Arm Launches Cortex-A320 for Edge AI Applications
Arm is expanding its embedded platform with the Cortex-A320 CPU core, the company’s first 64-bit Armv9 CPU core specifically designed for edge workloads. This move aims to equip devices with the processing power required for increasingly complex neural network applications.
Paul Williamson, senior vice president and general manager for Arm’s Internet-of-Things line-of-business, highlighted the rapid evolution of AI, using the example of a smart doorbell. It has transformed from a simple buzzer to a device capable of identifying individuals.
The continued demand for hardware to efficiently execute larger networks is pushing memory size requirements, so systems with better memory access performance are becoming really necessary to perform these more complex use cases,” Williamson said.
The Cortex-A320 is designed to be coupled with the Ethos-U85, Arm’s embedded neural processing unit (NPU) accelerator, in edge AI system-on-chip (SoC) designs. It can be configured in clusters of four cores to scale up and offer a range of performance options.
The A320, described as Arm’s “smallest Armv9 implementation,” features an AArch64 instruction set. It’s a single-issue, eight-stage core with up to 64KB of L1 cache and up to 512KB of L2 cache.
This new pairing offers more than eight times the machine-learning performance of last year’s platform, according to Williamson, and is capable of running large AI models with over a billion parameters. He added that Cortex-A processors are better at memory management, supporting more addressable memory than Cortex-M based platforms.
This processor is also said to be the most energy-efficient in the Armv9 family, using half the power of the Cortex-A520, the high-efficiency core used in some reference designs.
The architecture brings security features, including memory tagging extensions to catch memory exceptions, along with Scalable Vector Extensions (SVE2) and BFloat16 data type support for AI processing.
Arm’s Kleidi libraries will provide support for the new edge hardware, including Kleidi AI, a set of compute kernels for AI frameworks, and Kleidi CV for computer vision applications. These developments are integrated into AI frameworks such as llama.cpp, ExecuTorch, and LiteRT.
The Cortex-A320 supports real-time operating systems like FreeRTOS and Zephyr, as well as Linux. Although licensees will manufacture the chips, Arm expects silicon availability next year but hasn’t disclosed specific partners or products.
Besides network-edge applications, the low-power design makes it suitable for smartwatches, wearables, and baseboard management controllers in servers and infrastructure, according to Williamson.
