It’s been so long since I’ve seen something move out of the infamous “three to five years away” category and into actual volume production that I can’t actually remember the last time it happened. But Freescale is making it happen now with magnetoresistive RAM (MRAM).
Freescale just announced that they’ve gone into volume production of a 4Mbit (256K x 16-bit) MRAM chip. The newly announced MR2A16A MRAM part, which has been in development for a decade and which sampled back in 2004, already has attracted a number of buyers and should start showing up in products before too long.
Back before Freescale was spun off from Motorola, we covered a couple of Motorola demonstrations of working MRAM chips. At the time, IBM, Infineon, Motorola, and the legion of twenty or so other companies that have been working on this technology were projecting that the first MRAM products would hit the market in 2004. Freescale has beat the rest of the pack, albeit two years behind schedule, and I expect to see a steady march of MRAM product announcements in the next year as the rest of the memory industry begins bringing their own products to market.
The big picture for MRAM
I went fishing about for datasheets on NAND flash, and for devices organized as 16-bit words I found access times (read cycle) in the 50ns to 90ns range. This makes the new 35ns MRAM part quite a bit faster than current NAND flash. However, it’s nowhere near the 3ns to 5ns access time of DDR2 DRAM; 35ns is more on the order of old EDO SDRAM.
So while MRAM’s current access time and capacity won’t make it an immediate replacement for either DRAM or NAND Flash, both of these numbers will scale to make it a major long-term player in the storage market. For large categories of devices that are more sensitive to power than raw performance, MRAM will replace DRAM and Flash altogether. For the longer-term performance picture, MRAM well replace DRAM in all but some niche applications. IBM’s research shows that they can get access times down into the 3ns range, and a paper from earlier this year shows that they expect to see MRAM access times between 5ns and 20ns.
As for hard disk technology, I think it’s pretty clear that MRAM isn’t going to become the default medium for mass storage any time soon. I imagine that many types of portable devices, from laptops to media players, will eventually use a combination of MRAM and hard disk technology. A single large MRAM pool could combine the functions of both main memory and the kinds of backing store/caching technology that’s starting to make its way to market in the form of Intel’s Robson. This fast memory pool would be coupled with hard disk-based mass storage.
MRAM’s biggest medium-term impact will be on the Flash market. MRAM is superior to Flash memory in pretty much every way—it’s faster, and unlike Flash, it can take an unlimited number of reads and writes. As the rest of the industry starts to move into volume production of MRAM, the new technology is going to start moving into niches currently occupied by Flash and begin squeezing it out in some places.