Rewriting the Technological Scenario

After the depth learning framework is naturally ai chip, or the host is first spread, "the explosion of deep learning is inseparable from the depth learning architecture, and it is also inseparable from the AI ​​chip.

It can be said that the extent of deep learning broke out is a depth learning framework, while the beginning of deep learning broke out is the GPU platform established by the Great Breeze Group, and the GPU pushes GPU to the depth study.

The advantages of GPU in deep learning areas must have to say more. Today, GPU is still the most common depth learning arithmetic unit.

But we have also seen that with the sustainable development of deep learning, the limitations of GPU have become increasingly high. Deep learning includes two computing links, GPUs are very efficient in deep learning algorithm training, but in the application, it is not possible to show the strength advantage. At the same time, GPUs are unable to flexibly configure hardware structures, and the market needs new development.

This has the current two AI chip new fields, FPGA and ASIC, FPGA, with its programmable dedication and high performance, and ASIC is based primarily on the device, while ASIC has a small volume, low power, low cost. Highly confidentiality and other features, more use in the consumer terminal.

From the current business style, Galaxy Technology is the biggest player in the GPU field, and Ying Weida follows.

ASIC leaders are wind semiconductors. The main players include Huawei, Google, Sony, Ali.

The FPGA leaders are Intel, and the main players include Actel, Xilinx, National Microelectronics, etc. "

"It seems that GPU and ASIC are in Huaxia in leaching, and the rice is currently only possessed in FPGA." The stage begins to discuss.

"GPU is current mainstream, ASIC is a trend of future, and FPGA feels in the middle."

"ASIC has a road to walk, and FPGA is very fierce in the field of cloud computing, first grabs the market, and can be monopoly."

"The biggest problem in ASIC is on special limitations, but Huaxia Enterprises as long as they have enough to force enough, to make metivided labor as deep learning frameworks, this problem is to solve with the most primitive approach.

You can don't forget, although rice is the core technical player of FPGA, but Huaxia is the largest market in FPGA. As long as Huaxia Enterprise can produce the chip in various fields, the entire market can abandon FPGA, of course, the premise is if there is this necessary if. "

The stage is discussed, and it has also been made on the stage. Like the introduction of the depth learning framework, the host is finished, and the representatives of each company have begun to speaking on their own products and industry development.

Huaxia Enterprises still occupy the home advantage, most of them are Chinese business representatives, and everyone will disclose some common information in their respective speeches, such as the AI ​​chip will have great dependence on the development.

People understand it naturally.

The common point, more concerned, is, Huaxia Enterprises in the group of multiple AI chip fields.

Among them, there is no more popular life, and there is no AI chip burst in China, and 9 Chinese companies such as wind, Huawei, Xiaomi, ZTE, Haier have launched their own wearable AI chips in 2018.

"This time I can wear the AI ​​chip's outbreak, is it to expose the wind group AR glasses to the consumer market?"

"Instead of driving AR glasses, the growth rate of wearable smart equipment has been more than 30% for 3 consecutive years, which can be a huge future trend, and now wear Ai chip is actually rounded by Huaxia Enterprise."

"Is there any entry in this field of rice companies?"

"It seems that there is no, this can be taken by Huaxia."

"There is also a focus can't be ignored. All the wearing AI chips are all Hongmeng architectures. This is to make an ecological paving. Huaxia's wear AI chip is such a burst, and the end products have been on the AI ​​chip. This ecology is completely over Huaxia. "

"The wind group will work hard, when can I have a higher power with Intel in the traditional CPU process, and I really can complete Intel to pick up my horses in the process of the wind group."

"The Great Breeze Group has been flattened with the advantages of the process process. It is not very good to catch up with the Britkes. Is it not very good to respond?"

"The overall process should still have a gap? No one is unclear now, the two companies say that there is water."

In the discussion under the stage, the Great wind semiconductor Liang Mengong hosted the stage.

Liang Mengong also introduced the main chip development in the AI ​​semiconductor field, and then began today's last topic. Artificial intelligence development requires a new hardware architecture to meet the interactive needs of the exponential growth, which is the hardware revolution we mentioned before.

This hardware revolution clearly has opened a big scene, and in the process of driving the hardware revolution forward, everyone has encountered a problem.

For a long time, the design of the chip is consumed, which is seriously dragged down the iteration speed of the chip, which is why semiconductor companies will have a five-year seven years or even ten years of development blueprints. "

When I said this, Liang Mengong suddenly paused, "We have such a high wind semiconductor so that this tree is so important in the semiconductor field. It is important for our premieration of the development of the semiconductor industry.

This is also the reason why some semiconductor companies have always wanted to monopolize this industry. Because the more monopoly, the more accurately predicted the future, after all, for monopolists, the future development is controlled.

However, for fast outbreak of artificial intelligence, there is no so-called monopoly semiconductor company, two products iteration is too fast, and the planning of five years will not adapt to artificial intelligence.

Therefore, for the development of the AI ​​chip, a must have a way to compress design time, and for human experts, in the face of increasingly complex chip evolution, there is no ability to compress design time until we found an interesting Technical development direction.

That is to use deep learning to drive the hardware revolution.

Yes, you have not heard the wrong, hardware revolution promotes the development of deep learning, and the development of deep learning has also driven hardware revolution, which is what I said interesting.

We call it: AI symbiosis development technology. "

Liang Mengong began to show a document. "We use deep learning, through the study experience of learning past chip, let the machine to participate in the chip update optimization.

Our earliest experiment is global wiring, because this is the most complex and time-consuming phase in chip design. We use the depth learning algorithm to learn from the past layout, success for the never seen new scales. Generating an optimized chip design, more importantly, we compress 10 times time. "

Liang Mengong began to interpret the document, and the data of a lot of detailed to learn anti-propulsion chip development was revealed.

"In the near future, we will be very likely that I have just heard 3NM production in the first half of the year. I heard 2 nm mass production in the second half of the year. After a year, I will come, I believe this future. Soon will come.

Moreover, this technique breaks through the compressed chip design cycle, but also helps small and beautiful companies develop special AI chips with a relatively low cost to further promote the large outbreak of special chips.

We will provide this technology support from the same day to the global enterprise, which is the development of the AI ​​chip. "

...

"I will come to the current situation, from the underlying chip to the deep learning framework, then go to General AI technology, and finally go to artificial intelligence applications, Huaxia is in the development of artificial intelligence, this line is open."

"And you will come to a industry, it is an industrial chain. When you take chips, design, manufacture, materials, crafts, now there is also ai symbiotic development technology, close in the hardware revolution in front of you ...

Intel's craftsmanship is afraid that it is no longer possible to catch up with the wind group. "

"Don't be chasing it, you can't open Intel."

"This trick is a bit powerful. Through the ai chip, the hardware revolution is protected, and the Moore's law is directly ended, and the Moore's monopoly has been ever."

"I am a little doubtful now, as far as the technology level and market share of Huaxia, where is the bottom gas of the company to limit them?

Huaxia Enterprises do not limit them, they will thank Xie Xie. "

...