2D Material Based Charge Trap Flash Device for Synaptic Operation
- 주제(키워드) Neuromorphic computing
- 발행기관 고려대학교 대학원
- 지도교수 성태연
- 발행년도 2021
- 학위수여년월 2021. 2
- 학위구분 석사
- 학과 대학원 신소재공학과
- 원문페이지 52 p
- UCI I804:11009-000000235634
- DOI 10.23186/korea.000000235634.11009.0001158
- 본문언어 영어
- 제출원본 000046072037
초록/요약
Researches in Artificial Intelligence (AI) has received a great deal of attention in the 4th industrial revolution. AI computing based on von Neumann structure, in which the memory and processor are physically separated, has been widely used. However, this conventional structure is inefficient for processing a large amount of information such as visual and speech recognition. This is because the high power consumption becomes the problem of the von Neumann based device. Therefore, neuromorphic computing which mimics human brain that consumes approximately 20watts becomes the solution to overcome the von Neumann bottleneck. The purpose of this paper is to provide a framework for hardware-based spiking neural network (SNNs) device. According to SNNs, neuron and synapse devices are required in order to mimic the human brain. Neuron acts as a spike generators and synapse acts as a regulators of signals as well as role of weight storage operation. In SNNs, high performance is essential for synaptic devices since the number of synapses in neural network is greater than the number of neurons. Here, two-dimensional (2D) material based charge trap flash (CTF) memory was used to show the synaptic behavior in neuromorphic systems; which demonstrates potentiation process, depression process, retention process and spike-timing-dependent plasticity (STDP). We showed the linear characteristics of potentiation and depression which is the most striking features among the synaptic behaviors. Also, for the emulation of the time dependence of the synapse, a biological process of STDP that modulate the strength of connection between neurons was examined.
more목차
Chapter 1 Introduction.......................................................1
1.1 Overview of the Neuromorphic Computing...................1
Chapter 2 Theoretical Background ..................................... 3
2.1 Components of a Biological Neural Network.................3
2.1.1 Biological Neuron Model........................................... 3
2.1.2 Biological Synapse Model.......................................... 4
2.2 Flash Memory Device....................................................6
2.3 Materials used for Flash Memory...................................8
2.3.1 Channel Material ...................................................... 8
2.3.2 Insulator Materials ................................................... 10
2.4 Operation Mechanisms of the Flash Memory.................11
2.4.1 Carrier Tunneling .......................................................11
2.4.2 Coupling Ratio ..........................................................11
2.4.3 Band Diagram .......................................................... 14
2.4.4 Carrier Trap (Program/ Erase) ................................... 16
2.4.5 Spike-Timing-Dependent Plasticity (STDP).................. 18
Chapter 3 Experimental Methods ........................................20
3.1 Device Characterization and Fabrication ........................20
Chapter 4 Results and Discussions.......................................25
4.1 Effects of Coupling Ratio ..............................................25
4.2 Electrical Characteristics................................................25
4.2.1 Output and Transfer Curve ....................................... 25
4.2.2 Potentiation and Depression Curve............................ 29
4.2.3 STDP ........................................................................ 32
Chapter 5 Conclusions........................................................36
Bibliography ......................................................................37
