New (guest) blog post: FINN Matrix Vector RTL Backend

[maltanar]

You may be aware that High-Level Synthesis (HLS) is one of the core technologies we use to make FINN possible -- indeed, we have a whole library of FINN HLS components for this. But for some applications, it's worth coming up with RTL implementations for key components for that extra bit of performance and resource efficiency.

This is what FINN contributor @asadalam did for the Matrix-Vector-Activation Unit, which is a vital FINN component used for implementing both dense convolutions and fully-connected layers. You can read more on the blog post just published here, or go straight to the repo here.

While the component itself isn't integrated into the FINN compiler yet, we hope to provide this integration in the near future, and in the meantime this can be used as a manual drop-in replacement for the HLS Matrix_Vector_Activate layers.