Introduction to New JiwonAt present, the dilemma of “lack of core and soul” still exists in China’s chip industry. Risc-v is undoubtedly a “dawn” in order to solve the “bottleneck” problem faced by China’s chip field. Recently, Zhongguancun Haidian Park united with new wisdom yuan jointly held to “RISC-V industrial application and ecological construction” as the theme of the closed-door forum, to listen to the views of experts.

At present, China’s chip industry is booming, technology has achieved multi-point breakthroughs, industry chain has been comprehensively upgraded, but the core product innovation is weak, “lack of core and soul” dilemma still exists. In fact, semiconductor companies such as Nvidia and TSMC benefited from open source chip projects in their start-up stages. During the semiconductor war with Japan in the 1980s, the United States launched a project called MOSIS project in DARPA, which lowered the threshold of chip design and enabled universities to make chips and streaming chips. Nvidia and Qualcomm were incubated on MOSIS. Chip development takes time and money. From EDA, server, IP purchase to streaming video, research and development expenditure of hundreds of millions, large investment, high risk, general companies can not afford, especially need open source project support.

Under the demand of global industry and Chinese integrated circuit market, RISC-V is like a “dawn”, “open ecology, good technical foundation and scalability” provides a solid technical foundation and independent and controllable development opportunity for the research of domestic chips, RISC-V has gradually become the focus of the industry. Not only Intel and other technology giants have announced their support for RISC-V, universities have also used RISC-V for teaching, and a large number of start-up technology companies related to risC-V have emerged, injecting fresh vitality into the independent and controllable development of information industry, in order to better establish a set of ecosystem based on a new instruction set. In order to coordinate the industrial application of RISC-V and boost risC-V industrial ecological construction, Zhongguancun Haidian Park and New Zhiyuan jointly held a closed-door forum themed “RISC-V Industrial Application and Ecological Construction” on August 13, 2019.

Giudice lc
Shu BiLei
BaoYunGang
Alvin chan
Peng ouyang
Huang Chang
Rock Yang
Zhou Haojie

On the forum, guests shared their views on RISC-V and related open source technology cutting-edge trends, let’s listen to what guests say. ** Led by the closed-door forum, ** pushes RISC-V forward

First of all, Shu Bilei delivered a speech, he said: Beijing is very concerned about RISC-V, but to support the region including the whole Beijing RISC-V ecological building can be done? What do you do? Will it be affected? What can our own enterprises do? Can the RISC-V alliance do something public and build ecology? This forum is just to listen to everyone’s opinions, with the closed-door forum as the introduction, I hope there will be some follow-up measures to support this matter.

The root cause of China’s chip “bottleneck” problem lies in the serious shortage of talent reserve

Bao yungang shared the two key words behind RISC-V, “open source” and “Agile development”. Today, Moore’s Law is slowing down. Moore’s Law allows us to keep increasing the number of transistors on chips, but we don’t use it to its full capacity.

A team at MIT has done a study, matrix multiplication is written by an ordinary programmer in Python, its performance is 1, if written in C language can be improved 47 times, and then written by a parallel person can be improved 7 times, if you know a little bit of memory structure, memory optimization can be improved 20 times, With parallel instructions and so on, the multiplication can be improved by 9 times, and the program written by a person who really knows the architecture will be 63,000 times worse than that written by an ordinary programmer. This shows that the programmer has not really given full play to the performance of the whole chip.

On the other hand, if you understand the architecture of chips, if you can give full play to the efficiency of transistors, then today’s chips still have a lot of room to explore. The chip gives so much power that the average programmer doesn’t make the most of it. How to solve this problem? Bao shared two approaches: the first is to hire very experienced programmers, or to build expert knowledge into the hardware; The second approach is to lower the bar for chip design.

The first
The second
The third
Three steps
The first step
The second step
The third step

Serve. v has been fully open source on the iHUB hosting platform of Pengcheng Lab, interested friends can try. (code.ihub.org.cn/projects/15… ** Zhijian Chen: ** Brother Flathead takes a commercial approach to reduce the cost of the entire chip development

Different from PC and smart phone in Internet era and mobile Internet era, chips in AIoT era are typical fragmented non-standard products, and only chips deeply integrated with applications are competitive chips in AIoT era. At the same time, AIoT applications are very sensitive to the price of chips. Only PRatt & Whitney AIoT can be widely used in all aspects of life to promote the digitalization and intelligence of existing equipment. Chen shared what Buzz is doing to position itself as the infrastructure provider for Pratt & Whitney chips in the ERA of AIoT.

Then why do you want to make “black iron”? “If you compare ARM’s product line with risC-V’s existing product line, you can see that in the middle and high-end segment, RISC-V is missing,” Chen said. Risc-v tends to give the outside world an impression that it can only do low-end, which is detrimental to the commercial application of RISC-V. Therefore, He hopes to make RISC-V to lower the threshold for users to obtain high-performance CPU cores. He hopes to expand the application boundary of the whole RISC-V architecture through this move, and make up for the blank of RISC-V in high-end applications. ** AI chips need to be flexible and energy efficient

Ouyang Peng shares reconfigurable computer chips and their applications. Ai computing is gradually sinking from the cloud to edge devices and iot nodes, forming an integrated intelligence of cloud and end. Shipments of deep learning chipsets are forecast to grow from 164.9 million in 2018 to more than 2.9 billion in 2025. The global market for deep learning chipsets will reach $72.6 billion by 2025.

The first
The second
The third

In the second half of the conference, Chaired by Ms. Yang Jing, the participants had a heated discussion on the ecology and layout of RISC-V in China, the ecology of self-developed chips in China, the role of open source and other issues.

I would like to consult you.

Yy: Chuangda makes the operating system, and we are also in the RISC-V alliance. The operating system is the closest layer of software to the chip, and many of our customer partners are chip companies. The biggest challenge of RISC-V is the software ecosystem, which still needs to be accumulated for a new architecture. If many companies use RISC-V chips, software applications can do their own vertical integration, as far as possible can be done by themselves, RISC-V opportunities are relatively large. Risc-v started with small chips for the Internet of Things, so now is probably the best opportunity for RISC-V. * * * *

Now all the other countries in the world chip companies are fewer and fewer, only the Chinese chip design companies are more and more

Second, the promotion of RISC-V should not be limited to chip companies, but to application companies. In the early stage of product planning, performance planning and function planning by terminal manufacturers and application manufacturers, RISC-V should be taken into account so as to boost the demand of chip manufacturers to choose RISC-V.

** Yang Jing: Can **RISC-V play a certain role in self-research chips or the autonomous ecology of chips?

** We are one of several companies in China to make our own chips. Loongson is relatively early. It is true that we take out chips and deliver them for use. The back of the flathead brother also up, Huawei also has the Haisi series. But we do chip is military engineering, three years of generation installed, unlike many companies.

In this era of RISC-V, there is such a large instruction set ecology, which will greatly promote the chip ecology in China. The trouble is also here. Several domestic manufacturers make autonomous controllable processors, but the instruction set of SMIC and Loongson is limited and they do not upgrade instructions. ARM has commercial support to upgrade instructions. Although different users collectively called RISC-V system, but different users on the function, performance, power consumption difference is very large.

** Yang Jing: ** Just now Bao mentioned that risC-V has an open source EDA. From the point of view of algorithms, can open source EDA solve the problem?

** Direct replacement of open source EDA is certainly unlikely in the short term. But people realized that EDA could be open source. There’s a professor at Stanford who says that the ultimate form of an open source chip is not a chip, that what comes out of an open source chip is just a bunch of software, and ultimately it’s a set of tool chains that help chip design, more like scaffolding for building a house. How to make these tools more accessible, that’s one aspect. Quality issues can be solved over time, and there are no purely technical challenges.

Second, another problem with open source EDA is that companies like RedHat will come along in the future and integrate open source stuff into it. For example, Linux, we all say that you can use or do all kinds of development in the enterprise, but the fortune 500 enterprises put so much data on an open source software, the whole business is built on the open source software, there is a little worry. But RedHat lets you buy services directly from it, and RedHat is growing fast. The open source business model also has ways of ensuring quality.

** Yang Jing: **** Now there is a worry, not only China to form a different technical system, RISC-V also allows users to add their own instructions, will it be fragmented? Is there a standard for software?

** China must have companies to compete with SiFive. It’s a war in two dimensions. It has commercial companies on its side, and we should have our own to compete with it. It’s a frontal battlefield. It’s been trying to get inside. In the United States, chip engineers are all in their 50s and 60s. In China, they are all in their 20s and 30s. The future must be in China.

End the “seven countries and eight systems” and jointly build RISC-V ecology

At the end of the forum, Ms. Yang Jing made a summary: RISC-V needs to establish an ecosystem, and developers need to see the commercial potential, while the most decisive aspect of the market-oriented ecosystem is the application scenario and user demand, which is rigid and pushes back to the AI hardware and software industry chain, and drives the surrounding developers. The only commercial success with Linux is Red Hat, and it took Arm 30 years to build a software ecosystem to prove it. Risc-v has just started in China, and the ecosystem of “seven countries and eight systems” among different stakeholders needs to be converged. The vitality and competitiveness of the whole ecosystem also need to be tested by the market.

The long march has just started. We look forward to more enterprises and developers paying attention to RISC-V and building an open source ecosystem of Chinese AI core together.