Though neuromorphic architectures are interesting, the Loihi is essentially an ASIC for Spiking Neural Networks [1]. The lower power consumption than GPUs is due to the limited nature of computation that can be performed on these chips (which is generally true for ASICs).

[1] https://en.wikichip.org/wiki/intel/loihi

I don't quite understand brain simulation. Most article talks about how neuron works, how they are connected. It seems I can write a simulation easily. https://medium.com/@bennashman/modeling-your-brain-on-a-comp...

What I don't get is, the interconnected neurons don't seem to be differentiable. How do they learn? How to provided feed back to the system? How to train them like we do to ANN?

I look forward to my brain leaking memory to other threads.