NeuPro-M: A Highly Scalable, Heterogeneous and Secure Processor for High-Performance AI/ML in Smart Edge Devices

AI capabilities have become an integral part of many products across a wide range of markets and applications. Increasingly, system developers prefer to implement AI workloads at the edge. This leads to a key challenge: meeting the growing computational demands of AI algorithms with hardware suitable for the edge, where power consumption is constrained. In this presentation, we will introduce the NeuPro-M, CEVA’s latest AI processor architecture, and show how it addresses today’s main AI processing challenges, such as adapting to the latest DNN topologies and reducing risk and design time via a full-featured software development kit. By examining some of the NeuPro-M’s heterogeneous processing features, we’ll show how it achieves its exceptional ML performance, high TOPS-per-watt ratio and scalability—which enable it to meet requirements of a diverse set of use-cases for a wide range of end markets including automotive, mobile, consumer, robotics, industrial, telecommunication, defense and IoT.