Microscale optoelectronic reservoir networks of halide perovskite for in-sensor computing

TL;DR

This paper presents a novel multimodal optoelectronic reservoir network using halide perovskite devices capable of processing voltage and light inputs, achieving high classification accuracy for images and videos, and suitable for in-sensor computing.

Contribution

The work introduces a microscale halide perovskite-based reservoir network that processes multimodal inputs and demonstrates high accuracy, advancing neuromorphic in-sensor computing technology.

Findings

Achieved up to 95.3% accuracy for image classification.

Achieved up to 87.8% accuracy for video classification.

Networks outperform linear classifiers significantly.

Abstract

Physical reservoir computing is a promising framework for efficient neuromorphic in and near-sensor computing applications. Here, we demonstrate a multimodal optoelectronic reservoir network based on halide perovskite semiconductor devices, capable of processing both voltage and light inputs. The devices consist of micrometer-sized, asymmetric crossbars covered with a MAPbI3 perovskite film. In a network, we simulate the performance by transforming MNIST images and videos based on the NMNIST dataset using 4-bit inputs and training linear readout layers for classification. We demonstrate multimodal networks capable of processing both voltage and light inputs, reaching mean accuracies up to 95.3 p/m 0.1% and 87.8 p/m 0.1% for image and video classification, respectively. We observed only minor deterioration due to measurement noise. The networks significantly outperformed linear classifier references, by 3.1% for images and 14.6% for video. We show that longer retention times benefit classification accuracy for single-mode networks, and give guidelines for choosing optimal experimental parameters. Moreover, the microscale device architecture lends itself well to further downscaling in high-density sensor arrays, making the devices ideal for efficient in-sensor computing.