Flexible and Disposable Paper-Based Optoelectronic Physical Reservoir Devices

Wearable sensor devices that use physical probes such as light and electric potential to monitor health conditions are widely used in the biomedical field. To classify and predict health conditions, the acquired data needs to be transmitted to an AI system in the cloud. The communication of such a large amount of data involves issues such as network load and communication speed. One of the methods to solve these issues is that we allow the sensor itself to have AI functions, such as physical reservoir (PR) devices. We believe that in-sensor optoelectronic PR device, which has (1) response times that match the time scale of biological signals (0.1-100 s), (2) flexibility, and (3) disposability, is one of the ideal frameworks to realize intelligent sensors. However, because the response times of the optoelectronic PR devices reported so far are several tens of milliseconds, it difficult to handle biological signals [1,2]. Moreover, flexibility and disposability have also not been addressed.<br/>In this study, we fabricated an optoelectronic PR device that satisfy three requirements mentioned above. Our device is composed of ZnO nanoparticles (NPs)-embedded cellulose nanofiber (CNF) freestanding film. The film was prepared by spray deposition of CNF aqueous solution dispersed with ZnO NPs (average diameter: 25 nm)[3]. Patterned Au electrodes were formed on the film surface. We demonstrated a pattern recognition using ultraviolet (UV) light input pulses. To verify that it works as a PR device, the film was exposed to UV pulsed light encoded with 4-bit patterns (pulse width: 50-500 ms, wavelength: 365 nm), and the photocurrent was measured.<br/>Under the UV pulsed light irradiation, the film showed the rise and decay time constants of 6.4 s and 3.4 s, respectively, indicating that our films have a time scale compatible with biomedical applications. Furthermore, the maximum photocurrent was increased with increasing the number of pulses when the UV light was repeatedly irradiated. This phenomenon indicates that our films exhibit the short-term memory required for reservoir devices. To demonstrate the AI performance, handwritten digit recognition was conducted using MNIST databases. Our device showed accuracy of 88%. Therefore, it was demonstrated that our films can be used as a PR device. In addition, handwritten digit recognition using our film with a bending radius of 9.5 mm and 16.1 mm showed almost no change in accuracy compared to the flat condition. Thus, it was demonstrated that our film can be used in the bent state. In this session, we will describe the results of pattern recognition in the bent state and introduce the possibilities of disposability.<br/><br/>[1] Y. Sun et al. Adv. Intell. Syst. <b>5</b>(2023)2200196<br/>[2] Z. Zhang et al. Nat. Commun. <b>13</b>(2022)1<br/>[3] H. Komatsu et al. Nanomaterials. <b>12(</b>2022)940