Machine Learning and Non-volatile Memories

This book explores the intersection of machine learning and non-volatile memory, discussing how the latter can help solve storage problems. It covers two main application fields: analog neural networks and solid-state drives, highlighting the challenges of implementing neural networks in digital hardware and the benefits of using non-volatile memories for analog implementation. The book also discusses machine learning techniques for optimizing SSD performance and pro-active reliability management.

Format: Hardback
Length: 161 pages
Publication date: 26 May 2022
Publisher: Springer International Publishing AG

This book delves into the fundamentals of both non-volatile flash storage and machine learning, providing a comprehensive exploration of the storage challenges that machine learning can help to address. At first glance, machine learning and non-volatile memories may seem vastly different, with machine learning involving complex mathematics, algorithms, and extensive computational processes, while non-volatile memories are solid-state devices used for storing information, boasting the remarkable ability to retain data even when power is not available. However, this book aims to bridge the gap between these two domains, showcasing how they can collaborate to bring significant value to each other.

In particular, the book covers two primary areas of application: analog neural networks (NNs) and solid-state drives (SSDs). Chapter 1 provides an overview of machine learning, covering the basics of neural networks, their computational requirements, and the challenges they face in mimicking the human brain. Chapter 2 then delves into the specific computation called vector-by-matrix (VbM) multiplication, which is crucial for neural networks but highly power-hungry. In the digital realm, VbM is implemented using logic gates, which dictate both area occupation and power consumption. This combination poses significant challenges to hardware scalability, limiting the size of neural networks, particularly in terms of the number of processable inputs and outputs.

Fortunately, non-volatile memories offer a solution to this problem. Phase-change memories, resistive memories, and 3D flash memories are discussed in Chapters 3, 4, and 5 and Chapter 6, respectively. These memory technologies enable the analog implementation of VbM, known as "neuromorphic architecture," which can outperform its digital counterpart in terms of speed and energy consumption. SSDs and flash memories are closely intertwined, as 3D flash scales, there is a significant amount of work that has to be done to optimize their performance and compatibility.

Overall, this book provides a valuable resource for anyone interested in understanding how machine learning and non-volatile memories can work together to solve complex storage problems. It offers a comprehensive exploration of the fundamentals of both domains, highlighting their strengths and weaknesses, and providing practical insights into their applications. Whether you are a researcher, engineer, or industry professional, this book will help you gain a deeper understanding of the potential of these two technologies and how they can be leveraged to drive innovation and progress in various fields.

Weight: 453g
Dimension: 235 x 155 (mm)
ISBN-13: 9783031038402
Edition number: 1st ed. 2022