2021 BRN Discussion, page-29294

This is the Intel blurb for Loihi 2:

Don't understand their 3-factor learning.

Loihi 2 has changed to accommodate multi-bit actuations.

A key difference is that Loihi2 uses 6 processors per chip, and they perform essential functions across the neuromorphic domain.

In contrast, Akida's ARM Cortex processor is only used for set-up/configuration.

Having processors involved in the NN operation will slow performance and increase power consumption.

https://download.intel.com/newsroom/2021/new-technologies/neuromorphic-computing-loihi-2-brief.pdf



...
Support for three-factor learning rules. Loihi 2 programmable neuron models may now manipulate synaptic input received from its dendritic compartments with arbitrary microcode and assign the results to third factor modulatory
terms available to a neuron’s synaptic learning rules. This reatly generalizes the support Loihi offered for a reward-based broadcast mechanism that was intended for the same purpose of modulating learning rules. Loihi 2’s modulatory factors may be mapped uniquely per post-synaptic neuron, such as to represent errors that apply to individual neurons in support of supervised learning algorithms.
...
Loihi 2 increases the number of embedded processors per chip to 6 from 3 in Loihi. These processors are programmed with conventional C or Python code and perform many essential tasks related to encoding and decoding data across the neuromorphic domain, as well as management and housekeeping operations. The increased processor count helps to prevent these conventional processing tasks from bottlenecking overall application performance, as occasionally occurred in Loihi.