The Development Of Software Also Promotes The Innovation Of Hardware

      The development of software also promotes the innovation of hardware.

      With the application and development of AI technology, CPU is more and more difficult to meet the demand of computing power, and AI chips emerge spontaneously. In recent three or four years, there has also been a boom of AI chip entrepreneurship in China, and a number of gazelle enterprises have been emerging.

Generally speaking, AI chips refer to chips that have been specially designed to accelerate AI algorithms. They are also known as AI accelerators or computing cards. They are specially used to process a large number of computing tasks in AI applications (other non computational tasks are still handled by CPU). At present, AI chips are mainly divided into GPU, FPGA and ASIC.

Currently, there are two world-class GPU "big factories" in the world, namely, NVIDIA, the "number one" and AMD, the "second" respectively. Their core GPU R & D team also remains in the United States. At the same time, because NVIDIA's main customer group is not in China, their service and support for domestic customers are still imperfect, which also leaves room for domestic innovative enterprises.

At present, there are many competitors in the domestic AI chip track. Although they were established only a few years ago, most of them have obtained tens of millions, hundreds of millions and even billions of accumulated financing. Some enterprises have mass production of chip products, while others have not been seen. Most of them are still striding over the "Death Valley" and growing towards the target of gazelle enterprises.

AI chip is still in its infancy

Financing is not difficult for the enterprises that develop AI chips and GPU chips. Money is pouring into the track.

With a break, tiantianzhixin, Denglin technology, Biren technology, Suiyuan technology, Cambrian, Kunyun technology, Yitu technology have gathered in the track, and they are also seed players of gazelle enterprises in the track.

In March this year, days smart chip completed the C round of 1.2 billion yuan of financing. In the same month, Bi Ren technology completed the B round of financing, with a total financing amount of more than 4.7 billion yuan since its establishment in 2019. Muxi also announced the completion of the pre a + round of financing in March.

Earlier, in February, Denglin technology has completed a + round of financing. In January, Suiyuan technology announced the completion of 1.8 billion yuan of c-theory financing. In August last year, Kunyun technology completed a + round of financing with tens of millions of yuan. In addition, the Cambrian science and technology innovation board was successfully listed last year.

In the opinion of Zhang Xiaorong, President of the Institute of deep science and technology, AI chip market is relatively special, many of which belong to professional chips. Only when combined with specific algorithms can they be put into the market. He told the 21st century economic report that at present, the industrialization and marketization ability of AI technology is relatively weak, and there is no actual large-scale use, and the AI chip is also in the primary development stage.

At present, the companies that have mass produced AI chips mainly include Huawei, Cambrian and Suiyuan technology. Among them, Huawei's AI chip did not iterate to push new products.

In February this year, the Cambrian cloud smart chip Siyuan 290 made its debut after mass production. Siyuan 290 is also the first AI training chip in Cambrian, and the AI inference chip was mainly introduced in Cambrian. AI training chip has higher technical requirements than reasoning chip.

Suiyuan technology has completed the iteration of two generations of products (including training and reasoning chips) more than three years after its establishment. In September 2020, Suiyuan technology announced that the first generation of artificial intelligence training accelerator card "yunsui T10" based on the version 1.0 of Shensi chip has officially entered the commercial stage. In addition, it also released version 2.0 in July this year, which is expected to be in mass production by the end of the year.

Semiconductor industry is known as "the boss eat meat, the second drink soup, the third dry stare". According to Zhao Lidong, founder and CEO of Shanghai Suiyuan Technology Co., Ltd., the survival rate of start-up enterprises can reach 1 in 10, which is a very high proportion. It is inevitable that the waves will scour the sand. He told the 21st century economic reporter that the tide will recede in the next two or three years and there will be a turn of dawn.

Who is the real gazelle in the track when the bubble is fading away? The answer will come out.

Three directions in the future

It is understood that some experts in the industry believe that cloud chips have reached the stage of delivery to customers, and the pressure of latecomers will be great. If you want to make the first chip solid, it usually takes two or three years; If you do something casually, when you see the real move, it will be more and more difficult to finance later.

For AI chip racetrack companies, it is more important than financing figures whether they can come up with products to speak, which will determine whether they can cross the "Death Valley" and enter the rapid growth period of gazelle enterprises.

At present, the products of Cambrian, Suiyuan technology and other companies have been mass produced for commercial use, and the chips of tiantianzhixin and other companies have been taped, and it is announced that the products will be mass produced this year. There are also some enterprises that have not yet produced chips, so there is not much time left for them at present, unless the performance or cost performance of their products can achieve a great breakthrough.

In fact, Zhao Lidong mentioned a syllogism of enterprise development“ The first stage is to make the chip according to the plan. The threshold of chip technology is very high, and it is not easy to realize the first stage, but it is actually the simplest of the three stages - everything is controlled within the company. ". The second stage, he added, is whether anyone is willing to pay for the chip; The third stage is how to scale up the business and ultimately achieve profitability. "Now many companies are still in the first stage, and we are in the early stage of the third stage.".

How to find and expand the market of AI chips is the concern of enterprises at the front of this track, and it is also related to whether they can successfully leap over the "valley of death" and become the "Gazelle" to achieve self blood production. Next, regions with market potential will also become destinations for these enterprises, such as the Pearl River Delta, Yangtze River Delta and Beijing, which are still popular choices.

Where is the AI chip market? Taking Suiyuan science and technology as an example, Zhao Lidong summed up three directions. He told the 21st century economic report that the first direction is Internet enterprises. "So far, more than 70% of the market for our products (whether training or reasoning chips) is still composed of Internet enterprises, but the markets in the other two directions are expanding more rapidly.".

The second direction is vertical industry. For example, the financial industry has begun to apply AI technology in daily business, as well as transportation, medical care, education, communication and other industries. With the application of AI, the market has begun to "rise".

"A lot of traditional industries haven't started using [AI chips], which is a good thing for us." Zhao Lidong said that this also means that they are not using other products. At this time, we don't need any migration costs to choose us. We can also "start from scratch" with our customers and even make customized solutions.

In the third direction, Zhao Lidong focuses on the "new infrastructure" in a broad sense, including the computing power center of urban construction, the brain of smart city, and so on. In some fields, it is best to use domestic products. This is a blue ocean for the innovative enterprises on the domestic AI chip track, and a great opportunity for the seed players of gazelle enterprises.