Chinese semiconductor thread II

[tokenanalyst]

GigaDevice's third quarter report: revenue and net profit both increased year-on-year​

GigaDevice released its third quarter report for 2024, saying that the company's total operating revenue was 5.65 billion yuan, up 28.56% year-on-year, and its net profit attributable to the parent company was 832 million yuan, up 91.87% year-on-year. According to the single quarter data, the total operating revenue in the third quarter was 2.041 billion yuan, up 42.83% year-on-year, and the net profit attributable to the parent company in the third quarter was 315 million yuan, up 222.55% year-on-year. During the reporting period, GigaDevice had a large amount of accounts receivable, and the current accounts receivable accounted for 145.07% of the net profit attributable to the parent company in the latest annual report.
The various data indicators published in this financial report performed well. Among them, the gross profit margin was 39.46%, an increase of 14.76% year-on-year, the net profit margin was 14.73%, an increase of 49.25% year-on-year, the sales expenses, management expenses, and financial expenses totaled 443 million yuan, and the three expenses accounted for 7.85% of the revenue, an increase of 44.69% year-on-year, the net assets per share were 24.02 yuan, an increase of 3.09% year-on-year, the operating cash flow per share was 2.79 yuan, an increase of 89.54% year-on-year, and the earnings per share were 1.26 yuan, an increase of 93.85% year-on-year.

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Atoman Semiconductor: Driven by MBE technology innovation, becoming a pioneer in the internationalization of China's compound semiconductors​

The P1400 large hybrid MBE equipment independently developed by the company and its partners can accommodate three 8-inch wafer substrates at the same time. By combining an ammonia source with a plasma source, the growth of epitaxial wafers has both the excellent performance and high-quality growth characteristics of MBE, and the growth rate is significantly improved compared to traditional molecular beam epitaxy equipment. This innovative achievement provides advanced support for the development of the compound semiconductor industry and also provides a strong guarantee for the company's epitaxy business.



Aitman Semiconductor, a company that focuses on MBE technology and combines MOCVD technology to master the mass production capabilities of two epitaxial growth technologies, has been committed to becoming an internationally leading compound semiconductor equipment and application service provider since its establishment in January 2020. As a key listed reserve enterprise and a specialized "little giant" enterprise in Fujian Province, Aitman Semiconductor has achieved outstanding achievements in the MBE field and successfully filled the key gaps in the domestic industrial chain.

Atman Semiconductor has planned and built epitaxial material production and R&D bases with leading technology and production capacity in Jimei, Xiamen and Zhuji, Zhejiang, focusing on the production and research and development of compound semiconductor epitaxial wafers, which has effectively promoted the progress of the domestic and foreign epitaxial wafer manufacturing industry and provided solid support and guarantee for my country's strategic emerging fields such as microwave radio frequency, high-speed optical communication, and quantum technology.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

 

[mst]

Huawei's homegrown AI chip examined — Chinese fab SMIC-produced Ascend 910B is massively different from the TSMC-produced Ascend 910​

Please, Log in or Register to view URLs content!

 

[tokenanalyst]

Given that the nodes for this fab are 28nm and 55nm I think that means that most or all the equipment and materials in this fab are domestic. Probably we are going to see more mostly or all domestic fabs in the future.

 

[iewgnem]

d

Huawei's homegrown AI chip examined — Chinese fab SMIC-produced Ascend 910B is massively different from the TSMC-produced Ascend 910​

Please, Log in or Register to view URLs content!

It is believed that the compute chiplet of the Ascend 910B (let's call it Virtuvian B) is made by SMIC using its N+1 fabrication technology, which is believed to be a 7nm-class production node. Given the considerably larger die size of the Virtuvian B compared to the original Virtuvian, either SMIC's N+1 process technology has significantly lower transistor density compared to TSMC's N7+ (which would not be surprising), or HiSilicon substantially enhanced DaVinci cores and even sacrificed their number for these enhancements, which would have made the die size considerably larger (an unlikely scenario).

It's 7nm process, but with larger die size, conclusion, -> low transistor density.
Someone doesn't know what 7nm means.

 

[Awenumick]

Given that the nodes for this fab are 28nm and 55nm I think that means that most or all the equipment and materials in this fab are domestic. Probably we are going to see more mostly or all domestic fabs in the future.

View attachment 140264

What is this?

 

[tphuang]

Please, Log in or Register to view URLs content!

this is another update from Loongson just saying that their next 3B6600 desktop CPU is 30% better than 3A6000. And single core performance is really good.

3C6000 is now doing taping out & plan to complete production and start mass production in Q2 so that they can launch it. 3C/D/E6000 are planned. It seems to me that they are constrained on R&D. So is slow here.

9A1000 expecting to have 32TOP computation and expect to tape out first half of next year.

 

[tphuang]

Huawei's homegrown AI chip examined — Chinese fab SMIC-produced Ascend 910B is massively different from the TSMC-produced Ascend 910​

Please, Log in or Register to view URLs content!

btw, Ascend-610 is probably designed for MDC-610, which is Huawei's chip for ADAS.

Who knows what the actual process for 910B is. But if it is N+1, then they definitely have significant improvement left for 910C once they get to N+2 or N+3. Larger die certainly helps.

 

[JPaladin32]

Did some rough calculations on transistor density of Kirin 9020, based on info and rumors from various sources.

All Cortex-A510 little cores have been replaced with homegrown new little cores. Many sources stated this is a new OoO 2-wide core with performance matching Cortex-A73. I looked at Kirin 960 which is A73 + A55 and A73 is roughly double the area of A55. A510 is basically just A55 with Armv9 so I don't expect much die area difference. So, I estimate that the new Kirin little core is roughly double the area of the A510 on Kirin 9000s and 9010, which is a total die area growth of about 2.8mm2 for 4 little cores.

Then, some sources say the GPU has gone from 4CU to 6CU. This matches the 50% uplift of scores in GPU benchmarks so I tend to believe it. Each CU is gigantic and visually matches the size of the big core on 9000s which is 4mm2. So, 2 more CUs give a die area of 8mm2.

There are other confirmed changes like upgrading the modem from 5G to 5G-A. The die area needed for this is unknown.

I would guess the total area growth should be about 12mm2. Kirin 9000s SoC is about 107mm2 in total, so there needs to be a 11.2% increase. But, if somehow the area is kept the same, then there must be a quite big transistor density increase of more than 10%, which I think is very unlikely because then it almost exceeds the density of Intel 4. I would guess there is a mix of die growth and density increase, and maybe 5% increase in each.

 

[tokenanalyst]

CIOMP Hybrid Laser patent was authorized