Embedded edge devices are getting smarter and more efficient

July 3, 2024

Plot

Edge is nothing new. Edge networking, Edge computing, and more recently Edge Intelligence are areas where many companies have been innovating and expanding capabilities and product lines for years.

There are many reasons, and it’s not the sole home of innovation, but there’s a convergence of historically separate horizontal layers that are driving a significant portion of innovation at the edge. These layers are IoT, Embedded, and AI, and they’re coming together in a single, connected horizontal technical layer that will likely inform every deployment or update at the edge for the foreseeable future. It’s simply because enterprises and end users want their applications to be unified and simple to use and manage, but also powerful, functional, and efficient. That means they need embedded processing power and efficient power management, Sensor Fusion for IoT connectivity and data collection, and automation of command and control from AI and ML algorithms. All of these things are being required by customers to flow through a single user interface, and that’s what’s driving this horizontal convergence.

Let’s take a look at some of the new edge innovations that are contributing to this evolution.

Intelligence

Ceva recently announced the expansion of its Intelligent Edge IP with new TinyML-optimized NPUs for AIoT devices. The Ceva-NeuPro-Nano NPUs are designed to be ultra-low-power and deliver high performance in a small, edge-friendly footprint for consumer, industrial, and enterprise products, the company said.

The TinyML market is growing with horizontal technology convergence. ABI Research predicts that by 2030, more than 40 percent of TinyML shipments will be powered by dedicated TinyML hardware rather than general-purpose microcontrollers. Ceva has help as it looks to lead the field, and it’s starting now with the new NeuPro-Nano NPU.

“OEMs are trying to cram more features into SoCs, but they’re running out of compute,” said Chad Lucien, vice president and general manager of Ceva’s sensor and audio business. “We still run into situations where a customer is using an MCU that may not be performing the function.”

The company’s new Embedded AI NPU architecture is said to be fully programmable and execute neural networks, feature extraction, control code, and DSP code, and also supports most advanced data types and machine learning operators, including native transformer computation, scattering acceleration, and fast quantization. This brings intelligence to any edge device without sacrificing power efficiency, as its optimized, self-contained architecture is designed to deliver higher power efficiency with a smaller silicon footprint.

“Ceva-NeuPro-Nano opens up exciting opportunities for companies to integrate TinyML applications with low-power IoT SoCs and MCUs and builds on our strategy to equip intelligent edge devices with advanced connectivity, sensing, and inference capabilities,” said Lucien. “The Ceva-NeuPro-Nano family of NPUs enables more companies to bring AI to the edge, resulting in intelligent IoT devices with advanced feature sets that deliver greater value to our customers.”

Ceva-NeuPro-Nano NPUs are available for licensing today. Click here to learn more.

IoT

The most popular and in-demand IoT application today is vision, whether through LiDAR, motion detection, or most often cameras. With this in mind, STMicroelectronics has unveiled a new image sensor ecosystem for advanced camera performance called ST BrightSense.

As recently announced, BrightSense technology is designed to enable faster, smarter design of compact, energy-efficient products for applications in factory automation, robotics, augmented reality (AR), virtual reality (VR), and medicine – applications that are arguably the most important use cases for edge (non-consumer) devices.

The company also released a set of plug-and-play hardware kits, evaluation camera modules, and software at the same time, reportedly to facilitate development with BrightSense global shutter image sensors. BrightSense image sensors sample all pixels at once, ST said, meaning the global shutter sensors can capture images of fast-moving objects without distortion and reduce power when paired with a lighting system, unlike a conventional rolling shutter.

ST said its CMOS sensors enable backside-illuminated pixel technology and high image sharpness to capture fine details, even in motion. Applications include barcode reading, obstacle avoidance in mobile robots, and facial recognition. Since form factor is always a key issue at the edge, the company has adopted a 3D-composition design to enable a small sensor area, making it easier to integrate in places where space is limited.

Click here to learn more.

Set

Computing and power management are at the heart of embedded systems and are the engines that drive both IoT and AI at the edge. Sticking with vision, robotics in manufacturing and warehousing, and ADAS and other automated drive systems, vision systems are becoming critical infrastructure, and the processing needs are incredible, while form factors need to be compact and rugged for these mobile edge devices operating in sometimes intense environments.

One of the latest innovations in this area comes from independent company Semiconductor, which recently announced the iND880xx product line, designed specifically to address the unique requirements of advanced driver assistance systems (ADAS) and vision sensing applications.

The company said it has noticed that initialization and processing delays are hampering ADAS camera systems and preventing them from achieving real-time safety capabilities. That’s why Indie Semiconductor says its iND880xx family is built specifically to combat latency with proprietary technology that reportedly supports low-latency processing of four independent sensor inputs simultaneously, which can provide throughput of up to 1,400 megapixels per second.

Check here.

Of course, these are not the only examples, just some of the latest. It is important for engineers and developers to consider all three pillars of IoT, AI, and Embedded when designing new products or improving existing lines. It is becoming increasingly clear that keeping these layers in silos is going the way of the dodo and Blockbuster. It is time to think about how to get to Blu-ray.