Artificial Intelligence Energy Efficiency in Low Power Applications

Abstract

In the direction of independent on-device AI .By deploying AI to edge devices, on-device AI may power a variety of functions in our daily lives, such as search and rescue with unmanned aerial vehicles, health care in robots, and augmented reality (AR)/mixed reality (XR) glasses (UAVs).However, it can be difficult to implement DL on edge devices and use it in practical applications. Real applications of on-device AI are not possible because the computational and energy costs of model inference are excessively high for edge devices with constrained computing power and battery capacity. Additionally, pre-trained models may not be accurate for new input instances because they cannot dynamically adapt to the real world after being deployed to edge devices. Two projects are carried out in order to achieve effective and adaptive on-device AI. A machine-learning-based analogue circuit regression model offers an alternate propose methodology for dealing with swiftly increasing invent complexity. The more modern technology structures are proposed, such as SOI or FinFET, the more robust calculation engine is needed to meet various design specifications while assuring operative resilience.