Chinese semiconductor thread II

[tphuang]

Is this really right?

Huawei‘s upcoming Kirin PC chip. Taishan v130 ( 8 Core CPU) (GPU)codenamed “Ma Liang 920.”(10-core GPU)
Similar performance bracket as Apple’s M2 and the Intel’s i7-13700H processor.

Please, Log in or Register to view URLs content!

Nobody knows for sure, this has been a rumor going around chinese social media for a long time.

Since the chip itself has not been released yet, I would say we should wait until when it comes out

As with all else, let's wait until actual news come out.

 

[BlackWindMnt]

If they can get even close to m1 chip performance it would break free Huawei prosumer laptop lineup from Intel chipset in a couple of years once the software ecosystem has been created. The twitter account also mentions ascend big + lite setup not sure what that means for AI performance but it could be interesting.

 

[gelgoog]

in the best case that SMCS succeeded in porting all 14nm down to 7nm, we can expect a 35K wpm 7nm max capacity today.

35K wpm is just SMSC SN1. They doubled the floorspace at SMSC last year by building SN2. They probably started moving machines in and I wouldn't be surprised if they started manufacturing chips at the fab expansion this year. The max capacity of SMSC with the expansion is supposed to be 70k wpm. See Chinese semiconductor thread II for more details.

 

[tokenanalyst]

Please, Log in or Register to view URLs content!

This was already posted in the thread, but a few extra things to note. It's interesting that the phase 2 experiments had to postphoned due to covid and had to be conducted at the Metrology Light Source in Berlin and not in a Chinese synchrotron. Makes it feel like the project doesn't have high priority, you would think something of this importance, they would have pressured Berlin to allow experiments to continue even during covid times or had modified an existing synchrotron for the SSMB project or even built a brand new synchrotron for this project.

A lot of things where postponed due Covid including work in SMEE immersion machine, that choice for many was between living in the workplace for week or months or staying at home, tough choices.

2ndly, weird that a project of this importance would still be allowed for cooperation between Germany and China. You would think that America would shut it down to try to slow down China's progress in EUV or that China would want to keep their progress in EUV a secret and not share important details with a Western power, espically after the heightened tensions of 2020.

You are giving too much credit to D.C stooges by thinking that many of them know anything about the things they want to legislate.
Either way I don't think they are no collaborating anymore, Tsinghua is working on their own in the project and that is reflected in the SSMB patent

3rdly, this implies that the design stage is still in early stages if phase 2 R&D efforts just started up in 2023, which doesn't match the rumors of the SSMB EUV having started construction in Xiongan in 2023.

For start R&D still require the construction of facilities, Tsinghua SSMB EUV I think is based Shanghai, but there other efforts including a Free Electron Laser EUV project in Shenzhen.

 

[tphuang]

35K wpm is just SMSC SN1. They doubled the floorspace at SMSC last year by building SN2. They probably started moving machines in and I wouldn't be surprised if they started manufacturing chips at the fab expansion this year. The max capacity of SMSC with the expansion is supposed to be 70k wpm. See Chinese semiconductor thread II for more details.

I think 50-60k wpm by end of next year is reasonable goal assuming domestic tools continue to improve

The big question for end of next yr is EUV process

 

[gelgoog]

There is a new RISC-V chip company called SpacemiT which has launched the K1 processor. 8-core RISC-V RVA22 compatible.
It uses their own design X60 core. Claimed to have 30% more performance than an ARM Cortex A55. So it should have similar performance to the A510. The chip also comes with an Imagination Technologies GPU.
This seems to be the first commercially available chip with RISC-V Vector 1.0 support.

There is a developer laptop with the chip:

Please, Log in or Register to view URLs content!

There will also be a Banana Pi board with the chip:

Please, Log in or Register to view URLs content!

This is the website for the CPU company:

Please, Log in or Register to view URLs content!

The SpacemiT company founders come from Alibaba T-Head and Allwinner technology.
The company has received financing from Lenovo Capital and Legend Capital:

Please, Log in or Register to view URLs content!

The X60 core seems to be a dual-issue in-order 9-stage pipeline core. RISC-V 64GCVB with 256-bit vector compute.
They also have the X100 core. Which is a quad-issue out-of-order core. This seems to be the product Lenovo is interested in.

 

[latenlazy]

3rdly, this implies that the design stage is still in early stages if phase 2 R&D efforts just started up in 2023, which doesn't match the rumors of the SSMB EUV having started construction in Xiongan in 2023.

The Xiong’an facility is a research facility, not a facility for production.

 

[interestedseal]

Please, Log in or Register to view URLs content!

Raintree shipped out several metrology/inspection tools. Chinese toolmakers catching up fast in this field. Just like EDA, over 30 startups in this field were founded recently.

 

[sndef888]

35K wpm is just SMSC SN1. They doubled the floorspace at SMSC last year by building SN2. They probably started moving machines in and I wouldn't be surprised if they started manufacturing chips at the fab expansion this year. The max capacity of SMSC with the expansion is supposed to be 70k wpm. See Chinese semiconductor thread II for more details.

How does 35k wpm translate to real life? How many Pura 70 is that every month?

 

[gelgoog]

It depends on yield. Let's assume the die size is 10mm x 10mm.



At 70% yield (which I think is a decent assumption) you would get 404 good dies per wafer. If you produce 35k wafers per month that is 14 140 000 good dies per month. i.e. 14.14 million.

The thing is back when it was said the fab had 35k wafers per month capacity it was still using the 14nm process. It remains to be seen what is the capacity with 7nm process which requires more passes. And like I said we don't know the yield. Plus it is not like this is the only thing the fab will be doing.