Embedded chipset vendors are increasing their focus on Neutral Processing Units (NPUs) for Internet of Things (IoT) applications thanks to the architecture’s efficient execution of neural network workloads. NPUs will take an increasing share of overall shipment numbers at the expense of the established Microcontrollers (MCUs) as implementers seek ever greater insights and intelligence at the far edge. According to ABI Research, a global technology intelligence firm, this will contribute to chipset revenues from AI-dedicated silicon for IoT-focused applications reaching over US$7.3 billion by 2030.
“NPUs for TinyML applications in Personal and Work Devices (PWDs) are already well established. However, they are still nascent outside of this device vertical, and major vendors ST Microelectronics, Infineon, and NXP Semiconductors are only just introducing this type of ASIC to their embedded portfolios,” says Paul Schell, Industry Analyst at ABI Research. “By screening PWDs, we provided greater insight into our modeling for IoT applications, which spans 15 verticals, including the most significant, namely Smart Home and Manufacturing.”
On the software side, comprehensive MLOps toolchains are now table stakes for vendors big and small, including start-ups like Syntiant, GreenWaves, Aspinity, and Innatera. As with bigger form factors, the investment into the software offering often matches hardware R&D, which has paid off for vendor Eta Compute in their partnership with NXP to license their Aptos software platform. Such innovations also democratize the deployment of TinyML by reducing the need for in-house data science talent.
Including highly performant architectures like NPUs and some FPGAs into embedded devices will expand the offering of applications able to run on-device from object detection to simple object classification for machine vision use cases, as well as some NLP for audio-based analytics. “Along with the trend in larger edge form factors such as PCs and gateways, this will contribute to AI’s scalability by reducing networking costs and the reliance on cloud. As such, we expect the TinyML market to grow as it capitalizes on these innovations, spurred largely by major industrial sites upgrading their IoT deployments, the growing intelligence of vehicles, and smart home devices.”
These findings are from ABI Research’s Artificial Intelligence and Machine Learning: TinyML market data report.
