Everspin and Globalfoundries are announcing pMTJ ST-MRAM for GF’s SOI process. SemiAccurate thinks this is important because of its relationship to flash or at least embedded flash.
The short version is that Everspin’s perpendicular magnetic tunnel junction (pMTJ) Spin Torque MRAM is now in production as discrete chips at GF. Just for fun Everspin built a PCIe SSD out of those 256Mb chips to show off their capabilities. It may be only 1GB in size but this PCIe SSD, effectively an FPGA and memory, was pushing some pretty astounding numbers. How does 1.5M IOPS at 4K block size strike you? What if I told you it was 100% writes?
Now that part is old news, these are production devices we are talking about, what is the new news? GF is now licensing Everspin’s tech for use as embedded memory in IOT type devices on SOI processes. GF has a 22nm SOI process and recently announced a 12nm variant. Everspin is only commenting on the 22nm version but there are no fundamental roadblocks to shrinking it to 12nm, it should scale just fine unlike some other technologies that depend on a capacitor or similar charge trap.
This is important because if you look at the availability of embedded flash, it is only starting to appear on 40nm nodes now, smaller, well plan on waiting. If you want a lower geometry process with embedded memory, you have a choice to make, usually which external memory you want to use. This is costly and can be power-hungry, especially for IoT type devices. SOI may be expensive up front but if you can add your memory on die, it fundamentally changes the economics for the device.
Everspin says their MRAMs can be optimized for the SRAM space, DRAM space, or flash space, and by space we mean performance space, not physical area. If you want you could replace any of the three types of embedded memory with MRAM, but would that be a good idea? The discrete chips are at DDR3 performance levels so pretty good there, and they are 1T devices too. SRAMs are 6T or 7T devices, flash has some high voltage components, and other technologies have their own building blocks. Everspin is confident they can offer the same storage capacity in the same or smaller areas as embedded memories like SRAM or flash.
That is why this announcement with Globalfoundries is so interesting, if the two companies can offer ST-MRAM as an off the shelf component for 22nm, they have an offering no one else can touch. If you need embedded NVRAM or even flash you are stuck on 40nm or higher process nodes, 22nm is a lot smaller than 40. SOI and smaller nodes are more expensive than 40nm but 22SOI leaks a lot less. This is the long way of saying it is a multi-dimensional tradeoff that will work really well for some and not at all well for others. No matter what happens, it will be a unique offering for Globalfoundries.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018