Description
State-of-the-art photonic neural networks (PNNs) demonstrate ultrafast processing speeds and broadband operation by exploiting the intrinsic properties of light. However, conventional integrated optical matrix–vector multiplication (MVM) schemes rely on lossy coupling schemes and off-chip nonlinearities, which significantly reduce the overall energy efficiency and scalability of the system. To address this limitation, we present the co-integration of passive nonlinear activation functions and linear operators on the same chip. By employing periodically poled thin-film lithium tantalate (PPLT) waveguides for nonlinear activation via second-harmonic generation (SHG) and integrating linear operators through high-speed electro-optic modulation, we demonstrate a pathway toward compact, energy-efficient, and scalable photonic computing architectures. Thin-film lithium tantalate (TFLT) is selected over lithium niobate because of its superior resistance to photorefractive damage and its stable nonlinear and electro-optic performance under high optical power, making it particularly well suited for integrated visible and near-infrared photonic computing. The proposed approach enables monolithic neuromorphic photonic integration, reducing device footprint and optical loss while enhancing efficiency and functional scalability. A key focus of this work is the development of efficient EOMs with low driving voltages to ensure compatibility with modern high-speed electronic drivers. Furthermore, the critical challenge of fiber-to-chip coupling is addressed through direct laser-written, out-of-plane polymer lenses optimized for near-infrared and visible wavelengths. Compared to conventional coupling schemes such as grating couplers, these polymer couplers exhibit significantly lower loss over a broad bandwidth and enable wavelength-division multiplexing for parallelized photonic computation. The demonstrated platform thus provides a versatile foundation for next-generation, low-power, and high-speed photonic neural networks.
| I am the presenting author | Yes |
|---|