Arm released their latest architectures today, which are cortex-m55, the latest member of Cortex-M series for embedded market, and ethos-u55, the supporting micro AI acceleration processing unit architecture.

Nowadays, the practical application of machine learning is no longer a new thing. It has begun to penetrate into our life and plays an increasingly important role in various industries and fields. Arm sees the future of AI terminal market, and it is likely that it will be the next explosive growth field. The concept of the new cortex-m55 architecture and its supporting ethos-u55 micro NPU architecture is to provide better AI processing capacity for devices, and they will become a powerful promoter of AI technology for terminal devices.

Arm releases new NPU architecture to help the development of embedded machine learning

Cortex-m55 is the first processor core architecture based on ARM helium technology. This technology is actually a new vector execution unit designed for Cortex-M series kernel. Cortex-M series kernel has the ability to execute SIMD instructions for the first time. This makes its digital signal processing capacity increased by five times, and the performance of machine learning increased by 15 times, which is very exaggerated. In addition, because machine learning has a high demand for memory bandwidth, M55 is also optimized in memory.

Let’s take another look at arm’s first micro NPU architecture, ethos-u55, which is an AI processor architecture for low-power areas. It needs to be used with Cortex-M series microprocessors. The size of its processing unit is also scalable, with a minimum of 32 MAC engines and a maximum of 256 MAC engines. Arm did not give more specific design details, but we can know that its design points are control area and power consumption.

Using this new architecture combination can greatly improve the performance of AI, and also fill the vacancy of the original arm product line in machine learning. The new cortex-m55 and ethos-u55 will play an important role in various embedded systems, and they will help AI applications enter more terminal devices.


Leave a Reply

Your email address will not be published. Required fields are marked *