Everspin's latest MRAM is designed to fit on a standard DDR3 DRAM DIMM (see Magentic DRAM Arrives). It replaces Everspin's original MRAM employed spin-torque write technology (see Magnetic Cores To MRAM: Nonvolatile Tipping Point?) with its newer spin torque write technology (Fig. 1).
Figure 1. The DRAM MRAM chips employ Everspin's spin-torque write technology that replaces the earlier toggle write approach.
Everspin wanted to get the new technology into the hands of those that needed it and could afford the tradeoffs in price and power. This turns out to be the enterprise space where flash storage is hot. Flash is moving on-board via PCI Express and is even providing non-volatile backup for DDR3 DIMMs as with Viking Technology’s ArxCis-NV (see The Fundamentals Of Flash Memory Storage). The DDR3 DRAM DIMM form factor is ubiquitous so that is what Everspin came up with (Fig. 2), a DDR3 DRAM DIMM composed of MRAM chips.
Figure 2. This is no ordinary DRAM DIMM. Those are MRAM chips.
Remember when a 64 Mbit memory chip was cutting edge? DRAM has moved past this point but a 64 Mbit MRAM in a JEDEC DRAM form factor is still significant. Unlike Viking's approach that requires a supercap, MRAM does not require a backup power source. MRAM does not have the speed tradeoff or lifetime of flash memory.
At this point, MRAM is still more expensive than flash and DRAM but it is as fast as DRAM with the non-volatility of flash. There are lots of enterprise applications that can take advantage of MRAM even though it requires more power than flash.
MRAM takes on flash storage by being more efficient "per operation." A 64 Gbit flash memory delivers about 800 IOPS at 80 mW. The 1 Gbit Everspin MRAM delivers 400K IOPS at 400mW. MRAM is still 50 times more expensive at this point but is more cost effective overall.
Everspin's MRAM DRAM chips use the standard JEDEC DRAM pinouts. They are available in 16Mbit by 4, 8Mbit by 8, and 4Mbit by 16 configurations.
DRAM still has the capacity edge and is supported by more vendors but MRAM and non-volatile approaches like Viking Technology's are changing the way developers of operating systems and applications think. In a sense, this is moving backwards to when core memory was dominant. It was possible to turn off a mainframe and have it back up an running in seconds. This was due to the fact that everything in main memory remained intact when power was turned off.
It was not simply a matter of handling power failures. It was the way things worked.
Designers make certain assumptions about their environment. Volatility changes the way one thinks and what types of tricks can be used in applications.
So will we ever get back to the days of core memory where everything was non-volatile? Maybe but it will be a few more years yet. Plenty of time for people to adjust.
In the meantime, MRAM will make inroads and not just in the enterprise. Embedded systems is ripe for this approach. Flash memory has changed the way designers work. It can be put on a board eliminating old rotating media. It makes mobile devices with decent battery life possible. MRAM is just the next step.
