TLDR: Arm has launched its groundbreaking Edge AI Platform, Armv9, designed to bring powerful artificial intelligence capabilities directly to edge devices. This platform, featuring the Cortex-A320 processor and Ethos-U85 NPU, enables on-device processing of AI models with over a billion parameters, enhancing privacy, security, and efficiency for a new generation of smart applications.
Arm has officially introduced its innovative Edge AI Platform, built on the Armv9 architecture, marking a significant stride in bringing advanced artificial intelligence capabilities closer to the data source. This new platform is engineered to facilitate the deployment of sophisticated AI models directly on edge devices, eliminating the constant need for cloud connectivity.
At the core of the Armv9 Edge AI Platform are two powerful components: the highly efficient Arm Cortex-A320 processor and the specialized Arm Ethos-U85 Neural Processing Unit (NPU). This formidable duo is capable of executing AI models containing over one billion parameters directly on the device. This on-device processing capability is a game-changer for applications requiring real-time intelligence, such as smart cameras that can interpret visual information, smart home devices that learn user habits, and robots that interact through vision, voice, and gesture.
Paul Williamson, who leads Arm’s IoT business, emphasized the strategic importance of this development, stating, ‘The next wave of AI innovation will happen at the edge – in the devices, interfaces, and systems that bring intelligence closer to where data is created.’ A key advantage of this approach is the significant enhancement of privacy and security. By processing data locally, sensitive personal information remains on the device, rather than being transmitted to external servers.
Arm President and CEO Takayuki Yokoyama further elaborated on this vision, noting, ‘What we are announcing this time is not cloud-side AI, but edge-side AI. As AI has developed to this extent, it is becoming very difficult to run all AI on the cloud side.’ He stressed the growing importance of running ‘the right AI workload in the right place’ to ensure continued progress in the rapidly evolving AI revolution.
The platform boasts impressive performance improvements. Compared to the Cortex-M85-based platform introduced in 2024, the new Armv9 Edge AI Platform delivers an eight-fold increase in machine learning (ML) performance. The Cortex-A320 CPU, the most compact processor utilizing the Armv9 architecture, leverages features like Scalable Vector Extension (SVE) 2 to achieve a 10x improvement in ML performance and a 30x boost in scalar performance compared to its predecessor, the Cortex-A35. This efficiency also extends to power consumption, with the Cortex-A320 performing the same tasks as the higher-end Cortex-A520 with approximately 50% less power.
Masashi Nakajima, Director of the Applied Technology Department at Arm, highlighted the platform’s ability to handle large parameter models due to its 64-bit architecture, crucial for the accelerating trend of transformer-based generative AI. He also pointed out that the Armv9 architecture significantly enhances security for IoT devices through features like Pointer Authentication Code (PAC) and Memory Tagging Extension (MTE), which help mitigate software vulnerabilities.
To foster innovation, Arm is making its powerful Edge AI platform accessible to startups through its ‘Flexible Access’ program. This initiative offers upfront, low-cost, or even no-cost access to a broad spectrum of Arm technology, tools, and resources. This ‘try before you buy’ model allows chip designers to experiment and iterate freely, only incurring license fees for the technology integrated into their final products. This program has already proven to be a ‘catalyst for innovation,’ contributing to approximately 400 successful chip designs over the past five years, with notable users including Raspberry Pi, Hailo, and SiMa.ai.
Looking ahead, Arm plans to expand edge use cases by enabling the execution of large language models (LLM) and small language models (SLM) specifically tuned for agent-based AI applications. The company is also extending its virtual platforms to Zena CSS, allowing software teams to develop and deploy software before the physical silicon is available, optimizing performance through digital twins of the hardware.
Also Read:
- Axelera AI Unveils Europa AIPU: A New Standard for AI Accelerator Performance, Efficiency, and Cost-Effectiveness
- VusionGroup Launches EdgeSense AI to Infuse Generative Intelligence into Physical Retail Spaces
This platform is not only transforming consumer devices but also making significant inroads into the automotive industry, powering in-vehicle AI experiences that merge voice, vision, and gesture for natural and conversational interfaces, further reducing latency and protecting privacy within the car.
