Arm has announced the Cortex-A320 processor, its first ultra-efficient Cortex-A product based on the Armv9 architecture. The chip is aimed at edge devices and IoT equipment with support for AI functions.
The new product is based on the Armv9.2-A (Harvard) architecture with support for QARMA3, SVE2, MTE, RAS extensions and cryptographic functions. The first-level cache can be 32 or 64 KB. An L2 cache with a capacity of 128 to 512 KB is optionally available, while L3 cache is not provided. In addition, work with RAM has been significantly accelerated.
The Cortex-A320 is a single-core processor with 32-bit instruction sequential fetching. Numerous microarchitectural updates, such as branch prediction and prefetching, achieve a 50% efficiency improvement over the Cortex-A520 and a 30% performance improvement in SPECINT2K6 over the Cortex-A35.
Image Source: Arm
According to Arm, the Cortex-A320 processor delivers 10x the machine learning performance of the Cortex-A35 (on GEMM int8) and 6x the performance of the Cortex-A53, the world’s most popular Armv8-A architecture product. The Cortex-A320 is the most efficient Cortex-A solution for machine learning workloads today.
Based on Cortex-A320, clusters of up to four cores can be formed. The vector processing unit with NEON and SVE2 technologies can be individual for each core or used by a bundle of two cores, including in a quad-core configuration. Thanks to the DSU-120T (an optimized version of the DynamIQ Shared Unit), it is possible to form clusters exclusively with Cortex-A320 cores. In addition, the new products support the Ethos-U85 NPU, which allows automatic transfer of processing of unsupported data types and instructions to the SIMD blocks of the Cortex-A320. In general, it is said that it is possible to launch models with more than 1 billion parameters on the new chips.
The Cortex-A320 processor can be used in a wide range of applications, from smart speakers and smart surveillance cameras to autonomous vehicles and service robot controllers. In addition, the new processor is suitable for battery-powered microcontrollers and devices running real-time operating systems (RTOS). Advanced security features are implemented, including Secure EL2 (Exception Level 2).