Question about feature 'resnet_dt' and 'useBN'

[Franklalalala]

Going through papers and docs, I knew that, deepmd builds embedding net and fitting net with resnet style MLP.
Reading the src code deepmd\utils\network.py, I found a button called 'resnet_dt' and notation says 'Using time-step in the ResNet construction', which confused me. Here is the code starting in line 206:
if resnet_dt :
xx += hidden * idt
else :
xx += hidden
I've read the blogs about residual net, but still couldn't understand the notation. What's the meaning about time-step? It seems like a weight between identity and mapped identity, since it's a randommly initialized constant.
On the same file, I see the button 'useBN', apparently it has been used. It's well known that a nomalization layer can avoid gradient vannishing etc... Is there any plan to reconsider the module? Or rebuild the structure of the training nets, like cnn, rnn? Just out of curiosity.