After the layoffs, I became a tech godfather

Chapter 213 Chip Industry Summit Day - Two Types of Chips: Smart Thinking and Spatial Computing

The Linzhou High-tech Zone Innovation and Entrepreneurship Center covers a total area of nearly 3,000 acres, with green mountains and waters and winding rivers.

This was originally a garden-style enterprise entrepreneurship park planned by Linzhou City.

However, a garden is a sacred place in a densely populated and economically developed area. In the suburbs or rural areas, it is a barren land.

Just a few months ago, there was no life here. There were only a few companies in the entire park. Together with the surrounding new towns, it was called a ghost town.

This park was lucky, as it had just set up a small computing center at the time, which solved the urgent needs of Baiju Technology's initial entrepreneurship and attracted it here.

Today, three months later, the entire park has been taken over by Baiju Technology as its first R&D center and is named: Xiaosha Future Town.

From then on, every blade of grass and every tree here seemed to be more energetic, swaying in the sunshine.

The clear pond inside has also begun to gurgle, gushing out with infinite vitality.

The butterfly-shaped building next to the Qingquan Stone Platform is the town's comprehensive lecture hall that can accommodate 800 people at the same time. At this moment, the lecture hall is bustling with people.

"Everyone here today is our family. There is no outsider, so let's get straight to the point." This is Hao Cheng.

Unlike a public press conference in front of all consumers, today is the first day of the chip industry summit, and all the people present are old acquaintances.

Before the Chinese New Year, Hao Cheng met with the heads of leading chip industry companies such as Shanghai Micro, SMIC, Huawei, Moore's Law, and Unisoc, and reached a basic consensus.

Hao Cheng’s opening keynote speech at today’s summit came with four straightforward words: The Core of the Future.

On the large conference screen, a straight line runs across the center, dividing the entire screen into two parts: the upper part: the metacognitive chip, and the lower part: the traditional semiconductor chip.

Hao Cheng said: "The metacognitive framework was first announced at a winter press conference before the New Year. After the press conference, I met with my predecessors and had some exchanges with them.

"At that time, someone once asked me, 'Brother Cheng, the metacognitive chip is about to be launched. Will traditional semiconductor chips be completely useless after that? Will we all have to switch to designing chips based on metacognitive architecture and metacognitive instructions?'"

Hearing what Hao Cheng said, Zhang Jian, who was sitting in the audience, grinned and felt quite embarrassed.

He was the one who asked the question. It was nothing. Everyone had this question in their mind. However, he was fifteen years older than Hao Cheng, but he called Hao Cheng "Brother Cheng". This was a bit...

Hao Cheng also smiled: "I didn't answer him at that time, not for any other reason, but because he was 48 years old and called me brother, I didn't dare to respond!"

This was a joke of course, but at the same time, such a relaxed opening speech was used to set the tone for the summit over the next few days, telling everyone: relax, relax, and relax again.

During the discussion before the meeting, Hao Cheng discovered a problem with these companies: they were nervous.

Logically, all the people present here have seen the world, and are leaders in the field of science and technology, especially in their own professions. There are not many times when they feel nervous.

However, it is different for Baiju Technology. The chip with metacognitive architecture alone has the ability to shatter all their previous efforts.

And if they, especially state-owned enterprises, cannot occupy a place in Baiju Technology's next plan, the results that await them will not be good.

Last time, when we were tackling the software ecosystem construction problem, a large number of state-owned enterprises came with high arrogance, but returned dejected and defeated. Everyone here saw their final outcome.

Therefore, some business leaders are indeed very nervous.

But Hao Cheng's current attitude is obviously different from his attitude towards state-owned enterprises in the software industry. Some state-owned enterprises in the software industry are really mud. And in the chip industry, it's half and half.

Half of the people are working hard, while the other half are lagging behind at various times.

Hao Cheng is still willing to pull this half that is willing to fight and try again.

So, at the beginning, the tone was set: that is, cooperation is the basis, and resolving differences comes later. So, he started with a half-hearted joke, and finally added:
"This 48-year-old man is Zhang Jian, the founder of Moore Threads, Brother Jian!"

Looking at Hao Cheng, who looked like the boy next door, everyone laughed and felt much more relaxed.

"I was just joking, don't blame me, Brother Jian. Now, let's get to the point. What kind of relationship will metacognitive chips have with traditional chips in the future?

“Here, let me tell you: the power of metacognitive chips lies in their native compatibility with AI training, which can cultivate AI that is better at thinking and more sophisticated in thinking, such as Xiaosha and Xiaosha’s future.

"But if we only compare computing power, the metacognitive chip cannot compare with the traditional chip."

"Hmm?" Many people immediately said "hmm", which was a sound of incomprehension: This is not right, computing power, isn't that the basis of AI? If computing power is not comparable, how can AI surpass it?
"I see the confusion in your eyes," Hao Cheng continued, "In a word: everyone has been fooled too much by NVIDIA!

"You may not understand what I mean, but let me give you an example: between humans and computers, who has better thinking ability? In other words, who is more intelligent? It must be humans, right?

"But, who can calculate 10,000 times, 100,000 times, or 100 million times faster? This is obviously a machine, right? Even if a person spends his entire life, he may not be able to calculate 100 million times! This is the simplest and most plain logic!

"Traditional semiconductors are obviously better at computing, especially ultra-large-scale computing.

"Take the graphics card for example, its initial operation is graphics operation, which calculates the arrangement of pixels on our monitor one by one to form the images and pictures we see.

"In the future, we will make three-dimensional display cards. The coordinate calculation of spatial three-dimensional display technology requires not only metacognitive chips to provide smarter AI, but also the computing power of traditional semiconductors." What Hao Cheng said is easy to understand, but it subverts the past cognition a little bit -

Traditional cognition believes that artificial intelligence and computing power are strongly correlated. The emergence of deep learning and artificial neural network computing, and AI big models based on probabilistic computing and feedback regulation, directly and strongly correlate intelligence levels with computing power.

And with NVIDIA's further rendering, these two things are almost bound into one concept.

But Hao Cheng now says: These two are completely different things. At most, they have a little intersection in the field of traditional AI. He also uses Xiaosha as an example. Who can refute this?
"With strong computing power, the pixels and coordinates on a plane or in space can be calculated faster, so the display device can have a higher refresh rate, a higher frame rate, and better 3D and ray tracing effects."

Hao Cheng continued: “But this thing has nothing to do with AI.

"If AI wants to be more powerful, it needs a metacognitive chip. The computing speed of a metacognitive chip may not be that fast, but each calculation is more like an extension of thinking.

“So, computing power and AI have gone their separate ways since then.”

The conference screen, a very short animation, described these relationships vividly and introduced the two branches of three-dimensional processors in the field of computing power: intelligent thinking computing power and spatial computing computing power.

"Just like the current CPU and graphics card?" Seeing these two concepts, some people immediately linked them to the current concept of computers.

"Yes and no!" Hao Cheng explained:
"Spatial computing is indeed similar to current graphics cards, especially gaming cards. It calculates spatial coordinates and then shows us a three-dimensional, augmented reality, virtual reality, or even an immersive reality that can be felt by all five senses in the future. It is indeed equivalent to current graphics cards or GPUs."

Hao Cheng paused for a moment and continued, "This is the same part, but the ability of traditional GPUs to train AI is completely different for metacognitive chips. Pure calculations cannot train further AI, and three-dimensional chips cannot do that either. There is an upper limit to calculations!
"Therefore, in the future AI field, in addition to three-dimensional computing power, simple three-dimensional intelligence is also needed. In other words, these two types of chips will jointly achieve a leap forward in the future AI field."

"So, the difference between the two is that one is a game card and the other is an AI inference training card?" Some people insist on linking it to today's traditional chips, but this statement only sounds a bit similar, but in fact it is not the case.

Hao Cheng scratched his head and said: "Let's put it this way: in the field of metacognitive chips, the concepts of traditional computers such as CPU, graphics card, memory, hard disk, etc. will be blurred to a certain extent and will no longer be so clear.

"I won't explain the principle, but let me give you an example: Is your brain divided so finely? Which part is responsible for thinking, which part is responsible for memory, which part is responsible for spatial thinking, and which part is responsible for mathematical operations?
"Maybe there are parts, but each part does not exist independently, and the metacognitive chip will theoretically move in this direction."

"Buzz!" The brains of a group of people buzzed at the same time. They understood, but what was the meaning of using the human brain as an example? Was it just a random example or was there some deeper meaning?
If it were any other company, they wouldn’t think too much about it, and might even think that this person is showing off or bragging without thinking. But this is Baiju Technology, could it be that they are really doing some research on the brain?
Hao Cheng certainly didn't know what they were thinking, and he didn't explain. He continued with his original rhythm and began to introduce the metacognitive chip and the traditional chip field:
"Let's put the metacognitive chip aside for now. In the next two days, chip design companies, manufacturing companies, manufacturing equipment and industry chain companies will have special sub-venues to carry out their respective authorization and division of labor. Now, let's talk about the future of traditional chips under this line."

Traditional chips naturally refer to today's silicon semiconductors. Even people who know nothing about this industry have at least heard of the term "photolithography machine" thanks to America two years earlier.

Those who have a deeper understanding, following the development of the domestic chip industry and Huawei's efforts to conquer chip production, coupled with the publicity from various media from top to bottom, should be familiar with chip processes such as 14nm and 7nm, and know about EUV and DUV, advanced packaging, and multi-layer packaging.

Today, everyone sitting here is an industry insider, and they all know that a key big thing has been successfully developed! That’s right, it’s the EUV lithography machine!
When it comes to traditional chip manufacturing, this guy is absolutely indispensable.

Before this, there was only one company in the world that manufactured this thing: ASML in the Netherlands.

Of course, behind this is the support of parts from almost all Western powers such as America, Germany, England, including light sources, lenses, materials, cavities, etc.

Some people even assert that it is impossible for a single country to build an EUV lithography machine, let alone China, not to mention America!
However, this big guy is now lying quietly on the experimental production line of a chip factory in the southern region, and the trial production of chips has begun.

Everyone's eyes were focused on the Huawei representatives who came here today, Hu Kun, Yu Dong and He Gang.

Originally, the EUV lithography machine was not that fast, and it was estimated that it would take at least a year and a half to put it into mass production. However, with Xiaosha's help, all the problems were solved quickly.

In just one month, most of the problems were solved and trial production of the chip was achieved.

Although there are still some problems with the production line, such as instability, low yield, and small production capacity, as long as the first step is taken, the next step is just to follow the example and then optimize and expand production according to the Xiaosha Industrial Model.

The chip that is currently in trial production is the domestically produced EUV enhanced version of HiSilicon Kirin's 9020, which Huawei internally named 9020Super.

Compared with the version engraved by DUV lithography machine, the performance of 9020Super is almost doubled.

This is just the effect of the trial production version. Once everything is stable, the performance can be further improved.

The planned computer CPU chip can also be put on the agenda.

Of course, this was the previous plan, and now we have to look at the relevant plans of Baiju Technology and then decide what to do.

Hao Cheng continued his opening speech, the theme of the future core:
"According to Xiaosha's assessment, in the future, traditional semiconductor chips will still occupy half of all chips. We cannot, do not need to, and should not use metacognitive chips for all chips. Even if metacognitive chips are very smart, we still need to make a trade-off between intelligence and extreme speed.

"Therefore, the combination of metacognition and traditional instruction chips will be the norm in the chip industry for a long time in the future."

Hao Cheng then proposed how to better design our CPU, GPU, storage, memory, chiplets, sampling analog chips, etc. with the help of Xiaosha in the future.

This is also the part that Loongson, Zhaoxin, Unigroup Zhanrui, Moore Thread and Huawei are most interested in at this time - because the metacognitive chip will definitely be dominated by Baiju Technology itself, while the more powerful traditional chips will be left to them.

Especially when the EUV lithography machine test has been successful and the conditions for mass production are already in place, according to Hao Cheng, there is great potential for traditional semiconductors, which account for half of the future.

There is one more important thing - who says traditional chips are really unchanging?

What they need to do is to use the support of our metacognitive chip and the assistance of Xiaosha to innovate the traditional chip, and then achieve the ultimate and take it into their pockets!

(End of this chapter)