An annotation regarding the 3.5 years above: This number is pure calculatory with little practical importance. In a real-world application, endless, consecutive accesses at maximum speed to one and the same cell do not occur. Thus, the limited number of access cycles is more or less of academic value.
Generally, I think that any niche should be big enough for specialized manufacturers to continue with their plans and that it would indeed be nice, if smaller companies could resist to sell themselves for a couple of stacks. I also think Ramtron has built a solid foundation with their products to continue but there is a lot of work to do to develop FRAM technology further in order to overcome its drawbacks that somehow restrict its use. Will that happen? Companies are in business to make money and those transaction discussions regarding Ramtron make me think they will not continue their current operations long enough to refine their excellent products.
Embedded FRAM can surely play an important role. But I think how they get paid for it is just one thing.
As far as I know, TI is producing for Ramtron so it seems to be a keen thing for TI to integrate FRAM into their own MCUs. This might also be the reason that Ramtron no longer offers their own line of 8051 compatible MCUs that came with FRAM integrated (for quite a long time so the idea of an FRAM based MCU is neither new nor revolutionary). I do not think that FRAM will leave its niche in the near future. There are technical constraints and the trend to even use high speed memory in applications where it is not mandatory that seem to restrict a widespread use of FRAM, embedded or stand-alone.
FRAM is quite slow, with access times around 50-60 ns, and precharging stretches access cycles over 100 ns. Even in slower systems, this may require special attention (this is what we learned using parallel FRAM). The same fact restricts Ramtron's state savers to maximum frequencies of 1 MHz which restricts their use. Next, FRAM does wear out. A cell can last over 1e15 access cycles so practically, this might be of no concern. Our systems work with an 8 MHz bus and can use FRAM in place of SRAM. The devices meet bus timing requirements very well, data is immediately non-volatile and before the FRAM fails, the system could work for more than 7,900 years. Apparently excellent, especially when taking FRAM's radiation resistance into account. But speed can matter: An MPU working in the same application at maximum access time cuts the FRAM memory system MTBF down to 3.5 years. This may represent problems so the use of FRAM is restricted to more or less slower devices or those that take its characteristics into account. Looking closer at TI's Wolverine µC also shows up this fact.
Reliable it may be, but integrating FRAM is not necessarily trivial or cheap, so I think the market will always be a niche and using FRAM in the manner TI does might not be feasible for others.
I think the stand-alone FRAM market is a niche, but embedded FRAM as a non-volatile memory for microcontrollers appears to have some appeal (see TI's Wolverine).
This is probably why Cypress is interested.
The question is how does Ramtron get paid for embedded FRAM?