Recap: Nvidia CEO Jensen Huang addressed issues concerning the firm’s demand for superior packaging from TSMC earlier this week, clarifying that whereas their technological wants are evolving, total demand stays strong. Particularly, Nvidia is transitioning from CoWoS-S to CoWoS-L, representing a major development in its chip structure in addition to a significant shift for TSMC.
Talking on the sidelines of an occasion hosted by chip provider Siliconware Precision Industries in Taichung, Taiwan, Huang defined the transition in Nvidia’s chip packaging necessities. “As we transfer into Blackwell, we’ll use largely CoWoS-L. After all, we’re nonetheless manufacturing Hopper, and Hopper will use CoWoS-S. We may even transition the CoWoS-S capability to CoWoS-L,” he said.
Huang emphasised that this shift doesn’t point out a discount in capability however reasonably a rise in capability for CoWoS-L expertise. “So it is not about lowering capability. It is really rising capability into CoWoS-L,” he stated.
CoWoS-L (Chip-on-Wafer-on-Substrate with Native Silicon Interconnect) represents a major development over CoWoS-S when it comes to efficiency and effectivity for high-end computing functions like AI and HPC.
The primary distinction between the 2 lies within the incorporation of Native Silicon Interconnect (LSI) chips for die-to-die connections. This seemingly small change permits a dramatic enhance in interconnect density, instantly translating into improved bandwidth capabilities. The expertise can even probably assist as much as 12 HBM3 modules, surpassing the capabilities of CoWoS-S.
Huang’s feedback had been in response to current hypothesis about Nvidia probably lowering orders from TSMC. Tech analyst Ming-Chi Kuo had reported at one level that Nvidia was shifting its focus to the newer CoWoS-L expertise, which might have implications for suppliers. Moreover, some Taiwanese media retailers steered that Nvidia was chopping CoWoS-S orders from TSMC, probably impacting the chip foundry’s income.
These issues might have some benefit. Whereas Nvidia’s total demand from TSMC stays strong, the shift to CoWoS-L is prompting TSMC to adapt its manufacturing capabilities.
TSMC is anticipated to double its CoWoS manufacturing capability by 2025, with Nvidia projected to occupy greater than half of this capability. Nonetheless, Nomura Securities initiatives that Nvidia’s discount in CoWoS-S orders might result in a 1% to 2% lower in TSMC’s income, and a few analysts predict that Nvidia may reduce CoWoS-S orders at TSMC by as a lot as 80%.
In the meantime, different gamers in Nvidia’s provide chain can also really feel the consequences as demand for CoWoS-S packaging decreases.
Within the broader context, Nvidia’s shift to CoWoS-L is more likely to speed up the adoption of superior packaging applied sciences throughout the semiconductor trade, setting new benchmarks for high-performance AI chips.
Recap: Nvidia CEO Jensen Huang addressed issues concerning the firm’s demand for superior packaging from TSMC earlier this week, clarifying that whereas their technological wants are evolving, total demand stays strong. Particularly, Nvidia is transitioning from CoWoS-S to CoWoS-L, representing a major development in its chip structure in addition to a significant shift for TSMC.
Talking on the sidelines of an occasion hosted by chip provider Siliconware Precision Industries in Taichung, Taiwan, Huang defined the transition in Nvidia’s chip packaging necessities. “As we transfer into Blackwell, we’ll use largely CoWoS-L. After all, we’re nonetheless manufacturing Hopper, and Hopper will use CoWoS-S. We may even transition the CoWoS-S capability to CoWoS-L,” he said.
Huang emphasised that this shift doesn’t point out a discount in capability however reasonably a rise in capability for CoWoS-L expertise. “So it is not about lowering capability. It is really rising capability into CoWoS-L,” he stated.
CoWoS-L (Chip-on-Wafer-on-Substrate with Native Silicon Interconnect) represents a major development over CoWoS-S when it comes to efficiency and effectivity for high-end computing functions like AI and HPC.
The primary distinction between the 2 lies within the incorporation of Native Silicon Interconnect (LSI) chips for die-to-die connections. This seemingly small change permits a dramatic enhance in interconnect density, instantly translating into improved bandwidth capabilities. The expertise can even probably assist as much as 12 HBM3 modules, surpassing the capabilities of CoWoS-S.
Huang’s feedback had been in response to current hypothesis about Nvidia probably lowering orders from TSMC. Tech analyst Ming-Chi Kuo had reported at one level that Nvidia was shifting its focus to the newer CoWoS-L expertise, which might have implications for suppliers. Moreover, some Taiwanese media retailers steered that Nvidia was chopping CoWoS-S orders from TSMC, probably impacting the chip foundry’s income.
These issues might have some benefit. Whereas Nvidia’s total demand from TSMC stays strong, the shift to CoWoS-L is prompting TSMC to adapt its manufacturing capabilities.
TSMC is anticipated to double its CoWoS manufacturing capability by 2025, with Nvidia projected to occupy greater than half of this capability. Nonetheless, Nomura Securities initiatives that Nvidia’s discount in CoWoS-S orders might result in a 1% to 2% lower in TSMC’s income, and a few analysts predict that Nvidia may reduce CoWoS-S orders at TSMC by as a lot as 80%.
In the meantime, different gamers in Nvidia’s provide chain can also really feel the consequences as demand for CoWoS-S packaging decreases.
Within the broader context, Nvidia’s shift to CoWoS-L is more likely to speed up the adoption of superior packaging applied sciences throughout the semiconductor trade, setting new benchmarks for high-performance AI chips.