| 학회 |
한국재료학회 |
| 학술대회 |
2021년 가을 (11/24 ~ 11/26, 경주 라한호텔) |
| 권호 |
27권 2호 |
| 발표분야 |
A. 전자/반도체 재료 분과 |
| 제목 |
Highly linear and symmetric weight modification for spiking neural network system using memristive device |
| 초록 |
To achieve improved synaptic behavior of memristive device, various operational methods and material modifications were performed for the spiking neural network (SNN) based neuromorphic system. The conventional Deep Neural Network (DNN) should use the various strength of stimuli to maintain the signal transfering accuracy between layers of neural network during its back propagation process. However, because this process is very energy consumative and slow process, the SNN attracts a considerable attention as a promising alternative. To realize the SNN in a hardware based device, the memristive synaptic device should have the highly linear and symmetric potentiation (P) and depression (D) characteristics with an identical electric pulse stimulation. For this, methodologically, the materials and structure of memristive synaptic device were optimized. In addition, the modification of the applied electric pulses was also performed. It should be noted that the developed method in this research can be adopted to general memristive devices if the device has some of important operational conditions. The detailed experimental processes, electrical measurement results, and microscopic chemical analyses will be explain in poster session. |
| 저자 |
Jin Joo Ryu1, Hyunchul Sohn2, Gun Hwan Kim2
|
| 소속 |
1Korea Research Institute of Chemical Technology(KRICT), 2Yonsei Univ. |
| 키워드 |
Inference; Linearity; Memristive devices; Weight modification
|
| E-Mail |
|
