Armv9: Elevating Security and AI Performance

Arm has unveiled Armv9, its first new architecture in a decade, designed to fortify security with a confidential compute architecture (CCA) and accelerate AI workloads through a scalable vector extension (SVE) upgrade.

Armv9‑A builds on Armv8‑A, delivering a suite of enhancements that will roll out over the coming years. The roadmap introduces CCA—developed in partnership with Microsoft—to isolate code and data within hardware‑based secure realms, even from privileged software.

CCA’s Realms provide a separate, protected region for applications, shielding commercially sensitive information during use, storage, and transmission. A recent Pulse survey found that over 90% of enterprise leaders believe confidential computing can lower security costs and unlock greater engineering investment.

Microsoft’s Henry Sanders explains, “The complexity of edge-to-cloud use cases demands more than a one‑size‑fits‑all approach.” Sanders, corporate vice president and CTO for Azure Edge and platforms, added, “Heterogeneous compute is becoming ubiquitous, requiring deeper synergy between hardware and software. ArmV9’s confidential compute features, forged in close collaboration with Microsoft, accelerate this ecosystem and foster open innovation on an architecture powering billions of devices.”

CCA extends Arm TrustZone’s system‑wide isolation, ensuring data remains encrypted while in transit, at rest, and during processing—even for virtualized CPUs sharing physical cores.

Arm also introduces Memory Tagging Extension (MTE) in first‑generation Armv9‑A processors. MTE protects against memory‑corruption attacks by tagging data and enforcing lock‑and‑key access through pointers, a critical safeguard as many high‑profile breaches stem from memory exploits.

SVE2 – Powered by a Partnership with Fujitsu

AI demand is soaring, with Statista estimating over eight billion AI‑enabled voice devices by the mid‑2020s and 90%+ of on‑device apps incorporating AI features. To meet this, Arm collaborated with Fujitsu to evolve the scalable vector extension (SVE), foundational to the world’s fastest supercomputer, Fugaku.

Armv9’s SVE2 lets designers choose vector lengths in 128‑bit increments up to 2048 bits, dramatically boosting parallel compute. This flexibility supports high‑performance computing, digital signal processing, and emerging XR workloads—from genomics to computer vision.

SVE2 enhances 5G, VR/AR, and on‑CPU machine learning, while Arm plans further AI‑centric matrix‑multiplication improvements in CPUs, Mali GPUs, and Ethos NPUs over the next few years.

Simon Segars, Arm CEO, notes, “As Armv9 matures, v9‑M profiles will unlock new application domains. We’re equipping developers with advanced tools and a model‑based platform to eliminate lag between hardware release and software readiness.”

Arm has launched virtual prototypes supporting SVE2 and integrated compiler support into GCC and LLVM. C++ intrinsics and auto‑vectorization tools ease rapid porting of workloads to virtual v9 targets.

Richard Grisenthwaite, SVP and chief architect, states, “The need for secure, specialized processing fuels AI demand. Armv9 bridges hardware and software, enabling trusted compute platforms that balance rapid time‑to‑market with cost control and custom solutions.”

Partner quotes highlight Armv9’s impact: Cadence highlights performance and security gains; Google sees MTE as vital for mitigating memory vulnerabilities across connected devices; Samsung underscores security and ML improvements for its next‑gen Exynos processors; Synopsys points to new possibilities in 3D IC, ML, and smart process monitoring; VMware notes enhanced security and performance for hybrid cloud via Arm‑based SmartNICs in Project Monterey.

Armv9 positions the industry for secure, AI‑driven innovation across cloud, edge, and mobile ecosystems.