With the support of the National Natural Science Foundation of China (Grant numbers: 61974050, 61725505, 61905266),  Prof. Lei Ye (Huazhong university of science and technology) and Prof. Weida Hu (Shanghai institute of technical physics of Chinese academy of sciences), as well as their colleagues, made progress on a novel homogeneous transistor-memory device. The research article was published in Science on August 19, 2021, entitled “2D materials-based homogeneous transistor-memory architecture for neuromorphic hardware”. (https://doi.org/10.1126/science.abg3161).

Novel neuromorphic electronic circuits are promising and appealing for a wide range of applications including auto vehicle, image recognition, voice processing, smart prediction, brain-computer interface, internet of things, etc. However, the performance of these circuits is limited by the integration compatibility between memory arrays and their peripheral circuits. For example, new memory arrays such as memristors, phase change memories, magnetic memories, ferroelectric memories, and peripheral circuits using traditional CMOS-based field effect transistors (FETs) or bipolar junctional transistors (BJTs) exhibit huge variances in device structure and fabrication technology, which results in the physical separation of these two modules and further restricts improving the integration density and lowering the power consumption for novel neuromorphic electronic circuits.

Focused on this problem, the research group took a different approach to construct a novel homogeneous transistor-memory architecture, which is based on the physical mechanism of proximal coupling between two-dimensional (2D) materials and ferroelectric (FE) substrates. By precisely controlling the polarization characteristics of the FE substrate, the innovated device architecture realized the functionalities of FETs, BJTs, and memories. The multifunctional and reconfigurable device architecture provide the foundation to integrate memory module and peripheral circuits on a single chip (see Fig. 1). On the one hand, they constructed BJTs via proximity-induced doping effect of highly polarized FE substrate. On the other hand, non-volatile memory array was constructed by modulating the build-in barrier height, realized by switching the polarization of FE domains. The peripheral circuits and memory array were effectively integrated based on the same device architecture, and the hardware was used to achieve binary classification in a neural network and novel ternary content addressable memory (TCAM) cells. They further discovered that the FE-modulated multifunctional device can realize optoelectronic sensing under external field modulation, which provides a new scheme for the design of integrated sensing-memory-computing devices.

This research work is promising to further promote the development of novel neuromorphic hardware based on 2D materials, showing valuable academic significance and application prospects.

Fig. 1. Homogeneous transistor-memory architecture based on 2D materials. (A) The schematic of the BJT structure. (B) The mechanism of the FE proximity-induced doping effect. (C) The schematic of the memory structure. (D) The mechanism of resistance modulation in a memory cell. (E) The schematic of the integrated system based on the homogeneous transistor-memory architecture. (F) The training accuracy of the binary classification algorithm in the neural network. (G) The cost of the binary classification algorithm in the neural network.