Do the math, people!
"the company is moving from a fab to an intellectual-property (IP) model" + "Unity is in the final stage of completing a deal with a major memory company" = "we are selling our IP to Micron for next to nothing."
Leakage current plagues all switching devices: memories, diodes, transistors. Even stored charge leaks. You have to operate orders of magnitude above leakage, preferably. Another interesting consideration, is whether voltage or current is the cause or the effect.
Having worked in this area for many years, I feel most of the next-generation memories have serious challenges to implementation. Widespread adoption will take at least 5-10 years to happen, if at all. Have listed a few challenges to each memory type below (these challenges are well-known in the industry).
(1) Unity CMOx - Retention is well-known to be a challenge (10 years 85C or even 70C has not been demoed by Unity yet at 1-3uA reset current), getting large array sizes required for competing with NAND needs a bidirectional selector which is again very challenging, Unity uses some pretty exotic fab-unfriendly materials like Pt and PCMO. On the positive side, I have never seen 1uA reset current demonstrated by anyone else.
(2) Phase change memory - Reset current numbers are greater than 100uA at 45nm, and these numbers don't scale with area if you want to keep thermal disturb a non-issue (Ireset scales with linear dimension). Essentially, when you scale to 16nm, wiring resistance shoots up exponentially, but reset current is still very high, so IR drop budget increases. Let's say you have 40kohm wiring and periphery resistance at 45nm, then IR drop budget is 40kohm * 100uA = 4V!!!! This restricts array size, and makes it hard to compete with NAND, which is the biggest market for NVM today. On the positive side, it is a fairly mature technology, and if executed very well, could give NOR a run for its money.
(3) RRAM - Requires bidirectional selector, there is a tradeoff between retention and reset current that is difficult to manage, reset current doesn't scale (for the same filament width and retention performance). On the positive side, doesn't require exotic materials.
The IP model requires extensive legal expertise and very deep pockets. Tessera, Rambus, Interdigital & Tivo all have big scares to show. Multinationals (apart from Japanese) may opt to try to bleed to death a startup by draging it through the US court system instead of taking a license. With legal costs at up to 100k/day, they easily succeed with 95% of the adversaries and can eventually sign the surviving 5% with little penalty.
Timing is critical. As long as the perception is DRAM and Flash have some ways to go, the chance to replace these directly is next to impossible. But these other memories are excellent in their own right (higher speed, higher endurance, for example), they could drive some other applications in some niches here and there. RRAM has an advantage in requiring virtually no new materials.
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...