International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024
p-ISSN: 2395-0072
www.irjet.net
ADVANCED FPGA DESIGN: HIGH-SPEED, AREA-FFICIENTAPPROXIMATE PARALLEL PREFIX ADDER P. Sudarshan1, P. Subha Sree2, P. Ganesh3, P. Raja Shekar Reddy4 , Mr. N. Samba Murthy5 1,2 ,3,4Student,5 Assistant Professor, Department of Electronics and Communication Engineering, Seshadri Rao
Gudlavalleru Engineering College, Gudlavalleru, Krishna Dt., Andhra Pradesh. ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract—Addition units are important in the field of computational kernels for error-tolerant applications such as signal, image, and video processing and machine learning. They function as independent units as well as essential building blocks for a number of mathematical operations, including division, multiplication, subtraction, comparison, and squaring. Parallel prefix adders are among the quickest adder designs among them. Prefix operators , often referred to as carry operator nodes, make up the parallel prefix graph structure that PPAs utilize. The speed of carry generation and propagation is increased by this design, which maximizes their parallelization.This research integrates approximations inside the POs to introduce approximate PPAs in a novel way. With this method, AxPPAs that provide a balance between accuracy and performance may be created. We specifically introduce four AxPPA architectures: Ladner-Fischer, approximate Brent–Kung, approximate Kogge–Stone, and Sklansky. The performance of these AxPPAs is compared against energy-efficient approximation adders, such as Copy, error-tolerant adder I, lower-part OR adder, and Truncation, in order to evaluate our approach.We tested our AxPPAs in both stand-alone and embedded settings within two important signal processing application kernels: a finite impulse response filter kernel and a sum of squared differences video accelerator kernel. Interestingly, the findings we obtained show that AxPPA-LF presents a new Pareto front with comparable energy-quality performance. Keywords–Computational kernels, Error-Tolerant , Ladner-Fischer, Brent–Kung, Kogge–Stone, Sklansky, LowerPart OR Adder, Truncation.
1.INTRODUCTION Processing systems that are more complex are being housed on VLSI circuits as integration sizes rise. These systems are designed to support signal processing applications that need a large amount of computing power as well as energy. Achieving high performance and optimizing power consumption are the main goals of system-level or circuit-level design. In computational kernels designed for error-tolerant uses such as machine learning, image, video, and signal processing, addition units are essential components. In addition to being independent objects, they are also necessary parts of certain mathematical operations, such as squaring, division, multiplication, subtraction, and comparison. Parallel prefix adders are one of the quickest addition unit designs available.Prefix operators, often referred to as carry operator nodes, make up the parallel prefix graph structure that PPAs utilize. The speed of carry creation and propagation is increased by this architecture, which maximizes their parallelization. In this study, however, we want to go above the limits by including approximations into these prefix operators, presenting a unique idea of approximate PPAs. Appropriate prefix operators can combine approximations with performance to achieve a compromise between accuracy and performance. In particular, we provide the approximation Brent–Kung, Ladner–Fischer, Kogge–Stone, and Sklansky AxPPA designs. We compare the performance of these AxPPAs against energy-efficient approximation adders, such as Copy, error-tolerant adder I, lower-part OR adder, and Truncation, in order to assess the efficacy of our technique.We thoroughly tested our AxPPAs in embedded and standalone contexts, as well as in two key signal processing application kernels: a video accelerator kernel that sums squared differences and a finite impulse response filter kernel. Interestingly, we find that AxPPA-LF exhibits similar energy-quality performance and introduces a new Pareto front. AxPPAs' practical consequences are examined through an assessment that covers both standalone situations and embedded applications within major signal processing kernels. Interestingly, the results highlight how AxPPA-LF can open up new possibilities in energy-quality performance trade-offs and highlight how it might improve computational efficiency in error-tolerant systems.Essentially, this work explores the incorporation of approximations into PPAs, providing a viable path towards improving the effectiveness and performance of addition units in computing kernels that are essential for a range of applications.
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1125