SK Hynix Plans On Integrating GPU & Memory Semiconductors Into a Single Package, Defying Industry Trends

wccftech.com:

The Korean memory manufacturer SK Hynix has decided to break from the industry's tradition of semiconductor development, as they plan to integrate memory and logic semiconductors on the same die.

According to reports by Korean outlets, SK Hynix believes there is a possibility of a more "efficient" implementation of semiconductor placements in the market, which the firm believes they can achieve with next-gen HBM4 memory. It is said that SK Hynix has hired a considerable number of design experts for logic semiconductors, which in the first place didn't make sense since the company solely focuses on memory, but with today's development, it seems like SK Hynix wants it to take on the semiconductor markets. What has prompted SK Hynix for a "rapid" change in the ecosystem? Well, we will discuss this later on.

Regarding current-gen products, advanced memory semiconductors such as HBM are achieving efficiency by attaching them as closely as possible next to GPU chips, which are logic semiconductors. Individual computations are separated through a dedicated chip, which, in broader terms, does look like an inefficient method. Packaging techniques such as the CoWoS come in handy to bridge the gap between the semiconductors. However, some sort of "disparity" is still present, which is now intended to be catered by integrated memory and logic semiconductors into a single piece.

Moving on to why SK Hynix desires such implementation, it is simple. Currently, the semiconductor ecosystem order is divided into chip design (fabless), consignment production (foundry), and memory/logic. This not only involves sophisticated equipment for each manufacturing process, but in most cases, the work has to be outsourced to different firms specializing