After proving its strength in the cloud

At the beginning of this century, with the progress of network and Internet technology, the Internet of Things (IoT) came into being and rose rapidly, which greatly promoted the popularity of edge computing devices with stronger processing power and connectivity, making data processing closer to the data source. With the development of ML and AI technology, smart devices can not only perform tasks, but also learn and adapt.

"Recently, with the development of Transformer and large models, AI model universal, multi-modal support, and model fine-tuning efficiency have made qualitative breakthroughs, coupled with low-power AI accelerators and dedicated chips integrated into end devices, edge intelligence is becoming more and more autonomous and powerful.""In subsequent development, the system has become more powerful and its complexity has increased. Software and hardware must work together to unlock the full potential of AI processing."

This is exactly what has been working on in AI for the past few years.

Deep farming edge AI, more than 10 years
Edge AI is not a recent phenomenon; it was implemented five to ten years ago in applications such as simple voice assistants, as well as in applications such as the more familiar intelligent vision system. Most of these edge AI deployments are based on -based AI computing platforms.
Has also continued to invest in edge AI over the past decade. In addition to optimizing CPU processors, including Cortex-M and Cortex-A, to enhance support for AI and machine learning (ML) on the edge side, the company has also introduced the Helium vector instruction set for the Cortex-M. Delivers significant performance improvements for digital signal processing (DSP) and ML for embedded devices.
The important features of edge AI are bandwidth savings, more security, reduced data transmission, improved response speed and reliability, but it also faces some design challenges, that is, there are relatively strict restrictions on energy efficiency and cost. And that's what I've been good at and focused on for years.

Edge AI accelerators have been developed over the years to meet the growing demand for inference workloads on both edge and end sides. Two previous successful NPU products, Ethos-U55 and Ethos-U65, bring high-performance, energy-efficient solutions for edgeside and end-side AI applications.
Ethos-U55 is typically deployed in heterogeneous systems based on Cortex-M. The Ethos-U65 extends the applicability of the Ethos-U family to Cortex-A based systems and brings a two-fold increase in performance to on-device machine learning (ML) capabilities. Both products offer a unified tool chain that simplifies development and supports common ML neural network operations, including convolutional neural networks (CNNS) and recurrent neural networks (RNNS).
Keep up with The Times when it comes to safety concerns. Firstly, the TrustZone technology is introduced into the processor to ensure that the processor is more secure and better protects the privacy of user information. At the same time, security also requires a concerted effort across the entire ecosystem
PSA is a platform security architecture that has evolved over the past few years and is Certified by the ecosystem. To date, more than 200 products have been Certified by PSA. Going forward, we will continue to enhance safety in our products and leverage the power of the entire ecology, with more products Certified by PSA, which will make it easier for more partners to pass regional or global safety standards compliance requirements.
It is also working with a range of software algorithm and tool partners to ensure that edge AI system developers are provided with the tools and support they need. Only with enough technical breadth and experience can the entire edge computing ecosystem be strongly supported to seize the AI opportunity.
But not satisfied with this, in recent days brought a new generation of NPU accelerator.