GlobalFoundries Infuses 22FDX Platform With ST-MRAM Memory

[Guest]

GlobalFoundries announced that it is producing embedded magnetoresistive nonvolatile ST-MRAM for use in its 22FDX platform.

GlobalFoundries Infuses 22FDX Platform With ST-MRAM Memory : Read more

 

[heffeque]

This looks interesting...

 

[blazorthon]

MRAM has long (like decades) been a viable memory technology. The biggest problem with it is just that most companies that work on MRAM don't work together, so it tends to not go far. It has been used in a lot of small embedded designs, especially those where non-volatility and high tolerance for heat and radiation are important. Mainstream usage has been low despite it's many advantages. It can be made far denser than SRAM while being similarly fast, non-volatile, having lower power consumption, and not being too difficult to produce, but like most memory technologies that didn't quickly hit mainstream, it never really got enough fab allocation. The companies don't really like spending the big money on products that they aren't 100% certain they can sell, so they mostly just go on producing products they're already selling, like DRAM.

 

[mavikt]

I'm not sure I get it... how sensitive are these MRAM's to external magnetic fields?

 

[alextheblue]

I want to hear more on the 22nm and future 12nm FD-SOI processes.

 

[photonboy]

I'm not sure this is going to compete against Intel's 3D XPoint very well though.

 

[gondor]

"300nm CMOS wafers"

Now those are some tiny wafers!

 

[aldaia]

I'm not sure this is going to compete against Intel's 3D XPoint very well though.

MRAM and 3D XPoint are not competing products, they fit in diferent segments.
MRAM is fater than DRAM, some say its as fast as SRAM, 3D XPoint is 5-8 times slower than DRAM
3D XPoint is about 1000 times more durable than NAND, while MRAM is "tens of millions more times endurant than NAND"
MRAM has similar density to DRAM, while 3D XPoint is expected to have packing density 8-10 times greater than DRAM

Short term I see MRAM as a replacement for SRAM and DRAM, while 3D XPoint can be a replacement for NAND
Long term, if MRAM can be made more dense and 3D XPoint can be made faster, then , yes i can see them competing. But I bet more on a new futuristic technology that replaces them both.

 

[blazorthon]

I'm not sure I get it... how sensitive are these MRAM's to external magnetic fields?

https://www.everspin.com/frequently-asked-questions

So, this got me going. First, I looked up a data sheet of one of their current products:
EST00352_MR4A16B_Datasheet_Rev11.4 031116.pdf

Scrolling down, the max magnetic field strength is listed as 8000A/m. I fond a website saying that a strong neodynium magnet has a maximum field strength of about 12,000 Oe. Converting units gives 8000A/m = about 100Oe. So, like with hard drives, don't stick strong magnets near this stuff. A fridge magnet is probably safe, but any decent magnet should be kept away.

Other MRAMs products might have more or less shielding. I'm not going to look through all of their data sheets to find an average or anything, so take these results with a grain of salt. I can't say whether the MRAM made on the newer process in the article will be significantly different regarding sensitivity to external magnetic fields when compared to the product from the data sheet I found.

Other links I used:
http://www.wikihow.com/Determine-the-Strength-of-Magnets
http://www.translatorscafe.com/cafe/units-converter/magnetic-field-strength/oersted-%5Boe%5D-to-ampere-turn-per-meter-%5Bat/m%5D/

 

[Kewlx25]

I'm not sure I get it... how sensitive are these MRAM's to external magnetic fields?

Electrons from adjacent atoms are probably exerting magnitudes more electro-magnetic stress on the MRAM than an external magnetic field.

I would love for someone with a physics background to give a more definitive answer.

 

[summakor]

MRAM is hard to do because it's not electrical. SRAM, DRAM and Flash are fundamentally all based on silicon transistors with more or less capacitance between the elements, depending on how fast you want it to be, how much power you're willing to use and how long you want the data and the device to endure. MRAM is truly different. Half a century of breakneck progress on Si transistors doesn't help with the physics of tiny magnets. The parent industry of MRAM is hard drive read heads, an industry that's been losing market share, R&D investment and staff for two decades. Magnets - how do they work??

 

[blazorthon]

MRAM is hard to do because it's not electrical. SRAM, DRAM and Flash are fundamentally all based on silicon transistors with more or less capacitance between the elements, depending on how fast you want it to be, how much power you're willing to use and how long you want the data and the device to endure. MRAM is truly different. Half a century of breakneck progress on Si transistors doesn't help with the physics of tiny magnets. The parent industry of MRAM is hard drive read heads, an industry that's been losing market share, R&D investment and staff for two decades. Magnets - how do they work??

Direct quote from the article that debunks everything you said here:

MRAM comes in many flavors, but Everspin uses Spin-Torque (ST-MRAM) technology. MRAM stores data magnetically, which is in stark contrast to electron-based data storage techniques. MRAM retains all stored data when the device loses power (which means it is a persistent non-volatile storage medium). A perpendicular magnetic tunnel junction (pMJT), which HDD vendors use in read heads, serves as the memory cell. It uses standard CMOS transistors. The design consists of two ferromagnetic plates with an insulating dielectric layer between them. One plate is set to a known polarity (fixed layer), and the other plate's polarity can be changed (free layer). If both plates have matching polarity, the cell will be in a low-resistance state (1), and if they have different polarity, the cell will be in a high-resistance state (0). The device measures the resulting electrical resistance to determine if the cell holds a "1" or "0." The cell does not have to be erased before it is rewritten, which simplifies management and increases performance.

What the hard drive industry is doing is irrelevant. The hard drive industry could all go bankrupt and that wouldn't directly matter because the hard drive read head is not a needed component for making MRAM; it is merely a product that uses a similar concept. Both my car and a little RC helicopter drone use motors and have batteries, but either car or RC helicopter industries could go out of business without harming the other.