Memory Technologies - overview
IBM RAMAC hard disk drive circa 1956
IBM played a pioneering role in developing traditional memory technologies, and is now actively pushing the limits of emerging memory technologies. IBM developed the first magnetic hard disk drive and invented DRAM, and we are now developing next generation options like magnetic random access memory (MRAM), resistive random access memory (ReRAM), and phase change memory (PCM).
Traditional Memory Technologies
IBM was a pioneer and made fundamental contributions in the invention and development of dynamic random access memory (DRAM). The simplicity, low cost and low power consumption of DRAM, when combined with the first low-cost microprocessors, opened the door to small personal computers. Today, all PCs, notebook computers, game consoles and other computing devices are loaded with DRAM chips. DRAM also powers mainframes, data center servers, and the machines that run the Internet. IBM Fellow Robert Dennard was awarded the US National Medal of Technology in 1988 by President Ronald Reagan for his invention of DRAM.
Emerging Memory Technologies
IBM continues to innovate and drive advances in memory technology. These forward looking technologies include magnetic random access memory (MRAM), resistive random access memory (ReRAM), and phase change memory (PCM).
These assorted technologies have wide ranging applications across existing and emerging technology sectors. With unique combinations of power consumption, speed, and non-volatility, emerging memory technologies have the potential to replace DRAM, SRAM, or flash memory in certain applications such as ultra-low-power devices for internet-of-things (IoT) applications. MRAM could eventually serve as a fast, dense, cache memory. ReRAM and PCM show potential as storage class memories. The near-analog range of states available through ReRAM and PCM are also promising candidates for incorporation into complex neuromorphic computers.
Magnetoresistive Random Access Memory (MRAM)
What is MRAM?
MRAM is the generic term for a solid-state memory that uses ferromagnetic materials to store information in a way that can be addressed with traditional electrical currents and voltages. This is a subset of a larger field called spintronics, which is the name for devices that aim to make use of the quantum mechanical spin degree of freedom of an electron in the operation of the device.
There are some distinct advantages of MRAM over traditional memories. MRAM memory elements are predicted to scale to smaller sizes than traditional DRAM memory elements. Through the use of ferromagnets, MRAM adds non-volatility as a key advantage to blend the speeds of DRAM with normally-off storage of flash memory. Additionally, the use of ferromagnets leads to an inherently radiation-hardened memory element.
IBM's J. Slonczewski and D. Worledge at the 20th anniversary STT-MRAM symposium.
IBM's J. Slonczewski discovered the possibility of Spin-Transfer-Torque(STT)-MRAM with his prediction that ferromagnets can be manipulated using electrical currents. Since then, IBM has driven innovation within the MRAM community including leading edge research on using magnetic materials with out-of-plane oriented magnetization, and pushing the extremes of how small an individual memory element can be.
MRAM relies on the concept of magnetoresistance (MR) as a way to read out the two states needed for a memory. This process uses two separate ferromagnetic layers in a single structure separated by a thin nonmagnetic spacer layer. The relative magnetic orientation of these layers, either parallel or antiparallel, gives two different electrical resistances. In the parallel state, the resistance is low, and in the antiparallel state the resistance is high. This mechanism was first discovered in all metallic structures where it has the name giant magnetoresistance (GMR) and received the Nobel Prize in Physics in 2007. In GMR, the resistances are all very small. Device resistances are a few Ohms and resistance changes are a few percent, or milliohms.
The prospects for MRAM increased with the discovery of tunneling magnetoresistance (TMR). In TMR structures, called magnetic tunnel junctions, the spacer layer is an ultrathin insulator such as aluminum oxide or magnesium oxide that is only 1-2 nm thick. The memory elements in this case have resistances of a few kOhms up to hundreds of kOhms. In these tunneling structures, the change in resistance also is much larger, with changes of 50% up to 600% in the best case (relative to the smaller resistance). This increase in separation of the two resistance levels, and resistances compatible with traditional CMOS circuitry led to increasing interest in creating an operating memory array with magnetic elements.
When research started on MRAM, the only known way to manipulate the magnetic orientation of ferromagnets was by using an externally applied magnetic field, either generated by being in close proximity to another ferromagnetic material or by generating a field by flowing electrical current through a nearby wire.
Using carefully timed current pulses, the magnetic orientation of the storage layer of an MRAM bit can be reversed while the reference layer stays in the original orientation. However this approach turns out not to be scalable—the current required to switch the bits increases as the bits are made smaller. Hence work on development of field-switched MRAM has now ended, with attention turning to Spin-Transfer-Torque MRAM.
Spin-Transfer-Torque MRAM (STT-MRAM)
Example STT MRAM Bit-Cell.
STT-MRAM is a leading competitor to be the next big change in the evolution of computer memory. In STT-MRAM, a traditional electrical current can be used to change the magnetic state of the tunnel junction. By flowing electrical current through the two ferromagnetic layers, the STT effect can be used to set the memory element into the parallel or antiparallel relative magnetic orientation without the use of an externally applied magnetic field.
Phase Change Memory (PCM)
In PCM, the memory element changes physical state between an amorphous and a crystalline phase that have different electrical conductivity properties. The memory element can change between these two states in a controlled way using properly designed electrical current pulses.
The PCM cell is programmed with fast current pulses which heat the material to crystalize or melt-quench the phase change material. a) PCM cell in the low resistance, poly crystalline phase. b) PCM cell in the high resistance state showing an amorphous dome of material covering the lower electrode. c) PCM layers embedded in a CMOS technology. d) Zoom-out TEM image of PCM cells. a) and b), Phase Change Materials: Science and Applications, Ed. S. Raoux and M. Wuttig, 2008. c) and d), G. Close et al., VLSI 2011.
IBM Research has been a pioneer in the field of Phase Change Memory for over 15 years and continues to have a strong effort in refining the PCM device itself and also in exploring applications both inside and outside of the traditional memory applications. From materials research and process and integration development, to circuit and systems design and applications, IBM's efforts span the full range of atoms to applications in this field. IBM research labs from around the globe are combining efforts to continue to break new ground on hardware based neuromorphic applications tapping into the analog-like nature of the range of resistance states achievable by PCM.In PCM, the memory element changes physical state between an amorphous and a crystalline phase that have different electrical conductivity properties. The memory element can change bewteen these two states in a controlled way using properly designed electrical current pulses.