MRAM technologies: from space applications to unified cache memory?
imec-int.com
imec-int.com
The Ambiq Apollo 4 (Cortex-M4F) includes 2 MB of MRAM and achieves quite a bit of power savings (4 uA/MHz) from the Apollo 3. No idea what proportion of savings comes from new NVM tech vs smaller transistors, better peripherals, or finer voltage thresholding ("SPOT"), and I never checked the claims that MRAM has infinite endurance (as opposed to FRAM) or that computations can be performed from MRAM in the nanowatt regime.
This is though, way, way better than flash. To get to DRAM level it needs 10^17, or ideally 10^18, so it can be used as cache.
It is also not the most power efficient memory, but still way better than any DRAM.
SOT MRAT promises more performance, and endurance, and VCMA promises more power efficiency, and logic compatibility.
The reason I know about this is because back at HP we helped to pioneer the process/technique.
We found that editors and writers could not possibly develop relevant stories given the amount of research required and their publishing deadlines with their budgets.
We realized we could "prime/frame" the stories they published by providing 100% publish-ready stories/articles to them. The person who came up with this idea had worked in the publishing industry for 20 years so she knew reality on-the-ground.
So we identified both applications and product technologies for everything we sold and then created a minimum of 3 "articles" for each combination.
Then when we inevitably got a call from a publisher like EEweb asking for comment/pointers/info about a given market or product, we'd whip out one of the articles and give it to them.
99% of the time, they publish what we gave them without significant/any edits. That was a little creepy but hey it's THEIR due diligence, not ours.
This was started in the late 1990s. Since then ALL NEWS pretty much works this way now which the addition of "Think Tanks", NGOs and "Action Committees" writing the articles instead.
There is NO SUCH THING as "corporate MSM journalism" anymore. It died starting in the 1990s and now 20+ years later it's 100% dead. Only nontraditional VLOGS, BLOGS and such actually do anything that can be legitimately called objective journalism.
BTW I worked on MRAM in the 1980s for the military (I worked on "domain wall technologies" based on bubble memory back then but nothing ever 100% new - applying it to MRAM is an obvious thing) and worked with MRAM vendors in the 2000s, so the history and the content is quite accurate - you can NOT fake the laws of physics, thankfully. But non-technology stuff absolutely can be faked.
Say your embedded design will have a 10 year service life, won't have a network connection, and is required to keep an elapsed time indicator of the powered-on time of some expensive peripheral with a 1 second resolution.
With a flash-based EEPROM, you may have 100K writes before cells start failing from write fatigue. If you update some cells once per second, you'll hit that threshold in about a day. If you relax your requirement to writing once every hour, your flash will last for 10 years, but you're susceptible to inaccuracies if the device loses power before a write can be recorded. You could also integrate a wear leveling scheme at the cost of additional development and test time.
Instead of working around flash's deficiencies, you could replace that EEPROM with MRAM. MRAM lacks enough of a write wear out mechanism for the write interval to matter, so you can safely update a value several times a second for decades without destroying it. This can be done easily from the kind of uC that lacks the memory to support a filesystem or network stack.