Power Management Applied to Neuromorphic Sensor Inputs

In this work group, we will explore how to exploit activity fluctuations in neuromorphic sensors for power management on the SpiNNaker2 prototype. The idea is that at lower input activity the core supply voltage can be reduced, saving energy consumed per spike. We have a DVS connected and support for live I/O, allowing to try out low-level vision processing algorithms with neuromorphic power management. As well, we are interested in inputs from other sensors.

We will bring the first prototype chip for the second-generation SpiNNaker system, with a DVS connected to it. The chip has four ARM cores that can each be operated at three different supply voltage levels and corresponding operating frequencies. A basic framework for aquiring DVS input and getting live output from the chip is available, allowing to observe the activity fluctuations and reaction of the cores to them. We want to try out some low-level vision processing and see how to exploit the different supply voltage levels for saving energy in these tasks. We also want to create acitivity statistics of different sensors as a basis for assessing the potential of power management for neuromorphic hardware.

Go to group wiki

Timetable

Day Time Location
Wed, 25.04.2018 14:00 - 15:00 Lab/Disco
Wed, 25.04.2018 19:00 - 19:30 Lab/Disco
Tue, 01.05.2018 10:00 - 11:00 Outside Disco

Moderator

Christian Mayr
Johannes Partzsch

Members

Tobi Delbruck
Charlotte Frenkel
Richard George
Germain Haessig
chen liu
Christian Mayr
Johannes Partzsch
Yexin Yan
Ali Zeinolabedin