TXN has been making some interesting presentations since the beginning of their FRAM/RFID development powered by their MSP430 mpu product. This is taken from one earlier this year given before the 6/30 announcement of the 8Mb F-RAM:
FRAM Technology Snapshot
•FRAM
Is a Random Access Memory: Each bit read/written individually
Similar to DRAM except data stored in crystal state –not charge
Is non-volatile –retains data when power is turned off (like Flash/EEPROM)
Has read/write access and cycle times similar to DRAM (30-50ns)
Operates at low voltage (1.5V)
•Market Status
Volume production ongoing over last 8 years (Ramtron, Fujitsu: < 1Mb)
Moving to high-density (TI, Samsung, Toshiba, Fujitsu, IBM)
•TI focus:
Embedded FRAM (2 mask adder to digital process flow)
FRAM as a true NVRAM technology to replace cache SRAM, DRAM, Flash / EEPROM
1.5V operation for low power
focus.ti.com
I don't know how much of that is just PR spin, but I do know that Samsung has been selling embedded FRAM in a LSI suitable for RFID apps for some time:
samsung.com
Samsung also has produced a 4Mb FRAM and has been a part of a consortium working on nano-logic IP with IBM, Chartered, Samsung,IFX:
samsung.com
More interesting to me is this Samsung abstract on an embedded MFM embedded FRAM in a stacked structure which can not only operate at the 800MHz speeds of the new EPC Gen2 UHF smartcard standard, but do so using less die space than required by traditional FLASH or EEPROM IP despite the old IP's use of more advanced process nodes:
adsabs.harvard.edu
I haven't found that Samsung paper itself... but perhaps more importantly... I did find details in this Hynix paper from 2008 which describes a stacked MFM structure with identical characteristics:
etrij.etri.re.kr
If you go back two years, EPC Gen2 products were being produced at anywhere between 0.90um and 0.40um process nodes like those referenced in this article focused on Impinj NV-memory IP:
rfid-world.com
Unless these presentations are simply vaporware... it appears that companies are now ready to move RFID ICs to a range of design rules from 0.25um to 0.130/0.90um and many of the larger producers are at least "thinking" of doing so with F-RAM based designs.
It appears that Impinj IP was less than adequate to the task at any design node. Its a private company but this site indicates its revenues at only $750K/year and it sold its NV-memory IP to Virage in 2008 for only $5.2M:
bizjournals.com
Whether FRAM/F-RAM is "adequate" to the task... and whether the market size is sufficient to allow so many large IC companies to make serious money without driving each other to the poor house is the basic question.
It appears the Fab companies are only pushing the RFID market as far as their fab equipment depreciation schedules will allow. It doesn't sound very much like a high profit product line.
The most important aspect of all this may simply be the MFM stack structure that we now know at least the Korean IC makers companies are promoting. It may well be best suited for RFID apps at this point but put it together with the NV-logic work Rhom and RMTR have done and it is clear that the ability to stack F-RAM directly on/in logic and analogue circuits at the FEOL on a CMOS compatible fab line is being accepted by fab engineers. That suggests that TXN's "proposal" of F-RAM as a SRAM cache replacement isn't so far fetched as it might seem.
Certainly in mobile apps, the densities already exist for that job and if F-RAM IP can be pushed to a cycle endurance >10^14... well... a NV product that can compete as a low power system cache (mobile or not) should be worth some real money... real big money. Neither MRAM nor PCMRAM are likely to ever be manufactured at the FEOL in a CMOS factory.
But that's a huge disruption in a market where the old IP is still holding all the slots at the cutting edge design rules.
I think I'll have to see it to believe it.
0|0 |
