Advanced Design and Fabrication Techniques for Nanoscale Semiconductor Devices in Next-Generation Electronics
Abstract
The advent of smart edges and increased demands in real-time data processing systems has made energy-efficient semiconductor architectures a necessity in Edge AI and Internet of Things (IoT) end-to-end (edge-to-edge) applications. The systems are run in resource limited applications where power efficiency, speed and accuracy must concur. The optimization strategies addressed in this study are used to improve the efficiency of the hardware that carries out AI inference at the edge. Conventional techniques which include quantization, de-pruning and model distillation are examined and they are shown to be limited in terms of scalability and flexibility. To mitigate them, a new architecture is presented that integrates dynamic voltage and frequency scaling (DVFS) technologies and special AI processors and near memory computing designed together to minimize the overhead of moving data. The reduction of energy use by 35 percent and increment in computational speed by 28 percent was recorded in the results of the experiments whereas precision was retained at the same level as that of traditional systems. These advantages facilitate real-time AI capabilities in a wearable, surveillance or autonomous node devices. The suggested solution will allow incorporating high-end artificial intelligence characteristics into low-power applications to retain performance levels. It develops the framework to create sustainable and high-performance edge AI/IoT systems, which fit the current requirements of intelligent energy-aware computing.