Nvidia, in collaboration with TSMC and ASML, has made a significant breakthrough in chip manufacturing. They have developed a new algorithm that speeds up one of the most crucial steps in chip manufacturing by 40 times. This could potentially allow TSMC to produce more advanced chips without upgrading the equipment.
The process of chip manufacturing requires extreme precision. Currently, we are at two-nanometer transistors, which means we can fabricate transistor features of the size of a human DNA. This is made possible by using ASML UV machines, which are used for lithography, a process similar to a high-tech printer that projects tiny patterns through masks.
However, there are challenges. Computing these masks takes a lot of time, often weeks, and requires dedicated data centers. As transistors get smaller, it becomes more difficult to create these tiny patterns on a mask. This is done through a process called inverse lithography, which is a significant advance in the last two decades.
Nvidia has presented a new solution: a GPU-based algorithm that accelerates computational lithography by 40 times. This algorithm can parallelize the task among many GPUs, reducing the time needed for computing a photomask from several weeks to an overnight job.
According to Nvidia, with this new algorithm, 500 Nvidia DJX supercomputers do the work of 4,000 CPUs. A single reticle currently takes two weeks to process on CPUs, but with the new GPU-based litho, it can be processed in a single eight-hour shift, reducing power from 35 megawatts to just 5 megawatts.
This new breakthrough will help TSMC increase its throughput and produce more of these beautiful stunning silicon Wafers. It will also speed up our progress towards one nanometer and beyond. The new library allows TSMC to create curvy polygons on a mask, resulting in a greater depth of focus. This could reduce the total number of masks needed, which will reduce costs and improve yield.
This technology will be used by TSMC in production starting from June this year, and other companies like Synopsis, a huge EDA software company, and ASML plan to support these new libraries.
1. Nvidia, in collaboration with TSMC and ASML, has made a breakthrough in chip manufacturing.
2. They have developed a new algorithm that speeds up one of the most crucial steps in chip manufacturing by 40 times.
3. This breakthrough will increase TSMC's report and potentially allow TSMC to produce more advanced chips without upgrading the equipment.
4. The precision required in chip manufacturing is extreme, with current technology able to fabricate transistor features of the size of a human DNA.
5. ASML UV machines are used for the lithography process, which is akin to a high-tech printer projecting tiny patterns through masks.
6. For modern chips, about 100 photo masks are needed, and the number of masks acquired depends on the complexity of the chip.
7. The more complex the chip, the more masks are needed, which takes a lot of time to compute.
8. The process of creating these tiny patterns on a mask is called inverse lithography and is one of the biggest advances of the last two decades.
9. The calculation of the photomasks involves solving lots of Maxwell equations related to the physics of light, which is a very computationally expensive task.
10. Companies use huge data centers to compute such lithography masks, using tens of millions of CPU hours every year.
11. Nvidia has designed a new GPU-based algorithm along with a new library which accelerates computational lithography by 40 times.
12. This algorithm parallels the task among many GPUs, reducing the amount of time needed for computing a photomask from several weeks to an overnight job.
13. Nvidia's new technology runs on their own video GP racks and speeds up the fabrication of their future chips and GPUs.
14. According to Nvidia, with this new algorithm, 500 Nvidia DJX supercomputers do the work of 4,000 CPUs.
15. The new breakthrough will help TSMC increase its throughput and produce more of the beautiful stunning silicon Wafers.
16. The new library allows the creation of curvy polygons on a mask, resulting in a greater depth of focus.
17. This could reduce the total number of masks needed, which will reduce costs and improve yield.
18. The technology will be used by TSMC in production starting from June this year, and other companies like Synopsis, a huge EDA software company, and ASML plan to support these new libraries.
19. Nvidia is not the first company who used GPUs to try to accelerate computational lithography.
20. There was another startup in California called D2S or Design System which had the same AGM and started to build a GPU-based supercomputer for computational lithography already in 2019.