Optical computing has gradually demonstrated its efficiency in handling increasingly complex computational demands, attracting widespread attention. Optical switches can effectively control and modulate optical signals, providing flexibility and efficiency for optical computing systems. However, traditional optical switches face performance issues such as power consumption, switching speed, and compactness, severely limiting the implementation of large-scale photonic integrated circuits and optical neural networks. This paper proposes an innovative design structure for a non-volatile multi-level adjustable optical switch by combining a plasmonic slot waveguide with segmented phase-change materials. Modulation of waveguide light transmission is achieved by adjusting the phase state of Ge2Sb2Te5(GST). At a wavelength of 1550 nm, a low insertion loss of 0.5dB has been achieved, with approximately an 85% difference in optical transmittance between amorphous state (aGST) and crystalline state (cGST). The high transmittance difference contributes to achieving a wide range of weight variations and supports precise weight updates. Based on this design, we successfully implemented a handwritten digit recognition task with an accuracy of 95%, laying the foundation for future more efficient memory computing neural morphic networks.
Read full abstract