Matrix MAC Unit is 8-bit Multiply and Accumulate core for 4 X 4 size matrices.
Generally MAC units are the base of DSP's like FIR filters.
However to do calculation of Deep Neural Networks(DNNs), the matrix ability has to be added to the design.
The design supports 4 X 4 size matrices that are standard size in deep neural network. The reason why the design is 8-bit is that for deep learning computation hardware needs to be 8-bit in width at least. Bit width can be changd by changing the number 8 in code for anything suitable. The code is
localparam DATA_WIDTH = 8Matrix size can be changed by changing
localparam M_SIZE = 4and changing the loop so it has more combinations inside the initilaization from 1D to 2D and vice-versa.