Chapter 289 Technical Route

Jingu Xiqing Science and Technology Development Zone.

This is one of Huaxin International's foundries, and it is also the first foundry to complete the upgrade and transformation of the 28nm high-yield production line.

Wang Minjiang, led by the general manager and chief engineer of the branch here, came to the storage place of the final product. The spacious room was filled with the unique smell of PCB boards.

There were several black shelves in the warehouse. The worker took a tray from one of the small grids and handed it to Wang Minjiang.

"Mr. Wang, this is your Y1611A official version, the 'Yawu A1000' chip. It has been less than 20 hours since it was packaged and placed here. Take a look."

Wang Minjiang pried out a processor about 6cm square and put it under the light for careful observation.

The front of the chip is completely covered by a nickel-plated copper top cover, and the back is 3900 pure gold contacts, that is, gold fingers.

The first batch of chips from Yellow River Semiconductor is still based on X32115 to deeply optimize and improve the desktop side. It is a new generation of "Yaowu" processors designed with the "Taishan" architecture. The internal code names are Y1611A, Y1611B, and Y1611C.

Y is the first letter of the English letter of Yellow River, 16 represents the year, the third number is the number according to the series name of the year, and the fourth number is the order of the version iteration.

For example, the Y1601 used to support the Great Wall Computer, 0 is the code name of the "Qiongqi" series.

A, B, and C represent the performance level respectively. A is the workstation level of the consumer market, B is the fever level, and C is the economical level. There may be other letters in the future, and the numbers following are the specific distinctions of CPUs at all levels.

Y1611A is the flagship CPU released for the desktop market, but after its release, it will be called TPU, and T is the first letter of the English letter of ternary.

Because it is very different from the ternary architecture and the traditional architecture, the functional division is also very different.

The original computer was mainly composed of three parts: processor, memory, and hard disk. The screen output was hard-decoded by the CPU. Later, independent graphics cards, GPUs, were introduced to enhance graphics computing capabilities.

The ternary architecture directly does two jobs. The same architecture and design ideas can provide considerable computing power support for a variety of scenarios.

The core area of ​​​​Yuewu A1000 is nearly 120% larger than that of the mainstream CPU on the market. The rated power consumption is also not low, reaching 350W. Although it is indeed very high, because it does not require a graphics card, it can actually save more power than a normal computer with an independent graphics card.

The ​​Yuewu A1000 has 6 cores and 18 threads, which is comparable to Intel's flagship i7 6950X. At the same time, the graphics computing capability is super strong, and it can be close to the level of GTX1080ti when it comes to pure graphics computing.

If you want to achieve the ultimate performance in both aspects at the same time, you need the automatic overclocking program equipped with the A1000. Overclocking releases about 600w of power, and can reach a maximum of 800W under good heat dissipation conditions.

It is high, but it is indeed very useful, and this is still compared in traditional projects. It is even stronger with XOS and professional software, and there is AI hardware support that mainstream manufacturers have not even touched upon. This is a major plus.

The main purpose of the A1000 is to show off its muscles on the consumer side and sell it to enthusiast users and workstation users. Of course, the latter are all incidental.

Now the domestic ternary ecosystem has just begun to be cultivated. Huawei will release the world's first ternary mobile phone mate10 in a few days, and at the same time, Huanghe Semiconductor will officially promote the program editor of the XOS system to domestic companies.

The Science and Technology Commission has organized all the translations of basic office software, social and life software, but it is very difficult to translate huge entertainment software, especially games. The Science and Technology Commission plans to use 1 year to vigorously promote the domestic software industry to enter ternary development.

For this reason, the authorities have even made up their minds to issue relevant laws on software in the new year, vigorously supporting the ternary system while resisting foreign pressure.

This is not just the business of Huanghe Semiconductor, but the work of the entire domestic semiconductor software and hardware industry chain.

So Wang Minjiang has no worries about hardware. Xinyuan Company has already produced X32045, which is more advanced than the system X32035, and X32055 is also being designed. Each generation is stronger than the previous one. Huanghe Semiconductor only needs to squeeze a little bit from time to time, and X32035 will be enough for ten years.

Putting the TPU back in his hand, Wang Minjiang said:

"We are not worried about the hardware. Can the cost of solid-state drives be reduced?"

"This..."

The chief engineer of Huaxin was a little embarrassed, but still told the truth.

"Binary NAND flash memory is 2.8 yuan per GB, and ternary flash memory is 8.9 yuan per GB. At least we have no choice."

Wang Minjiang: ...

One of the current weaknesses of ternary computers: storage.

Now it is the end of 2016, and the mainstream storage medium in the market is still mechanical hard disks, and solid-state drives are expensive and have small capacity.

Whether it is a traditional mechanical hard drive or a solid-state drive, the ternary computer reading must be translated, which will greatly increase the delay and reduce the read/write speed.

NAND flash memory solid-state drives are okay, after all, they are fast. There will be a little performance loss after the transfer, but it is still fast, and the loss is acceptable. Mechanical hard drives are too uncomfortable. Not only is the delay sky-high, but the already low read and write speeds are also discounted.

The dedicated ternary particle solid-state drive is suitable for computers with fast response, fast reading and writing, and long lifespan, but the price is naturally more exaggerated.

In the era of cars and machines, if you can use ternary memory or solid-state hard drives, you don't care about the cost of hundreds or thousands of dollars, but this is not the case on the consumer side.

The storage space of mainstream computers is more than 500G, which means spending more than 1,000 to buy a hard drive? I'm afraid most people won't be able to accept it, but the experience of mechanical hard drives is indeed a bit impatient.

Unless NAND flash memory technology advances or ternary memory particles are mass-produced in seven or eight years, it will be possible to drive down the price of storage equipment.

Wang Minjiang has no way to do this. The cost of chips in Xinyuan's business is very low, and there has been no relevant research.

Just when he was at a loss what to do, the chief engineer of Huaxin said something unintentionally:

"Mr. Wang, don't be so anxious. Nothing is perfect, and current silicon wafers are not suitable for ternary processing."

Wang Minjiang:!

"What do you mean? You are looking for other technical routes."

The chief engineer nodded:

"It is too wasteful to use two binary components to simulate a ternary field effect transistor. From this point of view, there is no way to compromise. In order to adapt to the existing system, we must find a substrate that has three states as a native support for ternary." chip material.

The Chinese Academy of Sciences has been studying it for a long time. Their current favorite is silicon carbide. Both elements are relatively common and it is easier to obtain high-purity elements. They also meet the requirements of having three states.

We and Modu Microelectronics are all involved. Silicon carbide manufacturing chips need to restructure the industry line. Currently, they plan to complete this road within five years. In a few days, they should bring you in to work together. . "

After the chief engineer finished speaking, he found that Wang Minjiang's eyes were very strange, and the expression on his face was that he was struggling to say something but stopped. Finally, he patted the chief engineer on the shoulder and said solemnly:

"Your innovative spirit is very good. Huanghe Semiconductor will definitely support it fully. Silicon carbide is good, so let's do this."

The chief engineer of Huaxin was a little confused: "Actually, it's not bad, because in the past few years, silicon carbide IGBT high-power semiconductors have been developed, so I naturally thought of it. Most of them are existing ideas."

Wang Minjiang: "That's right. Since silicon carbide chips already have a foundation, I suggest starting from this. The progress should be much faster."