Issue link: https://resources.mouser.com/i/1481902
Intelligence at the Edge 28 Abstract Designers are increasingly moving Artificial Intelligence (AI) from the center of their designs to the edge for a few important reasons. Reducing the current consumption, increasing the security of data transmission, lowering the latency, and streamlining algorithm implementation are some of the advantages of migrating the AI to the edge. Sensors play an essential role in developing the latest AI solutions at the edge in various applications that require real-time data from the environment and equipment. Thanks to the continuous advances and improvements in sensor technology and performance, the latest sensors offer new levels of capabilities for using AI at the edge of the edge, which means adding intelligence to the silicon die of the sensor. Machine Learning (ML) and AI at the edge encompass several implementation techniques based on the context and target applications. Depending on the complexity of an AI algorithm, the solution can be implemented in a processing unit that is part of the sensor node subsystem or integrated into the sensor chip. Intelligent Sensors at the Edge of the Edge Jay Esfandyari, Ph.D., STMicroelectronics Introduction There have been continuous improvements and embedded innovations in MEMS-based sensors in the last two decades. Embedded FIFO (First-In-First-Out), free- fall and wake-up interrupt generators, high resolution, low noise, high bias stability, an always-on feature, and ultra-low current consumption are just some of the improvements over the last few years. The first series of smart features embedded in MEMS sensors started with on-chip functions, including step detector, step counter, tilt, and significant motion detection. However, the true intelligence in sensors started with compact and powerful Finite State Machines (FSMs) and a Machine Learning Core (MLC). The idea behind adding this first step of intelligence from an external computing unit into the sensor module was to simplify the implementation of applications, lower the current consumption required to run the application, and to reserve the computing budget of the external unit for more complex tasks. Also, the embedded intelligence This article will discuss the capabilities and the advantages of moving AI into the sensor. We'll also discuss some of the AI applications that can be implemented in sensors.