Home Page> Industry Information> Stacked Wafer Technology Of Tsmc Can Double The Performance Of Nvidia And Amd Gpu.

Stacked Wafer Technology Of Tsmc Can Double The Performance Of Nvidia And Amd Gpu.

Form: laoyaoba.com 2018/5/4 Browse:7695 Keywords: TSMC NVIDIA AMD GPU

TSMC is currently holding the twenty-fourth Annual Technical Symposium in Santa Clara, which has just released a technology Wafer-on-Wafer (WoW, stacked wafer) that can bring revolutionary changes to the graphics card. As the name suggests, WoW works in a vertical stack, rather than placing them on a circuit board, just as the 3D NAND flash is stacked in a modern solid-state drive. This means that Nvidia and AMD GPU can improve their performance without increasing their physical size or reducing the manufacturing process.


2_2018050406215014ewo (1).jpg

2_2018050406215028nPE.jpg



Wafer-on-Wafer (WoW, stacked wafer) technology contacts each other through a 10 micron hole connected by a silicon through hole (TSV). Cadence, the co - partner of TSMC, explains that the Wafer-on-Wafer (WoW, stacked wafer) design can be placed on the intermediate layer, routing one connection to another connection and creating a double crystal cube. Even WoW can be used to vertically stack more than two wafers.


The technology will allow more cores to be stuffed into a package, and means that each chip can be very fast and communicate with each other at a minimum delay. It is particularly interesting that a manufacturer can use WoW to put two GPU on one card and publish it as a product update, creating a basically two GPU without displaying it as a multiple GPU setting of the operating system.


WoW the biggest problem now is the production of wafer. When they are glued together, if only one wafer is broken, even if the two wafers are fine, they must be discarded. This means that the process needs to be used on production nodes with high yield, such as the 16 nanometer process of TSMC, in order to reduce costs. However, the company's goal is to use WoW technology in future 7Nm and 5nm manufacturing process nodes.

Next:

Google: The Existing Chip Needs To Be Optimized By Ai, So Far No Successful Ai Chip Has Been Developed.