///

Ultra-reliable AI memristor with strain engineering

How does flexoelectricity open new possibilities for semiconductors? A study reveals its potential to boost control and improve ferro-ionic performance.

Start Reading

The rise of AI memristor device

Memristor devices are gaining attention as promising components for the next generation of artificial intelligence (AI) systems. These devices offer potential advantages in non-Von Neumann architectures, particularly for in-memory computing and efficient weight storage at low power levels. However, ensuring the reliability of each memristor is crucial for implementing large-scale neural computing.

A key focus of research is the formation of filaments within the switching medium of memristors, as their behaviour significantly affects device performance. The migration of active ions, such as silver, oxygen, and oxygen vacancies, can lead to unpredictable filament growth and positioning. This unpredictability results in considerable variability between cycles and across different devices. Thus, presenting a significant challenge for the practical application of memristive switching in neuromorphic systems.

Figure 1. Schematic of rat neurons on CMOS nanoelectrode array.
Credit. Samsung Electronics

How do we solve the stochastic limitation of AI memristor?

Researchers are examining the flexoelectric effect at the nanometre scale to address this stochastic limitation. Recent studies suggest that mechanical strain can serve as an effective method for switching ferroelectric domains in various materials. Previous findings indicate that the flexoelectric field can be maximised when an external force bends the lattice structure of a material. It leads to the generation of internal polarisation and electric fields.

When a localised inhomogeneous strain is applied to ferroelectric materials using an atomic force microscope (AFM) tip, the mechanical bending induces a flexoelectric field, resulting from a net charge shift within the lattice unit cell. As the degree of mechanical bending is inversely proportional to the length scale, the flexoelectric field demonstrates pronounced effects in nanoscale materials. However, to selectively activate phase changes and ion migration at specific locations within flexoelectric materials, significantly larger lattice bending is necessary than that observed in previous research.

Flexoelectricity: Universal platform for spatial ion control

To enhance the flexoelectric field beyond previous findings, researchers applied vertical shear stresses to free-standing two-dimensional materials using an AFM tip. This method selectively amplified the flexoelectric field and its corresponding downward polarisation in specific regions. As a result, the team observed the targeted growth of conductive filaments through the localized flexoelectric field.

Furthermore, they succeeded in reversibly controlling the phase transition threshold voltage by modulating the bottom ferroelectric polarization, enabling active spatial control of conductive filaments at the nanometre scale. As a result, 5.76 × 102-fold maximum current (Imax) enhancement and 6.25 × 102-fold increased Imax, with strain / Imax, without strain ratio was experimentally observed with vertical shear strain 720 nN with 25 nm tip radius.

Since the upward polarized bottom In2Se3 suppresses the undesirable ionic conduction in the suspended junction area, a topographical Cu+ ion extraction was selectively derived within 232.42 nm width and 58.98 nm height. In conclusion, this research presents a novel flexoelectric approach for the nanoscale localisation of Cu⁺ ions in free-standing two-dimensional ferro-ionic memristors, paving the way for reliable neuromorphic computing and ultra-low energy consumption systems.

Figure 2. Schematic of free-standing two-dimensional ferro-ionic memristor.
Credit. Author

Solution for next-generation AI device

1. Promote open access and collaboration in advanced AI research: Governments should incentivise cross-border collaboration in memristor and AI research. This can be done by establishing open-access platforms that allow scientists and engineers globally to share data and findings. By democratising access to cutting-edge knowledge, we can reduce disparities in technological advancements. It will ensuring that developing nations can equally participate in the growth of neuromorphic computing.

This research overcomes the stochastic limitations of conventional ferro-ionic materials and offers a new perspective on ion movement based on the flexoelectric effect from a structural perspective. It is expected to significantly enhance the performance and reliability of semiconductor devices by enabling precise spatial control of ions.

-Prof. Taesung Kim

2. Create global standards for ethical AI innovation: Industry bodies and international regulatory authorities must formulate a comprehensive ethical framework for developing AI systems, including memristor-based technologies. These standards should ensure that a few do not monopolize advancements but benefit the broader global community. Ethical AI guidelines will also safeguard against the misuse of such powerful technologies, ensuring their role in promoting equality and fairness in society.

3. Support investment in sustainable semiconductor manufacturing: Governments and industry should prioritize funding for sustainable production processes in the semiconductor industry, specifically those that reduce environmental impacts. Memristor devices, known for their low energy consumption, offer a path to greener AI technology. Promoting eco-friendly practices in their mass production can help mitigate climate risks and foster a more just global economy where technological progress does not come at the cost of environmental degradation.

4. Ensure technological inclusivity in global AI systems: Policymakers must work to bridge the digital divide by ensuring that advancements in memristor-based AI technologies are inclusive. This can be done by supporting initiatives that provide access to next-generation AI devices for underserved regions, ensuring equitable access to technologies that have the potential to transform industries and improve living standards globally.

This research was supported by the National Research Foundation of Korea (NRF) and the Institute for Basic Science (IBS).

🔬🧫🧪🔍🤓👩‍🔬🦠🔭📚

Journal reference

Lee, J., Woo, G., Cho, J., Son, S., Shin, H., Seok, H., … & Kim, T. (2024). Free-standing two-dimensional ferro-ionic memristor. Nature Communications, 15(1), 5162. https://doi.org/10.1038/s41467-024-48810-3

Tags:

Jinhyoung Lee is currently enrolled in a combined M.S. and Ph.D. course in the Department of Mechanical Engineering at Sungkyunkwan University. His research focuses on exploring the microscopic nature of quantum materials and their device applications. He combines the novel implementation of 2D van der Waals heterostructure devices with scanning probe measurements, designed to access the atomic-scale optoelectrical and magnetic properties of 2D materials.

View all

Gunhoo Woo is currently a Ph.D. candidate in the Department of Nanoscience and Technology at Sungkyunkwan University. His research focuses on the fabrication of van der Waals devices, including photodetectors, physically unclonable functions, memristors, flash memory, and transistors. He conceptualises next-generation memory devices and systems using interfacial electron tunnelling and electronic band engineering, enabling access to the optoelectrical properties of van der Waals materials.

View all

Taesung Kim received his B.S. degree from Seoul National University in 1994 and his Ph.D. from the University of Minnesota in 2002. He is currently a professor in the Department of Nanoscience and Technology, the Department of Mechanical Engineering, and Semiconductor Convergence Engineering at Sungkyunkwan University. His research interests include wafer-scale 2D material synthesis via plasma-enhanced chemical vapour deposition, fundamental mechanism studies, aerosol technology, semiconductor manufacturing processes and equipment, and nanodevice applications, including photodetectors and neuromorphic computing.

View all

You might be interested in