In the 2021 Gemmini paper (
https://arxiv.org/pdf/1911.09925) the total area is presented to be 1 029 000 um^2 for a 22nm 1GHz default (16x16 systolic array) Gemmini configuration.
However this is(if I'm not mistaken) for the INT8 precision for inputs and weights.
My question is: what is the area of a 16-bit floating-point configuration, keeping everything else default?
So using the same systolic array size(16x16), same scratchpad sizes, same Rocket single-core CPU, but just changing the systolic array to support 16bit floating point numbers instead of 8bit integers?
I couldn't find any papers that report on area for a floating-point Gemmini configuration.
If there are papers, or people who have an estimation on the area increase, feel free to point me in the right direction.
Best regards,
Edward