Arm launched its first major chip architecture upgrade in a decade on Tuesday called Armv9 to first appear in mobile devices in 2022 after sampling later this year.
It has taken 30 years for Arm-designed chips to reach 180 billion devices. Arm estimated it will take only a decade to reach devices with the next 300 billion Arm-based chips. Arm chip designs are used by 1,000 partners that create their own chips for a wide variety of computers from smartphones to high performance servers running in the cloud.
The new v9 architecture will have the capability to move into specialized compute functions involving AI and IoT with a strong emphasis on security and 30% faster processing. Versions are expected yearly for the next decade.
To bolster security, Armv9 includes a Confidential Compute Architecture that protects portions of code and data from access or modification while it is being used, even from authorized software, by putting computation into hardware-based security.
CCA works on a concept called Realms, similar to containers, that work in a region separate from secure and non-secure environments. That means Realms protects sensitive data and code used in healthcare or financial settings from the rest of a system whether it is in-use, at rest or in transit.
A recent Pulse and Arm survey of 250 engineering and security executives found that 91% believe that Confidential Computing could reduce the cost of security and enable them to invest more in engineering innovation.
Arm said it also created Scalable Vector Extension 2 technology for Armv9 to enable advanced Machine Learning and Digital Signal Processing across a wider array of applications. The original SVE technology was developed with Fujitsu and is used in Fugaku, the world’s fastest supercomputer.
SVE2 is designed to enhance processing in 5G, virtual and augmented reality, and ML work running in CPUs, including image processing and smart home apps.
To support the significance of Armv9, Arm tapped 20 partners that already work with Arm, including Nvidia, to endorse the move. Nvidia is also in the process of making a $40 billion purchase of Arm from Softbank, first announced last September.
Nvidia said in a statement that Armv9 will deliver a wide range of computing possibilities as AI moves deeper into gaming, autonomous vehicles, enterprise data centers and embedded devices.
The increased performance of Armv9 and the AI extension will make Arm more attractive to Nvidia as the acquisition gets reviewed by government entities, said Jack Gold, an analyst at J. Gold Associates. However, Gold believes Nvidia will not be able to surmount regulatory hurdles to ultimately consummate the deal.
Both AI enhancements and Confidential Computing in v9 are important enhancements, Gold said. “V9 chips will enable more companies to create AI and ML apps,” Gold said. “This won’t replace high end Nvidia and other AI systems, but it will enable many mobile and IoT devices to include AI and ML algorithms that will enhance usability and features.”
Bob O’Donnell, president of Technalysis Research, said Armv9 is a recognition of the need for more custom chips for a variety of jobs. “Arm is setting out a big vision for a whole new level of devices that require faster compute, while also laying the groundwork for a much more secure means of compute and extending the range of AI and ML,” he said.
Arm also recognizes the need to provide ML capabilities across CPUs, GPUs and NeuroProcessing Units (NPUs), O’Donnell added.
In an online interview with reporters, Arm executives said the changes for partners building chips with Armv9 will be sweeping but manageable, as silicon makers adopt tightened security and AI capabilities. In one example, operating system changes will be required “all through the stack” to accommodate the Realm Manager for 64-bit systems, said Peter Greenhalgh, Arm vice president of technology.
However, “the nature of the OS changes is not so drastic that you throw everything away,” CEO Simon Segars added. “It evolves. There’s work to go through, but the lift to adopt these features is not huge or requires disproportionately large investment in engineering capability or development. This is part of a continuum and people want enhanced security. This is just going to be one of those things. It’s not exceptional in any way from the cost point of view.”
Segars added that the cost of overhead for new silicon to implement Armv9 depends on the scalability and range of implementations made by chipmakers. “It is really up to partners how much they spend on pushing performance,” he added.
Patrick Moorhead, an analyst at Moor Insights and Strategy, said Arm is making it easier for chip designers to integrate ML into their chips. However, he added, “chip designers will still need to architect something performant.”
The security changes in v9 are “dramatically improving,” and can’t arrive soon enough, Moorhead added. “If we had these technologies fully enabled today, it could ward off most of the known attacks.”