An ultrafast bipolar flash memory for self-activated in-memory computing

Xiaohe Huang; Chunsen Liu; Zhaowu Tang; Senfeng Zeng; Shuiyuan Wang; Peng Zhou

doi:10.1038/s41565-023-01339-w

An ultrafast bipolar flash memory for self-activated in-memory computing

Nat Nanotechnol. 2023 May;18(5):486-492. doi: 10.1038/s41565-023-01339-w. Epub 2023 Mar 20.

Authors

Xiaohe Huang^#¹, Chunsen Liu^#^{2

3}, Zhaowu Tang¹, Senfeng Zeng¹, Shuiyuan Wang¹, Peng Zhou^{4

5}

Affiliations

¹ State Key Laboratory of Integrated Chips and Systems, School of Microelectronics, Zhangjiang Fudan International Innovation Center, Fudan University, Shanghai, China.
² State Key Laboratory of Integrated Chips and Systems, School of Microelectronics, Zhangjiang Fudan International Innovation Center, Fudan University, Shanghai, China. chunsen_liu@fudan.edu.cn.
³ Frontier Institute of Chip and System, Shanghai Key Lab for Future Computing Hardware and System, Fudan University, Shanghai, China. chunsen_liu@fudan.edu.cn.
⁴ State Key Laboratory of Integrated Chips and Systems, School of Microelectronics, Zhangjiang Fudan International Innovation Center, Fudan University, Shanghai, China. pengzhou@fudan.edu.cn.
⁵ Frontier Institute of Chip and System, Shanghai Key Lab for Future Computing Hardware and System, Fudan University, Shanghai, China. pengzhou@fudan.edu.cn.

^# Contributed equally.

PMID: 36941359
DOI: 10.1038/s41565-023-01339-w

Abstract

In-memory computing could enhance computing energy efficiency by directly implementing multiply accumulate (MAC) operations in a crossbar memory array with low energy consumption (around femtojoules for a single operation). However, a crossbar memory array cannot execute nonlinear activation; moreover, activation processes are power-intensive (around milliwatts), limiting the overall efficiency of in-memory computing. Here we develop an ultrafast bipolar flash memory to execute self-activated MAC operations. Based on atomically sharp van der Waals heterostructures, the basic flash cell has an ultrafast n/p program speed in the range of 20-30 ns and an endurance of 8 × 10⁶ cycles. Utilizing sign matching between the input voltage signal and the storage charge type, our bipolar flash can realize a rectified linear unit activation function during the MAC process with a power consumption for each operation of just 30 nW (or 5 fJ of energy). Using a convolutional neural network, we find that the self-activated MAC method has a simulated accuracy of 97.23%, tested on the Modified National Institute of Standards and Technology dataset, which is close to the conventional method where the MAC and activation operations are separated.