A Visit to the Neuromorphic Zoo

As Deep Learning (DL) enters embedded devices, developers wonder what hardware they should use for these algorithms. Besides CPUs, embedded GPUs have appeared recently, but the market of dedicated ASICs for DL showed a real revolution in the last two years. Mostly American and Chinese companies - big players like Intel and Google and many start-ups - have developed chips that outperform embedded CPUs and GPUs by far. Moreover, new terms appear in the articles. Spiking Neural Networks (SNN) are about to be used in first practical applications, and analog electronics including memristors/ReRAMs claim a place in the zoo of DL hardware. All of these species are refered to as neuromorphic hardware. Our presentation will explain why dedicated chips are better suited for DL than CPUs and GPUs, how DL can be carried out in analog electronics and what is actually an SNN. Moreover, we will present some examples of commercial chips and ongoing research activites and projects.