Approximate Arithmetic Circuit Architectures for FPGA-based Systems

This book presents novel architectures for FPGA-optimized accurate and approximate operators, their accuracy and performance analysis, and application-level analysis for error-resilient applications. It focuses on multiplication, which is computationally expensive and commonly used in error-resilient applications.

Format: Hardback
Length: 176 pages
Publication date: 28 February 2023
Publisher: Springer International Publishing AG

This book presents a comprehensive exploration of innovative architectures for FPGA-optimized accurate and approximate operators, encompassing their intricate accuracy and performance analysis, diverse techniques for modeling approximate operator behavior, and thorough application-level evaluation to assess the impact of approximations on final output quality and performance metrics. Given the prevalence of multiplication as a computationally intensive operation in diverse error-resilient applications like digital signal and image processing, machine learning algorithms, and more, this book specifically concentrates on this operation. The book commences by delving into the diverse sources of error resilience and potential avenues for approximations across various layers of the computation stack. It then presents a detailed overview of the state-of-the-art approximate computing-related works, highlighting their limitations and gaps.

Chapter 1: Introduction
This chapter provides an introduction to the book, outlining its objectives and scope. It highlights the significance of FPGA-optimized accurate and approximate operators in error-resilient applications and the need for efficient and effective methods to model and evaluate their behavior.

Chapter 2: Error Resilience and Approximations
In this chapter, the authors delve into the concept of error resilience and its importance in various applications. They discuss the various sources of errors and their impact on the final output quality. The chapter also explores the opportunities available for approximations on different layers of the computation stack, including the logic, arithmetic, and memory subsystems.

Chapter 3: Approximate Computing Techniques
This chapter presents a comprehensive overview of the various approximate computing techniques, including rounding, truncation, quantization, and interpolation. The authors discuss the principles behind each technique, their advantages and disadvantages, and their suitability for different applications. They also provide examples and case studies to illustrate the practical application of these techniques.

Chapter 4: Modeling Approximate Operators
In this chapter, the authors explore the techniques for modeling approximate operators, including finite difference methods, Taylor series expansions, and neural networks. They discuss the challenges associated with modeling these operators and the parameters that need to be optimized to achieve accurate results. The chapter also provides examples and case studies to demonstrate the effectiveness of these modeling techniques.

Chapter 5: Application-Level Analysis
In this chapter, the authors conduct a thorough application-level analysis to evaluate the impact of approximations on the final output quality and performance metrics. They discuss the trade-offs between accuracy and speed, the selection of appropriate approximations, and the design of efficient FPGA-based implementations. The chapter also provides examples and case studies to illustrate the practical application of these analysis techniques.

Chapter 6: Conclusion
In this chapter, the authors summarize the key findings and conclusions of the book. They summarize the contributions made by the authors and highlight the potential applications and future research directions in the field of FPGA-optimized accurate and approximate operators.

Weight: 465g
Dimension: 235 x 155 (mm)
ISBN-13: 9783031212932
Edition number: 1st ed. 2023