In August 2014, the 20nm process technology, which was not progressing well, finally completed the adjustment in the new batch of Aries 800 trial production after the success of tape-out but could not solve the yield problem and caused mass production to skip tickets many times, and began to achieve the process of mass production of economic production yield.

The performance of the Aries 800 with the new process technology has been greatly improved, which stimulated the Aries project team to directly confirm the launch of a small model version of the Aries 810 with the original architecture.

In this way, they no longer have to worry about whether to choose the P1500 core of MIPS32 or the Q6442 core of MIPS64, the model based on the MIPS64 core will be the Aries 900!

Starting from Aries 800, the future will be a new cycle of R&D iteration with 2 years as a next generation and a year and a half as a small version of the asynchronous update model.

In this way, after waiting for the new generation of processors to come out, the slightly modified version of the older generation will automatically enter the mid-range market, imitating Qualcomm to form a high-end and high-end matching situation.

Of course, in the low-end market, Sun Mobile Computing will not get involved now and in the future, and that piece is left to Liu Ruoyi's MTK to toss.

However, the breakthrough of 20nm process technology is not only this benefit, but also the greater benefit is that from the perspective of process development, 20nm process technology has a strong and smooth transition to 16nm in terms of technical potential.

Therefore, after the completion of the mass production of the 20nm process technology, the research and development progress of the 16nm process technology has also become smooth, with Liang Jisong's optimistic judgment, as early as the end of 2014, the 16nm process technology of the Dragon factory is very likely to be successfully taped out.

In 2015, there will be an opportunity to complete the mass production of 16nm process technology!

At present, the main key equipment of the 20nm process process of the Dragon factory basically supports this upgrade, but the process needs to be readjusted to a certain extent.

The research and development of 14nm process technology and lower 10nm and 7nm process technology need to wait for the latest batch of equipment booked to carry out critical debugging and verification, which will be at least after the arrival of new equipment in 2015 to start these new process research and development, and can only do some preliminary technical pre-research work.

This progress is almost exactly the same as that of its competitor TMS.

Therefore, for semiconductor foundries, the general pattern of competition has not changed.

After TMS suffered a loss in the production capacity of the high-end process, they supplemented the capital expenditure and urgently booked a batch of advanced process technology main equipment, which will also arrive in 2015, when the production capacity of the high-end process technology, the two companies will be leveled on the same running line.

Overall, there should only be a few months left for the Dragon factory to occupy the advantage of high-end process capacity.

It's a pity that Zhang Rujin has not yet cut into the core customers of competitors, Qualcomm and Broadcom are not counted, when these two families have not grown up, they are only second-class customers in competitors.

Now they have grown up, and in the eyes of the outside world, they have grown up with the dragon factory of Zhongxin International, which has been refining in the 25nm process for a long time, and now they also have the right to use the 20nm process technology of the dragon factory for the first time.

In this regard, the Dragon Factory did not favor the Snapdragon 800 and Aries 800, and also ate the 20nm process high-end chip production orders of the two companies!

It shows the neutral position of Dragon Factory in terms of OEM - of course, it is also based on the position of making money, and now Qualcomm order OEM is a more profitable big business than Aries, after all, they sell more.

After the Snapdragon 800 was introduced into the 20nm process technology, the same performance and power consumption have made great progress, which makes the Aries 800 and Snapdragon 800 run together again!

Of course, for external mobile phone manufacturers, this progress has too many benefits for mobile phone manufacturers, and after switching to a new generation of process technology mass production chips, the reduction in power consumption will at least not be complained about using high-end mobile phones like hand warmers.

The high performance of high-end mobile phones must come at a cost in terms of heat dissipation and power consumption, whether it is the Aries 800 or the Snapdragon 800, there are basically no exceptions!

However, the benefits of mobile processor chips are far less than the benefits obtained by the chips in the north and south that are now jokingly called in the mainland.

Wu Zuhua, who followed up with M64-1620, changed to a 20-core design, and the GS 4A-4000 design goal was 20-core 2GHz main frequency, but it encountered mass production of 20nm process technology after it was taken out this year.

Gaosen Semiconductor, which only got the money after financing it, was originally designed according to the 25nm process process, but now it is directly throwing money into the 20nm process technology! Their amount of energy is less than that of mobile processors, and the cost of tape-out is indispensable.

However, it is very worthwhile to replace the new process technology, the design of the highest frequency is 2.3GHz chip, the result is directly soaring to 2.6GHz is still running steadily, and the power consumption is nearly 10% less than the original design power consumption!

Of course, this also shows that the scalability of the MIPS64 new core architecture is indeed excellent.

This allows Gaosen Semiconductor to obtain a low-power processor with mainstream performance. Although it is still not as good as the SPARC T3 in terms of performance, it is better because of its low power consumption, and the addition of DSP for signal processing and hardware encryption and decryption.

There may still be some gaps in performance in ordinary commercial applications, but for special industries and equipment applications, GS 4A-4000 can be said to be the only choice in China, far surpassing SPARC T3 in signal processing.

With GS 4A-4000, it can now be said that Gaosen Semiconductor has products to support high, medium and low in terms of general-purpose processors.

In contrast, Shenwei Technology is more sad.

After Sun Mobile Computing open-sourced SPARC T2, the main research and development of the chip is the research institute of Shenwei Technology and Toyo's FJT, and finally got the SPARC T3 with 16 cores and 128 threads.

However, at this stage, if the follow-up is a simple reactor core, both sides feel that there is little meaning, and the performance improvement space is limited, so the two families began to work together to decide to improve the kernel part, after all, after several years of exploration, Shenwei Technology has been familiar with the number of SPARC roads, and now there is also in-depth research and development towards the kernel design.

FJT is an old bird of the SPARC system, and the cooperation with Sun Mobile Computing technology can be traced back to more than ten years ago, and the two are now interested in working together to develop new kernels, which is naturally stronger than one.

Preparing for the newly developed SPARC core, the Toyo side is named Yaqi, and Shenwei Technology is called Fuxi internally, and both sides of the codename are unified S3 core, because this core is still ready to use a single-core super large number of 8-thread design, so take the S of SPARC, and the 3rd power of the traditional binary name 2 as the new kernel codename.

A new high-performance processor core is naturally not so well designed, and it took both parties almost a year to do this in terms of initial specifications and architecture alone, neither side is an original design, and the FJT side is better, they have been immersed in the SPARC architecture for a longer time and understand more clearly.

Although the theoretical knowledge of processor design is very deep, the experience in specific core design is quite scarce, which makes the design take a lot of detours at the beginning.

If it weren't for Zhao Mengguo's strong pressure and refusal to turn back, many people in the project team would want to continue to pile up to 20 or 24 cores on T3!

Anyway, from the perspective of SPARC's architecture, there is no problem at all according to the double-digit reactor core, and the access between the cores is peer-to-peer and peer-to-peer, so the performance effect can be seen immediately after the reactor core.

But Zhao Mengguo resolutely resisted the possibility of the project team going back, and in his opinion, it was not just because the T3 architecture had come to a way where the head could be seen at a glance.

More importantly, without the polishing of a new core design, the project team will not really grasp the design essence of the processor.

It just so happens that the Oriental side is interested in doing this together, and although the chip technology department of Sun Mobile Computing is not the main force involved, it also sends capable R&D to work together for the new core, and there is naturally Tony Tsai's promotion and participation behind Sun Mobile Computing's own commercial interests.

Now that SPARC is open, the market has not shrunk but has expanded, and Sun Mobile Computing will not easily give up this technologically advantageous market.

Zhao Mengguo feels that this is a once-in-a-lifetime good opportunity for Shenwei Technology, and if he misses this time, he may not have such a good opportunity again in the future! Working with the world's top processor design team is not an easy encounter.

"Even if you're crying, move forward!"

Zhao Mengguo mobilized the members of the project team.

After getting through this, we can really say that we are the ones who design processors.

I stumbled all the way and finally got the S3 core core, 256KB high-speed L2 cache and 4MB high-speed L3 cache, the main frequency started at 2.5GHz, and the maximum can reach 3.5GHz, and the theoretical value of single-core floating-point arithmetic performance exceeded 45GFLOPS.

This performance can be put out to develop single-core processors can enter the mainstream level.

Therefore, although the process of Shenwei Technology this time is very hard, the result is really very rewarding, and a group of R&D that was tortured to death is now all alive.

According to the plan, the first SPARC T4 based on S3 cores will be available in 4-, 8-, and 12-core models, with 8 S3 cores being the main model, and its total on-chip L3 cache will reach a staggering 32MB.

Of course, compared to the 80MB L3 cache of POWER7+ and the 96MB L3 cache of POWER8, SPARC T4 with the new S3 core is still a good win.

However, the design of up to 64 threads has a tendency to dominate multitasking, with a single core and 8 threads, which is unique to Sparc.

The final effect will not be felt until the beginning of 2015.

However, the maturity of the 20nm process technology provides excellent external processing conditions for the difficult process of SPARC T4, and the design of T4 with more than 1.5 billion transistors may be difficult to process well if the process technology is not enough.

If you want to achieve the best results, I'm afraid the 16nm process technology is the best partner for T4.

However, now it seems that the first batch of tape-outs cannot catch up with the 16nm process of the Dragon factory, and the Dragon factory has not yet completed this high-end process.

In any case, the 20nm process technology of the Dragon Factory has begun mass production, and the T4 of Shenwei Technology is at least very promising.

If this processor is successful, Shenwei Technology can basically catch up with IBM's footsteps in one step.

At that time, Shenwei Technology will not only be a staking in the domestic market, but also a step out of the country to fight with IBM.

As long as the performance is equal to the opponent, with the cost performance of Shenwei Technology, the price of a server with the same performance may be only a fraction of the price of the opponent, and even half of it cannot be reached.

If you go to the extreme, the pricing of the two sides can basically ignore the change in the exchange rate.

Basically, IBM uses servers with US dollar prices, and Shenwei Technology's servers with the same performance can be sold at the same price in RMB and make money!

With such a cost performance, it is not possible to say in the future that Shenwei Technology and Gaosen Semiconductor's server ecological company on the Huaxia side will make the server the same as the price of cabbage.

Now everyone knows that to be truly reliable, you must have your own hard power, and now Shenwei technology research has reached a new bottleneck, and the breakthrough is a new world, and together with Toyo's FJT, it has become a real new level force in the RISC camp.

For this reason, the two main forces are gritting their teeth and struggling, and the massive R&D expenses, in addition to the support of the two companies, will also be amortized in the future by charging a certain amount of fees for processors in the production of compatible SPARC servers of the Open Alliance.

This was a collective decision that was discussed by the SPARC Open Alliance, and it was clear to everyone that since they entered this door, it was a situation of both prosperity and loss.

Therefore, those alliance members who are temporarily insufficient in strength and cannot get involved in chip design and directly contribute can only pay later.

Fortunately, the money does not need to be given in advance, but when purchasing processors, the price of processors is increased in disguise and a proportional way is extracted and handed over, relatively speaking, the members of the alliance do not have to bear the loss of research and development failure, which is already quite cost-effective.

If the 20nm process of the Dragon factory is mature and mass-produced, as long as the production yield exceeds 50%, then the foundry cost of a SparcT4 processor will be mass-produced at the level of millions, and no matter how much a single processor will not exceed the production cost of 300 US dollars.

The $200,300 increase in this production cost is also far lower than the cost of competitors starting at $1,000.

That's a lot cheaper than competing processors that cost a few thousand dollars, so the title of the best price/performance ratio for the same performance of RISC is firmly on SPARC servers.

The server landscape has changed greatly, and it is difficult to say how it will evolve in the future, at least Huaxia has begun to have automatic derivation capabilities in terms of processors, and it is still difficult to say whether Toyo will be a top three in the future.

It's just that the days of the exclusive monopoly of the server processor Yankees have come to an end.

In the semiconductor industry, Liu Ruoyi set off this whirlwind that has changed because of me, from memory to mobile processors to server processors, from peripheral analog logic devices to display screens, the Chinese circle has become a majestic system, and the future will be the PC desktop.

Only, how far away will that day be?