Chinese semiconductor thread II

[FairAndUnbiased]

Honestly given that TSMC also struggled with 3 nm on a FinFET architecture I'm beginning to wonder if the shrinking game is still tenable as a driver for performance. Makes me wonder if it wouldn't be a better strategy to stay at a larger node and focus innovation on transistor architecture.

it is changes in transistor architecture like planar->FinFET->GAA enable die shrinks at the very advanced nodes. It's not like planar die shrinks anymore where just making things smaller is OK, now every shrink needs its own new architecture. Changes in transistor architecture also make the processes much harder, especially fully 3D architectures like GAA.

That's why I think advanced packaging and maybe new materials is the next big thing. The big part is that for more advanced nodes, the percentage of the chip that is dark silicon is rising, thermal budgets are being more constrained, etc. These are problems that are IMO easier solved by advanced packaging than even more die shrinking. Stuff like microfluidic IC cooling.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Saimet received Series C+ financing, and CIM has been verified in multiple 12-inch wafer fabs​

Saimet recently received Series C+ financing. The investor in this round is Ceyuan Capital (leading investment). After the financing, Saimet plans to build the Chengdu subsidiary into the western headquarters project. Through the merger and acquisition of domestic and foreign semiconductor industry software companies, it will focus on high-end manufacturing directions such as integrated circuit manufacturing in the west, and conduct research and development and sales of CIM industrial software for intelligent manufacturing in the semiconductor industry. .
Founded in 2017, Saimet is a domestic intelligent manufacturing software solution provider that focuses on providing overall solutions and services to customers in the semiconductor and pan-semiconductor industries, and is committed to providing one-stop domestic CIM system integration solutions. The industrial intelligent manufacturing software system provided by Saimet can provide customers with a one-stop system integration solution from CIM system to equipment interface. This solution has been verified by multiple customers and multiple production lines to meet the systematization, standardization, and requirements of the manufacturing industry. The development needs of automation and intelligence.
According to information from Shengchi Capital, Saimet has created independent and controllable semiconductor intelligent manufacturing CIM system industrial software, filling the gap in this field in China. It also has the most technical personnel, the most complete product line, and the 12-inch 12-inch among domestic semiconductor CIM manufacturers. The domestic system software provider with the richest production line cases has broken the relevant technology blockade and accelerated the process of domestic substitution. Saimet's self-developed CIM solution has been verified in multiple 12-inch wafer fabs, helping 12-inch wafer manufacturers solve challenges such as high process, high cost, high yield, and high output.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

 

[Overbom]

5nm production coming??

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

China on cusp of next-generation chip production despite US curbs​

SMIC and Huawei plan to make new 5nm processor, supporting Beijing’s goal for advanced semiconductors

China’s national chip champions expect to make next-generation smartphone processors as early as this year, despite US efforts to restrict their development of advanced technologies.

The country’s biggest chipmaker SMIC has put together new semiconductor production lines in Shanghai, according to two people familiar with the move, to mass produce the chips designed by technology giant Huawei.

According to two people with knowledge of the plans, SMIC is aiming to use its existing stock of US and Dutch-made equipment to produce more-miniaturised 5-nanometre chips. The production line will make Kirin chips designed by Huawei’s HiSilicon unit and destined for new versions of its premium smartphones.

If production is judged successful enough for smartphones, Huawei’s most powerful artificial intelligence processor, the Ascend 920, will also be produced at 5nm by SMIC, the two people said

---

Meanwhile, SMIC has increased its current 7nm production capacity to make more Kirin chips and AI GPUs, said two people familiar with the matter.

7nm production capacity has been increased

The move to create more advanced chips has incurred additional costs, however. Three people close to Chinese chip companies said that SMIC was having to charge 40 to 50 per cent more for products from its 5nm and 7nm fabrication nodes than Taiwan’s TSMC does at the same nodes.

However, SMIC’s yield — the number of chips considered good enough to ship to customers — is also less than one-third of TSMC’s.

Rumours about costs and yield

 

[ansy1968]

5nm production coming??

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

---

7nm production capacity has been increased


Rumours about costs and yield

You know bro, we had this discussion before circa 2020, everybody was sceptical, then we had been given an insight when Liang Mong Song in his resignation letter mention the 5nm and 3nm had finished their development and are waiting for the necessary equipment to produce it. Well he doesn't disappoint, like I posted before SMIC will able to match the TSMC ARIZONA fab 5nm by 2025 and TSMC 3nm in 2027. I was wrong on the first by a year and I hope I'll be wrong on the second as well (2026). Anyway whatever happen we had leapfrog the US by a generation and have all Chinese occupying the top.

Liang Mong Song: SMIC's 7nm node has already completed full development, and is preparing for mass production in April 2021. Designs and most of the development for 5nm and 3nm nodes have also already been completed, only awaiting EUVL delivery at this point.​

 

[gelgoog]

SMIC (SMSC) SN1 fab design concept with the SN2 fab expansion:



Aerial view from Google Maps showing SN2 under construction:

Address: No.18 Zhangjiang Road, Pudong New Area, Shanghai, China

Looking at Sentinel-2 satellite pictures the roof construction on SN2 seems to have been finished last December. It is hard to tell for sure because of the resolution though:



I doubt the SN2 factory is operational yet. But the building is there. Typically it takes a year to move in machines and start trial production. This area should allow the doubling of FinFET production. Each fab (SN1, SN2) can do 35,000 wpm. So it will be 70,000 wpm total.

 

[measuredingabens]

Rumours about costs and yield

I think that depends on the chip. Yields for the Ascend chips probably aren't great based on the defect rate due to the size of the dies, but the ones for the Kirin chips should be fine.

 

[AndrewS]

5nm production coming??

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

---

7nm production capacity has been increased


Rumours about costs and yield

Let's go with SMIC making 7nm and 5nm chips (for Huawei) which are 40-50% more expensive than the TSMC equivalent.
But this is still profitable for SMIC and Huawei. Analysis below.

---

Look at the NVidia financials

1) $11 Bn = Cost of goods (TSMC chips, packaging etc)
2) $26 Bn = Revenue

So on average, NVidia sell their product at a 136% markup to their cost. That's a huge profit margin.

---

So even if a SMIC equivalent chip costs 50% more, then Huawei would still have a 62% markup to work with.
This is still very profitable and worth doing.

---

In the case of Qualcomm chips, it's difficult to do such a comparison because Qualcomm does a lot of licensing. But I recall a semianalysis piece where they estimated a SMIC/Huawei chip would be cheaper or comparable to an equivalent sold by Qualcomm.

---
EDIT. I see NVidia profit margins for 2023Q3 are even higher eg. For every $10K of product, they sell for $30K

If you rework the figures, you end up with Huawei chips being competitively sold at a 106% markup. That's basically selling for double what it costs to make.
So there's even more incentive for SMIC/Huawei to ramp up production of advanced chips

 

[tphuang]

菊菊轻薄小折叠，之前说过延续前代设计加大镜头模组和副屏，紫色素皮和灰色玻璃颜值不错，艺术版依旧是3D浮雕工艺。4520mAh+K9000S 5G。

三折确实也在路上，菊菊折叠屏软硬技术确实领跑……

according to this, the thin flip phone that Huawei is coming out in the next month also uses K9000S with 4520mAh battery

And to they are also working on a 3 fold phone

And I'm pretty sure P70 to be launched in 2 months also uses K9000S

So the production level and yield must be going up.

They are probably are selling up to 4 million phones a month that requires Kirin chips

let's say 3 million right now, but will reach 4 million by April, so they need to produce that many per month.

if we assume something like 400 good chips/wafer if including the binned K9000S, that's still 10000 wafers/month

if we add in maybe 20000 Ascend chips/month & 20 good ones per wafer, that's another 1000 wafer

of course, there is also requirement for chips to be used in the AITO ADAS systems & other inference chips.

 

[tacoburger]

It's hard to imagine that EUV will come out within this decade while SMEE has been struggling with DUVi for years without much result. I read in 2020 that their 28nm DUVi machine was going into mass production next year. Rinse and repeat for the last 3 years without any real result. If making DUV machine is so hard, I can't imagine how long China has to wait before mass production of EUV comes online.

 

[measuredingabens]

It's hard to imagine that EUV will come out within this decade while SMEE has been struggling with DUVi for years without much result. I read in 2020 that their 28nm DUVi machine was going into mass production next year. Rinse and repeat for the last 3 years without any real result. If making DUV machine is so hard, I can't imagine how long China has to wait before mass production of EUV comes online.

SMEE isn't responsible for the EUV effort, that's a collection of Chinese institutions. Unlike with DUVi, Chinese companies have far greater incentive to get an EUV machine into production as soon as possible. SMEE's DUVi wasn't going to receive any buyers with ASML around until the sanctions kicked in and spurred domestic firms to actually work with SMEE.