Future generations of computer memories could consume drastically less power if a DARPA program run by the Agency's Microsystems Technology Office (MTO) bears fruit. The Defense Advanced Research Projects Agency is trying to come up with memory technology based on a mechanism called spin torque transfer.
Spin-based electronics, or spintronics, is based on the manipulation of electron spin. The electrons in question are the current carriers for electronic circuits. In particular, spin torque transfer-random access memory, or STT-RAM, seems especially promising and could be just two or three years away from widespread commercial use, according to Dr. Dev Shenoy, DARPA SST-RAM program manager.
The energy savings potential of STT-RAM technology is significant. SRAM/DRAM memories currently use approximately 5 pJ/bit for each switching action. The target for STT-RAM is approximately 100x less. Of course, the amount of energy savings a given STT-RAM system will see depends on how often the memory is accessed. In addition, conventional SRAM/DRAM use between 1 nW/cell (SRAM and some DRAM) to 10 fW/cell (DRAM) of power to retain each bit. Thus a typical computer having more than a gigabyte of volatile STT-RAM memory might consume 10 W less power than its conventional equivalent.
STT-RAM could also compete with non-volatile Flash memory. The attraction here is that a Flash write operation takes approximately 10 nJ/bit of energy. In this case, the STT-RAM program target is a level 20,000x lower.
However, to hit such goals, DARPA will have to develop optimized materials and processes. In addition, Shenoy says process yields will have to improve if STT memories are to be economically competitive.
“What we're aiming for, a ‘universal memory,’ one that is very small, fast, and low energy, nonvolatile, and can be fabricated onto a logic chip,” explains Shenoy.
One factor bolstering the use of spintronics is the body of knowledge built up around its commercial uses. Most of today's disk drives use GMR (giant magnetoresistive) read/write heads, which employ changes to the spin characteristics of electrons within layers of the read head as a sensing mechanism. Another spintronic technology, magnetic tunneling junctions (MTJ), has led to development of, among other things, magnetic random access memory (MRAM). MTJ sandwiches a thin insulating tunnel barrier layer between layers of magnetic material. It stores a bit of information as the magnetization of the layers relative to each other. MRAM is nonvolatile — layers retain their field orientations without power — and STT can be used with this structure to electrically change the magnetic orientation of a material. “During STT, a spin-polarized current is used to exert force on a magnetic layer, which results in the layer switching its relative orientation,” says Shenoy.
Similarly, DARPA's STT-RAM program aims to develop materials and processes to exploit STT switching of nanomagnet orientation as a means of creating nonvolatile magnetic memory structures that don't need external magnetic fields to operate. Researchers hope to create a universal memory element able to not just cut power requirements, but also to perform better in multiple electronic systems. The first goal is to develop prototype memory arrays using processes compatible with high yield manufacturing.
Once perfected, STT-RAM could usher in such advances as:
Enterprise data centers where STT-RAM use would reduce the energy needed for cooling by more than 75%
STT-RAM-equipped cell phones exhibiting dramatic improvements in performance and battery life
Instant-on PCs thanks to STT-RAM that replaces both the internal memory as well as the hard disk drive
Memory in space: STT-RAM could be radiation-resistant yet fast.
DARPA isn't the only organization eyeing STT-RAM. According to Shenoy, several semiconductor and technology companies have STT-RAM projects, including IBM, Everspin Technologies Inc., Qualcomm Inc., and MagIC (a division of TDK) in the U.S., and Toshiba, Renesas Technology America, NEC, Sony, South Korea's Hynix, Samsung, and TSMC in the east. Some of these companies, such as Sony, Hitachi, Toshiba, and MagIC, have made significant progress in STT-RAM development, with papers at major international conferences such as IEDM and ISSCC.
|Write energy||0.06 pJ/bit|
|Write/read speed||5 ns/bit|
|Cell area||0.12 µm2|
|1 MB memory|
|Thermal stability: Average stability of the bit to randomly switching states (equivalent to >20 years)|
|Endurance: Minimum number of times each bit can be changed from a 1 to a 0 or back before the bit fails|
Besides spin torque transfer, DARPA's MTO group has an entire portfolio of intriguing low-energy electronics programs in the works. A few of the programs include work on 3-D integrated circuits, hybrid NEMtronics, and Threads. Hybrid NEMtronics focuses on eliminating leakage power in electronics to enable longer battery life and lower power requirements for computing, while Threads is an acronym for technologies for heat removal from electronics at the device scale. For more information, visit www.darpa.mil/MTO.
HOW STT WORKS
Spin transfer torques occur when the flow of spin angular momentum through a sample is not constant, but has sources or sinks. This happens, for example, when a spin current (created by spin filtering from one magnetic thin film) is filtered again by another magnetic thin film whose moment doesn't coincide with the first. While filtering, the second magnet absorbs a portion of the spin angular momentum carried by the electron spins. Changes in the flow of spin angular momentum also arise when spin-polarized electrons pass through a magnetic domain wall or any other spatially non-uniform magnetization distribution. In this process, the spins of the charge carriers rotate to follow the local magnetization, so the spin vector of the angular momentum flow changes as a function of position.
In either of these cases, the magnetization of the ferromagnet changes the flow of spin angular momentum by exerting a torque on the flowing spins to reorient them. So the flowing electrons must exert an equal and opposite torque on the ferromagnet. This torque that is applied by non-equilibrium conduction electrons onto a ferromagnet is what is called spin transfer torque. Spin currents can flow within parts of devices even where there is no net charge current. Consequently, a spin transfer torque can also be applied to magnetic elements that do not carry any charge current.
Sources: Dan Ralph, Cornell University and Mark Stiles, NIST