Recent progress in artificial intelligence (AI) has dramatically transformed the electronic industries, leading to the development of sophisticated products capable of executing intelligent tasks. Specifically, the integration of AI with Internet-of-Things (IoT) technology has enabled the creation of intelligent IoT edge devices that readily provide AI services to the broader public. Nonetheless, the current growing trend towards larger neural network parameters for advanced AI services necessitates energy-intensive data transfer between the processor and off-chip memory in traditional von Neumann architectures. To develop energy-efficient neuromorphic computing device, I demonstrated that the poly(1,3,5-trivinyl-1,3,5-trimethyl cyclotrisiloxane)(pV3D3)-based memristor and 3D fin-shaped field-effect transistor, also known as FinFET, with a poly-Si/SiO2/Si3N4 gate stack on a single-crystalline silicon channel (SONS) devices can be operated as an electronic synapse device featuring analog conductance update. In addition to synapse application, I developed the memristor-based neuron device to emulate Rectifying Linear Unit (ReLU) activation function in order to address vanishing gradient problem in deep neural network. Also, I will introduce my recent research that can produce diverse, realistic medical images through the fusion of generative networks and semiconductor technology.
Furthermore, 3D NAND Flash memory, which is the main stream of the current flash memory market, can be powerful candidate for off-chip neuromorphic system, because 3D NAND Flash memory has the highest bit density among existing semiconductor devices. In order to utilize 3D NAND Flash memory as a neuromorphic system, it is important to improve reliability, such as high-intensive read disturbance and retention characteristics. To improve the reliability of 3D NAND flash memory, I will conduct research on new materials and novel memory device structures. Due to the cell current shortage phenomenon that occurs as the number of stacks of 3D NAND flash memory increases and the device structure in which the Si3N4 trap layer between adjacent devices is connected, there are the interference and reliability issues between adjacent devices. Now, the global memory semiconductor companies are trying to solve the reliability problem of 3D NAND Flash memory through the introduction of new materials, but due to the insufficient new product development schedule, it is only developed only within the limited material library. Furthermore, I will briefly talk about the technology trend of 3D NAND Flash memory. The high-density 3D NAND Flash memory is becoming important as the current growing trend of neural network parameters requires energy-efficient storage in data centers. Therefore, I will briefly talk about the technology trend of 3D NAND Flash memory. The high-density 3D NAND Flash memory is becoming important as the current growing trend of neural network parameters requires energy-efficient storage in data centers.