Benchmarking the Epiphany Processor as a Reference Neuromorphic Architecture

This short article explains why the Epiphany architecture is a proper refer-ence for digital large-scale neuromorphic design. We compare the Epiphanyarchitecture with several modern digital neuromorphic processors. We showthe result of mapping the binary LeNet-5 neural network into few modernneuromorphic architectures and demonstrate the efficient use of memory inEpiphany. Finally, we show the results of our benchmarking experimentswith Epiphany and propose a few suggestions to improve the architecturefor neuromorphic applications. Epiphany can update a neuron on average in120ns which is enough for many real-time neuromorphic applications