Abstract: Artificial Intelligence (AI) is experiencing a paradigm shift. AI began with large, computationally and energy-intensive server farms doing training and then moved to the cloud to distribute inference. But in the next generation, the compute resources needed for real-time, low-power, localized analysis and decision-making is shifting to the edge. This is enabled by a wide variety of new semiconductor technologies including domain-specific hardware accelerators, custom chips built for efficiency and novel architectures leveraging Many Integrated Core, chiplets, neural processing units, and in-memory compute architectures. The emergent capabilities make possible new services and use cases in the wireless networks, the devices in relation to the networks, and the applications that run on the devices. However, these use cases introduce exponentially higher demands for throughput, latency, training, and power efficiency compared to previous generations. To be effective, semiconductors must support energy efficient low latency inference processing, decentralization of training workloads, and proper device and network integration that recognize and have solutions for the device to data, data to model, and model to data transitions.
This document presents several focus topics spanning wireless networks, chip technology for devices, systems and applications for edge AI. These components themselves are but a subset of the next generation challenges, but they provide a good roadmap to drive focus and prioritization. Thematically as AI transitions to the edge, what does that mean in terms of AI services and workloads? What technological capabilities and requirements drive progress in use cases? What devices need be created? And how do we innovate and create solutions in a cooperative way at the pace required to realize these progress and capabilities?
Keywords : Global semiconductor market, growth forecast, chip industry expansion, integrated circuits, AI chips, 5G semiconductors, automotive electronics, consumer electronics, industrial IoT, advanced packaging, silicon wafer demand, fabrication capacity, semiconductor manufacturing, chip shortage recovery, global supply chain, foundry services, semiconductor innovation, R&D investment, chip exports, emerging markets, leading semiconductor companies, technological advancement, market dynamics, semiconductor trends, microelectronics, global demand drivers.
|
DOI:
10.17148/IJARCCE.2022.111258
