So basically ASML under the pressure of US
stop DUV immersion and after sale service to CHina and only the dry DUV and Krf allowed.
from 40% revenue in china to 20%, dropped half.
Chinese semiconductor thread II
- Thread starter vincent
- Start date
So basically ASML under the pressure of US
stop DUV immersion and after sale service to CHina and only the dry DUV and Krf allowed.
from 40% revenue in china to 20%, dropped half.
CNC machining for optical components of DUV and EUV lithography machines.
Structure design of four-axis linkage precision machine tool for hydrostatic motion pair
Large-aperture optical components are important parts in the optical systems of high-end equipment such as lithography machines, high-energy lasers, and high-resolution cameras. With the continuous improvement of the performance of high-end equipment, higher requirements are placed on the processing aperture and precision of optical components [ 1-3 ] . For example, the size of the large-aperture optical components required by the National Ignition Facility in the United States is ≥400 mm×400 mm [ 4 ] , and the surface profile accuracy of the components is better than λ /3 ( λ = 632.8 nm) [ 5 ] ; the optical system of the extreme ultraviolet lithography machine includes multiple high-precision, large-aperture aspheric and flat mirrors, and its low-frequency profile error accuracy is at least 1 nm (RMS value) [ 6 ] .
Since the end of the 20th century, the processing of large-diameter optical components has gradually developed into a combination of "traditional processing technology" and "ultra-precision processing technology". First, the optical component blanks are processed using processes such as flexible cutting; then ultra-precision processing methods such as CNC grinding and CNC polishing are used in turn to converge the contour error of the component to the highest nanometer-level precision level, and the surface roughness reaches the effect of a mirror.
Grinding, as a typical processing technology for large-aperture optical components, lies between traditional cutting and ultra-precision grinding and polishing processes. It has also achieved rapid development and requires a processing space of meter-level dimensions, micron-level processing accuracy, and high processing efficiency. Domestic and foreign scholars have conducted extensive research on grinding theory and engineering applications and have achieved fruitful results [ 8-9 ] . For example, the OAGM2500 CNC grinder developed by Cranfield University in the UK is a typical ultra-precision machining machine. The maximum processing space of the machine is 2500 mm × 2500 mm × 610 mm. It can complete ultra-precision grinding of meter-level aperture planar optical components, and the contour accuracy of the component surface reaches 1 μm (flatness RMS value) [ 10 ] . It can be seen that larger processing apertures and higher processing accuracy are the main development directions of ultra-precision grinding machine tools.
Aiming at the grinding processing demand of φ 900 mm caliber optical components, this paper designs a four-axis linkage ultra-precision grinding machine tool using liquid hydrostatic kinematic pairs. The main configuration of the machine tool is analyzed, the precision of the main moving parts is allocated, and the design of the four-motion axis system based on liquid hydrostatic kinematic pairs is completed. Finally, the optical component grinding process experiment is carried out.
Since the end of the 20th century, the processing of large-diameter optical components has gradually developed into a combination of "traditional processing technology" and "ultra-precision processing technology". First, the optical component blanks are processed using processes such as flexible cutting; then ultra-precision processing methods such as CNC grinding and CNC polishing are used in turn to converge the contour error of the component to the highest nanometer-level precision level, and the surface roughness reaches the effect of a mirror.
Grinding, as a typical processing technology for large-aperture optical components, lies between traditional cutting and ultra-precision grinding and polishing processes. It has also achieved rapid development and requires a processing space of meter-level dimensions, micron-level processing accuracy, and high processing efficiency. Domestic and foreign scholars have conducted extensive research on grinding theory and engineering applications and have achieved fruitful results [ 8-9 ] . For example, the OAGM2500 CNC grinder developed by Cranfield University in the UK is a typical ultra-precision machining machine. The maximum processing space of the machine is 2500 mm × 2500 mm × 610 mm. It can complete ultra-precision grinding of meter-level aperture planar optical components, and the contour accuracy of the component surface reaches 1 μm (flatness RMS value) [ 10 ] . It can be seen that larger processing apertures and higher processing accuracy are the main development directions of ultra-precision grinding machine tools.
Aiming at the grinding processing demand of φ 900 mm caliber optical components, this paper designs a four-axis linkage ultra-precision grinding machine tool using liquid hydrostatic kinematic pairs. The main configuration of the machine tool is analyzed, the precision of the main moving parts is allocated, and the design of the four-motion axis system based on liquid hydrostatic kinematic pairs is completed. Finally, the optical component grinding process experiment is carried out.
It's easier to replace desktop and server chips than phone SoC. You can do 7nm with the former.
Keep in mind that they can continue to buy from AMD. And they still have TSMC fabb'd ones that are self designed.
VeriSilicon's AI heterogeneous computing fusion solution may be the way forward for NPU development.
The correct answer to NPU development
As a mature core business of VeriSilicon, GPU IP has been widely used in the field of graphics rendering. VeriSilicon's GPGPU IP also performs well in deep learning training and reasoning. The coordinated development of NPU, GPU and GPGPU can make up for each other's shortcomings and form a more powerful computing platform.
In complex AI tasks, the NPU can focus on executing specific operations in deep learning models, while the GPU or GPGPU is responsible for processing graphics rendering and other computing tasks. This resource complementarity enables the system to perform better in a multi-tasking environment, thereby improving overall efficiency.
As shown in the figure above, VeriSilicon's GPU, GPGPU, and NPU are closely integrated, and the collaboration of the three can provide greater flexibility. Developers can choose the most appropriate computing resources according to actual needs, making the system more scalable. This flexibility is particularly important when dealing with diverse AI applications, especially in edge computing and real-time reasoning scenarios.
Dai Weijun emphasized that the close integration of the three is mainly reflected in the fusion at the instruction level. This fusion means that the three can share instruction sets at the hardware level, thereby achieving more efficient collaborative work.
Taking VeriSilicon's VIP9X00CC as an example, it realizes the instruction-level fusion of GPGPU modules and NPU modules, and adopts SIMT (single instruction stream, multiple data stream) parallel computing architecture, allowing a single instruction to operate on multiple data elements at the same time. This architecture is usually used in GPUs, and the use of SIMT architecture enables more efficient parallel processing when performing AI computing tasks.
VIP9X00CC also supports DLP/ILP/TLP parallelism, ensuring that NPU and GPGPU can flexibly handle different types of parallel computing tasks.
Dai Weijin further pointed out that the design of VeriSilicon NPU includes many years of optimization experience, especially in data organization, compression, migration and computing, and VIP9X00CC combines the design concept of NPU with the parallel computing capability of GPU, greatly improving the overall AI computing efficiency.
Key US lawmakers are pressing the Biden administration to block Huawei Technologies Co. suppliers from buying American chipmaking gear, escalating efforts to prevent the sanctioned Chinese telecom giant from making progress on semiconductor manufacturing.
Small number of tool companies? The biggest suppliers in the world? The biggest semiconductor companies that US have apart from Nvidia and Intel? The companies that if they get severely financially hit will have downsize creating a chain effect that could lead to serious shortages on and price increases?
It would be interesting to see the faces of these stooges if one day US tools makers go under or their market share becomes very reduced and the US then have to rely on imported semiconductor equipment to build their fabs. I have an idea how their faces look like.
And that's not just the title, it's the entire summary in bright bold and big letters like a big announcement, breaking news, the real deal that's been announced a million times before or Maybe they guy thought if it wasn't big enough no one would see it or care.