Home » AI »
The revolutionary approach of edge AI promises to transform the way embedded systems manage processing and workloads, moving AI from the cloud to the edge of the network.
Artificial Intelligence (AI) has revolutionized many industries, enabling applications that seemed unlikely just a few years ago. At the same time, the exponential growth of data and the need for real-time responses have led to the emergence of a new paradigm: edge AI. This type of technology is essential for the implementation of distributed systems, where data processing must occur as close to the point of origin as possible to minimize delays and improve security and privacy. The revolutionary approach of edge AI promises to transform the way embedded systems manage processing and workloads, moving AI from the cloud to the edge of the network.
Traditionally, AI has relied on the cloud to process large amounts of information, since complex models require significant computational resources that are often not available on edge devices. In a classic architecture, data collected by sensors or other embedded devices is sent directly to the cloud, where it is processed by sophisticated models. The results of this processing are then transmitted back to edge devices to make decisions or perform specific actions. This approach, while effective, also has some important limitations. First, the latency introduced by data transfer between the device and the cloud can be significant, especially in critical applications such as healthcare monitoring or autonomous driving, where every millisecond counts; second, sending data to the cloud raises privacy and security concerns, as sensitive data can be vulnerable during transfer or storage. Edge AI aims to overcome these limitations by bringing processing closer to the source, directly on embedded devices, dramatically reducing latency as data no longer has to travel back and forth between the device and the cloud, and improving privacy and security. Instead of sending large amounts of raw data to the cloud, these systems can process and analyze sensitive data locally without ever leaving the device. According to estimates, global spending on edge computing is expected to exceed $200 billion in 2024, up 15.4% from the previous year. Embedded devices like microcontrollers don’t have the computing power of a data center, but with advances in AI algorithm efficiency and specialized hardware, it’s now possible to run models on these devices. New chips designed specifically for edge AI, such as neural processing units (NPUs) integrated into microcontrollers, are making it increasingly possible to implement models in embedded systems. Edge AI not only reduces latency and improves security, but also has the potential to reduce operating costs. Cloud processing comes with significant costs associated with bandwidth, storage, and computational power. By moving some of the processing to the edge, it’s possible to reduce the load on the cloud and, therefore, costs, which is especially beneficial in applications involving large numbers of distributed devices, such as industrial sensor networks or smart cities, where the cost of sending data to the cloud can become prohibitive. Another area where edge AI is having a significant impact is the Internet of Things (IoT) where millions of interconnected devices collect and transmit data in real time. Edge AI enables these devices to make autonomous decisions without having to rely on the cloud for every single operation. For example, in an environmental monitoring system, sensors can analyze data on-site to detect anomalies or dangerous conditions and send only the relevant information to the cloud for further analysis, which benefits in terms of reducing the volume of data transmitted, but also allowing faster reactions to critical events. The automotive sector is another example where edge AI is making a difference. In autonomous vehicles, processing speed is crucial, and edge AI allows vehicles to process data from sensors, such as cameras and lidars, directly on board without having to send it to the cloud for centralized processing, thus reducing latency and allowing the vehicle to react quickly to unexpected situations. All of this significantly improves the safety and reliability of the system.
In developing edge AI solutions, Broadcom focuses on developing components and infrastructure that enable data processing and analysis directly in the field, rather than sending it to a central data center, an approach essential for applications that require low latency, high responsiveness, and real-time processing, such as the Internet of Things (IoT), intelligent surveillance, robotics, and autonomous vehicles. In particular, Broadcom enables companies to support edge AI workloads by simplifying their deployment and management, and by providing embedded device solutions that integrate improved computational capabilities, such as AI-specific processors and high-performance networking chips. Broadcom devices are optimized to handle large volumes of data generated by sensors in the field, while supporting scalability and energy efficiency. Broadcom stands out for its commitment to technology innovation, with products such as the Jericho3-AI that significantly improve the network infrastructure needed to support edge AI applications.
PCIM Asia 2024 to Strengthen Power Electronics Ties Across Asia
Italian-French technology company STMicroelectronics (ST) develops embedded devices for edge AI, offering solutions that integrate AI capabilities directly into devices. ST develops advanced microcontrollers such as the STM32 series, which include integrated AI accelerators and are designed to run machine learning algorithms directly on the device, enabling fast processing and reducing latency. ST offers complete development platforms that include software tools to train and deploy AI models on embedded devices, such as the STM32Cube.AI library that allows developers to convert trained neural networks into code that can run directly on STM32 microcontrollers.
The ST Edge AI Suite is a set of tools for integrating AI capabilities into embedded systems. It supports STM32 microcontrollers and microprocessors, Stellar automotive microcontrollers, and MEMS smart sensors, and includes resources for data management, optimization, and deployment of AI models.
ST also produces intelligent sensors that integrate advanced processing capabilities that can analyze data in real time, such as speech recognition or vibration monitoring, making them ideal for applications such as predictive maintenance or voice assistance. STMicroelectronics’ approach to integrating AI directly into embedded devices enables a wide range of innovative applications, improving operational efficiency and enabling new capabilities for intelligent devices.
While it is one of the most advanced frontiers of digital transformation today, edge AI also presents some significant challenges. Deploying AI models on resource-constrained devices requires significant optimization, both at the software and hardware levels. Models must be compressed without losing too much accuracy, and hardware must be powerful enough to run these models in real time, but also energy efficient, especially in battery-powered devices.
Additionally, the diversity of embedded devices and their hardware configurations means that there is no one-size-fits-all solution. Developers often need to customize AI models and processing infrastructure to fit the specific needs of the device and application. Another challenge is related to data security and privacy. While edge AI can improve privacy by processing data locally, edge devices are often more vulnerable to attacks than centralized servers, requiring the implementation of robust security measures, such as end-to-end encryption and strong authentication, to protect both the data and the AI models themselves from unauthorized access. Looking ahead, it is clear that edge AI will play an increasingly central role in the world of embedded systems. As the technology evolves, we expect to see an increase in the computational capabilities of embedded devices, making it possible to run increasingly complex models directly at the edge.
Additionally, the emergence of new communications technologies, such as 5G, that offer ultra-low data rates and latency will make edge AI even more efficient and widespread. Overall, edge AI represents a game-changer for embedded systems, providing a solution to the limitations of cloud computing and opening up new possibilities for autonomous and real-time applications. By reducing latency, improving security, and reducing costs, edge AI is quickly becoming a critical component in intelligent devices. While there are still challenges ahead, advances in hardware and algorithms are accelerating its adoption in embedded systems, bringing us one step closer to a future where AI is ubiquitous and seamlessly integrated into our daily lives.
English 
Français