| Jan 22, 2024 |
(Nanowerk Information) We’re tasking our computer systems with processing ever-increasing quantities of knowledge to hurry up drug discovery, enhance climate and local weather predictions, prepare synthetic intelligence, and way more. To maintain up with this demand, we want sooner, extra energy-efficient pc reminiscence than ever earlier than. |
|
Researchers at Stanford have demonstrated {that a} new materials might make phase-change reminiscence—which depends on switching between excessive and low resistance states to create those and zeroes of pc knowledge—an improved possibility for future AI and data-centric programs. Their scalable know-how, as detailed not too long ago in Nature Communications (“Novel nanocomposite-superlattices for low vitality and excessive stability nanoscale phase-change reminiscence”), is quick, low-power, steady, long-lasting, and will be fabricated at temperatures appropriate with business manufacturing. |
|
“We’re not simply enhancing on a single metric, corresponding to endurance or velocity; we’re enhancing a number of metrics concurrently,” mentioned Eric Pop, the Pease-Ye Professor of Electrical Engineering and Supplies Science & Engineering (by courtesy) at Stanford. “That is probably the most life like, industry-friendly factor we’ve constructed on this sphere. I’d like to think about it as a step in direction of a common reminiscence.” |
 |
| Cross-sections of phase-change reminiscence units within the high- and low-resistance states. The diameter of the underside electrode is ~40 nanometers. Arrows mark among the van der Waals (vdW) interfaces, which kind between layers of the superlattice supplies. The superlattice is disrupted and reformed between the high- and low-resistance states. (Picture courtesy of the Pop Lab) |
A sooner nonvolatile reminiscence |
|
In the present day’s computer systems retailer and course of knowledge in separate areas. Risky reminiscence—which is quick however disappears when your pc turns off—handles the processing, whereas nonvolatile reminiscence—which isn’t as quick however can maintain data with out fixed energy enter—takes care of the long-term knowledge storage. Shifting data between these two areas could cause bottlenecks whereas the processor waits for big quantities of knowledge to be retrieved. |
|
“It takes a variety of vitality to shuttle knowledge backwards and forwards, particularly with right now’s computing workloads,” mentioned Xiangjin Wu, a doctoral candidate co-advised by Pop and Philip Wong, the Willard R. and Inez Kerr Bell Professor within the College of Engineering and lead creator on the paper. “With any such reminiscence, we’re actually hoping to carry the reminiscence and processing nearer collectively, finally into one gadget, in order that it makes use of much less vitality and time.” |
|
There are a lot of technical hurdles to attaining an efficient, commercially viable common reminiscence able to each long-term storage and quick, low-power processing with out sacrificing different metrics, however the brand new section change reminiscence developed in Pop’s lab is as shut as anybody has come up to now with this know-how. The researchers hope that it’s going to encourage additional growth and adoption as a common reminiscence. |
|
The reminiscence depends on GST467, an alloy of 4 components germanium, six components antimony, and 7 components tellurium, which was developed by collaborators on the College of Maryland. Pop and his colleagues discovered methods to sandwich the alloy between a number of different nanometer-thin supplies in a superlattice, a layered construction they’ve beforehand used to attain good nonvolatile reminiscence outcomes. |
|
“The distinctive composition of GST467 offers it a very quick switching velocity,” mentioned Asir Intisar Khan, who earned his doctorate in Pop’s lab and is co-lead creator on the paper. “Integrating it inside the superlattice construction in nanoscale units permits low switching vitality, offers us good endurance, superb stability, and makes it nonvolatile—it might retain its state for 10 years or longer.” |
Setting a brand new bar |
|
The GST467 superlattice clears a number of vital benchmarks. Section change reminiscence can generally drift over time—primarily the worth of those and zeros can slowly shift—however their checks present that this reminiscence is extraordinarily steady. It additionally operates at under 1 volt, which is the purpose for low-power know-how, and is considerably sooner than a typical solid-state drive. |
|
“A couple of different varieties of phase-change reminiscence generally is a bit sooner, however they function at increased voltage,” Pop mentioned. “With all these computing applied sciences, there are tradeoffs between velocity and vitality. The truth that we’re switching at a number of tens of nanoseconds whereas working under one volt is an enormous deal.” |
|
The superlattice additionally packs a superb quantity of reminiscence cells right into a small house. The researchers have shrunk the reminiscence cells all the way down to 40 nanometers in diameter—lower than half the dimensions of a coronavirus. That’s not fairly as dense because it might be, however the researchers are exploring methods to compensate by stacking the reminiscence in vertical layers, which is feasible because of the superlattice’s low fabrication temperature and the strategies used to create it. |
|
“The fabrication temperature is nicely under what you want,” Pop mentioned. “Persons are speaking about stacking reminiscence in 1000’s of layers to extend density. One of these reminiscence can allow such future 3D layering.” |
