David Kanter (dkanter.delete@this.realworldtech.com) on November 19, 2018 1:40 pm wrote:
> Here's my second quick article from VLSI 2018. My first article focused on machine learning
> accelerators, while this second article is on emerging memories - specifically MRAM.

The technology might already be sufficiently mature to replace NOR flash in some microcontrollers (which use more mature process technologies both to get embedded flash and for cost reduction given I/O die size constraints). Execute-in-place, limited data logging and configuration storage would generally not require huge capacities.

In some applications, persistence is a significant energy efficiency consideration. (Sadly, in my opinion, persistence does not mean non-destructive reads. Some applications can tolerate destructive (or disturbing) reads and could benefit from better density, but low-energy execute in place would seem more tolerant of lower density if it allowed non-destructive reads. Currently, MRAM appears to provide non-destructive reads, but I would not be surprised if that might be sacrificed in the future for greater density or other feature.)

There probably are uses for small distributed persistent memories (perhaps localized, possibly more secure configuration storage?). If the lower density of small localized memories does not significantly impact chip-level density, there might be uses for such lower-density persistent memory.

(For some applications, routing rather than a binary (variable strength) signal might be useful. If the orientation of an anti-fuse/phase change material could be controlled, such a "memory" might be used in FPGAs and perhaps some other areas.)

Aerospace applications would also be concerned with radiation tolerance and magnetic field effects.

[snip]
> While MRAM is in production with Everspin, there is still a huge amount of research
> around the basics. I hope to write an additional article about MRAM in the future,
> since there are so many idiosyncracies around new memories.

Idiosyncracies can be opportunities for optimization in application targeting, system hardware design, and software design.

I am extremely ignorant in this area, but memory technologies do seem interesting.