma , Area And Delay Efficient GDI Based
Accuracy Configurable Adder Design, Microprocessors and Microsystems (2019), doi:
https://doi.org/10.1016/j.micpro.2019.102958
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
Abstract
Adders are one of the basic fundamental critical arithmetic circuits in a system and their performances
affect the overall performance of the system. Traditional n-bit adders provide precise results, where as
the lower bound of their critical path delay of n bit adder is Ὠ(log n). To achieve a minimum critical path
delay lower than Ὠ(log n), many inaccurate adders have been proposed. These inaccurate adders
decrease the overall critical path delay and improve the speed of computation by sacrificing the
accuracy or predicting the computation results. In this work, a fast reconfigurable approximate ripple
carry adder has been proposed using GDI(Gate Diffusion Logic) passing cell. Here, GDI cell acts as a
reconfigurable cell to be either connected with the previous carry value or approximated value in an
adder chain . This adder has greater advantage and it can be configured as an accurate or inaccurate
adder by selecting working mode in GDI cell. The implementation results show that, in the approximate
working mode, the proposed 64 -bit adder provides up to 23% ,34% and 95 % reductions in area, power
and delay ,respectively compared to those of the existing adder.
1 Introduction
The speed improvement and power dissipation reduction are the key objectives in
the design of digital signal processing units, particularly portable systems. Generally,
speed improvement can be achieved at the increased cost of more power consumption in
the exact processing units. One of the ways to improve the speed and power is to sacrifice
the computation accuracy .This strategy, which is approximating computing has been
used for error tolerant applications [1] .
Adder units, which are principle component in arithmetic unit of DSP systems, are
power starving and often form hot-spots on the die[2].These facts have been the
inspirations for realizing this block using approximate computing methods.
Normally,conventional full adder uses cascaded full adders of length equal to that
of number of input bits and hence , a critical carry propagation delay has been formed by
carry chain .To the best of the researchers knowledge, all the approximate adders rely
on the concept of overall carry propagation delay and it is less than complete length of
the adder. This has motivated the researchers to split the cascaded full adders into sub
adders or smaller disjoints for reducing carry chain delay [3]. In each smaller disjoints the
resultant bits are calculated separately and contributed to the final sum calculation and
the previous prediction bits are used to calculate output carry.
In this paper, reconfigurable ripple carry adder has been focused as well as the use
of GDI passing cell has been suggested for switching between the inexact and exact
working modes for both exact and inexact applications. The construction of the adder,
which is based on the conventional RCA adder, does not require any correction unit
externally for the exact addition operation. The proposed design during exact mode delay
and power is same as the conventional working one and it is considerably smaller in
inexact working mode. The rest of the paper has been structured as follows. In Section II,
some related works are pointed and reviewed .The structure of the proposed GDI
reconfigurable approximate adder is described in Section III. The results of the efficacy
of the proposed as well as their comparison with the conventional structures are discussed
in Section IV. Finally, the conclusion is depicted in Section V
2 Related works
Earlier researches on inexact adders have concentrated on two general strategies
such as error weight and error probability reductions (see [4]– [11]).The first design
strategy has used hybrid structure by combining two different parts, correct MSBs(Most
Significant Bit) and approximated Least Significant Bits(LSB) [6-11]. The accurate
values of MSB have been propagated to next stage and LSB has been formed by
summation. It restricts error weight to the weight of the carry input of accurate MSB part
and 70% of power reductions can be achieved [6]. In the second methodology, pure
inexact adder structures are utilized. For these adders, reductions of power consumption,
error probability and delay have been considered as key design parameters [10][13]]. It
may also need an error compensation unit which also has area ,power and time overhead
.To address these issues some image and video processing applications are required to
run time accuracy adjustments and hence, runtime accuracy altering designs have been
developed
Few Accuracy Configurable Adders(ACA) other than voltage over scaling have
been proposed using approximation schemes .At earlier [14],approximate adder has been
designed with error detection and correction unit. That adder can be configured to
varying accuracy levels or accurate levels based on the application. But,the error
detection and the correction circuit have further increased area overhead .In ACA and
GeAr (Generic accuracy configurable adder )[15] designs, error compensation starts from
Least Significant Bits (LSBs) that leads to accuracy improvement.Where as computation
time takes more cycles because of the connections of error correction circuits with the
pipelined fashion.
GDI logic has been proposed as an alternative to other logic styles and it offers
less number of transistor count for implementing boolean function by reducing swing
voltage at their outputs .The reduction in voltage swing has encouraged to design low
power circuits
Fig 1 GDI cell Fig 2 Carry Passing GDI Cell
N P G Out Output
0 In2 In1 In1’in2 f1
Table 1: Different logic
In2 1 In1 In1’+In2 f2
implementations by GDI
1 In2 In1 In1+in2 OR
In2 0 In1 In1In2 AND
In3 In2 In1 In1’In2+In1In3 MUX
0 1 In1 In1’ NOT
The structure of GDI cell is shown in Fig.1.The diffusion terminal acts as an external
input in the GDI cell and it helps to implement various Boolean functions such as OR,
AND, MUX, INVERTER, f1 and f2, as listed in Table 1. In the proposed work, N and P
terminals of GDI cell act as a selection line for choosing approximate mode or normal
mode operation and it is called as GDI Passing cell as shown in Fig 2. By setting NP
terminal as 10 ,the GDI passing cell can be configured as wire to pass an input terminal G
value to the output terminal. By setting NP as either 00 or 11 ,simply an output can be
set as zero or one.
Adder units are connected using GDI passing cell, as shown in Fig 3. 12 bit ripple
carry adder has been connected with GDI passing cells for the selection of accurate or
inaccurate mode.It selects carry-in from either the previous stage adder units or from a
fixed user defined carry inputs. For the accurate mode of operation NP input is set as 10
or 01 ,The entire GDI cell acts as a wire and passes previous carry values.Where as for
inaccurate mode ,NP input is set as 00 for passing zero as carry value or 11 for passing
one as a carry value. If all the GDI cells select the carry from the user defined inputs, the
delay to execute such an N-bit addition will be much smaller than the original delay of N-
bit conventional full adder,with some accuracy loss. Consequently, by setting the control
signals of GDI cells,the normal N bit addition can be executed without any accuracy
loss.
Table 3 Design parameters obtained for adder structures under different bit sizes
Structure Mode Bit Area POWER DELAY
size
RCA ------ 8 344.95 162.12uw 0.79
16 689.76 337.72uw 1.61
32 1379.39 682.91uw 3.22
64 2758.73 1370uw 6.48
RAP-CLA APPROX=0 8 328.46 17.52uw 0.57
16 656.79 348.23uw 1.08
32 1296.05 695.33uw 2.10
64 2630.37 1410uw 4.16
APPROX=1 8 401.97 196.85uw 0.58
16 803.79 399.69uw 0.58
32 1553.31 785.25uw 0.59
64 3184.38 1600uw 0.61
PROPOSED APPROX=0 8 344.94 162.12uw 0.79
16 689.76 337.72uw 1.61
32 1379.39 682.91uw 3.22
64 2737.67 1360uw 6.11
APPROX=1 8 302.62 130.50uw 0.34
16 562.76 241.94uw 0.34
32 1088.59 460.86uw 0.35
64 2123.73 892.13uw 0.34
5 Conclusion
In this work, a reconfigurable approximate ripple carry adder has been suggested
for high speed application. The adder has greater advantage of swapping between the
inexact and exact working modes by making it suitable for both error-tolerant and exact
applications. The structure is similar to conventional adder with some modification by
adding GDI cell. To evaluate the effectiveness of the proposed structure, its design
parameters have been compared to those of some conventional and reconfigurable
approximate adders. The parameters ,which include area, delay and power are evaluated
by using synopsis complier. The results have shown up to 23 %, 34% and 95 % reduction
in area,power consumption and delay, respectively
Conflict of interest: None
6 References