INI contributes to IBM largest spiking neural network

IBM's TrueNorth announcement of the largest spiking neural network IC signals a major advance in the developing field of physical neural computation. INI has been involved with IBM’s team as a subcontractor, via the INI spinoff iniLabs GmbH (www.inilabs.com). iniLabs has developed a high-performance spike-based dynamic vision sensor (DVS, http://inilabs.com/products/dynamic-vision-sensors) to interface directly to the TN hardware. IBM has used the DVS to develop working demonstrations of an eye tracker uses the DVS + TN hardware to track a user’s pupil, and a saliency detector using the sparse but highly informative DVS output to detect salient moving objects. By interfacing directly from a DVS to the TN hardware, IBM can avoid the power hungry conversion from standard camera frames to spikes and also can take advantage of the precise spike timing and coherence inherent in the silicon retina output. By combining the 10mW power consumption of the DVS with the 70mW power consumption of the TN, IBM could potentially build a sub-100mW visual processor that would outperform a conventional vision system consuming hundreds of watts.

© 2017 Institut für Neuroinformatik