Looking at the key terms in the current chip industry, RISC-V is definitely one of them.

Since the birth of computers, the instruction set architecture has been one of the core concepts in computer architecture. The two giants in the mainstream market are x86 and ARM, with the former basically monopolizing the PC, laptop, and server fields, and the latter dominating the smartphone and mobile terminal markets. In recent years, with the global demand for chip autonomy and control growing and the needs of fields such as the Internet of Things and edge computing expanding, RISC-V has received widespread attention and application in academia and industry, gradually becoming the third major instruction set architecture.

In 2023, the RISC-V architecture has taken root in more practical application scenarios, gradually moving from the Internet of Things devices and edge computing to the fields of AI computing and high-performance computing.

In the AI era, RISC-V is also facing a new opportunity for explosion. Silicon Valley chip legend Jim Keller said, "The potential of RISC-V is unlimited. In the future, we will usher in AI software applications that have never been seen before, and RISC-V is expected to create the next generation of AI engines."

Advertisement

So, what kind of magic does RISC-V have? How does it set off a technological storm in the AI era?

01

How does RISC-V apply to AI applications?

Entering 2023, AI has made the computing power of chips unparalleled. With the demand for hardware computing power of AI algorithms reaching an unprecedented level, traditional chip architectures are facing unprecedented challenges. Therefore, implementing specialized architectures and chip designs that are specifically optimized for AI algorithms have become a new trend to meet the high demand for computing power.

The instruction set is simple and easy to expand, and the cost is low.ARM and X86, as established instruction sets, have too many legacy instructions within them, making it difficult for subsequent engineers to update or add to the existing instructions. Many readers may have downloaded the official documentation from ARM, and even after reading thousands of pages, it is challenging to master, let alone innovate. The multi-instruction mode adds further complexity to the ARM architecture design. Therefore, in terms of instruction set expansion, the more dynamic RISC-V naturally has a significant advantage, after all, being open-source and streamlined is one of RISC-V's strong suits.

At the same time, being open-source also brings another important benefit, which is reducing the technical barriers and costs associated with RISC-V. Anyone can obtain the source code of RISC-V, customize and optimize it according to their own needs, without having to pay high licensing fees.

Aligning with Edge AI

RISC-V is not suitable for all AI models. It is clear that in the field of large AI models, due to their extremely high computational power requirements and the corresponding resource consumption, almost all RISC instruction sets are deterred. Taking large models like ChatGPT as an example, according to OpenAI's calculations, the total computational power consumption during the training phase of ChatGPT is about 3640 PF-days (equivalent to running at 1 PFLOP/s for 3640 days), and the training cost of GPT-3 is estimated to reach $5 million per training session. Edge AI cannot bear the high demand for computational power of large models, but it can focus on small models AI that target a single domain and have relatively lower computational power requirements, where RISC-V's opportunity arises. Especially in the field of the Internet of Things, RISC-V has already shown its advantages.

Matching with AI Algorithms

From the perspective of algorithms, the highly flexible RISC-V instruction set is well-matched with AI algorithms that require faster iteration and high computational power. As is well-known, RISC-V has high scalability and modularity, which also means that designers can tailor and customize it according to specific application needs, thus better adapting to different AI algorithms, including deep learning, neural networks, etc.

02

Betting and Layout by Major Companies

Driven by technological innovation and market demand, the development potential of RISC-V is fully demonstrated. Under this trend, major chip giants have all joined in, exploring the broad prospects of RISC-V together, hoping to gain a favorable position in future competition. This includes companies such as Apple, Intel, Qualcomm, Renesas, Infineon, NXP, Bosch, and others.In delving into the reasons why major chip manufacturers are placing bets on the RISC-V domain, what considerations and strategies are hidden behind the scenes?

For any major chip manufacturer, the choice of ISA (Instruction Set Architecture) is quite cautious. ISA not only determines the difficulty of chip development but also the software and development ecosystem bound to it, and may even affect the competition faced in the market.

Currently, the performance competition in the mobile chip market is intensifying. Regardless of whether it is Qualcomm, NVIDIA, Samsung, or Apple, they are all developed based on the Arm architecture and are members of the Arm alliance. To date, Arm has occupied more than 95% of the market share in the smartphone chip field, including tablet computers.

Over-reliance on technology from Arm, a subsidiary of SoftBank Group, may bring some risks. At the end of 2022, Arm sued Qualcomm over intellectual property issues.

Moreover, Arm technology is not only ubiquitous in smartphones but is also rapidly expanding into large devices, data centers, and automotive applications.

The automotive business is also one of Qualcomm's core businesses, especially the intelligent cockpit system. Now, other major semiconductor chip manufacturers, including Intel, have also indicated that they will support the development of RISC-V chips in the future. This may provide more motivation for Qualcomm. At the 2022 RISC-V Global Summit, Qualcomm executives announced that as early as the Snapdragon 865 SoC released in 2019, Qualcomm had already used RISC-V in the design of microcontrollers and has shipped hundreds of millions of RISC-V cores.

As one of the major MCU manufacturers, especially in the automotive field, Renesas has accumulated and developed over the years and has multiple MCU product lines under its belt. In 32-bit MCUs, there are the RX series based on self-developed CISC cores and the RA series based on the Arm Cortex-M architecture. Even so, Renesas is determined to develop a third product line, which is based on the RISC-V architecture.

Intel is also very interested in promoting RISC-V. In 2022, Intel strongly supported RISC-V by joining the RISC-V International Foundation as a Premier member. At the same time, it also established a $1 billion (approximately 6.4 billion RMB) IFS fund to support startups and established companies in innovating the foundry ecosystem, with a large part of it used for RISC-V.

The IFS plan provides a series of verified RISC-V IP cores and has optimized performance for different market segments, ensuring that RISC-V operates optimally across all types of cores (from embedded to high-performance) on IFS chips. At the same time, it will provide three types of RISC-V products: partner products manufactured based on IFS technology; RISC-V cores licensed as differentiated IP; and small chip building blocks based on RISC-V that utilize advanced packaging and high-speed chip-to-chip interfaces.

In fact, Intel has been active on the RISC-V path for quite some time. In 2021, Intel launched the Nios V processor based on the RISC-V architecture. Prior to this, Intel also intended to acquire RISC-V chip design company SiFive for $2 billion, but it did not ultimately succeed. However, this also showed the public Intel's attitude towards RISC-V.RISC-V CEO Calista Redmond stated: "The open collaboration of RISC-V has sparked a profound transformation in the semiconductor industry, and this partnership will accelerate innovation in open computing."

Vice President of Intel Labs and Dean of Intel China Research Institute, Song Jiqiang, said that the Intel China Research Institute specifically included the open-source instruction set RISC-V in its research field because China's influence in this area is quite evident. Among the senior members of the RISC-V International Foundation, nearly half are Chinese enterprises or schools. Additionally, there is guidance from the Chinese government, which wants RISC-V to become a focus for future independent innovation. Therefore, China is certain to do very well in the development of RISC-V in the future.

Next, let's take a look at China's layout in the RISC-V race.

03

China is an Important Base for RISC-V

According to statistics from RISC-V International, in 2022, the global shipment of processors using the RISC-V architecture exceeded 10 billion, completing in just 12 years what traditional architectures took 30 years to achieve. It is projected that the adoption rate of RISC-V will grow at a compound annual growth rate of 40% in the coming years, and by 2030, RISC-V architecture chips are expected to exceed 16 billion. Of these 10 billion shipments, nearly half come from China, and China's vast market will become an important base to support RISC-V. Moreover, among the members of the RISC-V Foundation, Chinese enterprises also account for nearly half.

Academician of the Chinese Academy of Engineering, Ni Guangnan, said at the first Xuan Tie RISC-V Ecosystem Conference held on March 2nd: "It is no exaggeration to say that today, RISC-V is the most popular architecture in China's CPU field. China's chip industry and the entire chip ecosystem will increasingly focus on the RISC-V architecture, and China's vast market will become an important base to support RISC-V."

It is reported that currently, more than 300 companies in China are paying attention to RISC-V or developing with the RISC-V instruction set, including main content such as IP cores and chip design.

Nowadays, a RISC-V Industry Alliance has been established in China, bringing together many well-known enterprises including Huawei, Alibaba Cloud, Unigroup, and Huami Technology, to jointly promote the construction of RISC-V's ecosystem in China. The alliance has accelerated the implementation and application of RISC-V technology in China through building cooperative platforms, holding seminars, and promoting the establishment of standards.Chinese research institutions and enterprises are actively engaged in the research and development of RISC-V chips, achieving a series of key technological breakthroughs. Renowned academic institutions and research organizations such as the Chinese Academy of Sciences and Tsinghua University, as well as startups like PingTouGe Semiconductor and Chiplai Technology, have demonstrated impressive capabilities in RISC-V processor design, software ecosystem construction, and application solution development.

Chinese companies have also developed a rich array of processor products based on the RISC-V architecture, covering various fields including embedded systems, the Internet of Things (IoT), edge computing, and data centers. For instance, Alibaba's PingTouGe has launched the XuanTie series of processors, and Chiplai Technology has introduced several MCU chips tailored for the IoT sector.

In addition to the aforementioned companies, numerous manufacturers such as VeriSilicon, Bluetrum, GigaDevice, StarFive, Allwinner Technology, SmartSemi, Innosilicon, Eswin, Thine Electronics, Telink Semiconductor, Loxone Technology, Sino Wealth Semiconductor, Airoha Technology, GigaDevice, Hengshan Chip, GigaDevice, ChipRise Technology, Bolyu Intelligence, Nasuda, SmartSemi, and FC Microelectronics have all shown strong R&D capabilities and excellent products.

Telink Semiconductor's RISC-V chip products are primarily applied in the IoT sector, such as smart wearables and smart homes, where their RISC-V processors excel in connectivity, power consumption, and performance, providing users with a stable wireless connection experience. VeriSilicon has introduced a range of high-performance processor IPs in the RISC-V domain, which can be widely applied in IoT, smart homes, wearable devices, and other fields; Bluetrum's RISC-V chip products are mainly targeted at the Bluetooth audio market, including Bluetooth earphones and speakers; GigaDevice has a solid foundation in storage and MCU areas, and its RISC-V chip products combine the advantages of storage and computation, suitable for industrial automation, smart homes, and other fields; StarFive offers a series of processors and SoC products based on the RISC-V instruction set, which are applicable to various application scenarios ranging from edge computing to cloud computing...

04

Commercialization of RISC-V

Looking at the current shipment situation, RISC-V chips were initially more concentrated in the low-computational-power MCU market and the IoT sector with low ecosystem dependency. For example, Wukong, in addition to launching WiFi data transmission chips, had already introduced and mass-produced RISC-V-based power line communication chips, which can be applied in smart meters, energy management systems, automotive-PEV and charging stations, remote meter reading devices, solar energy, industrial applications, and other fields, with shipments of this chip model disclosed to have exceeded one million units in 2019.

Loxone Technology, holding a leading position in the WiFi MCU field, has been using its self-developed RISC-V instruction set-based IPs in all new products since 2020.

Telink, with low-power Bluetooth SoC as its main product, has adopted the RISC-V architecture in its latest generation of products, the TLSR9 series, which has been certified.In addition to the IoT market, RISC-V also has development opportunities in the fields of servers, automotive, and more. For instance, the RISC-V CPU SG2042 server cluster by Suan Neng can provide support for next-generation cloud computing, artificial intelligence, and data analysis. Superwise Technology has planned three series: the A series (high performance), AE (high reliability), and E series (high energy efficiency). The A series focuses on performance and is suitable for desktop, server, and other application scenarios, which can be applied in cloud computing, HPC, AI, and smart cockpit fields.

At the Xuan Tie RISC-V Ecosystem Conference, the Institute of Software at the Chinese Academy of Sciences showcased an open-source laptop based on RISC-V called "Ruyi BOOK". This laptop has a comprehensive battery life of up to 12 hours and integrates a third-party large model AI assistant that supports voice dialogue and text interaction. Ruyi BOOK is equipped with the Xuan Tie C910 processor and the openEuler operating system, capable of smoothly running large office software like DingTalk and Libre Office, and for the first time, it has connected the entire RISC-V chain from the underlying chip to the operating system to commercial software.

In the automotive field, leading customers have already adopted Xuan Tie, and the Academy of Damo will make more arrangements for reliability, including high real-time processors.

In the communication field, Bio-Gee has applied RISC-V in 5G baseband chips. Its first high-performance, low-power 5G small base station physical layer SoC chip, PC802, has been mass-produced.

In the robotics field, Yibo Intelligence has promoted RISC-V's first entry into the ROS robotics field, developing a SLAM mapping and obstacle-avoidance navigation cart based on the Xuan Tie processor's Mlik-V Meles development board.

In the ICT field, the demand for reducing computational power consumption and cost has brought incremental space for cloud-network devices based on customized RISC-V. The China Telecom Research Institute has used Xuan Tie RISC-V to develop new hardware such as cloud desktops and AI edge boxes, creating a "end-edge-cloud" full-chain application.

Furthermore, in the power industry, the State Grid Smart Chip Company has developed an AI chip for industrial applications based on the Xuan Tie processor, used in scenarios such as smart patrol of substations and coordinated scheduling of source-network-load-storage.

05

Accelerating Integration with AIAccording to data from SHD, the RISC-V Foundation had only 17 members in 2015, but by the end of 2023, this number had grown to 4,037. Concurrently, SHD Group predicts that from 2022 to 2030, the compound annual growth rate of RISC-V in the mainstream market will be as high as 40%. The annual shipment of chips based on the RISC-V instruction set architecture is expected to increase from 600 million to 16 billion units. The proportion in mainstream applications such as MCUs, laptops, 5G, industry, consumer AI, and automotive AI will reach about 30%.

In the past year, the RISC-V ecosystem has further developed. Therefore, 2023 is also considered a breakthrough year for RISC-V.

In May 2023, Meta launched its first AI inference accelerator based on the RISC-V architecture. In June 2023, the global RISC-V software ecosystem initiative "RISE," initiated by 13 companies including Google, Intel, Nvidia, Qualcomm, and Alibaba, was officially launched. The aim is to accelerate the construction of the RISC-V software ecosystem and the commercialization of applications. Members will jointly promote the marketization of RISC-V processors in areas such as mobile communications, data centers, edge computing, and autonomous driving. Alibaba is the only board member from mainland China. In October 2023, Qualcomm announced a collaboration with Google to launch a smart wearable chip based on the RISC-V architecture and supporting the Wear OS system, which will be commercially promoted in the global market.

At the same time, it is evident that these leading chip companies are not only increasing their investments in the RISC-V field but also paying more attention to the deep integration of RISC-V with AI.

In the consumer electronics field, the Academy of Damo announced a new iteration of the XuanTie processor family - the XuanTie C907, which for the first time achieves matrix operation (Matrix) expansion, providing more options for future AI accelerated computing and will be integrated into other XuanTie processors. The next-generation flagship processor C930 will also continue to evolve towards higher performance, significantly enhancing AI computing processing capabilities, and is expected to be launched within the year.

Manufacturers such as YSW Computing have introduced AI chips for edge computing based on RISC-V, with powerful video encoding and decoding capabilities and a rich set of interfaces to meet the needs of various application scenarios. YueFang Technology and RISE Technology are also leaders in the AI field of RISC-V, offering high-performance processor IP cores and customized chip solutions.

Recently, Samsung has also taken action. Samsung's SAIT (formerly the Advanced Institute of Technology) has established an Advanced Processor Lab (APL) in Silicon Valley, specializing in AI chip design. This newly strengthened R&D team will fully leverage the openness and flexibility of the RISC-V architecture to design and optimize a new generation of AI chips. The ultimate goal of Samsung's APL research institute is to design its own AI chips based on RISC-V. Historically, Samsung has designed cutting-edge semiconductor chips such as Exynos based on ARM products.

It is reported that Samsung's strengthening of its R&D team aims to enhance its R&D capabilities in the AI chip field, further promote the development of AI technology, and also to meet the challenges of competitors such as Nvidia. Nvidia holds a leading position in the AI chip market, and Samsung hopes to break the existing market pattern by launching AI chips based on the RISC-V architecture.Abundant opportunities, challenges as well

Whether it's the rising tide of RISC-V worldwide or the attitudes and strategies of global chip giants, it's clear that RISC-V is an essential choice.

David Patterson, the father of RISC-V and Turing Award winner, also stated: "RISC-V is a global phenomenon, with developers from over 60 countries studying it, and in a few years, RISC-V will be everywhere!" He even asserted that by the end of this century, the dominant ISA for future product development will be the open RISC-V architecture.

This is not just a trend; it's more like the eruption of a technological revolution.

SHD Group estimates that by 2030, the revenue from RISC-V-based SoCs is expected to reach as high as $92 billion, with a compound annual growth rate of 47%.

However, for RISC-V to achieve widespread market application, there are many challenges to face, such as its ecosystem still lagging behind other mainstream instruction set architectures like ARM and x86. To achieve widespread market application, RISC-V needs to strengthen its ecosystem in terms of software compatibility, development tools, and operating systems. Secondly, in some high-performance computing scenarios, its technological maturity is still insufficient. To gain an advantage in the more intense market competition, RISC-V needs further optimization in terms of performance, power consumption, and cost. In terms of business models and cooperation, due to the openness of RISC-V, companies also face certain business challenges when using and releasing RISC-V products.

Let's conclude with the words of Roger Espasa, founder and CEO of Semidynamics: RISC-V started from the bottom, and now it has reached the mid to high-end, including servers; of course, it will take some time to grow and develop. Eventually, it will enter into HPC, GPUs, and achieve full coverage.