Researchers develop scalable approach to integrating ultrafast 2D flash memories

The widespread use of artificial intelligence (AI) tools designed to process large amounts of data has increased the need for more powerful memory devices. Data storage solutions that could help meet AI’s computational demands include so-called high-speed memories, technologies that can increase the memory bandwidth of computer processors, thereby speeding up data transfer and reducing power consumption.

Currently, flash memories are the most widespread memory solutions, capable of storing information when a device is turned off (i.e. non-volatile memories). Despite their widespread use, the speed of most existing flash memories is limited and does not allow them to best support AI operation.

In recent years, some engineers have therefore tried to develop ultra-fast flash memories capable of transferring data more quickly and efficiently. Two-dimensional (2D) materials have shown promise for making these more efficient memory devices.

Although some long-channel flash memory devices assembled from exfoliated 2D materials have been shown to exhibit ultrafast processing speeds, scalable integration of these devices has so far proven challenging. This has so far limited their commercialization and large-scale deployment.

Researchers at Fudan University have recently developed a new approach for the scalable integration of ultrafast 2D flash memory devices. The approach, described in a paper published in Natural electronicshas been effectively used to integrate 1,024 flash memory devices with an efficiency of over 98%.

“Two-dimensional (2D) materials could potentially be used to create ultrafast flash memory,” Yongbo Jiang, Chunsen Liu, and colleagues wrote in their paper. “However, due to interface engineering issues, ultrafast nonvolatile performance is currently limited to exfoliated 2D materials, and there is a lack of performance demonstrations with short-channel devices. We present a scalable integration process for ultrafast 2D flash memory that can be used to integrate 1,024 flash memory devices with an efficiency of over 98%.”

To fabricate their ultrafast flash memory array, the researchers used a combination of processing techniques, including lithography, electron beam evaporation, thermal atomic layer deposition, a polystyrene-assisted transfer technique, and an annealing process. In their recent study, they applied their proposed approach to fabricating memories with two distinct memory stack configurations, both of which achieved high yields.

“We illustrate the approach with two different configurations of memory stack tunneling barrier (HfO₂/Pt/HfO₂ and Al₂O₃/Pt/Al₂O₃) and using monolayer molybdenum disulfide grown by chemical vapor deposition,” the researchers wrote.

“We also show that the channel length of ultrafast flash memory can be reduced to less than 10 nm, which is below the physical limit of silicon flash memory. Our sub-10 nm devices provide non-volatile information storage (up to 4 bits) and robust endurance (more than 10⁵).”

Early tests by Jiang, Liu, and their colleagues demonstrated the potential of their approach for scalable integration of high-performance, ultrafast flash memories. The researchers successfully reduced the channel length of their flash memories to less than 10 nm and found that these sub-10 nm devices still exhibited ultrafast speeds, storing up to 4 bits and maintaining their nonvolatility.

Further studies could use the team’s proposed integration process to fabricate flash memory arrays based on other 2D materials and with different memory stack configurations. These efforts could contribute to the large-scale deployment of future ultrafast flash memory devices.

© 2024 Science X Network