Center for Functional Nanomaterials Seminar

Recent advances in artificial neural networks (ANNs) attracts magnificent interests from a variety of areas. Efforts to optimize a hardware platform for such a data-intensive paradigm have led to the hybridization of sensor-, memory-, and processor modules. A notable example is in-memory computing platforms that are being extensively explored with integrated circuits (ICs) of analog resistive memories, shaping the field of analog neural network computing.

On the other hand, two-dimensional (2D) materials are attractive material systems with a wealth of electronic and optoelectronic properties. Persistent photoconductivity (PPC) is an optoelectronic memory effect, prevalently reported from 2D material-based optoelectronic devices. Such an exciting device property opens new opportunities in optoelectronic analog neural network computing. However, the low integration level of 2D optoelectronics has thus far limited their use in ANNs.

Here, we present a 2D IC, where 1,024 MoS2 phototransistors are integrated, demonstrating unprecedented system functionality among 2D electronics/optoelectronics. Firstly, the 2D IC can perform as an in-sensor image memory, where an optical image is converted and stored into a matrix of electrical conductance. Secondly, this IC can be implemented as an analog ANNs in that the conductance matrix, representing synaptic weights of the network, is formed through a fully parallel, image-based optical programming scheme. Finally, it can serve as an in-sensor video processor, physically compressing information of sequential images into a single image data. The demonstrations are convergences of sensor, memory, and processor; thus, they are in-sensor memorizing, in-memory computing, and in-sensor computing, respectively. The first leverage of unique 2D materials' properties at the system level opens a new pathway for novel material systems to be implemented in various ANN applications.