利用機器學習於單一電晶體上實現多重模態感測效果

Abstract

由於物聯網的發展，次世代的感測器研究正在發生典範轉移。如今，多模態感測器已經在物聯網及人工智慧應用中扮演不可或缺的角色。為了更進一步降低功率及感測器尺寸，多模態感測元件將是未來感測技術的重要趨勢。因此，在本論文中，我們利用雙閘極離子感測場效電晶體成功實現多模態感測元件，成功同時量化酸鹼度，光及溫度三種物理變量。並提出虛擬感測器空間的模型來解釋多模態感測原理。首先，本論文提出利用時序產生虛擬感測器的理論，並解釋虛擬感測器空間與多重感測效應的關聯性。隨後，我們成功實現並驗證酸鹼度/光度雙重感測器以及酸鹼度/光/及溫度三重感測器，並且利用感測器的量測數據驗證我們所提出之虛擬感測空間的模型。利用本論文提出的虛擬感測器空間與多模態感測原理，我們可以將多模態感測器實現於單一感測元件中，並應用於感測器融合，物聯網以及人工智慧辨識等應用。

Internet-of-thing (IoT) has driven a paradigm shift to the next-generation sensor architecture. Nowadays, multi-modal sensors are essential for various IoT and AI applications. To achieve ultra-low power and size reduction, single-device-multi-sensor could be a promising direction of sensor technologies. Therefore, we have realized a machine learning-assisted multi-modal sensing device based on a dual-gate ion-sensitive field-effect transistor (DG-ISFET) and proposed a virtual sensor space model to effectively explain the fact of multi-sensing. In the first part, we propose a sequential method to generate virtual sensors and a virtual sensor space model to connect the physical sensing device and multi-sensing functionalities. Afterwards, we successfully realized and validated a pH/light dual-modal sensor and a pH/light/temperature 3-dimensional multi-modal sensor. In addition, the virtual sensor space model is also validated. With the proposed virtual sensor space model, multi-modalities sensing can be achieved on single device. It has great potential applying to sensor fusion and AI recognition technologies.