Recently, the mass media has been presenting a variety of interesting solutions in various fields. Among them, artificial intelligence (AI) is frequently mentioned as the most advanced and cutting-edge technology.

In fact, AI is a concept that was conceived as early as the 1950s, and any technology that enables computers to mimic human behavior is AI. This concept is very general and covers a wide range of research and solution areas.

A subfield of AI is machine learning, which is a set of algorithms and methodologies that improve over time by learning from data.

Moreover, its subfield, deep learning, is the most complex advanced AI algorithm because it uses not only data but also a neural network-like structure imitating the human brain to make decisions.

On the other hand, defining these algorithms is complex, as there are too many variables and creating an effective application model is extremely difficult. This is where machine learning opens up new opportunities.

Currently, AI technology is being used in various fields. Self-driving cars are the most well-known implementation of AI, but the applications of AI are endless in every sector that collects data: smart homes that detect intrusions using impact measurement algorithms, smart industries that implement predictive maintenance, smart buildings that adjust their environments based on predicted and actual occupancy to save energy, and advertising engines that analyze customer preferences, interests, and behaviors.

But this AI also has limitations.

For a centralized approach that uses AI algorithms running in the cloud, data is fed from sensors to the cloud without intermediate analysis.

This raises security concerns because data is transmitted across multiple devices, gateways, networks, and companies.

Responsiveness is also poor because the latency from node to cloud is relatively long.

Efficiency issues also arise because sensor nodes have high bandwidth and consume significant amounts of power.

The solution to overcome this limitation is ‘Deep Edge.’

Deep Edge decentralizes computing and services from the cloud to the network edge.

Increasingly, intelligence is being embedded within the host microcontroller or sensor nodes themselves, enabling them to make decisions and become more responsive. This has the following benefits:

First, it consumes less bandwidth. Since all raw data is not transmitted directly to the cloud, bandwidth requirements are significantly reduced, maximizing efficiency.

Second, information protection is easy. Data does not necessarily have to pass through other networks and is processed locally, ensuring security.

Third, latency is reduced. Real-time communication without centralized communication shines in critical situations where every second counts. This responsiveness is crucial for the successful execution of real-time applications.

Fourth, the current consumption of sensors or sensor nodes can be locally adjusted. This feature is particularly useful in battery-powered applications.

Global shipments of these deep-edge AI devices are expected to reach 2.5 billion units by 2030.

Deep edge AI technologies are growing rapidly, especially in standalone, low-power, cost-effective embedded solutions.

STMicroelectronics, a leader in deep edge AI, anticipates continued growth in deep edge AI across three segments: condition monitoring, audio and sensing, and computer vision.

On the other hand, the requirements for AI architecture and computing performance used in each sector are very different.

Therefore, practical implementation of AI requires a wide range of solutions and expertise, from smart sensors to key tools and ecosystems.

This deep-edge technology will be a key way to gain an edge in the semiconductor market, where every nanometer is being fought over.

※ This article is a contribution by Matteo Maravita, Director of the AI Competence Center at STMicroelectronics.