You know that feeling. For the last few hours, you have been drafting a critical report that is due today. You have been surprisingly lucid and efficient in your writing and are anticipating the imminent relief that will accompany sending the report off. Then, to your horror, the screen freezes, leaving you with no choice but to shut your machine down.
This scenario inspires terror because of the type of memory embedded in your computer. Dynamic random access memory (DRAM) is a volatile memory technology, meaning it requires power to maintain the information stored inside it — hence the irretrievable loss when the power source is cut. Despite this inconvenience, the high speed and endurance achieved by volatile memory systems to date make them the primary choice for temporary data storage on personal computers. When you save a document, the information is then ‘written’ onto a more permanent but slower non-volatile storage option such as a hard disk drive that makes your files magically reappear on start-up.
The only non-volatile memory contender currently on the market is flash, which is limited to smartphones, tablets and cameras. But researchers at the A*STAR Data Storage Institute (DSI) are contributing to the emergence of two non-volatile memory systems — spin-transfer torque magnetic random-access memory (STT-MRAM) and resistive random-access memory (RRAM) — that could replace both DRAM and flash. These systems are also being integrated into large storage and computing architectures that may blur conventional distinctions between memory and storage.
Computers store and process information as binary digits, or bits, represented as one of only two values — true or false, yes or no, plus or minus, on or off, one or zero. STT-MRAM and RRAM both speak the same binary language.
In STT-MRAM systems, each unit or ‘device’ is made of two small magnets, whose orientations can be changed by passing a current through the device. When both magnets face the same direction, the resistance in the device is low, but the resistance significantly increases when the magnets are made to face in opposite directions. Information can be stored in the device by assigning a binary value of one or zero to the high and low resistance levels. RRAM uses oxides instead of magnets to establish the binary code. The resistance in the oxide can be decreased by a factor of a 1,000 by applying a voltage to the device or by reversing the direction of the voltage. Information is thus also expressed as binary, as either a high- or low-resistance unit.
Comparing the processing efficiencies of the two systems with flash requires a bit of multiplication. Both RRAM and STT-MRAM offer speeds 1,000 times faster than flash — speeds measured in nanoseconds instead of microseconds. Moreover, whereas flash can write and erase data between 10,000–100,000 times before becoming unreliable, RRAM can endure one billion write cycles, and STT-MRAM has an almost infinite endurance. The two systems have caught the eye of the semiconductor industry, being acknowledged by the International Technology Roadmap for Semiconductors as the most promising challengers to mainstream memory and storage technologies.
“STT-MRAM shares the same performance, speed and endurance as DRAM, and on top of that, it is non-volatile so you could save a tremendous amount of energy with it,” says Franck Ernult, manager of the Non-Volatile Memories division at the DSI. And while STT-MRAM could potentially replace volatile DRAM, RRAM is predicted to replace non-volatile flash.
The DSI is working closely with industry to address remaining barriers to the commercialization of these technologies.
Bit by bit
At the level of a bit, the performance of STT-MRAM is on par with that of volatile systems, making it ideal for use as memory. But it has been difficult to reproduce that singular victory across the billions of bits required to process information in a high-capacity memory chip. The DSI has therefore partnered with a US-based manufacturer of semiconductor devices to ensure stable and uniform performance at larger scales. The DSI team is also developing techniques to reduce the power consumption involved in writing and erasing data in STT-MRAM devices.
RRAM is more suitable as a technology for storage and is relatively simple to manufacture, but packing as many devices as possible onto a tiny chip without degrading its performance can be challenging. Researchers at the DSI are conceptualizing high-density stacking options, including setting stacks on top of each other. “Flash is already moving toward the integration of three-dimensional architectures, which require different device specifications than those currently used by the industry,” notes Ernult. “We are taking RRAM in a similar direction so as to maintain its competitive edge against flash technology.”
But the real concern with RRAM is that researchers are still debating its working mechanism, which makes it problematic to rely on. The DSI, in collaboration with the University of Cambridge in the UK, is therefore studying the process of writing and erasing data in RRAM systems to eventually improve control over them. The focus thus far has been on the materials used in the active oxide layer and surrounding electrode, through which current is injected into the device. “We have discovered that the electrode is quite important and that the choice of material for this electrode needs to be carefully selected so that it matches the performance of the oxide,” explains Ernult.
The performance specifications achieved by Ernult’s division are being incorporated into designs and technologies developed by the DSI’s Data Center Technologies division, with the ultimate goal of constructing data centers based on non-volatile memory systems. “The unrelenting proliferation of cloud computing has created an unprecedented amount of data and will place tremendous stress on data centers,” explains division manager, Khai Leong Yong. “Emerging non-volatile memory devices will enable data to be processed at much higher speeds than existing systems.”
Data centers today are designed to separate ‘main memory’, based on faster yet temporary volatile technologies, from ‘secondary storage’, based on slower yet persistent non-volatile technologies. This separation requires data to be retrieved and copied from storage to memory at every boot-up — a costly and time-consuming process. “New non-volatile memory devices, however, can serve as main memory, thus creating persistent memory systems that retain data even without power and support fast rebooting,” adds Yong.
The DSI is developing the hardware and software for such ‘active storage’ facilities that show a nine-fold improvement in data transactional performance over standard storage systems using DRAM. The division is also making headway in the area of security by enhancing the encryption and decryption of information stored in data centers.
“We take a holistic approach at the DSI, from device-level research into material structure, fabrication, signal processing and integrated circuit design, to system-level integration of non-volatile memory devices into new storage and computing architectures and systems,” says Yong. The state-of-the-art equipment and close industry ties at the DSI ensure that researchers solve problems that are relevant and applicable, and that may even spare you the agony of losing your hard-earned work.
About the Data Storage Institute
The A*STAR Data Storage Institute was established in 1992 as the Magnetics Technology Centre (MTC) and renamed as the Data Storage Institute in 1996. The research institute’s vision is to be a vital node in a global community of knowledge generation and innovation, nurturing research talents and capabilities for world-class R&D in next-generation storage technologies.