In this paper we analyze several simplicial piecewise-linear digital computation architectures that are good candidates to replace the multiply and accumulate (MAC) units in deep neural networks. We analyze them in terms of area and energy of a digital VLSI realization and compare the results with the classic MAC architectures.
|Name||2020 Argentine Conference on Electronics (CAE)|
|Conference||2020 Argentine Conference on Electronics (CAE)|
|Period||27/02/20 → 28/02/20|
- Linear combination
- piecewise-linear functions
- vector-vector multiplication