What kind of processor is Hisilicon Kirin 950? What are the disadvantages?

Judging the quality of a mobile SoC chip mainly depends on three points-architecture, manufacturing process and integrated GPU. Based on these three points, on 20 15 16543815, Huawei officially released a new generation of flagship chip Kirin 950, which can be regarded as a salted fish of Haisi.

Huawei said at the meeting that the Kirin 950 used the ARM A72+ A53 architecture for the first time in the industry, and finally replaced the ancestral Mali T628, which was used in June of 20 14, with MaliT880.

Most importantly, Hisilicon adopted TSMC's 16FF+ process technology, and finally successfully caught up with Samsung's and Qualcomm's 14nm FinFET process technology.

Throughout the development of Hayes, it seems that there is always a word "late"

Since the first "high-end" (at least officially claimed) smart phone chip K3V2 came out, each generation of Hisilicon products will be slower than other chip factories in terms of technology, architecture and GPU.

On June 20 12, K3V2 adopted TSMC's 40nm process technology. Although the technology was still in its maturity at that time, the mobile processor itself changed very quickly. Judging from the wrong situation, Hess chose to wait for TSMC's 28 nm process technology.

However, TSMC's initial capacity at 28nm was insufficient, and the orders were completely occupied by Qualcomm, NVIDIA and AMD, resulting in two years. In the past two years, the 28 nm processor market has been seized by Qualcomm and MediaTek. On the other hand, Samsung's processor decided to use 32nm technology first, and Galaxy S III and Note 2 immediately suppressed a number of 40nm processor products in the market. K3V2 also got the reputation of "ancestral chip".

That's not all. June 2065438 +0654381October +20654381October+65438 in 2005, it was precisely because of the insufficient capacity of TSMC's 20nm process that it was completely divided between Apple and Qualcomm that Hess could not get involved. In addition, 16nm bounced again and again, and Hess was stuck in the process of 28nm again. If you are wronged, Hess really has a fight with the hammer.

Then, will the tragedy of the ancestral SoC be repeated on Kirin 950?

In this year, independent research and development architecture will become a trend. Now the news shows that the exposed Qualcomm Xiaolong 820 has started to use its own 64-bit architecture "kryo", and Samsung's next-generation Exynos M 1 will also use its own architecture "Mongoose" for the first time.

Obviously, the research and development of independent architecture will become the core competitiveness of the next generation of mobile chips.

Although with the arrival of Kirin 950, Haisi finally caught up with the big army, but in order not to let history repeat itself, Haisi must independently develop the SoC architecture.

Having said that, if Haisi follows the example of Samsung or Qualcomm, it will have to pay many R&D costs. After all, the whole process of mobile phone research and development is integrated, which affects the whole body.

Take 950 as an example, Mali-T880 is still the ancestral quad-core. Assuming that Hisilicon will be upgraded to eight nuclear, the packaging area of SoC will increase, which will also lead to greater heat generation. Huawei needs to find a solution to the fever. Every step will increase the cost and require higher shipments to amortize.

If you want to reduce costs, you need to abandon the mode of self-production and self-marketing.

Then the problem shifts to how to increase the shipment of chips. There are only two ways to improve chip productivity. One is to emulate Samsung's self-built wafer factory. The other is to find a foundry to produce.

According to Huawei's official data, mobile phone shipments in the first three quarters were 77.4 million units, while according to IDC statistics, Samsung's sales in the third quarter alone reached 84.5 million units. In other words, Huawei's mobile phone shipments can only reach one-third of Samsung's. In addition, Hisilicon chips are only used on Huawei mobile phones, and Samsung also provides chips to friends such as Meizu, which will further widen the gap on chips.

According to Huawei's official data, the shipment of Hisilicon Kirin in recent years is only 50 million, and the R&D cost of sharing a single chip is high enough. If Huawei forcibly builds a wafer factory or buys a wafer factory, it will lead to a further increase in production costs, which also goes against Hess's original intention of improving profit margins.

So Huawei can only continue to find foundries to produce chips. At present, there are only two foundries with large shipments and advanced technology, Samsung and TSMC. Because Samsung and Huawei have conflicts of interest as mobile phone manufacturers, and Huawei's Ascend P 1 has been pitted on the screen before, TSMC has naturally become Huawei's best choice.

However, as mentioned above, in order to obtain higher income, whenever the production capacity is insufficient, TSMC will give priority to supplying chips to Apple, Qualcomm, MediaTek and other manufacturers.

Therefore, it seems to be the only choice for Huawei to reduce costs by giving up the mode of self-production and self-sales and selling Kirin chips instead to increase the order volume.

Hayes needs to improve its competitiveness.

If you want other mobile phone manufacturers to buy their own chips, you must be as competitive in price as MediaTek or as powerful as Qualcomm.

Huawei insisted on using K3V2 on mid-to-high-end models for two years, which proved that Huawei did not want Hisilicon's SOC positioning to be lower than that of Qualcomm, and Hisilicon took the high-end route from the beginning. But at the same time, Huawei needs a certain amount of shipments to share R&D costs, so Huawei has been incubating HiSilicon's SoC with its own mid-to high-end mobile phones.

The situation of X 10 processor in MediaTek has proved that the bargaining power of chip manufacturers is limited before they have absolute technical advantages.

At the X 10 conference, Zhu Shangzu, senior deputy general manager of MediaTek, once bluntly said that the Helio X series is positioned at the high end, with powerful witty computing power and excellent multimedia functions. Obviously, it wants to hit the high-end market with X 10. However, in less than half a year, I was mercilessly beaten by Red Rice NOTE 2 at a price of 799.

This kind of thing may also happen to Hess. More importantly, Huawei, as a mobile phone manufacturer directly equipped with Hisilicon chips, will face the embarrassing situation of one glory and one loss.

Although the performance and power consumption of HiSilicon Kirin 950 are comparable to those of first-line chips, it still adopts the architecture of ARM A72 and the GPU of ARM MaliT880. As far as the current situation is concerned, Hess does not have bargaining power.

The research and development of autonomous architecture will become the core competitiveness of the next generation of mobile chips.

What Hess urgently needs now is to improve the competitiveness of its own chips. Although there are many factors that affect the performance and power consumption of the chip industry, there are only two decisive factors-the innovation of architecture and process. With the arrival of Kirin 950, Haisi's manufacturing technology has finally entered the era of 16nm, but the architecture used by Haisi is still the ARM public version.

The public version architecture naturally has the advantages of saving time and research and development costs, but ARM is "big". LITTLE "architecture is still not enough to balance performance and power consumption. Last year, the power consumption and calorific value of Qualcomm Snapdragon 8 10 were very serious. In addition to TSMC's 20nm boiler, another reason is that Qualcomm abandoned its own architecture and used ARM's public architecture to make a 64-bit processor as soon as possible.

The reason why Apple's A9 is significantly ahead of its competitors in performance is also because it has a self-designed chip architecture. Being able to design the architecture independently means that you can maximize the performance or power consumption of any application.

So the problem goes back to the self-developed SoC architecture. Whether Huawei can increase R&D investment this year, independent design architecture will become the key.

To borrow Lao Luo's words, "the capital market will not give Hammer Technology a third chance", and the same world will not give Hayes a third chance. If Hisilicon Kirin 950 becomes an ancestral SoC again, I'm afraid it will eventually fade out of people's sight like Texas Instruments and Broadcom.