The integration of memory into CPU components directly is nothing new. AMD already does it with their 3D V-cache technology.
Recently, leading memory maker and developer SK Hynix reportedly wants to try to do the same by combining their HBM4 (High Bandwidth Memory 4) memory with processing components, and they are seen to have asked NVIDIA about this collaboration.
Typically, how the combination of processing chips and memory will happen is that on the same circuit board, your processing chip and memory chip will be connected to each other using an interposer or an electrical interface that allows these two components to send data to each other using electrical signal.
SK Hynix's idea is to not use this interposer component, and directly connect the HBM4 memory component directly to the processing component circuit. The advantage of using this development technique is that it reduces the cost of manufacturing complicated and expensive interposers, and simplifies the design of SoC components or processing chips.
However, such a design has its own issues, such as the temperature of the components when they operate at 100 percent. Apart from the processing chip components, HBM memory is also seen to consume a lot of power. Therefore, apart from the high power consumption, the temperature of the components is also seen as one of the issues of this idea.
Among the ideas given include the use of a liquid cooling system, or immersing these components in a cooling liquid that does not conduct electricity. However, it is said that for this method to really work well, the memory chip should be developed using the same process as the logic chip (CPU).
The problem is that at this point, creating and manufacturing such memory would cause memory costs to skyrocket, and for the foreseeable future it is not something to be seriously considered.
For now, it is reported that SK Hynix and NVIDIA will create an all-new chip as a concept to see if this idea is something that can be used in a real-world scenario.