Title: AI, Innovation and Global Transformation: Interdisciplinary Perspectives on Technology, Business and Society
Editors: Dr. J. Preetha, and Dr. Siddhartha Mehrotra
ISBN: 978-81-69857-64-2
Chapter: 19
DOI: https://doi.org/10.59646/809/19
Authors: Prof. Pankaj Ramdas Bhusari, Prof. Gaurav Ramkrushna Bhalekar, and Prof. Swapnil Rajesh Kadam
Abstract
Modern artificial intelligence algorithms, characterized by deep neural networks and billion-parameter transformer models, demand compute and memory performance exceeding traditional von Neumann scaling capabilities. As classic Dennard scaling and 2D Moore’s Law face severe physical, quantum, and thermal limits, next-generation semiconductor paradigms must bypass the traditional “memory wall.” This chapter investigates cutting-edge semiconductor technologies for AI-centric compute workloads, focusing on monolithic 3D integration, Gate-All-Around (GAA) nanosheet/forksheet transistors, advanced packaging, and Compute-in-Memory (CiM) using non-volatile memristive elements (RRAM, PCM, FeFET). We present a multi-scale analytical framework combining physical drift-diffusion, quantum transport, compact device modeling, and circuit-level macro simulation. Using a case study benchmarking quantized deep learning inference on a 2nm GAAFET/3D-RRAM heterogeneous macro, our findings demonstrate over a improvement in energy-delay-area product (EDAP) and a
increase in energy efficiency (TOPS/W) compared to leading-edge 5nm FinFET/SRAM, defining a blueprint for post-silicon AI architectures.
Keywords: Semiconductor Technologies, AI Acceleration, Compute-in-Memory, Gate-All-Around FET, Monolithic 3D Integration, Non-Volatile Memory.