Nowadays, with the rapid development of “big data” technology, the gradual failing of Moore’s law and the constraint of the Von Neumann bottleneck hinder the computing ability of “big data” applications with random-access memory and photolithography. As a result, it is highly desirable to seek ways to increase computing power, which presents a great challenge to the traditional von Neumann architecture. Compared with electronic computing systems, the human brain excels at learning, recognizing, and classifying, due to its unified memory and processing.

In natural neural network, the information is stored and processed in the change of weight. In an artificial neural network composed of neuromorphic computing units, the memristor, of which the resistance can be modulated dynamically and repeatedly under external stimuli plays as important a role as neurons do in the human brain. In this work, we utilize the double metal-cladding waveguide to generate light field distribution and drive the photorefractive effect, switching the weight of the all-photonic synapse.

With two incident beams, signal beam and erasing beam, the refractivity index is switched between low and high reflectivity state.

Figure 1. The reflectivity switched between low and high reflectivity state.

As shown in Figure 2, with different incident signal, the weight of all-photonic synapse can be switched to arbitrary state among 1-5. And this process is stable and repeatable.

Figure 2. The weights switched by light field.

In this work, we demonstrated an all-photonic synapse which can be switched steadily and repeatably by CW laser without heating.

This work was published in “Qiheng Wei, Hailang Dai, Hongrui Shan, Honggen Li, Zhuangqi Cao, and Xianfeng Chen, All-photonic synapse based on iron-doped lithium niobate double metal-cladding waveguides, Physical Review B, 104, 235308 (2021)”.

Link: https://journals.aps.org/prb/abstract/10.1103/PhysRevB.104.235308