Table 3 Training time per epoch for memristors with different time required for write cycles (10 ns, 100 ns and 1000 ns) and average power consumption per neural computation.

From: Memristive GAN in Analog

Update Configuration

Power (W)

Training time (s)

max

min

10 ns

100 ns

1000 ns

Parallel update of independent columns in layers and independent layers

0.28 μ

47.08 n

0.47

4.71

47.10

Update of memristors in 4 cycles of independent layers in series (rows and columns)

2.52 μ

0.408 μ

0.014

0.144

1.44