Multimode diffractive optical neural network

Abstract

On-chip diffractive optical neural networks (DONNs) bring the advantages of parallel processing and low energy consumption. However, an accurate representation of the optical field's evolution in the structure cannot be provided using the previous diffraction-based analysis method. Moreover, the loss caused by the open boundaries poses challenges to applications. A multimode DONN architecture based on a more precise eigenmode analysis method is proposed. We have constructed a universal library of input, output, and metaline structures utilizing this method, and realized a multimode DONN composed of the structures from the library. On the designed multimode DONNs with only one layer of the metaline, the classification task of an Iris plants dataset is verified with an accuracy of 90% on the blind test dataset, and the performance of the one-bit binary adder task is also validated. Compared to the previous architectures, the multimode DONN exhibits a more compact design and higher energy efficiency.

Realization of optical logic gates using on-chip diffractive optical neural networks

“…Leveraged on high-contrast-transmit-array (HCTA) metasurfaces [1], several on-chip diffractive optical neural networks on a silicon-on-insulator (SOI) substrate have been demonstrated in previous works [2][3][4][5][6][7]. Despite many benefits offered by on-chip diffractive optical neural networks like low-power consumption and light-speed parallel signal processing, challenges are faced because of deviations between the diffraction-based analysis methods and experimental/full-wave electromagnetic verifications.…”

“…Despite many benefits offered by on-chip diffractive optical neural networks like low-power consumption and light-speed parallel signal processing, challenges are faced because of deviations between the diffraction-based analysis methods and experimental/full-wave electromagnetic verifications. While this discrepancy was mainly attributed to the limited capability of the diffraction-based analysis methods in modelling the evolution of optical fields through the network [2,8] and several previous works attempted to unravel the problem by applying a relatively large distance between successive metasurfaces to maintain stable interference [3][4][5], restricting multiple consecutive meta-atoms to be the same in the metasurfaces to decrease the mutual coupling between the adjacent meta-atoms [3][4][5][6][7], etc.…”