The company Everspin is tempting our entrance into the world of spintronics by the introduction of its first commercial spin torque magnotoresistive random access memory (ST-RAM) chip. While the tech is still quite new and will probably not be integrated to wide spread applications immediately, the company is selling 64 Mb DIMMS to select manufacturers to explore. The DDR3 form factor of the samples shows how fast speeds and non-volatility can be best applied before its final production.
This device functions by manipulating the spin of electrons to store data in a magnetic state. The technology uses a small current to spin-polarize and funnel electrons through a magnetic tunnel junction (MTJ) (a dielectric tunnel) and one transistor, to transfer its angular momentum to the magnetic layer essentially flipping its polarity. Information is retrieved by measuring the resistance of individual cells.
The EMD3D064M 64Mb ST-MRAM is compatible with JADEC, DDR3 and offered in a wBGA package. The chip can perform 400,000 input/output operations per second (IOPS) and offers 3.2 GB/s bandwidth with nanosecond latency as opposed to 800 IOPS performed by common NAND modules and 120,000 IOPS for the 512GB OCZ Vertex 4.
This technology conserves power because it does not require individual cells to be recharged as with Flash and it uses considerably less power to write than MRAM. The tech is also suspected to be easily and more effectively scaled down in size, with gigabyte densities and increased speeds. Once memory density can reach around 8 Gb it could begin to replace SSD applications in caches, small amounts of memory and any other application that demands ultra-low latency reading/writing like DRAM, RAID storage, high-performance computing and supercomputer nodes.
At the moment, the 64 Mb version is envisioned to be used to compliment common memory storage systems. Everspin believed these ST-MRAM devices could also be used as cloud storage devices. Currently, Everspin is producing ST- MRAM chips on their 200mm production line in Chandler, Arizona but work is underway to produce tools necessary to make a 300mm line.
A big hurdle for the technology will be to find a place in computer systems that have already been established. Gaining a foothold here will ensure that the technology continues to be improved to the point where it could replace existing tech completely, but it will have to play catch-up for a long time.