Integration of periodically poled thin-film lithium niobate and electro-optic modulators for on-chip photonic neural networks

Photonic neural networks (PNNs) enable ultrafast and broadband computation using the intrinsic properties of light, but conventional matrix–vector multiplication (MVM) schemes suffer from lossy coupling and off-chip nonlinearities, limiting efficiency and scalability. We present a monolithic approach integrating passive nonlinear activation and linear electro-optic operations on a thin-film lithium niobate (TFLN) platform. Periodically poled TFLN (PPLN) waveguides provide nonlinear activation via second-harmonic generation, while low-voltage electro-optic modulators (EOMs) enable high-speed control. Efficient fiber-to-chip coupling is achieved using direct laser-written polymer lenses with low loss and broadband performance. This co-integrated architecture offers a compact, energy-efficient, and scalable foundation for next-generation photonic neural networks.