Artificial neural networks, which are layers of interconnected artificial neurons, are becoming increasingly popular in machine learning tasks such as speech recognition and medical diagnosis. Electronic computing hardware is nearing its capabilities, but the demand for more computing power is always increasing. Researchers chose the photonic processor over the electronic processor to transmit data at the speed of light.
Not only can photonic processors process information much faster than electrons, but they are also the foundation of today’s Internet, where it is critical to avoid the so-called electronic bottleneck (conversion of an optical signal into an electronic signal and vice versa).
The proposed optical neural network can recognize and process large amounts of data and images at speeds exceeding ten trillion operations per second. An optical frequency comb, a light source made up of many equally spaced frequency modes, was integrated into a computer chip and used as a low-power source for optical computing.
This device performs a type of matrix-vector multiplication known as convolution for image-processing applications. It achieves promising results for real-time massive-data machine-learning tasks such as face recognition in cameras and pathology identification in clinical scanning applications. Their approach is scalable and trainable to much more complex networks for demanding applications such as unmanned vehicles and real-time video recognition, allowing full integration with the upcoming Internet of Things in the not-too-distant future.
Related Content: Research Team Develops New Way To Make Integrated Photonics