Arm intros Armv9 architecture to propel ‘next 300 billion chips’ on wave of AI, IoT, 5G
Arm has unveiled its first new architecture in a decade to raise performance and security in the “next 300 billion Arm-based chips” to be delivered to the market over the coming decade. The firm said the new Armv9 architecture, replacing Armv8, is geared towards the move from general-purpose to “ubiquitous specialized” compute processing, notably as AI, IoT, and 5G gain momentum.
Simon Segars, the company’s chief executive, stated: “The Armv9 architecture signals a new era for our company; a globally-pervasive platform driving secure AI-driven computing that will enable our ecosystem of more than 1,000 partners well into the 2030s. Armv9 will be at the forefront of the next 300 billion Arm-based chips; driven by the demand for increasingly specialized, secure and high performance processing built on the economics, design freedom, and accessibility of general-purpose compute.”
The new Armv9 roadmap brings higher security, faster processing, and smarter analytics into the design of connected devices, the UK-based company said. It suggested it has some responsibility to keep the next generation of wireless computing secure and performant, based on the idea its reference designs make it the chief architect, or draftsman at least, for the incoming (‘fourth’) industrial revolution.
Over 100 billion Arm-based devices have shipped in the last five years, and, at the current run-rate, 100 percent of the world’s shared data will “soon be processed on Arm” – either at the endpoint, in the data networks, or the cloud. “Such pervasiveness conveys a responsibility on Arm to deliver more security and performance,” it said. Security is the “greatest challenge” for computing, it added.
Segars said in a blog post: “Hardware security isn’t a given. A single vulnerability could compromise an entire network, and we face new attempts to exploit Arm technology every day. To get a sense… Symantec detected almost 19 million attacks on its IoT honeypots in the first quarter of 2020. That’s a rate of more than 100 attacks per second, 13 percent higher than we saw towards the end of 2019.”
Among the innovations, Arm has introduced ‘confidential computing’, via the new Confidential Compute Architecture (CCA) in Armv9, to shield code and data while in-use. Where the device operating system (OS) traditionally has the highest authority, the CCA mechanics change that, so while the OS still decides what runs, and when, the applications are moved into a separate hardware-protected area of memory, isolated from the rest of the system.
The new Realms feature protects sensitive data and code from the rest of the system while in-use, at rest, and in transit; the new Arm Memory Tagging Extension (MTE) enables developers to lock strings of data using a ‘tag’, to stop hackers overwriting with malicious code. Data can then only be accessed with the right key. “Implementing lock-and-key access is a huge step in securing not only the code but the data it processes,” writes Segars.
Meanwhile, for higher-performance processing, for ultra low-latency AI applications, the new Armv9 architecture also features a new second-generation Scalable Vector Extension (SVE2) technology, developed with Fujitsu, for enhanced machine learning (ML) and digital signal processing (DSP) capabilities across a wider range of applications. SVE2 is being used in the Fujitsu A64FX chip in Fujitsu’s (“world’s fastest”) Fugaku supercomputer.
SVE2 enhances the processing ability of 5G systems, virtual and augmented reality (presented as ‘XR’), and ML workloads running locally on CPUs, such as image processing and smart home applications, said Arm. “Over the next few years, Arm will further extend the AI capabilities of its technology with substantial enhancements in matrix multiplication within the CPU, in addition to ongoing AI innovations in its Mali GPUs and Ethos NPUs,” it said.
The SVE2 upgrade will allow chip designers to choose a vector length in multiples of 128, up to 2048 bits. “That’s an enormous amount of parallel compute, and while SVE was initially developed for the HPC space, SVE2 in Armv9 extends SVE support for a range of specialized DSP and XR workloads, from genomics to computer vision,” said Segars. But he said much of the most interesting AI work is going on in the low-power IoT sensor space, with Arm’s low-power Cortex-M processors.
Segars stated: “By applying ‘total compute’ design principles across its entire IP portfolio of automotive, client, infrastructure and IoT solutions, Armv9 system-level technologies will span the entire IP solution, as well as improving individual IP. Additionally, Arm is developing several technologies to increase frequency, bandwidth, and cache size, and reduce memory latency to maximize the performance of Armv9-based CPUs.”