In-memory-vs-near-memory-computing
Models demonstrate various computers at a South Korean discount chain Home Plus. Many scientists work on futuristic chipsets, which would revolutionize semiconductor architecture and computer performance. Photo courtesy of Home Plus

Which is better future of semiconductors?

Many corporations and research institutes have worked on designing better semiconductors, and one of the alternatives has been memory-centric chip technologies.

The idea is to bring the memory closer to the processing tasks so as to deal with the bandwidth bottleneck issues and speed up the whole system.

There are two competing memory-centric technologies dubbed in-memory computing or near-memory computing.

The former is about merging memory and system semiconductors, just as a Samsung Electronics team announced this week through its paper on the peer-reviewed journal Nature (see the Korea News Plus article on Jan. 15).

Otherwise dubbed near-data processing, the latter is to incorporate the memory and logic chips in an integrated circuit package _ they are installed nearby but are not merged.

In the semiconductor architecture, data is supposed to move between a memory chip and a processor. The movement itself reduces the performance of chipsets by causing power consumption and latency.

In-memory computing is geared toward removing the movement, while near-memory computing is aimed at minimizing the movement. Both approaches have the same target of increasing the performance of chipsets.

For scientists working on in-memory computing, non-volatile memories like RRAM and PRAM have been actively used. But it has been difficult to use MRAM despite its advantages.

Samsung claims that it has come up with advancements in the MRAM research through the Nature paper.

Experts point out it remains to be seen which would become the better technology at the end of the day. Both might be commercially viable, or both might not be so.

“In-memory computing is a revolutionary path, while near-memory computing is an evolutionary path,” Prof. Han Tae-hee at Sungkyunkwan University said.

“Many have developed the memory-centric chip technologies, and there have been some breakthroughs. But the new technologies cannot be adopted overnight because they require a whole new ecosystem, including different operating systems and central processing units.”

KPUB-Banner-Ads