Deep learning is changing the way computers see, hear and identify objects -- but will it ever migrate into smartphones or vision SoCs for cars? Has anybody come up with SoCs optimized for neural networks?