5-Bit Electrically Programmable ITO Modulator Based on Photonic Crystal Nanobeam for Photonic Neural Networks

This is the authors' abstract. We don't add key points for this paper.

Photonic convolutional neural networks (PCNNs) demand compact, energy-efficient modulators to overcome the limitations of conventional thermo-optic devices in scalability and power consumption. Here, we present an on-chip thermo-optic modulator integrating a photonic crystal nanobeam cavity (PCNC) with an indium tin oxide (ITO) micro-heater. The ultra-small mode volume of PCNC enhances light-matter interactions, and the ITO heater enables efficient Joule heating with minimal optical loss. Experimentally, the modulator attains a shift of 2.04 nm/V at resonant wavelength and 56 states of programmable optical intensities, superior to micro-ring resonators (MRRs) and Mach-Zehnder modulators (MZIs) for tuning efficiency. Being implemented in a 5-bit PCNN, the device achieves the recognition accuracy of 98.90% on MNIST database (Modified National Institute of Standards and Technology database) of handwritten digits as well as low quantization error of 0.0107 on average, which demonstrates its performance and feasibility for high-precision photonic computing. The compact footprints and large free spectral range (FSR) of PCNC could further mitigate the impacts of the wavelength division multiplexing (WDM) crosstalk. Therefore, our design scheme has the potentials to advance the development of compact, scalable, and power-efficient photonic integrated circuits (PICs) for next-generation neuromorphic computing, bridging the gaps between applications of silicon photonic device and improvement of PCNN system performance.