Truth about edge AI in MCUs: is it worth it?

Writen by Junko Yoshida for the Ojo & Yoshida Report – NXP’s approach seeks to address the mismatch between long-lived embedded chips and evolving AI applications.

Despite all the rhetoric about the artificial intelligent of things (AIoT), implementing AI inference in MCUs is still a stretch. Edge AI is “economically constrained as much as… physically constrained,” according to a market analyst. At stake is the future of MCU suppliers and AIoT OEMs. How long are they willing to stay in the game and how much are they investing?

“Edge AI” in the microcontroller unit (MCU) will be on the tip of everyone’s tongue at Embedded World next week.

On the brink of the trade show in Nuremberg, Germany, NXP Semiconductors, the leading MCU supplier, threw its hat into the ring with its first homegrown neural processing unit (NPU). Designed to accelerate inference at the edge, the NPU will go into the MCX N series, a high-end version of NXP’s brand-new family of microcontrollers.

NXP has developed its MCX family— consisting of four series of MCU devices designated N, A, W and L — to address a broad swath of connected edge devices including industrial and Internet of Things (IoT) edge applications.

The pervasiveness of MCUs and sensors in the embedded market is building momentum, with high expectations for the coming era of edge AI. Every leading MCU supplier is preparing to seize the moment.

MCUs come with numerous constraints. As Tom Hackenberg, principal analyst for computing software at Yole Intelligence, pointed out, “Adapting to these resource constraints is a very large challenge for AI applications targeting MCUs.”

Blaming hardware is easy, but that’s not the whole story. Checking off the edge-AI box in an MCU’s list of features is easy. The tall order is doing it effectively — in software development, hardware design and AI processing — without burning a lot of power in a tiny chip… Full article

Source: ojoyoshidareport.com