Tsinghua University makes progress in the field of pre sensing optical computing

In the era of the Internet of Things, visual image sensors, as key devices in the intelligent society, are embedded in various devices such as mobile communication terminals, smart wearable devices, automobiles, and industrial machines. With the continuous expansion of applications, higher requirements have been put forward for the system power consumption, response speed, safety performance, and other aspects of sensors. In the traditional "sensory transmission computing" link, the access speed of memory and communication bandwidth have gradually become the main bottlenecks limiting system power consumption and speed. Moving the computing unit closer to the sensing unit has gradually become a powerful way to solve this problem, as it enables the near sensing end of the system to have certain data processing capabilities. Compared to other proximity computing methods, pre sensing light computing has the advantages of high speed, high bandwidth, and low power consumption. However, currently the vast majority of optical neural networks require coherent lasers as light sources, with bulky and complex hardware systems that can only perform linear operations and lack interlayer nonlinear activation, which limits the application of pre sensing light computing in edge scenes.

Figure 1. Paradigm of Near Sensor Computing in Machine Vision Link

Professor Chen Hongwei's team from the Department of Electronic Engineering at Tsinghua University proposed a compact passive multilayer optical neural network (MONN) architecture, which consists of a passive mask and a quantum dot thin film, to complete multilayer optical calculations with interlayer nonlinear activation under incoherent light illumination. The optical length of this architecture is as short as 5 millimeters, which is 2 orders of magnitude smaller than existing lens based optical neural networks. Experimental results have shown that this multi-layer computing architecture outperforms linear single-layer computing in various visual tasks, and can transfer up to 95% of computations from the electrical domain to the optical domain. This architecture has the advantages of small size, low power consumption, and high practicality, and is expected to be deployed in mobile visual scenarios such as autonomous driving, intelligent manufacturing, and virtual reality in the future.

Figure 2. Multi layer pre sensing optical neural network architecture and interlayer nonlinear activation function measurement

Meanwhile, the absorption and emission spectra of CdSe quantum dots overlap within the wavelength range. By designing, the absorption and excitation spectra of the front and rear quantum dots can be aligned, enabling cascading and expansion of existing three-layer architectures to more layers. The MONN architecture can also be combined with other proximity computing paradigms to complete complex computing functions.

Recently, the related research results were published in Science Advances under the title "Pre sensor Computing with Compact Multi layer Optical Neural Network". The Department of Electronic Engineering at Tsinghua University is the first unit of the paper, Chen Hongwei is the corresponding author of the paper, and Huang Zheng, a doctoral student from the Department of Electronics in 2020, is the first author of the paper. The research has received support from the National Natural Science Foundation of China and the Beijing Municipal Science and Technology Commission.

Source: Opticsky