Table 1 The comparison results of hardware implementation for neural network applications using memristive crossbar arrays
This work | Ref. 57. | Ref. 36. | Ref. 30. | Ref. 31. | Ref. 56. | |
|---|---|---|---|---|---|---|
Array configuration | 1 R (SRC) | 1S-1R | 1T-1R | 1T-1R | 1T-1R | 1T-1R |
Array density | 32 × 32 (1 kb) | 9 × 9 (<1 kb) | 128 × 64 (8 kb) | 54 × 108 (~5 kb) | 128 × 16 (2 kb) | 128 × 64 (8 kb) |
Array yield (%) | 100 | 100 | 98.9 | Not available | 99 | Not available |
Energy efficiency (TOPS/W) | 4.35 | Not available | 115 | 1.37 | 11 | 77.4 |
Application | MNIST classification | Pattern classification | MNIST classification | Pattern classification | MNIST classification | MNIST classification |
Accuracy | 100 | 100 | 89.9 | 94.6 | 96.19 | 92.3 |
Integrated device stack | Ru/HfSiOy/ Al2O3/HfSiOx/TiN | Pt/Ru /TiO2 /RuO2/Pt /HfO2 /TiN | Ta/HfOx /Pd | Pd/ WOx/Au | TiN/TaOx /HfOx /TiN | Pt/TaOx /Ta |
