Microsemi Licenses Crossbar ReRAM Non-Volatile Memory(anandtech.com)
anandtech.com
Microsemi Licenses Crossbar ReRAM Non-Volatile Memory
https://www.anandtech.com/show/12758/microsemi-licenses-crossbar-reram-nonvolatile-memory
9 comments
Hah, yeah that wasn't a substitution I'd seen before. They really should have just called out Fin-FET process compatibility since all of those are below 20nm.
1x, 1y, and 1z nomenclature is standard in the memory industry, where the foundries are less keen on giving out hard numbers about their manufacturing processes than the logic fabs are.
FinFETs don't have anything to do with ReRAM (or MRAM), since the memory is a back-end construction. When designing the memory, you don't have to care about whether the underlying transistors are planar or FinFET, you just need to know whether the memory can scale down to the same pitch. There will be 12nm planar FD-SOI chips taping out next year at GloFo, so 1x nm for a logic process doesn't imply FinFETs, and FinFETs aren't all smaller than 20nm.
FinFETs don't have anything to do with ReRAM (or MRAM), since the memory is a back-end construction. When designing the memory, you don't have to care about whether the underlying transistors are planar or FinFET, you just need to know whether the memory can scale down to the same pitch. There will be 12nm planar FD-SOI chips taping out next year at GloFo, so 1x nm for a logic process doesn't imply FinFETs, and FinFETs aren't all smaller than 20nm.
To be clear these (Resistive RAM and Magnetic RAM) are really competitors to Flash (EPROM) rather than DRAM or 3D XPoint from Intel. Their main advantages are process simplification (Flash requires a LOT of extra layers) and cost rather than speed. In the industry people are very careful with new Non-volatile memories, because the failures may have long (delayed) failure modes that cause horrible problems and liability.
I'm fairly positive on these two choices (Re&M-RAM) mostly because people have learned to be fairly sceptical and require a lot of testing and they are porting to mulitiple Fabs. Even with the license it will be at least 12 months before there could be customer samples from the first licensee, Microsemi.
I'm fairly positive on these two choices (Re&M-RAM) mostly because people have learned to be fairly sceptical and require a lot of testing and they are porting to mulitiple Fabs. Even with the license it will be at least 12 months before there could be customer samples from the first licensee, Microsemi.
MRAM has so far been primarily a competitor to SRAM and NOR flash, but definitely not to NAND flash. ReRAM as an embedded memory will be competing in a similar space, but as a discrete memory will probably be most comparable to 3D XPoint.
I assume MRAM and FRAM refer to the same thing? If so Texas Instruments have some pretty cool (I think) microcontrollers using FRAM allowing for a user defined spit between volatile and non vatime memory, the MSP430FRxx series, I don't think they have it in any ARM parts though.
No, FRAM is quite different from MRAM. A FRAM cell has a single ferroelectric element that is read destructively (just like magnetic core memory). A MRAM cell has a pair of magnetic elements and resistance through one of them varies depending on whether the magnetization is aligned with the other element. That resistance is measured non-destructively.
Microsemi primarily deals with the aerospace industry and manufactures radiation-tolerant FPGAs; Their latest product line is a series of Flash-based reprogrammable devices (RTG4). Perhaps these alternative memory technologies lend to better data integrity.
ReRAM is, but MRAM was definitely positioned as a competitor to eDRAM and SRAM for low end applications (it's generally unstable at higher frequencies)
I was confused for a bit.