DESIGN OF HIGH PERFORMANCE DYNAMICALLY

TRUNCATED APPROXIMATE MULTIPLIER FOR

VLSI APPLICATIONS
1
Mr .k.srinivasRao, 2B.Srilakshmi, 3B.Pushpalatha, 4D.visali, 5S.sumalatha
1
Assistant professor, Department: Electronics and Communication Engineering, DVR & Dr.Hs MIC
College Of Technology, Kanchikacherla, NTR District, Andhra Pradesh
2,3,4,5
Student, Department: Electronics and Communication Engineering, DVR & Dr.Hs MIC College
Of Technology, Kanchikacherla, NTR District, Andhra Pradesh

ABSTRACT
Multipliers serve as integral arithmetic functional units across various applications,
often necessitating numerous multiplications that contribute significantly to power
consumption. To mitigate this challenge, the adoption of approximate multipliers has
emerged as a promising strategy in applications tolerant to errors, offering a trade-off
between accuracy, energy efficiency, and performance. In this study, we present a
novel approach comprising an approximate 4-2 compressor of high accuracy, coupled
with an adjustable approximate multiplier capable of dynamically truncating partial
products to accommodate variable accuracy requirements. Furthermore, we introduce
a straightforward error compensation circuit to minimize error distances. Our
proposed approximate multiplier offers runtime adjustment of accuracy and power
consumption tailored to user specifications. Experimental findings showcase notable
reductions in both delay and average power consumption of the adjustable
approximate multiplier—27% and 40.33% (up to 72%) respectively—compared to
traditional Wallace tree multipliers. Moreover, we illustrate the adaptability and
versatility of our proposed multiplier within convolutional neural networks (CNNs),
demonstrating its efficacy in meeting diverse requirements across different network
layers. This multifaceted approach not only enhances energy efficiency and
performance but also underscores the flexibility and applicability of approximate
multiplication techniques in real-world applications.

INTRODUCTION crucial for arithmetic operations in

various applications ranging from digital
signal processing to machine learning.
In the realm of VLSI design, multipliers
stand as indispensable components,
However, the demand for high-speed the multiplier's performance to their
computation often comes at the expense specific application needs.
of significant power consumption. To
address this challenge, the utilization of Through extensive experimentation and
approximate multipliers has gained analysis, we demonstrate the superior
traction, offering a compelling solution performance of our dynamically
by trading off accuracy for reduced truncated approximate multiplier.
power consumption and improved Comparative results showcase notable
performance. reductions in both delay and average
power consumption, outperforming
In this context, our project focuses on traditional Wallace tree multipliers by
the design and implementation of a significant margins. Moreover, we
novel Dynamically Truncated illustrate the versatility and adaptability
Approximate Multiplier tailored for of our proposed multiplier within the
VLSI applications demanding high context of convolutional neural
performance. Our approach is networks (CNNs), highlighting its
multifaceted, comprising the potential to address diverse requirements
development of an approximate 4-2 across different network layers.
compressor with exceptional accuracy
and the integration of an adjustable By offering a comprehensive solution
approximate multiplier capable of that prioritizes energy efficiency,
dynamically truncating partial products performance, and adaptability, our
to accommodate varying accuracy project aims to contribute to the
requirements. advancement of VLSI design,
facilitating the development of more
Furthermore, we introduce a power-efficient and high-performance
straightforward yet effective error computing systems across various
compensation circuit aimed at application domains.
minimizing error distances, thereby
enhancing the overall accuracy of the
multiplier. The flexibility of our
proposed design allows for real-time
adjustment of accuracy and power
consumption, empowering users to tailor
 architecture compared to traditional
designs, showcasing reductions in delay
and power consumption while
maintaining acceptable levels of
accuracy. The versatility and
adaptability of the proposed multiplier
within convolutional neural networks
(CNNs) will also be explored,
highlighting its potential to address

OVERVIEW diverse computational requirements

The project focuses on the design and across different network layers.

implementation of a Dynamically
Truncated Approximate Multiplier PROBLEM STATEMENT
tailored for VLSI applications requiring In VLSI applications, multipliers play a
high performance. By leveraging critical role in arithmetic operations, but
approximate multiplication techniques, their high-speed operation often results
the project aims to address the trade-off in significant power consumption. The
between computational accuracy and project addresses this challenge by
power efficiency inherent in multiplier proposing a Dynamically Truncated
design. The proposed approach involves Approximate Multiplier, aiming to
the development of an approximate 4-2 balance computational accuracy with
compressor with high accuracy, power efficiency. The key problem lies
integrated with an adjustable in designing a multiplier architecture
approximate multiplier capable of capable of dynamically adjusting its
dynamically truncating partial products accuracy to meet the specific
to meet varying accuracy requirements. requirements of different applications
Additionally, the project introduces an and operating conditions. Additionally,
error compensation circuit to minimize ensuring that the approximate multiplier
error distances and enhance overall maintains acceptable levels of accuracy
accuracy. Through extensive while reducing power consumption
experimentation and analysis, the project poses a significant technical challenge.
seeks to demonstrate the superior The project also seeks to develop
performance of the proposed multiplier effective error compensation techniques
to mitigate inaccuracies introduced by
approximation. Furthermore, integrating
the proposed multiplier architecture
within CNNs presents additional
challenges, including compatibility with
existing network architectures and
efficient utilization across different
layers. Overall, the project aims to
overcome these challenges to provide a
comprehensive solution for achieving
high performance with reduced power
consumption in VLSI applications. CONCLUSION
In conclusion, the Dynamically
Truncated Approximate Multiplier
RESULT designed for VLSI applications
represents a significant advancement in
the realm of arithmetic units and digital
signal processing. By introducing
innovative techniques for approximate
multiplication and dynamic truncation,
the project addresses critical challenges
in power efficiency and computational
accuracy inherent in multiplier design.
Through extensive experimentation and
analysis, the proposed multiplier
architecture demonstrates superior
performance compared to traditional
designs, showcasing reductions in delay
and power consumption while
maintaining acceptable levels of
accuracy. The integration of an error
compensation circuit further enhances
the overall accuracy of the multiplier,
ensuring reliable operation across
diverse application scenarios. integrating the proposed multiplier
Additionally, the versatility and within larger system architectures could
adaptability of the proposed multiplier enhance overall system performance and
within convolutional neural networks efficiency. Collaborative efforts between
(CNNs) underscore its potential to academia and industry stakeholders are
revolutionize computation in various essential to drive further innovation and
domains, from embedded systems to facilitate the adoption of Dynamically
high-performance computing. Truncated Approximate Multipliers in
real-world applications, thereby
FUTURE SCOPE advancing the state-of-the-art in VLSI
design and digital signal processing.

Looking ahead, there are several

promising avenues for further research REFERENCES
and development in the field of
1.B. Moons and M. Verhelst, "DVAS:
Dynamically Truncated Approximate
Dynamic voltage accuracy scaling for
Multipliers for VLSI applications.
increased energy-efficiency in
Future endeavors could focus on
approximate computing", Proc.
optimizing the proposed multiplier
IEEE/ACM Int. Symp. Low Power
architecture to achieve even greater
Electron. Design (ISLPED), pp. 237-
energy efficiency and performance
242, Jul. 2015.
gains. This may involve exploring
2.D. Mohapatra, V. K. Chippa, A.
alternative approximation techniques,
Raghunathan and K. Roy, "Design of
refining dynamic truncation algorithms,
voltage-scalable meta-functions for
and investigating novel error
approximate computing", Proc. Design
compensation methods to further
Autom. Test Eur., pp. 1-6, Mar. 2011.
improve accuracy. Additionally,
3.K. Yin Kyaw, W. Ling Goh and K.
extending the applicability of the
Seng Yeo, "Low-power high-speed
proposed multiplier to other
multiplier for error-tolerant
computational tasks beyond
application", Proc. IEEE Int. Conf.
convolutional neural networks could
Electron Devices Solid-State Circuits
broaden its scope and impact.
(EDSSC), pp. 1-4, Dec. 2010.
Furthermore, exploring hardware-
4.M. de la Guia Solaz, W. Han and R.
software co-design approaches for
Conway, "A flexible low power DSP
with a programmable truncated Conf. Comput. Design (ICCD), pp. 33-
multiplier", IEEE Trans. Circuits Syst., 38, Oct. 2013.
vol. 59, no. 11, pp. 2555-2568, Nov. 11.P. J. Edavoor, S. Raveendran and A.
2012. D. Rahulkar, "Approximate multiplier
5.R. Zendegani, M. Kamal, M. design using novel dual-stage 4:2
Bahadori, A. Afzali-Kusha and M. compressors", IEEE Access, vol. 8, pp.
Pedram, "RoBa multiplier: A rounding- 48337-48351, 2020.
based approximate multiplier for high- 12.F. Sabetzadeh, M. H. Moaiyeri and
speed yet energy-efficient digital signal M. Ahmadinejad, "A majority-based
processing", IEEE Trans. Very Large imprecise multiplier for ultra-efficient
Scale Integr. (VLSI) Syst., vol. 25, no. approximate image
2, pp. 393-401, Feb. 2017. multiplication", IEEE Trans. Circuits
6.C. S. Wallace, "A suggestion for a fast Syst. I Reg. Papers, vol. 66, no. 11, pp.
multiplier", IEEE Trans. Electron. 4200-4208, Nov. 2019.
Comput., vol. EC-13, no. 1, pp. 14-17, 13.A. G. M. Strollo, E. Napoli, D. De
Feb. 1964. Caro, N. Petra and G. D. Meo,
7.A. Weinberger, "4:2 carry-save adder "Comparison and extension of
module", IBM Tech. Discl. Bull., vol. approximate 4–2 compressors for low-
23, no. 8, pp. 3811-3814, 1981. power approximate multipliers", IEEE
8.A. Momeni, J. Han, P. Montuschi and Trans. Circuits Syst. I Reg. Papers, vol.
F. Lombardi, "Design and analysis of 67, no. 9, pp. 3021-3034, Sep. 2020.
approximate compressors for 14.H. Xiao, H. Xu, X. Chen, Y. Wang
multiplication", IEEE Trans. Comput., and Y. Han, "Fast and high-accuracy
vol. 64, no. 4, pp. 984-994, Apr. 2015. approximate MAC unit design for CNN
9.Z. Yang, J. Han and F. Lombardi, computing", IEEE Embedded Syst.
"Approximate compressors for error- Lett., Dec. 2021.
resilient multiplier design", Proc. IEEE 15.O. Akbari, M. Kamal, A. Afzali-
Int. Symp. Defect Fault Tolerance VLSI Kusha and M. Pedram, "Dual-quality
Nanotechnol. Syst. (DFTS), pp. 183- 4:2 compressors for utilizing in dynamic
186, Oct. 2015. accuracy configurable
10.C.-H. Lin and I.-C. Lin, "High multipliers", IEEE Trans. Very Large
accuracy approximate multiplier with Scale Integr. (VLSI) Syst., vol. 25, no.
error correction", Proc. IEEE 31st Int. 4, pp. 1352-1361, Apr. 2017.
16.T. Yang, T. Ukezono and T. Sato, "A South Pacific Design Autom. Conf.
low-power high-speed accuracy- (ASP-DAC), pp. 235-240, Jan. 2020.
controllable approximate multiplier 19.B. Jacob, S. Kligys, B. Chen, M.
design", Proc. 23rd Asia South Pacific Zhu, M. Tang, A. Howard, et al.,
Design Autom. Conf. (ASP-DAC), pp. "Quantization and training of neural
605-610, Jan. 2018. networks for efficient integer-arithmetic-
17.I. Hammad, L. Li, K. El-Sankary and only inference", Proc. IEEE/CVF Conf.
W. M. Snelgrove, "CNN inference using Comput. Vis. Pattern Recognit., pp.
a preprocessing precision controller and 2704-2713, Jun. 2018.
approximate multipliers with various 20.P.-Y. Chen, F.-Y. Gu, Y.-H. Huang
precisions", IEEE Access, vol. 9, pp. and I.-C. Lin, "WRAP: Weight
7220-7232, 2021. RemApping and processing in RRAM-
18.C. Guo, L. Zhang, X. Zhou, W. Qian based neural network accelerators
and C. Zhuo, "A reconfigurable considering thermal effect", Proc.
approximate multiplier for quantized Design Autom. Test Eur. Conf. Exhib.
CNN applications", Proc. 25th Asia (DATE), pp. 1245-1250, Mar. 2022.