Fig. 4: Experimental results of PEIC for vision tasks.
From: Photonic edge intelligence chip for multi-modal sensing, inference and learning
a Experimental setup for image classification using the PEIC. b Structure and microscope image of the random scattering chip, which features waveguides etched with random duty cycle gratings. The chip utilizes multi-stage scattering and coupling to convert spatial image data into unique spectral signatures. c Measured optical intensity distributions of the random scattering chip at two input wavelengths (1542 nm and 1543 nm), highlighting the differences in the resulting speckle patterns. d Measured average normalized spectral correlation function of the scattering chip. e Schematic of the spatial-to-spectral mapping process, where the spatial distribution of light is transformed into spectral signatures. f Example of spatial inputs and the corresponding spectral samples after processing through the random scattering chip and subsequent AWG sampling. g Training curves of the vision task and the corresponding confusion matrix. h Comparison of confusion matrices for the image classification task, with (right) and without (left) the unsupervised hardware fine-tuning procedure.
