| Title |
Impact of 3D NAND Current Variation on Inference Accuracy for In-memory Computing |
| DOI |
https://doi.org/10.5573/JSTS.2022.22.5.341 |
| Keywords |
Compute-in-memory; deep neural network; 3D NAND; variation |
| Abstract |
3D NAND Flash has been proposed and investigated as a memory device candidate for the energy-efficient and ultra-high density compute-in-memory system. To achieve the acceptable accuracy for the inference applications, 3D NAND string current must be controlled precisely. However, there exist many challenging points which bothers the precise current control such as retention, temperature, pattern dependency in the cells of the 3D NAND string. In this work, we investigated the causes and effects of the 3D NAND string current variation and the resulted inference accuracy drop. The current variation drops the accuracy significantly so that the compensating design schemes must be implemented for the practical designs. |