The Universal Memory Dream Dies

The long-held industry dream of the universal memory is not the way the industry is actually going – instead the memory business is fragmenting into a horses-for-courses diversity.

“Everyone is looking for a universal memory but it’s more like the world is going for specialty memories – phase change, resistive RAM, nano-RAM – there’s not going to be one solution that fits all,” Phil LoPresti CEO of MRAM specialist Everspin tells me.

Tsugio Makimoto of Makimoto’s Wave fame, always said he’d present the Makimoto Prize to anyone who invented the universal memory which would fit all slots and be the best solution for all applications – fast to read, fast to write, non-volatile, low-power, high endurance and dense.

But now the industry is going the other way with a plethora of new technologies finding niche applications.

For instance the difficulty of getting embedded flash to scale – it’s still not shrunk below 40nm – means that companies release aggressive nodes and then have to embed memory made on a node that is several generations older.

To get round the anomaly, Everspin has inked a deal with GloFo to put spin-torque switching MRAM, which does scale aggressively, on GloFo’s 22nm and 12nm FD-SOI processes and 14nm finfet process.

As well as scaling well, MRAM has a 1T cell which gives it density advantages over 6T SRAM and flash which needs a large transistor for programming, while embedding flash on SOI is ‘almost impossible” says LoPesti.



  1. Of course you’re right Chris and of course eflash will scale lower, no question. Everspin’s point is simply that MRAM scales easier.

  2. Correction to above – TSMC already announced embedded flash on their 28nm node during their Symposia. Not in volume production yet admittedly (40nm is) but it is coming, contrary to the statement above and it is scaled correctly.

    There may also be further embedded flash nodes below 28nm.

Leave a Reply

Your email address will not be published. Required fields are marked *