With the advancements of new technology such as the Internet of Things (IoT), Artificial Intelligence (AI), and open-source processor architecture, as well as the popularity of new industrial trends like renewable energy, smart cars, AI big models and AIGC, China’s embedded system industry has entered a brand-new age. Previously, China have been studying and following foreign industry leaders; today, China can compete on even ground. The industry has also progressed from primarily focusing on hardware and applications to producing processor cores and fundamental software. There are three current notable trends in the industry.

On-device AI: the Coming of the Age of »Light« Intelligence
At the device of IoT, AI is quickly being integrated. An AI-enabled IoT device can be an AIoT (Artificial Intelligence of Things) terminal or an edge gateway/server. The integration of AI and IoT opens up a broad range of new applications, for instance, smart cleaning robots and predictive maintenance of industrial equipment. According to market researchers, AIoT devices will be widely used; the overall demand of chips will rise significantly to tens of trillions of units.

The age of “light” intelligence is booming thanks to the movement of AI towards the device side. MPUs (microprocessors) such as Cortex-A on devices are being replaced with MCUs (microcontrollers) like Cortex-M/RISC-V. Linux, which is often seen running on MPUs, is giving way to RTOS. According to the research done by the Key Laboratory for Embedded and Network Computing of Hunan University, while it takes the open-source Zephyr RTOS 0.13 second to run the Baidu Paddle Lite image categorization samples on the Rockchip RK3568 platform, Linux needs 0.42 second.

One of the major challenges of developing on-device AI is the lack of AI experts. A development team needs an AI expert to properly provide data and train models. Thus, tools-assisted development has taken hold. Developers can leverage tools provided by vendors like Edge Impulse, SensiML, and Sensory AI to configure device intelligent models and construct the algorithm. These tools solve the problem of lacking AI talent and help the teams to concentrate on AIoT product functionalities and AI features; this is important in shortening the Time to Market of AIoT products. Such tools need to achieve two major goals: improve the adaptivity of models and simplify data gathering and optimization. Global embedded system companies such as ST, Silicon Labs and Renesas are leading the pack in terms of these tools, while companies in China are only getting started. Chinese companies like Canaan, ChipIntelli, and SigmaStar have made progress in intelligent vision and voice wake-up; there is a lot of room to grow.

RISC-V: the Embedded Processor Industry is Flourishing

RISC-V is an open-source Instruction Set Architecture (ISA) that was initially developed at the University of California Berkley in 2010. In 2022, more than 10 billion RISC-V-based chips have been produced globally. Half of these chips have been from China, especially as part of MCU embedded products. For instance, BlueTrum’s RISC-V TWS earphone chip has reached one billion produced.

At the 2023 Shanghai Dishui Lake RISC-V industry summit, the organizer announced that several RISC-V products have reached major production milestones: 10 million for ChipIntelli AI speech chips; one million for XinSheng Technology (China Mobile) low-power RISC-V MCU, Boufallo Lab IoT chip, HPMicro super high-performance MCU respectively; two million for the Canaan K210 series. According to market research companies, MCU, wireless, and automotive will be the three major market for RISC-V.

RISC-V IP (Intellectual Property) has been growing rapidly in China. Nuclei has accumulated more than 150 licensed customers in the four years since its inception. Ali T-Head XuanTie series CPU has seen more than 2.5 billion units produced; they are utilized in applications such as smart speakers, dashcams, printers, and security cameras.

Software ecosystem development is a critical aspect of the development of RISC-V processors. Owing to the open-source movement, significant progress has been made in the RISC-V software ecosystem. In 2023, Google announced support for RISC-V in Android, which means most mainstream OSes now have support for RISC-V, including Debian, Fedora, Ubuntu, Anolis OS, UOS, openKylin, Coolita OS, FreeRTOS, RT-Thread, etc. Software on these OSes is also starting to support RISC-V. For example, the open-source community of the Debian Linux distribution started to support RISC-V in 2019; by 2022, 95% of available Debian software packages have been ported to RISC-V, making RISC-V a Tier-1 architecture in Debian.

RISC-V is attracting big players to invest more in research and development. In 2023, Qualcomm and four industry leaders in semiconductor founded a joint venture dedicated to RISC-V. SiFive predicted the total number of RISC-V chips will reach 80 billion by 2025 at its 2023 China summit.

Embedded Software: Multi-OS Hybrid Deployment

Benefiting from the swift development of integrated circuit (IC) technology and manufacture process, embedded systems are gaining stronger hardware capabilities: faster single-core speed, multi-core transformation, heterogeneous multi-cores, virtualization, and trusted execution environment (TEE). Applications of these technologies lay a solid foundation for executing multiple OSes on the same SoC .

In the meantime, requirements from applications like IoT, intelligent systems, functional safety, and security mean that it is becoming increasingly difficult for one OS to handle all functionalities. One solution is to leverage multiple OSes for different areas. For example, Windows for user interface, Linux for the network and communication stacks, and an RTOS for real-time performance and dependability. The arrangement must also be easy to develop, deploy, and extend; containers and virtualization are key technologies to achieve this goal.

OpenEuler is a free open-source Linux distribution maintained by The OpenAtom Foundation of China. An open community is responsible for developing the OS and the ecosystem around it. OpenEuler has a version specific for embedded systems, as in OpenEuler Embedded. The latest version of OpenEuler Embedded integrates support for OpenAMP. Through an active collaboration, OpenEuler and Zephyr have realized hybrid deployment. OpenEuler Embedded plans to support more RTOSes like UniProton, RT-Thread, Jailhouse (a partitioning Hypervisor based on Linux), and ZVM (an open-source virtualization project in Hunan University based on the Zephyr OS).

Embedded OS is thriving in China. Open-source OS is now mainstream; such as RT-Thread, SylixOS, Intewell, and OneOS have been deployed successfully in many areas, including consumer electronics, IoT, industrial and automotive.

Conclusions

Embedded system is a converging point of multiple academic and professional disciplines, including computer science, electronic and information technology, IC, IoT and AI as well. In China, institutions are joining forces to overcome engineering challenges in embedded system and embrace the era of intelligent connectivity.

About Author

Allan He is the Deputy Director China Software Industry Embedded System Associations，He has published 8 books in Chinese and English, including "Embedded Operating System-History and Future in the Internet of Things ", "Embedded Real-time Operating System" and " A Deep Dive Into RISC-V Programming", published more than 80 papers in journals and conferences, and taught in universities and enterprises. Allan he is a member of the Embedded World Program Committee and member of the Standing Committee of the Embedded System Special Committee of the China Computer Society.