Your Memory is Getting its OWN CPU ?? – WAN Program February 19, 2021 

Quoting samsung’s own material, “We plan to build upon this breakthrough by further collaborating with AI solution providers for even more advanced PIM-powered applications.”. Clearly it requires some application support. When integrated into AI processors is probably where user code needs not bother with it, and is probably where the “no software changes” thing comes from. The way it’s described sounds akin to an FPGA integrated into the memory itself, which would allow massively parallel processing of in-memory data while reading it, after configuring the FPGA matrix. Great for huge neural networks BTW, but also other applications. Many imaging chips handle data processing this way, so it’s not entirely new, but it’s certainly interesting if it were to be made mainstream, especially due to its programmability. Just how programmable it ends up being in its initial form is yet to be seen, but as GPUs started “merely configurable” and then shifted entirely into “programmable” land, I would expect this to follow a similar path if successful.

The chicken game: Moorhuhn (crazy english in countries outside Germany).

It was a huge fad in Germany for years in the early 2000s