Achieved a 30% increase in processing performance compared to Armv8.
CCA adds SVE2 technology to enhance security and AI performance.
MediaTek to launch Armv9-based chips in the second half of the year


On the 31st, Arm held an online press conference called 'Arm Vision Day' and announced 'Armv9', a new architecture that responds to market requirements for security, AI, and ubiquitous processing.

“The Armv9 architecture, which is secure, powerful, and well-suited for specialized processing, will play a critical role in the 300 billion Arm-based chips that will ship in the future,” said Simon Segars, CEO of Arm. The previous generation, the Armv8 architecture, was announced in October 2011. It also appears that the remarks were made in consideration of the fact that this was the driving force behind the shipment of more than 100 billion Arm-based chips from 2016 to the present.

Chips based on the Armv9 architecture are expected to be released in the second half of this year by MediaTek and others.

◇ Enhanced security by adding Arm CCA, which is inaccessible during computing.

With Arm-based chips now being installed in countless devices, Arm has focused on security in addition to the 30% performance improvement over Armv8 for the Armv9 architecture.

The newly added 'Arm Confidential Compute Architecture (CCA)' performs computing in a hardware-based secure environment, blocking access and modification of some code or data in the processing state, and even from privileged software.

Arm CCA will introduce the concept of "Realms," dynamically created segments separated from the secure and non-secure worlds, which can also be used by existing applications. In business applications, Realms are protected from all systems, even when sensitive data and code are in use, stored, or transmitted.

◇ Enhanced AI processing performance with SVE2 technology installed on supercomputers.

According to the 'IDC FutureScape' report published in 2020, by 2025, more than 90% of on-device applications are expected to have AI elements, including AI-based interfaces such as vision and voice. To achieve this, Arm developed Scalable Vector Extension (SVE) technology with Fujitsu, which was also adopted into the world's fastest supercomputer, Fugaku.

With the addition of Arm SVE2 technology, the Armv9 architecture enables enhanced machine learning (ML) and digital signal processing (DSP) capabilities across a wider range of applications.

SVE2 enhances 5G, virtual reality (VR), and augmented reality (AR) processing capabilities, as well as ML workloads running locally on the CPU, such as image processing and smart home applications. Arm is currently strengthening AI processing support with Mali GPUs, Ethos NPUs, and other products, and plans to enhance matrix multiplication performance within the CPU.

Furthermore, Arm will establish Total Compute design principles and drive system-level hardware and software optimizations to accelerate overall computing performance and enhance use-case-specific performance. This will enable the system-level Armv9 architecture technology to be extended to all IP solutions.

“We are developing a range of additional technologies to maximize the performance of Armv9-based CPUs by increasing operating frequency, bandwidth, cache size and reducing memory latency,” said Richard Grisenthwaite, senior vice president at Arm.

Meanwhile, an Arm representative stated that Nvidia had no influence on this architecture update or future product roadmap. The merger and acquisition between the two companies is still undergoing approval.