TSMC is preparing to revolutionize NVIDIA’s Feynman chips by switching to CoPoS packaging technology in 2028. Here are the critical developments in new generation chip production.
According to the information shared by the famous analyst Ming-Chi Kuo, semiconductor giant TSMC aims to put the new generation CoPoS (chip-on-panel-on-substrate) packaging technology, which is expected to create a revolution in artificial intelligence chip production, into mass production in the second half of 2028. This technology, of which NVIDIA will be one of the first adopters for the new generation Feynman artificial intelligence chips, aims to exceed the physical size limits of the current CoWoS packaging formula. This move by TSMC aims to provide a significant efficiency increase in the sector, especially in the production of ultra-large-scale packages. This new system, which will compete with Intel’s EMIB-T technology, is being developed to respond to the increasing process power needs of the artificial intelligence market.
CoPoS Technology Will Define Physical Borders Again
Current CoWoS technology could create technical bottlenecks in the production of very large-scale AI chips because it is limited by the size of silicon middle layers. CoPoS technology eliminates the need for a middle layer and allows direct deployment on the panel. In this way, the physical limitations of the lithography machine in the production lines are overcome.
TSMC’s new formula will pave the way for massive chip designs over 9.5x reticle size in the cut.
Glass Core Design Will Increase Performance
As detailed by analyst Ming-Chi Kuo, the new packaging system is built on a structure in which the glass core is placed in the middle of layers of ABF (Ajinomoto Buildup Film). This design allows chip components to be assembled in a more stable form, while also offering superior performance in terms of thermal management and signal transmission.
Offering a much more flexible production process compared to the CoWoS structure, CoPoS will play a critical role in meeting the demands of companies with high-performance computing requirements, such as NVIDIA.
NVIDIA and TSMC Cooperation Will Be Strengthened
Although there are claims in the market that Intel’s EMIB-T technology is being evaluated by NVIDIA, this new step of TSMC aims to preserve its strategic partnership with NVIDIA. As we move towards 2028, this change in the architecture of artificial intelligence chips promises a major transformation not only in size but also in cost efficiency. Analysts predict that if this technology is successful, TSMC will continue its dominance in the high-end chip packaging market.
What kind of fundamental changes do you think the transition to glass core technology in artificial intelligence chip production in 2028 will bring about in the hardware world? You can share your opinions and claims with us in the comments section.