| Nov 08, 2021 |
Materials advancement accelerates the realization of AI technology
(Nanowerk News) Researchers in Korea succeeded in developing a core material for the next-generation neuromorphic (neural network imitation) semiconductor for the first time in the country. This is a result of a research team led by Dr. Jung-dae Kwon and Yong-hun Kim of the Department of Energy and Electronic Materials of the Korea Institute of Materials Science, together with Professor Byungjin Cho’s research team at Chungbuk National University (Advanced Functional Materials, "Low Power MoS2/Nb2O5 Memtransistor Device with Highly Reliable Heterosynaptic Plasticity").
|
|
This new concept memtransistor uses a two-dimensional nanomaterial with a thickness of several nanometers. Memristor is a compound word of memory and transistor. By reproducibly imitating the electrical plasticity of nerve synapses with more than 1,000 electrical stimulation, the researchers succeeded in obtaining a high pattern recognition rate of about 94.2% (98% of simulation-based pattern recognition rate).
|
|
Molybdenum sulfur (MoS2), widely used as a semiconductor material, works on the principle that defects in a single crystal are moved by an external electric field, which makes it difficult to precisely control the concentration or shape of the defect.
|
|
To solve the problem, the research team sequentially stacked an oxidic layer of niobium oxide (Nb2O5) and a molybdenum sulfur material and succeeded in developing an artificial synaptic device having a memristor structure with high electrical reliability by an external electric field.
|
|
In addition, they have demonstrated that the resistance switching characteristics can be freely controlled by changing the thickness of the niobium oxidic layer, and that brain information related to memory and forgetting can be processed with a very low energy of 10 PJ (picojoule).
|
|
Currently, as artificial intelligence hardware consumes large amounts of power and costs in the form of Graphics Processing Units (GPUs), Field Programmable Gate Arrays (FPGAs), and Application Specific Integrated Circuita (ASICs), it is expected to generate explosive demand as the industry grows in the future. The wearable AI market is expected to reach $42.4 billion by 2023, at a CAGR of 29.75% from about $11.5 billion in 2018.
|
|
A research team led by Dr. Jung-dae Kwon and Yong-hun Kim at KIMS said, “Using a high-reliable, new-concept memristor structure-based AI semiconductor can greatly reduce the circuit density and driving energy. It is expected to be applied to low-power edge computing and wearable AI systems in the future.”
|
