Regarding the earlier launched Cortex-M55 and Ethos-U55 NPU architecturesXu Dayong, director of application engineering at Arm Taiwan, expressed his hope to expand artificial intelligence computing technology to more IoT devices as 5G network applications become more popular.
According to Arm Taiwan President Eric Tseng, the world has officially entered the 5G network application phase since telecom operators began promoting commercial 5G network services last year. This means that different computing modes can be performed through wireless network connection modes with greater data transmission bandwidth and low latency, and artificial intelligence computing applications can also be realized on IoT devices.
Samsung shipped a total of 670 million 5G-connected mobile phones last year, and is expected to expand the launch of more 5G-connected mobile phones, tablets, and even laptop devices this year. Many Chinese mobile phone brands have also confirmed that they will fully launch 5G-connected mobile phone products starting this year, which will significantly increase the penetration rate of 5G-connected mobile phones this year. It is also expected that 5G network services will be popularized and applied more quickly.
Therefore, IoT applications, which have been discussed since the 3G and 4G network eras, are expected to integrate greater data transmission bandwidth and low-latency performance in the upcoming 5G network era, thereby changing the original computing model and even integrating device-side artificial intelligence to accelerate computing effects.
Xu Dayong said that from the perspective of Arm's product and service provision, the main goal is to meet customer needs, while also enabling developers to write application services in a simpler way, such as using unified compatible coding, thereby allowing more application products to enter the market faster.
以此次推出的Cortex-M55、Ethos-U55 NPU架構設計為例，實際上後者能與Cortex-M33、Cortex-M7或Cortex-M4，甚至運算應用模式無需DSP元件輔助運算的話，依然可讓Ethos-U55 NPU搭配極低耗電的Cortex-M0打造產品。
As for the introduction of NPU design into computing devices from mobile phones to IoT devices, does this also mean that NPU components will be considered a necessary design in Arm's architecture design products in the future?
As Xu Dayong explained, it all depends on how partners integrate Arm's architecture design products, and Arm's position is to assist partners in achieving various computing application needs.
For example, the "Prime Core" design used in Qualcomm's flagship processors launched in recent years, while inherently part of a multi-core architecture, also emphasizes vector operation acceleration during the computational process. This is similar in form to the NPU designs that other industry players have added to their processors to accelerate vector operations. Developers and hardware vendors can also leverage existing coding techniques to seamlessly create software services that leverage the processor's computing power.
This is naturally due to the fact that Arm incorporated code integration, a universal database, and architecture compatibility into its architecture designs from the early stages. This allowed different partners to combine various architecture designs according to their own product requirements, thereby creating new computing models that can meet different market demands. Therefore, it is not believed that future architecture design products will necessarily include specific components.
