Chinese semiconductor industry

Skywatcher said:

Huh, looks like a 2nm chip design could cost $1 billion.

Click to expand...

The only IC's economically viable at <=5nm nodes are smartphone SoCs and even then only if the volumes are in the tens of millions. Besides Qualcomm, Samsung (Exynos), MediaTek and formerly HiSilicon, any new entrants into this space are taking a huge risk. The only new entrants on the horizon are Xiaomi's long delayed Surge S2 SoC they've been putting off for years and the recent rumors of the Oppo-Vivo in-house SoC development. Both Xiaomi and Oppo-Vivo (BBK Electronics) have over 100 million unit sales annually.

ansy1968 said:

14nm check next year and 7nm maybe late next year to 2023, that is for the domesticated line, right now SMIC is mass producing 14nm, N+1 (8NM) this year and maybe N+2 late this year. If the domesticated 14nm and 7nm line is finished, we can flood the market and let it run its course. From my thinking TSMC, Samsung and Intel is focusing on mass production of 5nm and below knowingly that it will take China at least 4 years (EUVL) to catch up, surrendering the 7nm, 14nm and 28nm processes to the Chinese. But it has its drawback to many producers chasing to few customers, those 5nm chips and below are niche products and the West policy makers is to preoccupied with lower nodes not knowing the significant of other processes and the latest chip shortages is a result of that policy.

Click to expand...

I think a lot of people are not seeing the full picture here. Right now it does not make sense to design for the most advanced process for a lot of chips because of several costs, like the cost of masks, or the cost per transistor of the manufactured chips. DUV with lots of multi-patterning steps or current EUV are expensive. But eventually over the next decade EUV will have increased power light sources, and mask pellicles will be in use. With an increased power light source the exposure time is decreased and the production rate will ramp up. The use of mask pellicles will decrease the amount of defects and will make the masks themselves last longer. This will further increase production, because less defects means more good chips, and reduce the cost of masks because you need less of them.

Eventually it is most likely the optimum design node won't be 28nm anymore.

The chart above explains why 28nm is an optimum node, and why the nodes in between 28nm and 7nm didn't last long in mass usage.


The chart above indicates that once EUV machines improve the optimum node might not be 28nm but 5nm. Because that is the limit of what you can get with single-patterning with EUV.

horse said:

Doesn't it seem like the Americans don't really have strategy? A coherent strategy? Wishful thinking is for dreamers and stoners. Seems like their specialty. You got to be ruthless in business. You got to play to win.

Click to expand...

From listening to American discourse, the strategy seems to be based on a combination of hubris and jealousy, sad as that may seem. In some respects, it's actually become extremely petty. This campaign of "maximum pressure" as American policy makers call it will not succeed against a country like China for a host of reasons. So, unless China bends down before America, count on more and more pressure, more desperate measures. This is where it's all heading because America's character is that of a hegemon, an imperialist, control freak. Keep this in mind when trying to figure out why they do the things they do.

krautmeister said:

From listening to American discourse, the strategy seems to be based on a combination of hubris and jealousy, sad as that may seem. In some respects, it's actually become extremely petty. This campaign of "maximum pressure" as American policy makers call it will not succeed against a country like China for a host of reasons. So, unless China bends down before America, count on more and more pressure, more desperate measures. This is where it's all heading because America's character is that of a hegemon, an imperialist, control freak. Keep this in mind when trying to figure out why they do the things they do.

Click to expand...

Chyna cant innovate
Chyna only steals IP from the West
Chyna is only good at memorisation, not critical thinking.
Chyna communism!
Too advanced for Chyna to make. It only makes cheap plastic toys

gelgoog said:

I think a lot of people are not seeing the full picture here. Right now it does not make sense to design for the most advanced process for a lot of chips because of several costs, like the cost of masks, or the cost per transistor of the manufactured chips. DUV with lots of multi-patterning steps or current EUV are expensive. But eventually over the next decade EUV will have increased power light sources, and mask pellicles will be in use. With an increased power light source the exposure time is decreased and the production rate will ramp up. The use of mask pellicles will decrease the amount of defects and will make the masks themselves last longer.

Click to expand...

Mask pellicles will decrease costs but even so, not enough to definitively drop the cost lower than single exposure DUV in the near term. LPP-EUV light source is being researched for >300w output but even with success, it's not the leap that's needed. The same plasma contamination to the mask also happens to the optics. Mirror contamination isn't a fully solvable problem due to the nature of LLP-EUV, which will suppress efficiency, a problem that doesn't exist with DUV. Maybe source power can be ramped up enough that this will become a moot point eventually but I don't it happening anytime soon. By the time it does happen, I think SSMB EUV would be a reality. Then mask pellicles wouldn't even be needed because there wouldn't be any plasma to deal with.

I guess China could aim to become a serious, serious supplier of chips over 28 or 14nm, like it did with solar panels and other items. Maybe I am wrong but I believe in 2021 China will be fully self sufficient both in design, EDA, materials, litography, etc... for 14 and above. China has everything to simply mince competition in those lines.

In the meantime, quietly, it should quietly and calmly develop EUV and the industrial lines thereof. The sooner they come, the better, but meantime, flood the market with >=14 and memory with YMTC and CXMT.

There is an awful lot to be done in those industrial segments.

jfcarli said:

I guess China could aim to become a serious, serious supplier of chips over 28 or 14nm, like it did with solar panels and other items. Maybe I am wrong but I believe in 2021 China will be fully self sufficient both in design, EDA, materials, litography, etc... for 14 and above. China has everything to simply mince competition in those lines.

In the meantime, quietly, it should quietly and calmly develop EUV and the industrial lines thereof. The sooner they come, the better, but meantime, flood the market with >=14 and memory with YMTC and CXMT.

Click to expand...

Very good point, and something I should have mentioned to Gelgoog about the sweet spot node process, per transistor cost. China only recently had its full design flow EDA with Empyrean completed to 28nm and I believe they are just now working out the kinks with ArF photoresists down to the 7nm process. So, I don't think China will achieve self-sufficiency to 14nm in EDA and materials until around 2023. Once China has created its domestic 14nm supply chain, once it reaches saturation in its domestic market, sometime after 2026, I think it will do to the IC market, what it did to a bunch of other market segments. Namely, it will mass produce low priced equivalents and flood the world markets, destroying the profit margins of >=14nm ICs.

ansy1968 said:

@Tyler bro step by step, 28nm check to be mass produced, 14nm check next year and 7nm maybe late next year to 2023, that is for the domesticated line, right now SMIC is mass producing 14nm, N+1 (8NM) this year and maybe N+2 late this year. If the domesticated 14nm and 7nm line is finished, we can flood the market and let it run its course. From my thinking TSMC, Samsung and Intel is focusing on mass production of 5nm and below knowingly that it will take China at least 4 years (EUVL) to catch up, surrendering the 7nm, 14nm and 28nm processes to the Chinese. But it has its drawback to many producers chasing to few customers, those 5nm chips and below are niche products and the West policy makers is to preoccupied with lower nodes not knowing the significant of other processes and the latest chip shortages is a result of that policy.

Click to expand...

What are they applications for the fastest chips, besides flagship smartphones, supercomputers and graphic cards for bitcoin mining and graphics intensive gaming?

@WTAN @foofy @Oldschool and other experts, can this be used as a future power sources? I have seen comments suggesting that It can be used for developing and verifying EUVL components such as mirrors?

Please, Log in or Register to view URLs content!

486 views51 minutes ago

To the experts, PingTouGe is equivalent to Huawei Hisilicon?

From CnTechPost

Alibaba's chip arm unveils new processor​

May 18, 2021
Semiconductor, the chip division of Chinese e-commerce giant Alibaba, today announced a new processor in its Xuantie series - the Xuantie 907.
The processor is optimized for the open-source RISC-V architecture and can be used in microprocessors, intelligent voice, navigation and positioning, storage control, and other applications. The processor has been licensed to several companies.
The Xuantie 907 operates at a maximum frequency of more than 1GHz and achieves a unit performance of 3.8 Coremark/MHz.

The processor implements the latest DSP instruction standard of RISC-V for the first time, and is suitable for real-time computing scenarios with high computational performance requirements such as storage and industrial control.

PingTouGe, a wholly-owned semiconductor chip business entity of Alibaba, was established in September 2018 to develop data center and embedded IoT chip products for the new architecture of next-generation cloud all-in-one chips.
Since its establishment in 2018, PingTouGe has launched several products, including AI chip Hanguang 800, RISC-V processor Xuantie 910, and one-stop chip design platform "Wujian", of which Hanguang 800 is Alibaba's first chip.
PingTouGe Xuantie series processors have shipped over 2 billion units.