R&D: On-Chip Learning in Vertical NAND Flash Memory Using Forward–Forward Algorithm

IEEE Transactions on Electron Devices has published an article written by Sung-Ho Park; Jonghyun Ko; In-Seok Lee; Ryun-Han Koo; Jeong-Hyun Kim; Department of Electrical and Computer Engineering and the Inter-University Semiconductor Research Center (ISRC), Seoul National University, Seoul, South Korea, Yeongheon Yang; Research and Development Division, SK Hynix Inc., Icheon, South Korea, Dongseok Kwon; Jae-Joon Kim; and Jong-Ho Lee, Department of Electrical and Computer Engineering and the Inter-University Semiconductor Research Center (ISRC), Seoul National University, Seoul, South Korea.

Abstract: “Unlike the previous ON-chip learning using backward propagation (BP) in vertical NAND (V-NAND) flash memory, a new approach utilizing only forward propagation (FP) is proposed. ON-chip learning has been successfully implemented in V-NAND flash memory using a forward–forward (FF) algorithm. FF eliminates the requirement for separating the source line or using a separate string array, thereby reducing parasitic resistance. Both supervised and unsupervised learning are implemented using FF, and high accuracy is obtained through simulations. Furthermore, a comparison of accuracy with OFF-chip learning using BP is also conducted, and ON-chip learning with FF demonstrated similar or higher accuracy.“