We ask computers to process ever-increasing amounts of data to speed drug discovery, improve weather and climate predictions, train artificial intelligence, and more. To meet this demand, we need computer memory that is faster and more energy-efficient than ever before. To this end, researchers at Stanford University have developed a new type of phase-change memory that can help computers process large amounts of data faster and more efficiently.
Innovation in memory technology
Stanford University researchers have demonstrated that a new material could make phase-change memory, which relies on switching between high and low resistance states to create the ones and zeros of computer data, an improved option for future artificial intelligence and data-centric systems. Recently, the journal Nature Communications detailed their scalable technology, which is fast, low-power, stable, long-lasting, and can be manufactured at temperatures compatible with commercial manufacturing.
"We're not just improving a single metric like endurance or speed, but we're improving multiple metrics at the same time," said Eric Pop, Pease-Ye Professor of Electrical Engineering and Distinguished Professor of Materials Science and Engineering at Stanford. "This is the most realistic and industry-friendly thing we've ever built in this area. I like to think of it as a step toward universal memory."
Improve computing efficiency
Today's computers store and process data in various locations. Volatile memory (fast, but disappears when the computer is shut down) is responsible for processing data, while non-volatile memory (not fast, but can hold information without constant input of power) is responsible for long-term data storage. Moving information between these two locations creates a bottleneck when the processor is waiting to retrieve large amounts of data.
Xiangjin Wu, co-first author of the paper and a doctoral candidate co-supervised by Pop and Philip Wong, the Willard R. and Inez Kerr Bell Professor in the School of Engineering, said: "Shutting data back and forth consumes a lot of energy, especially in today's computing workloads. With this kind of memory, we hope to bring memory and processing more closely together, and ultimately integrate them into a single device, thus reducing energy consumption and time."
There are many technical hurdles to achieving an efficient, commercially viable, general-purpose memory that can perform both long-term storage and fast, low-power processing without sacrificing other metrics, but the new phase-change memory developed by Pope's lab is the closest anyone has come so far with this technology. The researchers hope it will inspire further development and adoption of this universal memory.
The promise of GST467 alloy
The memory relies on GST467, an alloy composed of four parts germanium, six parts antimony and seven parts tellurium, developed by collaborators at the University of Maryland. Pop and his colleagues found a way to sandwich this alloy between several other nanometer-thin materials in a superlattice, a layered structure they had previously used to good effect in nonvolatile memory.
"GST467's unique composition makes it switch exceptionally fast," said Asir Intisar Khan, who earned his PhD in Pop's lab and is co-first author on the paper. "Integrating it into the superlattice structure of a nanoscale device allows for low switching energy, gives us good durability, very good stability, and makes it nonvolatile -- it can maintain its state for 10 years or more."
Set new standards
The GST467 superlattice passed several important benchmark tests. Phase-change memory can sometimes drift over time, with the values of 1 and 0 slowly shifting, but their tests showed that the memory is very stable. It also operates at less than 1 volt (a goal for low-power technology) and is significantly faster than regular SSDs.
"Several other types of nonvolatile memory may be faster, but they operate at higher voltages and consume more power," Pop said. "All of these computing technologies require a trade-off between speed and energy consumption. The fact that we can switch in tens of nanoseconds at voltages below one volt is very important."
Superlattices can also accommodate large numbers of memory cells in a small space. The researchers reduced the diameter of the memory cells to 40 nanometers, less than half the size of the coronavirus. This approach is possible because the superlattice is made at lower temperatures and uses advanced manufacturing techniques. The manufacturing temperature is much lower than required. Researchers are discussing stacking memories into thousands of layers to increase density. This memory could enable future 3D layering.
Compiled source: ScitechDaily