Chinese RISC-V startups have raised more than $1B

[Daniel Nenni]

China RISC-V summit will be held on Aug. 23-25th. The event will be held in Beijing and virtually. Register here: https://hubs.la/Q01Yk2Fm0 In a recent talk, Dr. Yungang Bao of the China RISC-V Alliance (CRVA) and Beijing Institute of Open-Source Chip (BOSC) discusses the impressive momentum of RISC-V in China from industry and academia. You can see the video here:

Major players
RISC-V is still new to China’s chip industry and most companies in the sector are startups. Here are four Chinese semiconductor companies that are working with the architecture.

Nuclei System Technology is a semiconductor design company and provider of commercial RISC-V processor IP.

Nuclei was founded in 2018 by Synopsys and Marvell veteran Hu Zhenbo. Hu told a forum (in Chinese) in 2018 that China had realized “self-sufficiency” in chip design, and the real problem that dogged China’s semiconductor industry was “the lack of ISAs.”

• “RISC-V offers a new opportunity for China. RISC-V is likely to become a new mainstream architecture worldwide in the future and it is completely open. It will enable China to truly achieve the production of a mainstream and homegrown processor core,” he said at the time.
• Nuclei has rolled out several generations of chip designs that can be used in IoT and manufacturing, according to its website. The company said its designs have been used to build trusted execution environments (TEE) for fintech giant Ant Group’s payment platform Alipay and financial services corporation UnionPay.

GigaDevice, one of China’s largest manufacturers of nonvolatile memory (NVM), in 2019 launched a general-purpose microcontroller based on RISC-V. A microcontroller is a device that is widely used in IoT.

• The company is China’s largest maker of microcontrollers for IoT devices. It shipped (in Chinese) more than 200 million microcontrollers in 2020. Not all of the company’s microcontrollers are based on RISC-V.

Rivai, a Shenzhen-based startup, is a fabless chipmaker that designs RISC-V processors for IoT devices and artificial intelligence applications like robots and smart speakers.

• The company said on its website (in Chinese) that it is backed by Chinese search engine Baidu and Kai-Fu Lee’s Sinovation Ventures, but it didn’t specify the amount or time of the investment.
• 
  
• Rivai was founded in 2018 by Tan Zhangxi. Tan is also the co-director of the RISC-V International Open Source Laboratory (RIOS Lab) at Tsinghua-Berkeley Shenzhen Institute.
• 
  
• Tan told Chinese media in an interview in January that RISC-V is “not a cheap alternative to Arm” because it can do a lot of things that can’t be done by Arm. “As the demand for customization increases, RISC-V is expected to generate a market for high-end IP customization that is different from the Arm era,” he said.

E-commerce giant Alibaba’s chip unit, T-Head, in July 2019 released its first RISC-V-based processor design, Xuantie 910. The company said it can be applied to the design of chips for fifth-generation (5G) wireless networks, artificial intelligence, as well as autonomous driving.nIn May, the company announced (in Chinese) its second RISC-V-based processor, Xuantie 907. The company said at the launch that its Xuantie-series processors had cumulatively shipped more than 200 million units by the end of 2020. Most of T-Head’s Xuantie-series processors are based on architectures other than RISC-V.

RISC-V’s future in China
RISC-V is not coming to smartphones and laptops any time soon, but Chinese experts say RISC-V could emerge as a major ISA within the next decade.

One of the biggest problems for the 11-year-old architecture is the lack of an ecosystem, both in hardware and software. That includes systems on a chip (SoCs), developer boards, design tools, and the operating systems running on the chips.

Wu, the analyst at Chasing Securities, used the Android operating system as an analogy to argue that an ecosystem will not be a problem for RISC-V.

“The RISC-V architecture is simple, efficient, free, and open, which also gives it a competitive advantage. In the face of the proprietary Symbian operating system, Android became one of the major OSs for mobile devices in the following decade by taking advantage of its open-source features,” he said, referring to a once-might mobile operating system on Nokia phones.

However, RISC-V is still a niche in China’s semiconductor market. Around 95% of Chinese-designed chips were based on the Arm architecture as of September 2020, according to a local media report.

Even if a RISC-V ecosystem readily arises for IoT devices, there are bigger obstacles to widespread RISC-V adoption. “The hardest market to establish an ecosystem for is actually mobile, followed by desktop and server,” Allan He, vice chairman of China Software Industry Association’s Embedded Systems Association, told local media in 2019.

It will take a concerted effort (in Chinese) by chip designers to build up RISC-V’s ecosystem, said Hu Kangqiao, chief executive of Hexin Hulian, a Beijing-based company that designs home appliance chips based on RISC-V. “When the number of manufacturers designing RISC-V chips is in the same order of magnitude as ARM, it means that the RISC-V’s ecosystem is mature,” Hu said in 2020.

“That will take approximately five to 10 years,” he predicted.

China's chip industry is embracing RISC-V · TechNode

RISC-V was long considered a “hobbyist architecture.” Now, China’s semiconductor industry is taking it seriously.

technode.com

 

[cliff]

We don't think it will take 5 years to catch on. We are putting together quad tiles of the OpenTitan Earl Grey RV32IMCB (RISC-V) with separate Dcache and Icache. The documentation on this Google/WD/etc sponsored project is amazing. Tile link allows the peripherals to communicate. Tenstorrent is moving rapidly on their RISC-V solution.

I would like @blueone and other's opinions. Should ARM be worried?

Edit: Here are some examples of what the open source community is doing. It seems like infrastrucure is growing fast:

Design - OpenTitan Documentation

opentitan.org

FreeRTOS for RISC-V RV32 and RV64

Using FreeRTOS on RISC-V Microcontrollers Preamble As noted on the Upgrading to FreeRTOS V10.3.0 page, the configCLINT_BASE_ADDRESS configuration setting has been deprecated and replaced by the configMTIME_BASE_ADDRESS and configMTIMECMP_BASE_ADDRESS settings described on this page. Legacy...

www.freertos.org

 

[blueone]

I would like @blueone and other's opinions. Should ARM be worried?

I think the main reason Arm should be worried is that they seem to have developed an adversarial relationship with their customers, or at least a lot of them. It reminds me of Intel of 25 years ago. I think a good part of their embedded core business is at risk, and it's all their own fault, IMO. Perhaps most of it is at risk. Since they're still a private company their financials are opaque, so assessing the damage so far is difficult. Arm has publicly said they're focusing on high-end mobile and datacenter designs, which is a natural evolution of their business model, which I also think is indicative of significant projected embedded market share loss to RISC-V.

RISC-V comes up a lot in conversations with former colleagues who work around the industry now. IMO, RISC-V is a very strange animal. Perhaps the strangest body part of that animal is the incredibly permissive licensing structure. (I joked with a colleague a while back that this body part is analogous to a biological reproductive system.) I would sum up the licensing model as "anything goes". (Companies can modify the spec and keep the result private, and build products based on the proprietary version without fees or compulsion to donate the enhancements back to the community.) So in the future it seems like there will be N+1 RISC-V specs, where N is the number of proprietary versions, and the 1 is open version defined by the RISC-V Foundation. I think a likely scenario is where RISC-V overall is like a tree, where the RISC-V Foundation version is like the trunk, and the numerous proprietary versions are like branches. And as the trunk grows often the branches may get episodically redefined to take advantage of some/all new open source features in the trunk.

I can't think of another major industry hardware spec that has a structure like this. I can easily argue both advantages and disadvantages at this point. In the end it looks chaotic. Spec & design evolution uniqueness and chaos make the future difficult to predict. Is RISC-V going to become analogous to the current status of the FreeBSD project? Alive, but only relevant to companies producing proprietary software, or remain the focal point of RISC-V architecture development? I think the latter is required for long-term RISC-V success.

For years I have divided industry specifications into two categories, Class A and Class B. Class A specs are developed in a single team with a strong leader and final technical decision-maker, often in one company or group. Class B specs are developed in committees which vote on inclusions. Industry (trade organization) specifications often start out in Class A, and then transition to Class B, so I sometimes call them Class A/B.

Industry standard specs (IEEE, IETF, ANSI) are always in Class B. Some specs which stay in Class A are Linux (via the Linux foundation) and now RISC-V, in the RISC-V Foundation, but the RISC-V licensing puts RISC-V in a category of its own, in a way. It is unclear to me if RISC-V's licensing openness ends up fracturing the design in too many ways for broad success. We'll see over the next few years.

(My bias is decisively towards Class A specs. I think Class B and A/B specs always end up overly-complicated collections of incompatible business or political agendas, and then the architectures become unwieldy and obsolete, and often get replaced.)

Regarding the RISC-V software ecosystem, Google is strongly behind it, as evidenced by having Amber Huffman chair the RISE BoD:

Rise: RISC-V Software Ecosystem – Linux Foundation Project

riseproject.dev

Amber is very competent, and led the development of NVMe in Intel, and led its technical evolution in the industry in the NVMExpress industry group. So I have high hopes for the development of the RISC-V software ecosystem.

 

[KevinK]

Like Arm, RISC-V’s path to volume and success probably lies in working their way up from the bottom leveraging licensing flexibility and freedom from royalties. A lot is going to depend on whether RISC-V finds a killer app/platform like mobile, PC or server, that fosters creation of a large central ecosystem, or whether the flexibility of RISC-V fragments into many small-scale commercial and near-proprietary ecosystems. The problem with the Android analogy is that Android targeted a killer app / platform. It offers a combination of a commercial sponsor, plus ubiquity of the killer app / platform (Arm-based mobile) plus flexible licensing, enabled Android adoption in many adjacent platforms - smart TVs, streaming boxes, tablets, etc. Gonna be interesting.

 

[cliff]

RISC-V’s path to volume and success probably lies in working their way up from the bottom leveraging licensing flexibility and freedom from royalties.

It seems to both the top and bottom, correct? Google, FreeRTOS, etc from the top (the software side), and what you are saying about the bottom up is certainly true in our case. It is crazy for us not to jump on this at the bottom. In fact, we are now defining our digital EDA tools around OpenTitan and SRAM (soon MRAM or ReRAM) tiles. We have our own standard cell generators, so what the heck. I put an army of MSEEs on this. It is far better than what they learned in school, and it costs us almost nothing. I just have to pay for one employee to bop them in the head with an empty paper towel cardboard tube when they watch tiktok on their cellphones.

I was expecting Mr. Blue, who I view as the representitive of the digital establishment, to shoot RISC-V down. He didn't. As a naive member of the chaos (wild-west) group, I find this encouraging.

We will allow the customers to instantiate these CPU tiles for free, so I would think this would catch on for the IoT crowd.

 

[blueone]

I was expecting Mr. Blue, who I view as the representative of the digital establishment, to shoot RISC-V down. He didn't. As a naive member of the chaos (wild-west) group, I find this encouraging.

Fascinating. I didn't even know there was a "digital establishment", and not only am I a member of it, I'm a leading indicator of the group's agenda! I'm going to put this on my resume.

 

[cliff]

You are the leading member/bureaucrat of it. You were deputized over a year ago. Sorry, I must have forgot to tell you. Now you know.

 

[blueone]

You are the leading member/bureaucrat of it. You were deputized over a year ago. Sorry, I must have forgot to tell you. Now you know.

Wait... bureaucrat? I refuse the nomination.