We are pleased to share our latest work, โ๐๐น๐น-๐ผ๐ฝ๐๐ถ๐ฐ๐ฎ๐น ๐ต๐ถ๐ด๐ต-๐๐ฝ๐ฒ๐ฒ๐ฑ ๐ฝ๐ฟ๐ผ๐ด๐ฟ๐ฎ๐บ๐บ๐ฎ๐ฏ๐น๐ฒ ๐ป๐ผ๐ป๐น๐ถ๐ป๐ฒ๐ฎ๐ฟ ๐ฎ๐ฐ๐๐ถ๐๐ฎ๐๐ถ๐ผ๐ป ๐ณ๐๐ป๐ฐ๐๐ถ๐ผ๐ป๐ ๐๐๐ถ๐ป๐ด ๐ฎ ๐๐ฎ๐ฏ๐ฟ๐โ๐ฃ๐ฒ๐ฟ๐ผ๐ ๐น๐ฎ๐๐ฒ๐ฟโ, which advances the development of fully optical, programmable nonlinear activation units for photonic neural networks.
In this paper, we demonstrate how a semiconductor laser under single and dual optical injection can act as a high-speed, reconfigurable nonlinear element. By achieving an excellent agreement between the theoretical model and experimental measurements, the study confirms the predictive strength and robustness of our model, which accurately captures the underlying physics of optical nonlinearities.
๐๐ฒ๐ ๐ต๐ถ๐ด๐ต๐น๐ถ๐ด๐ต๐๐ ๐ถ๐ป๐ฐ๐น๐๐ฑ๐ฒ:
โก Rapid nonlinear responses at high data rates
๐ง Programmability using optical control signals
๐ The ability to implement various activation profiles (e.g., sigmoid-like, saturating behaviors) entirely in the optical domain
๐ Very low energy consumption per nonlinear operation, making it suitable for integration into next-generation photonic systems
The work also presents an early-stage validation, demonstrating the feasibility of using FabryโPรฉrot laser diodes as building blocks for scalable, all-optical neural network architectures that operate at ultrafast speeds with minimal energy cost.
Read the full paper here