Chinese semiconductor industry

Status
Not open for further replies.

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Please, Log in or Register to view URLs content!

China Breaks Through Restrictions with Advanced Chiplet Strategy: 3D-IC Breakthrough in Chinese Ethereum Miner​


TechInsights Reveals Design of the Jasminer X4 with DBI Hybrid Bonding.
Please, Log in or Register to view URLs content!
to learn more.


TechInsights made a remarkable discovery on the Jasminer X4 Ethereum miner ASIC, developed by China-based crypto mining machine manufacturer Jasminer. This design marks the industry’s first-ever application of DRAM to logic direct bond interconnect (DBI).
TechInsights found hybrid bonding technology in Yangtze Memory Technologies Company’s (YMTC) Xtacking 3.0 architecture, where logic circuits were bonded to a separate wafer with storage cells in their 3D NAND products. Jasminer X4 uses this technology differently—it represents the first instance where the logic circuits have been bonded directly to DRAM.

The Jasminer X4 showcases the ability of a Chinese company, under trade restrictions, to combine matured technologies to create high-performance, cutting-edge applications creatively. A sizable 32 mm × 21 mm logic die employs the earlier XMC (Wuhan Xinxin Semiconductor Manufacturing Co.) planar 40/45 nm CMOS technology. The large die is patterned using two masks, which could represent a design workaround for a reticle constraint. The flip chip is assembled within the package.
Additional findings covered in the Disruptive Event Brief:
  • Xilinx Zynq FPGA on the board containing the Mining ASIC.
  • Jasminer X4 is the first Ethereum miner ASIC to utilize a storage mining design.
To learn more about these findings,
Please, Log in or Register to view URLs content!
and stay tuned for additional findings.
Please, Log in or Register to view URLs content!
This reminds me of this article.
2022年,达摩院与紫光把25纳米DRAM堆叠在55纳米逻辑芯片上,构建了神经网络计算以及推荐系统里的匹配加速等。系统带宽达到1.38TBps,性能上,相比CPU版本,速度提升9倍,能效比超300倍。
so 25nm DRAM stacked on 55nm logic chip to build neural network computing. compared to CPU version, speed is 9x and energy efficiency 300x.
可以认为3D堆叠技术已经成熟,国内企业已经具备成熟的实现方案。最近的杀手级应用,譬如Transformer,图计算,图数据库兴起对带宽延迟的刚性需求在兴起,或许预示着3D Stacked DRAM与逻辑的商业化时机已经到来。
so this should be considered mature technology by this point

Am i missing something here? This doesn't seem like revolutionary technology. Then again, I don't have access to the article.
 

pevade

Junior Member
Registered Member
I have my doubts, after a decade of stagnation all the sudden this startup that nobody has ever heard of before will reach the level that took years of trial and error to Samsung and TSMC who have way more resources and talent than this rapidus company. I hope they succeed because I am all for more competition and lower prices but I am a bit skeptical.
I mean research and development and successes in GAAFET in China and even subnanometer transistors is not something unheard before.


I mean, they do have pretty decent support and they have some semi industry veterans. They aren't just coming from nowhere.
Rapidus was established in August 2022 with the support of eight major Japanese companies:
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!
,
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!
, and
Please, Log in or Register to view URLs content!
.
On 10 August 2022, Rapidus was established by eight Japanese companies with total investment of 7.3 billion yen
Please, Log in or Register to view URLs content!
for domestic production of advanced semiconductors; with further investments totaling $36 billion expected over a decade.
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
Rapidus is led by Tetsuro Higashi (who formerly led
Please, Log in or Register to view URLs content!
)
Please, Log in or Register to view URLs content!
and Atsuyoshi Koike (who formerly led
Please, Log in or Register to view URLs content!
's Japanese subsidiary).
Please, Log in or Register to view URLs content!
In the same year, METI and the
Please, Log in or Register to view URLs content!
(NEDO) publicly solicited applications for advanced semiconductor development consignment projects related to strengthening the foundation of
Please, Log in or Register to view URLs content!
communication systems, which resulted in 70 billion yen of government subsidies for Rapidus.
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!


On 6 December 2022,
Please, Log in or Register to view URLs content!
announced it had signed a Memorandum of Cooperation with Rapidus, where both companies will set up a long-term and sustainable collaboration on advanced semiconductor technologies, while Rapidus would plan to mass-produce chips in 2-nanometer technology in Japan in the latter half of 2020s.
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!


On 13 December 2022,
Please, Log in or Register to view URLs content!
and Rapidus announced the development of 2 nanometer node technology, with production of the
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
(GAA FET) devices previously announced by IBM in 2021
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
to be done by Rapidus at its fab in Japan.
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!
The 2 nm semiconductor chips which Rapidus is aiming to produce are expected to have up to 45% better performance and use 75% less energy compared to 7 nm chips on the market in 2022.
Please, Log in or Register to view URLs content!
Please, Log in or Register to view URLs content!
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Please, Log in or Register to view URLs content!
SK Hynix is panic mode. Looking to cut back production to match demand.

As I said last night, YMTC is just massively undercutting Koreans here. This latest defender SSD is now going for 479 RMB and it's 2 TB! This is at least a 50% droop in prices vs last year
Please, Log in or Register to view URLs content!


As demand recover next year, YMTC should also be able to increase its 128-layer production (in fab 2 as well as replacing 64-layer production at fab 1) and I think leave prices permanently lower
 

tokenanalyst

Brigadier
Registered Member
Please, Log in or Register to view URLs content!
This reminds me of this article.

so 25nm DRAM stacked on 55nm logic chip to build neural network computing. compared to CPU version, speed is 9x and energy efficiency 300x.

so this should be considered mature technology by this point

Am i missing something here? This doesn't seem like revolutionary technology. Then again, I don't have access to the article.
Please, Log in or Register to view URLs content!
 

Phead128

Captain
Staff member
Moderator - World Affairs
while China semiconductor fabrication sector is growing, rest of the world is not stagnant either.

Japan’s ambition to establish 2nm production capability continues to chug along. If all goes well, they will have 2nm capability before China.
Please, Log in or Register to view URLs content!


so while we celebrate progress in China, just want to remind y’all to have a global perspective and not just flat out assume we are doing better than reality.
Rapidus are actually going for 1nm capability.
Please, Log in or Register to view URLs content!

Japanese gov't is giving $500 million USD and private partners have given $54 million USD, with additional $255 million USD spread out over 10 years. Who knew you can get 1nm/2nm capability for under $1 billion in investment in R&D. In reality, they need Tens of billions of dollars to reach 2nm/1nm.
 

siegecrossbow

General
Staff member
Super Moderator
The South Koreans are surrendering.

Interesting, they are surrendering to the Americans, and to the Chinese at the same time.

Haha! Haha!

If the South Koreans choose China and surrender to China, then they get to keep their business, and probably get a preferential deal from the CCP.

If the South Koreans choose the Americans, then they would have to surrender twice, like how Hynix is squeezed out, under duress, basically a surrender.

It wasn't that hard to figure out, surrendering once and getting a good deal, is better than surrendering twice and seeing someone cut your flesh off.

China does not believe there is a new cold war.

However, if one should develop, it could take a while. They must weaken themselves first before they can get stronger.

Geez, where have we heard that shit before?!

LOL! Haha! Haha!

:oops::D
They are about to find out what it means to have two dads. Who said that China is against LGBTQ+?
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Please, Log in or Register to view URLs content!
right, that's what I'm saying. Looks like they've been working on this for a while. And that example they listed match the example in the eet article. Maybe they found more improvement in how they linked it now or this is the first time people have seen it in application?


anyways, another interesting story. Shanghai has started China's first computation exchange. Looking to have 15 EFlops of computation by 2024 (had 9 at end of 2022). this will allow more efficient system to find computational resources I guess.
Please, Log in or Register to view URLs content!
上海超级计算中心(上海市人工智能公共算力服务平台)、腾讯长三角人工智能先进计算中心、阿里云华东智能算力中心、商汤科技人工智能计算中心、上海有孚临港云计算数据中心等五大智算中心已正式逐步投产,全部建成后,智能算力总承载能力超 10EFlops(IT之家注:1 EFlops 相当于每秒可进行一百亿亿次的浮点运算),为上海城市数字化转型提供坚实算力基座。
list of computational centers involved with this exchange. Currently have over 10 EFlops of computation
 
Status
Not open for further replies.
Top