ReRAM Goes 3D

　　Resistive random-access memories (ReRAMs) are a new breed of “universal” memory that could replace all other types, offering the speed of RAM but with the density and non-volatility of flash. To date, however, flash has managed to stay ahead of ReRAM by going 3D. Now the Moscow Institute of Physics and Technology (MIPT) says it has reengineered its ReRAM process to achieve a thin-film technique that is amenable to 3D stacking.

　　All ReRAMs work using memristors, in which migrating oxygen vacancies in the dielectric layer change the dielectric’s resistance to represent ones and zeros. In addition to MIPT, researchers from 4DS Memory Ltd., Crossbar Inc., HP Inc., Knowm Inc., and Rice University have created prototypes.

　　For 3D ReRAMs, “we needed not only to form oxygen vacancies in the dielectric layer, but also to detect them,” MIPT scientist Konstantin Egorov told EE Times. To do so, MIPT specialists used a method for observing the electron states in the bandgap of the dielectric that arise in the presence of oxygen vacancies.

　　“To study oxygen vacancies formed in the process of tantalum oxide film growth, we used an experimental cluster incorporating growth PEALD [plasma-enhanced atomic-layer deposition] and analytic XPS [X-ray photoelectron spectroscopy] chambers connected to each other by a vacuum tube with sample transfer manipulators. The cluster enabled us to grow and study deposited layers without breaking the vacuum,” Egorov said. “This is crucial because as soon as you take the experimental sample out of the vacuum, the nanolayer of dielectric oxidizes on its surface, which results in the annihilation of oxygen vacancies.”

　　Any semiconductor research lab could construct the unique ALD cluster by connecting the PEALD and XPS chambers and then adding robotic manipulators to transfer wafers between the chambers. For mass production, the cluster would not be needed, except to sample test wafers. A new assembly line, however, would have to be built that would compensate for the slow growth speed of ALD films.

　　If those efforts are successful, MIPT claims the resulting ReRAMs could be stacked vertically, yield a universal memory that would overcoming the limitations of 3D flash, which to date is restricted to 64 layers.

　　Deposition details

　　Though ALD is slow growing, it enables conformal coating of 3D structures, replacing the nanofilm deposition techniques used to date by MIPT and other research labs. The key difference is that ALD exposes a substrate, sequentially, to both a precursor material and a reactant material, and depends on the chemical reaction between the two to produce the active layer. MIPT’s technique also uses ligands—molecules chemically attached to a metallic precursor—to hasten the chemical reaction, but the ligands have to be removed before the active layer can be used in a device.

　　“Depositing oxygen-deficient films requires finding the correct reactants to both eliminate the ligands contained in the metallic precursor and control the coating’s oxygen content,” said Andrey Markeev, lead researcher at MIPT. “After much experimentation, we successfully used a tantalum precursor containing oxygen, and a plasma-activated hydrogen reactant.”

　　Next, the researchers plan to optimize the process and increase the ALD speed to enable mass production of 3D ReRAMs.

　　Funding for MIPT’s work was a provided by a Russian Science Foundation grant and MIPT’s 5-100 Program within the Russian Academic Excellence Project.