Synthesis of Power-Efficient Analog Neural Networks for Signal Processing

Konferenz: ANALOG 2020 - 17. ITG/GMM-Fachtagung
28.09.2020 - 30.09.2020 in online

Tagungsband: ITG-Fb. 293 Analog 2020

Seiten: 6Sprache: EnglischTyp: PDF

Aul, Florian; Hedrich, Lars (Institute for Computer Science, Goethe University Frankfurt, Germany)
Katsaouni, Nikoletta; Schulz, Marcel H. (Institute for Cardiovascular Regeneration, Goethe University Frankfurt, Germany)

This paper presents a concept and an implementation for an automatic, power-aware synthesis of neural networks as analog circuits. Our concept incorporates fully automatic design space exploration (synthesis) of powerefficient neuron-building blocks. Further, it implements techniques to have a programmable, and at the same time small neural network. First results on the schematic level for energy consumption of a small neural network applied to a peak detection dataset are presented.