First 64-Bit Armv9 CPU Designed for AI in Edge Workloads Debuts

Key Takeaways

Arm has introduced the Cortex-A320 CPU core, specifically designed for AI workloads at the edge, marking its first 64-bit Armv9 core.
The Cortex-A320, paired with the Ethos-U85 embedded NPU, is claimed to deliver over eight times the machine-learning performance of its predecessor.
This new architecture enhances memory access performance and efficiency, making it suitable for a variety of applications, including AI and IoT devices.

Advancements in AI Processing Capabilities

Arm, the UK chip design company owned by Softbank, is advancing its embedded platform with the introduction of the Cortex-A320 CPU core. This 64-bit Armv9 core targets edge workloads, responding to the rapid evolution of AI development. According to Paul Williamson, Arm’s senior vice president, AI workloads at the network edge have transformed from basic tasks like noise reduction to complex processes capable of facial recognition.

The Cortex-A320 is designed to pair with Arm’s Ethos-U85, an embedded neural processing unit (NPU). Together, they form a powerful system-on-chip (SoC) solution, configurable in clusters to meet varied performance needs. The Cortex-A320, touted as the smallest Armv9 implementation, features an AArch64 instruction set and offers significant enhancements over previous generations.

In less than a year since Arm released a reference platform combining the Ethos-U85 with the Cortex-M85 microcontroller, the new Cortex-A320 aims to tackle larger AI models exceeding one billion parameters. Williamson highlights that the demand for larger networks necessitates increased memory size and efficiency in memory access for complex AI applications.

This latest Cortex-A320 is said to be the most energy-efficient model in Arm’s history, utilizing half the power of its predecessor, the Cortex-A520. It incorporates enhanced security features like memory tagging extensions and supports Scalable Vector Extensions (SVE2) and the BFloat16 data type for optimized AI processing.

Software development remains critical, and Arm supports the new hardware with its Arm Kleidi libraries, which include compute kernels for AI and computer vision applications. These libraries integrate popular AI frameworks, ensuring robust support for developers.

Cortex-A320 supports real-time operating systems such as FreeRTOS and Zephyr, as well as Linux, expanding its applicability across various sectors. Besides its primary focus on edge computing, the low-power design of the Cortex-A320 is also suitable for smartwatches, wearables, and could serve as the ideal CPU for baseboard management controllers in server infrastructure. While specific product partnerships have not been disclosed, Arm anticipates the chip will reach silicon production next year.

The content above is a summary. For more details, see the source article.