On April 18, a research team in China published its latest research result (the first time to use machine-learning inverse design to achieve three-dimensional vectorial holography) in the top-notch academic journal, Science Advances. Science Advances, affiliated to Science, is an open and comprehensive scientific journal, covering all academic fields.
The breakthrough in the field of optical holography was completed by the International Laboratory of future optics led by academicians Zhuang Songlin and Gu min at University of Shanghai for Science and Technology. In their research, the inverse design, based on machine learning, can accurately and rapidly generate one or more arbitrary three-dimensional vector light fields. The research result is expected to be applied in various fields such as ultra wideband holographic display, ultra-secure information encryption, ultra-large optical storage, ultra-precise particle control, etc.
Light is one type of electromagnetic wave. It transmits in the medium accompanied by electromagnetic and magnetic field oscillations, which are called the vector characteristics of light. Because of the S-wave characteristics of light waves, the oscillation of light is usually limited to a two-dimensional plane perpendicular to its transmission direction. In recent years, scientists have found that the vibration of light can break the shackles of the traditional two-dimensional plane, and generate longitudinal light oscillation through interference, that is, the third-dimensional light vector.
However, to accurately generate any three-dimensional vector light field has always been a problem plaguing researchers around the world. In physics, a three-dimensional vector light field distribution can be obtained by solving the three-dimensional Maxwell equation, but three-dimensional vector light field distribution, thus produced, is uncontrollable.
The Research team led by Academician Gu solved this problem by using the machine learning inverse design of artificial intelligence, took the lead in realizing 3D vector holography, and could accurately control any 3D vector state of each pixel in the 3D holographic image.
Such control can encode, transmit and decode the information carried by each three-dimensional vector light, thus eliminating the limits of the traditional two-dimensional polarized light. "Through the new technology of artificial intelligence machine learning, we have realized the control of three-dimensional vector light for the first time, and extended the algorithm of machine learning into optical holography, " said Academician Gu.
Machine learning plays an increasingly important role in optical design. Dr. Ren Haoran, the first author of the paper, said: "our research has proved that the trained artificial neural network can effectively and rapidly generate any three-dimensional vector light field, achieving nearly 100% accuracy, and greatly improving the efficiency of light field control."
In addition, the invention opens up a new way for optical holography. It is the first time to prove that the three-dimensional vector state of light can be used as an independent information carrier to realize information coding and multiplexing. "This invention not only lays the foundation for the next generation of ultra-wideband, ultra-large-capacity, ultra-fast processing optical holography system, but also provides a new platform for the researchers to deepen their understanding of the interaction between light and matter (such as particle control)," Academician Gu said.