Arm unveils its latest v9 architecture as a solution for the AI- and security-focused future
Arm projects that there will be more than 8 billion AI-enabled voice-assisted devices in use on Earth by about 2025, geared towards voice- or display-based interfaces in 90% or more of their use-cases. However, at the same time, the often-vital chip-design firm also sees security as another rising priority, particularly for enterprise-grade users.
Therefore, Arm's CEO, Steve Segars, has touted the new v9 architecture as capable of informing next-gen DSPs, GPUs or NPUs as it is of ensuring data integrity. It will do the latter by allowing a given AI to use the new update's Confidential Compute Architecture (CCA) to create "Realms". Segars likened these to containers during their presentation, albeit ones theoretically better based on the hardware in question.
CCA Realms can apparently be created as needed in order to ring-fence valuable and sensitive corporate data, regardless of whether it is being actively transferred or is in long-term storage. On the other hand, Armv9 is also now slated to be powerful, flexible and versatile enough for applications more like ML.
It will apparently achieve this through its new SVE2, derived from the same scalable vector extension technology developed by Arm in conjunction with Fujitsu for that OEM's Fugaku supercomputer. This new generation is also to be applied to other demanding tasks such as the deployment of 5G, virtual reality and augmented reality under Armv9.
Therefore, its maker asserts that this new architecture is ready for the next wave (an estimated 300 billion) of AI-devoted chips based on ARM technology. They should start to emerge from 2022 on, and will be produced or taken up by partners including Xiaomi, VMWare, TSMC, Synopys, Samsung, and (obviously) NVIDIA.