**Abstract:** In the age of the Internet of Things (IoT), visual image sensors have become essential components in a variety of devices, including mobile communication terminals, smart wearable devices, automobiles, and industrial machinery. As the applications of these sensors continue to expand, there is an increasing demand for improvements in system power consumption, response speed, and security. Traditional sensor systems, which follow a "sense, transmit, compute" (STC) chain, are often constrained by the limitations of memory access speed and communication bandwidth, leading to higher power consumption and slower processing times. To address these challenges, integrating computational units closer to the sensing units has emerged as a promising solution, enabling the system's near-sensor end to handle certain data processing tasks. Pre-sensor optical computing, in particular, offers significant advantages such as high speed, high bandwidth, and low power consumption compared to other near-sensor computing methods. However, most current optical neural networks (ONNs) rely on coherent lasers as light sources, which result in bulky and complex hardware systems. Additionally, these ONNs are limited to linear operations and lack the ability to perform non-linear activation between layers, which restricts their applicability in edge computing scenarios. A team led by Professor Chen Hongwei from the Department of Electronic Engineering at Tsinghua University has developed a novel, compact, and passive multi-layer optical neural network (MONN) architecture. This architecture consists of passive masks and quantum dot films, enabling multi-layer optical computations with non-linear activation under non-coherent illumination. The optical length of the proposed MONN is as short as 5 millimeters, making it significantly smaller (by two orders of magnitude) than existing lens-based ONNs. Experimental results demonstrate that this multi-layer computational architecture outperforms linear single-layer operations in various visual tasks and can transfer up to 95% of the computation from the electrical domain to the optical domain. The unique properties of cadmium selenide (CdSe) quantum dots, which exhibit overlapping absorption and emission spectra over a wide wavelength range, allow for the cascading of quantum dots. This design facilitates the expansion of the current three-layer architecture to more layers, enhancing the computational capabilities of the system. The pre-sensor MONN architecture can also be integrated with other near-sensor computing paradigms to perform more complex computational tasks. The research findings have been published in the journal *Science Advances* under the title "Pre-sensor Computing with Compact Multi-layer Optical Neural Network." Tsinghua University's Department of Electronic Engineering is the first author institution, with Professor Chen Hongwei as the corresponding author and Huang Zheng, a doctoral student from the Department of Electronic Engineering, as the first author. The study was supported by the National Natural Science Foundation of China and the Beijing Municipal Science and Technology Commission. This innovative architecture holds significant promise for deployment in mobile visual applications such as autonomous driving, smart manufacturing, and virtual reality, where small size, low power consumption, and high practicality are crucial. The development of this compact, multi-layer ONN represents a significant step forward in the field of pre-sensor optical computing, addressing the limitations of current technologies and paving the way for more efficient and versatile sensor systems. **Key Events:** - Development of a compact, passive multi-layer optical neural network (MONN) architecture. - Demonstration of multi-layer optical computations with non-linear activation under non-coherent illumination. - Experimental validation showing superior performance in visual tasks and significant computational offloading. - Potential for expansion to more layers and integration with other near-sensor computing paradigms. - Publication of research findings in *Science Advances*. **Key People:** - Professor Chen Hongwei (Tsinghua University, Department of Electronic Engineering) - Huang Zheng (First author, doctoral student, Department of Electronic Engineering, Tsinghua University) **Key Location:** - Tsinghua University, Beijing, China **Time Element:** - July 30, 2024 (date of the news release) - Recent publication in *Science Advances* (specific date not provided)