Following last year’s proposal to address the demand for artificial intelligence computing on mainstream devices,Ethos-N57 and Ethos-N37 NPUsIn addition to announcing the launch of the new embedded processor Cortex-M55, Arm also announced the launch of the new Ethos-U55 NPU earlier, allowing more IoT devices to accelerate computing modes through artificial intelligence technology.
Compared to last year's Ethos-N57 and Ethos-N37 NPUs, which were designed with Cortex-A and Mali architectures, the newly announced Ethos-U55 is paired with the new Cortex-M55 embedded processor, allowing more IoT devices to adopt artificial intelligence technology for accelerated computing.
According to Arm, the combination of these two technology licensing architectures will enable existing embedded devices to achieve up to 480 times the machine learning efficiency. It is also expected to be gradually applied to various embedded devices in 2021, and will realize the edge computing effects previously only used in devices such as mobile phones, combined with artificial intelligence technology.
The Cortex-M55, a new processor architecture designed for embedded devices, can be paired with the simultaneously launched Ethos-U55 NPU to enable on-device AI computing, eliminating the need for additional DSP components required to implement AI computing. This also further reduces the size of IoT devices and lowers overall computing power.
Arm is also collaborating further with Google on the Cortex-M55 and Ethos-U55, making them compatible with Google's TensorFlow Lite computing framework, thereby connecting more artificial intelligence computing technologies based on the TensorFlow Lite computing framework.
In addition, the Cortex-M55 design adopts the Armv8.1-M instruction set and integrates the "Helium" vector acceleration design (officially named MVE (M-Profile Vector Extension)) to meet the needs of machine learning.
Compared with the previous Cortex-M processor, the combination of DSP components can improve machine learning computing performance by 15 times, and also bring a 5-fold computing performance improvement compared to existing embedded devices through DSP component computing mode.
The Ethos-U55 NPU is not only Arm's first NPU designed for embedded devices, but its MAC computing engine, with 32 to 256 cores, emphasizes low power consumption while also offering scalability.
In addition to working with the Cortex-M55 to optimize computing performance, the Ethos-U55 NPU can also be used in conjunction with existing processor designs such as the Cortex-M33 and Cortex-M7.
In terms of performance, compared to the Cortex-M7, the Cortex-M55 boasts approximately 6x higher computing efficiency. The combination of the Cortex-M55 and the Ethos-U55 NPU delivers 50x higher computing efficiency. The Cortex-M55 also reduces power consumption by approximately 7x compared to the Cortex-M7, and the Ethos-U55 NPU reduces this by an additional 25x. The proposed architecture complies with the Arm Platform Security Architecture (PSA) and the Corstone reference design, and can also be used with Arm TrustZone security technology.
Currently, the two architecture designs have been confirmed to be in collaboration with industry partners including Amazon, Alif Semiconductor, Bestechnic, Cypress Semiconductor, Dolby, Google, NXP, Samsung, and STMicroelectronics. They are expected to be applicable to various speakers and headphones that integrate voice-controlled smart assistant services, as well as other embedded devices that integrate artificial intelligence technology and edge computing models.
