Not only SMIC. Huawei joint venture with ICRD, Jiading second phase will start producing in next few months. ICRD have 14nm capability.
That would be great, do you have any links to this? I'd be curious to read. I think Huahong will start 14nm production soon. If ICRD can start 14nm also, then SMIC can move off 14nm and just concentrate on more advanced nodes in their SN1 plant.
Bro you mean the mythical Kirin 9100?
14nm process for smartphone CPU? That won't be very competitive. The best Huawei can do in the immediate is chips using SMIC's N+2 process. That should start soon.
Interesting dive into Biren BR100
"BR100 is a multi-die GPU with two tiles. Each tile has two HBM2E stacks, giving the GPU a total of four HBM2E stacks and 64 GB of DRAM. The tiles and HBM are mounted on top of a gigantic interposer using TSMC’s CoWoS (Chip on Wafer on Substrate) packaging. This advanced packaging technology lets Biren put a 896 GB/s link between the two dies"
It is not only the 7nm node, there is also TSMC proprietary advanced packaging technology here that is an integral part of their secret sauce to excel in benchmarks.
I understand they can't promptly switch to SMIC in case they are banned from TSMC, today or tomorrow. They will be stuck for some time.
I'm not sure the I agree with that. I'm sure SMIC has their own packaging technology. More importantly, BR100 is using a 7nm process here to get the same performance for AI computation as H100 got with 4nm TSMC process. The transistor density for BR100 is 71 million per mm^2. That's not very dense. Biren working with SMIC's N+2 process or N+2 improved process will be able to pack transistors a lot more tightly into the die.
The other thing is that Biren has been very fast in developing this first chip. It took them 3 years to push a product out to the market that is industry leading. Having recently invested in a promising startup, I can assure you that a good chunk of the first year of a tech startup is just spent getting funding, building relationships and hiring people. So in 3 years, they've managed to get all the necessary funding, attacking partnerships with the biggest players in China, hired up an entire team, developed all their own language (SUPA) for chips, the software support packages for it and developed an industry leading chip design and tested it out. Their next chip should not take 3 more years. I think 1 year (2 year max) would be the time they need to develop a second generation chip using SMIC process. After all, even Phytium was able to design a new server CPU will SMIC N+1 process in a year.
Let's hope they learnt their lesson from the recent news that working with TSMC for future chips is a bad idea and that they need to work with SMIC asap. Until then, they need to stock up as many BR100/104 as possible to not suffer from future sanctions.
