2 years of chips stocked up, but problem is that they don't have their own ai chip design team. They don't want to help Huawei and US sanctions crippled MT and Biren. So their domestic backup plan likely don't have the same capacity at SMIC
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Xinqinwei received A+ round of financing for the construction of chemical mechanical polishing fluid production line.
Zhejiang Xinqin Microelectronics Technology Co., Ltd. (hereinafter referred to as "Xinqin Micro") received Series A+ financing, and investors include Yida Capital. This round of financing will be used for production line construction and R&D investment in chemical mechanical polishing fluids.
Chemical mechanical polishing (CMP) is the most mainstream wafer polishing technology today. During the polishing process, the wafer factory will select the removal rate (MRR) and surface roughness (Ra) according to the processing requirements of the flatness of the wafer chip at each step. ) and other indicators to improve polishing efficiency and product yield.
Core Qinwei was established in 2021 and specializes in the R&D and production of functional chemicals in semiconductor wafer manufacturing, semiconductor advanced packaging and new display industries. The company's main products include chemical mechanical polishing fluids (used for silicon, silicon carbide and other substrates, integrated circuits and advanced packaging) and functional electronic chemicals (used for cleaning, etching, and photolithography stripping products).
At present, Xinqinwei 's customers have spread across multiple industries such as silicon wafers, silicon carbide, discrete devices, logic chips, and advanced packaging.
i usually ignore Globaltimes but this time, they quoted official source.
At a press conference on economic operations in Beijing in October , Liu Aihua, a spokesperson for the National Bureau of Statistics, said that new growth engines for the Chinese economy are expanding, pointing to rapid growth in the production of equipment in related fields.
Specifically, the added value of the manufacturing sector of special equipment for semiconductor devices jumped by 33.9 percent in October, according to Liu..
At a press conference on economic operations in Beijing in October , Liu Aihua, a spokesperson for the National Bureau of Statistics, said that new growth engines for the Chinese economy are expanding, pointing to rapid growth in the production of equipment in related fields.
Specifically, the added value of the manufacturing sector of special equipment for semiconductor devices jumped by 33.9 percent in October, according to Liu..
I am starting to think SMIC secretly have more advanced process nodes capacity and with higher yields than some people say they have.
On November 15, at the 25th China Hi-Tech Fair, Yuntian Lifei launched a new generation of independently controllable edge artificial intelligence (AI) reasoning chips, the DeepEdge10 series, with a maximum computing power of up to 48TOPS and support for D2D/C2C Mash interconnection. Expansion can meet the deployment requirements of large models with hundreds of billions of parameters.
According to reports, Yuntian Lifei DeepEdge10 is China’s first 14nm Chiplet architecture master-level SoC chip for large-model AI reasoning.
Specifically, DeepEdge10 integrates a domestic RISC-V CPU core with 2 large cores + 8 small cores, with a maximum frequency of 1.8GHz; it also integrates a GC8000L GPU with performance comparable to Arm Mali-G52, with a main frequency of 800MHz, and supports a maximum of 8K@ It has 30fps video encoding and decoding capabilities and a maximum JPEG encoding and decoding capability of 200 million pixels, and supports dual-screen differential display (2K@60fps); built-in Yuntian Lifei’s latest generation independent intellectual property NPU NNP400T, with int8 computing power up to 12Tops. At the same time, DeepEdge10 also has hardware-level security features and complete high- and low-speed peripheral interfaces.
As mentioned earlier, DeepEdge10 is specifically optimized for large model reasoning needs. Its internally integrated NNP400T NPU is a neural network processor that supports a new computing paradigm for large models. It not only supports FP16/INT16/INT8 and other data format, also supports multi-threading and Transformer network structure model.
In terms of manufacturing technology, due to limited access to more advanced manufacturing processes, DeepEdge10 chose to be based on the domestic independently controllable 14nm process. At the same time, it adopts the Chiplet advanced packaging architecture based on D2D (Die to Die) interconnection, which can support flexible expansion of computing power and meet the diverse needs of computing power and cost in different scenarios.
For example, through the advanced packaging technology of D2D Chiplet, by loading more DeepEdge10 Dies into one package, the performance can be doubled. Of course, for advanced packaging technology, it is necessary to ensure that high-speed interconnection, large bandwidth, low power consumption, low latency, high reliability of transmission, powerful routing, and unified memory can be achieved between Die and Die.
Yuntian Lifei releases large model inference chip: 14nm Chiplet architecture, the first in China!
On November 15, at the 25th China Hi-Tech Fair, Yuntian Lifei launched a new generation of independently controllable edge artificial intelligence (AI) reasoning chips, the DeepEdge10 series, with a maximum computing power of up to 48TOPS and support for D2D/C2C Mash interconnection. Expansion can meet the deployment requirements of large models with hundreds of billions of parameters.
According to reports, Yuntian Lifei DeepEdge10 is China’s first 14nm Chiplet architecture master-level SoC chip for large-model AI reasoning.
Specifically, DeepEdge10 integrates a domestic RISC-V CPU core with 2 large cores + 8 small cores, with a maximum frequency of 1.8GHz; it also integrates a GC8000L GPU with performance comparable to Arm Mali-G52, with a main frequency of 800MHz, and supports a maximum of 8K@ It has 30fps video encoding and decoding capabilities and a maximum JPEG encoding and decoding capability of 200 million pixels, and supports dual-screen differential display (2K@60fps); built-in Yuntian Lifei’s latest generation independent intellectual property NPU NNP400T, with int8 computing power up to 12Tops. At the same time, DeepEdge10 also has hardware-level security features and complete high- and low-speed peripheral interfaces.
As mentioned earlier, DeepEdge10 is specifically optimized for large model reasoning needs. Its internally integrated NNP400T NPU is a neural network processor that supports a new computing paradigm for large models. It not only supports FP16/INT16/INT8 and other data format, also supports multi-threading and Transformer network structure model.
In terms of manufacturing technology, due to limited access to more advanced manufacturing processes, DeepEdge10 chose to be based on the domestic independently controllable 14nm process. At the same time, it adopts the Chiplet advanced packaging architecture based on D2D (Die to Die) interconnection, which can support flexible expansion of computing power and meet the diverse needs of computing power and cost in different scenarios.
For example, through the advanced packaging technology of D2D Chiplet, by loading more DeepEdge10 Dies into one package, the performance can be doubled. Of course, for advanced packaging technology, it is necessary to ensure that high-speed interconnection, large bandwidth, low power consumption, low latency, high reliability of transmission, powerful routing, and unified memory can be achieved between Die and Die.
this seems like a direct outcome of the recent sanctions. Limited 7nm capacity and unable to compete on designs with Huawei, so you end up with firms coming up with SoCs or Edge AI chips for use on AIoT objects.
There is actually a huge market on this in China
They are offering both edge computing and server AI computing using a mesh architecture were they connect multiple chiplets to form a mesh.
I saying that because have seen a lot of Chinese chips and AI companies lately with plans of making AI and others chips in advanced nodes (14nm and below) which SMIC is the only domestic provider and because I think 14nm and 7nm are probably made in the same facilities, is eye catching that SMIC has not only been keeping with the demand from Huawei but also others Chinese players too.
I saying that because have seen a lot of Chinese chips and AI companies lately with plans of making AI and others chips in advanced nodes (14nm and below) which SMIC is the only domestic provider and because I think 14nm and 7nm are probably made in the same facilities, is eye catching that SMIC has not only been keeping with the demand from Huawei but also others Chinese players too.
any possibility of huawei obtain license production tech from SMIC or none at all?
- Status