Table 1 Comparison among various diffractive neural networks.
From: Reconfigurable magneto-optical diffractive neural network with enhanced optical phase modulation
1 st author [Reference] | \(\lambda\) (nm) | Diffractive surface (Hidden layer) | Accuracy (%) (MNIST digit) | |||||
|---|---|---|---|---|---|---|---|---|
Scheme | Layers | Neurons (per layer) | Size (µm) | Phase mod. (deg.) | Sim. | Exp. | ||
Lin7 | 750 k | Thickness | 5 | 200*200 | 400 | 0–360 | 92 | 81 |
Chen15 | 633 | Thickness | 5 | 1000*1000 | 4 | 0–180 | 92 | 84 |
Long17 | 532 | LC | 5 | 1000*1000 | 5 | 0–360 | 98 | 89 |
Fu20 | 4000 | PCM | 3 | 256*256 | 20 | 0, 180 | 92 | - |
Dong10 | 532 | Thickness | 2 | 1024*1024 | 8 | 0, 90 | 85 | 82 |
Zheng14 | 515 | LC | 1 | 200*200 | 8 | 0–360 | 97 | 93 |
This work | 532 | MO | 1 | 200*200 | 1 | \(\pm\) 1 ~ 5 | 98 | - |
112*112 | 4 | \(\pm\) 3.3 | 95 | 83 | ||||
