研究目的
Investigating the use of 2D NAND flash memory cells as high-density and reliable synaptic devices in multi-layer neural networks.
研究成果
The study demonstrates that NAND flash memory cells can be effectively used as high-density and reliable synaptic devices in multi-layer neural networks. The unidirectional conductance response of NAND flash cells is particularly suitable for this application, achieving learning accuracy comparable to that of ideal perfect linear devices. NAND flash memory's maturity and cell density advantages make it a promising candidate for implementing high-density multi-layer neural networks.
研究不足
The study focuses on the use of NAND flash memory cells as synaptic devices and does not explore other types of memory technologies. The experimental setup is limited to a 3-layer perceptron network trained on the MNIST database.
1：Experimental Design and Method Selection:
The study involves designing multi-layer neural networks using 2D NAND flash memory cells as synaptic devices, focusing on eliminating the waste of NAND flash cells and allowing analogue input values. An adaptive weight update method for hardware-based multi-layer neural networks is employed.
2：Sample Selection and Data Sources:
A 3-layer perceptron network with 40,545 synapses is trained on a MNIST database set.
3：List of Experimental Equipment and Materials:
Floating-gate 2D (planar) NAND flash cell strings fabricated with 26 nm technology are used.
4：Experimental Procedures and Operational Workflow:
The conductance response of NAND flash cells is measured to evaluate their suitability as synaptic devices. The study compares bidirectional and unidirectional conductance responses in terms of classification accuracy.
5：Data Analysis Methods:
The study uses an adaptive weight update method based on a unidirectional conductance response to achieve high learning accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
