This project offers an accelerator for neural network computations.
Instead of performing the calculations in software using GPU, we implement a set of neurons that can perform the same calculations faster and concurrently in hardware on FPGA.
The design offers flexibility, as at each iteration we can specify the active neurons that participate in the computation.
We implement a set of neurons, that can perform the computations concurrently.
Implementing the neuron in hardware, have two main benefits :
1. perform calculations in several clock cycles.
2. more parallel computations are achieved, that means that we can have many neurons that work concurrently .
In this project the numbers are in Fixed Point method.
The neural network’s algorithm runs on the ARM processor,
which uses the neurons implemented in the hardware for performing the calculation fast and concurrently.
