In October 2023, a research team led by Professor Lee Hong-sub of Kyung Hee University’s Dept. of Advanced Materials Engineering for Information & Electronics succeeded in developing high-reliability alkali-ion memristors (AIM). The achievement was published in American Chemical Society Nano, a prestigious international journal in the field of nanoscience and nanotechnology.

Memristor Technology

  The development of next-generation memory devices is considered an essential task for technological advancement, as existing semiconductors have begun to reach their limits. Memristors, also known as memristive synapse devices, are emerging as a promising solution. The term memristor is a combination word of memory and resistor, describing a non-volatile memory device that changes its resistance value based on electrical signals and retains this information.

  Prof. Lee explained that his research interest naturally gravitated towards memristors. During his graduate studies, he focused on resistive random-access memory (RRAM) as a next-generation memory device. At that time, the concept of the memristor, which dates back to the 1970s, existed only in theory. After Nature published an article claiming that RRAM and memristors operate on the same principle, Prof. Lee expanded his research to include memristors alongside RRAM.

  Despite its potential, existing oxygen-ion-based memristors faced a significant issue with unwanted doping, which involves the addition of unintended impurities to the semiconductor. This problem was caused by the materials used to bond the gold to the silicon substrate, leading to low reliability across the entire semiconductor. Reliability is crucial for semiconductors as it determines whether they can perform their intended function throughout the warranty period. Since the failure of a single element could cause the entire semiconductor to stop working, this issue was critical to its development and performance.

How AIM Operates

  Prof. Lee’s research team tackled the reliability issues of existing memristors by developing AIM using lithium (Li)-ions. By incorporating a Li-ion layer into the existing oxygen-based memristor, they not only resolved the unwanted doping issue but also enhanced its functionality.

  “Actually, the idea of doping Li-ions into memristor devices has been proposed before,” said Prof. Lee. He further explained that the property of Li-ions?actively moving in solids?was thought to aid in efficiently altering resistance. However, inserting Li-ions proved challenging, and maintaining the resistance changes caused by their movement was difficult. This difficulty arose because Li-ions tend to return to their equilibrium state. In short, there was an issue with the memory device’s retention ability, which refers to its capacity to store resistance data.

  The research team addressed the problematic doping by adding a Li metal layer to the specific site of unwanted impurities, thereby facilitating desired Li-ion doping. Moreover, they resolved the retention issue by using the Li metal layer as a reservoir, allowing Li-ions to accumulate effectively. As a result, the newly developed memristor is known to have improved speed and reliability by more than tenfold compared to existing devices.

Neuromorphic Technology and AIM

  The development of memristors is expected to contribute to the growth of neuromorphic computing technology. As the name suggests, neuromorphic technology refers to an approach to computing that imitates the neural network of the human brain. 

  Programs based on deep neural networks, such as the Chat Generative Pre-trained Transformer (ChatGPT), require neuromorphic hardware capable of processing complex calculations. However, current neuromorphic hardware consumes an enormous amount of power, making it impractical for everyday use. In fact, the power usage of worldwide data centers in 2020 totaled approximately 250 terawatt hours (TWh), compared to South Africa’s total consumption of 208 TWh. To address this issue, OpenAI, the creator of ChatGPT, explored various methods, including nuclear power, but found them neither sustainable as long-term solutions nor feasible for commercialization. 

  Fortunately, memristors are anticipated to offer a fundamental solution by alleviating the bottleneck phenomenon and improving the area-efficiency of existing hardware. Nevertheless, Prof. Lee noted that “it is difficult to expect daily use of neuromorphic computing in the near future” due to significant limitations in current technology. However, the AIM developed by his research team is anticipated to serve as an important stepping stone for the continuous development of neuromorphic hardware technology.

Future Plans of Research Team

  Prof. Lee’s research team plans to continue their efforts to develop their own neuromorphic hardware. The focus of their work extends beyond just developing memristors; it also includes ensuring their compatibility with actual hardware. Currently, the team’s priority is to assess the impact of the Li metal layer on the semiconductor process. Additionally, Prof. Lee revealed plans to identify the most efficient type of oxide-ion for use in memristors.

  Prof. Lee further mentioned his intention to create a neuron device that can be integrated with memristors for the complete fabrication of neuromorphic hardware. He said, “Some people consider memristors themselves to be a form of neuromorphic device, while others view neuromorphic hardware as a combination of neuron devices and memristors.” 

  Memristors are gaining attention as a crucial component in the future development of artificial intelligence (AI). In this context, the recent achievement of Prof. Lee’s research team is expected to significantly contribute to the efforts to commercialize this technology. Prof. Lee expressed his hope that more students will become interested in the field of advanced materials engineering, emphasizing its significance by stating, “the manufacturing field plays an important role in the future evolution of AI.”