NVIDIA teams up with SK Hynix to try 3D stacking HBM memory onto GPU cores
Views : 235
Update time : 2023-12-16 12:39:29
SK Hynix has begun recruiting designers for logic semiconductors such as CPUs and GPUs in the hopes of integrating HBM4 directly onto chips via 3D stacking, according to Joongang.co.kr. SK Hynix is reportedly discussing its HBM4 integration design approach with several semiconductor companies, including Nvidia.
The foreign media believe that Nvidia and SK Hynix will likely jointly design this integrated chip, and with the help of TSMC for foundry, and then through TSMC's wafer bonding technology will SK Hynix's HBM4 chip stacked on the logic chip. Joint design is inevitable in order to realize the synergy between memory and logic chips. If SK Hynix is successful, this could change the way the industry operates to a great extent, as it will not only change the way logic and memory new chips are interconnected, but also the way they are manufactured. At this stage, HBM stacking is primarily placed on an intermediary layer next to the CPU or GPU and connected to the logic chip using a 1024bit interface. SK Hynix aims to stack HBM4 directly onto the logic chip, eliminating the intermediary layer altogether. In a way, this approach is similar to AMD's 3D V-Cache stacking, where the L3 SRAM cache is packaged directly on the CPU chip, while HBM's allow for higher capacity and are cheaper (IT House note: HBMs are obviously slower than caches as well). One of the main factors plaguing the industry at the moment is that HBM4 requires a 2048bit interface, so the intermediary layer of HBM4 is very complex and costly. Therefore, it would be economically feasible to stack memory and logic chips together, but this raises another issue: heat dissipation. Modern logic chips, such as the Nvidia H100, generate hundreds of watts of heat while delivering huge VRAM bandwidth thanks to the huge HBM3 memory. As a result, it can take a very sophisticated approach to heat dissipation for logic and memory package assemblies, even considering liquid cooling and/or immersion heat dissipation. If the heat dissipation problem is solved in two to three generations, the HBM and GPU will be able to operate as one without the need for an intermediary layer," said Kim Jung-ho, a professor at the Korea Advanced Institute of Science and Technology (KAIST) Department of Electrical and Electronic Engineering. An industry insider told Joongang that "the 'rules of the game' for semiconductors could change in the next 10 years, and the difference between memory and logic semiconductors could become negligible".
The LDK220 series of low-dropout linear regulators (LDOs) is a high-performance device designed specifically for low-power consumption and high-precision voltage regulation, widely used in scenarios such as consumer electronics, industrial control, and po
Xilinx Spartan®-7 FPGA Family stands as a defining solution in the mid-range FPGA landscape, blending high performance, energy efficiency, and cost-effectiveness to redefine versatility for industrial, IoT, and consumer electronics applications. Built on
The Altera FLEX series was more than a lineup of FPGAs—it was a blueprint for how programmable logic devices could evolve to meet diverse market needs. The FLEX 8000 laid the architectural groundwork, the FLEX 10K redefined functionality with embedded mem
Among TI standout offerings, the LM4765 and LM4766 are dual-channel amplifiers designed to cater to diverse audio needs—from compact setups to high-fidelity systems. While sharing the same product lineage, these chips differ significantly in power output,
