Blending DSP and ML Features Into a Low-Power General-Purpose Processor ? How Far Can We Go?
July 29, 2020
New technology from Arm is blending DSP and ML features into a low-power general-purpose processor for the first time. What are the challenges? How far can we go?
With increasing signal processing requirements in various types of embedded systems, some companies designed chips that combine both a digital signal processor (DSP) and a general purpose processor to address the processing demands. While this suits some high-performance devices where silicon area and power are less of a concern, such devices could be difficult to program and can be power hungry. To address the challenge, Arm has been working on technologies that boost the signal processing and machine learning capabilities in the Arm Cortex-M55 processor. In this white paper we look at how Arm Helium technology compares to features found on traditional DSPs, and some of the fundamental differences between VLIW (Very Long Instruction Word) architecture and the Helium approach to the Cortex-M55 processor’s design. We will also explore how the processing requirements affect the processor’s level-one memory system design and the overall performance benefit of the Helium technology.