19/10/2023 1:15 pm
Topic starter
Notifications
Clear all
0
Hello! We've met 2 problems and would appreciate your response.
- We've trained a ResNet18 SNN. However, due to the limit of the Speck devkit, we're not able to test it on chip and want to test it on Specksim. While it's convenient to build residual shortcuts on chip by samna_cfg, is there a similar solution to build this sort of structure in Specksim?
- Many SNN ResNets entail a shared IF layer after two convolution layers to combine the shortcut and the mainstream output. We've tried hard to find this kind of support when using Speck but failed at last. Is there any method to realize this structure on Speck?
Looking forward to your reply!
