A 2048-Neuron Spiking Neural Network Accelerator With Neuro-Inspired Pruning And Asynchronous Network On Chip In 40nm CMOS

A 40nm, 2.56mm2, 2048-neuron globally asynchronous locally synchronous (GALS) spiking neural network (SNN) chip is presented. For scalability, we allow neurons to specialize to excitatory or inhibitory, and apply distance-based pruning to cut communication and memory. An asynchronous router limits the latency to 1.32ns per hop. The reduced traffic and lower latency allow the input channel to be parallelized to achieve 7.85GSOP/s at 0.7V, consuming 5.9pJ/SOP.

[1]  Zhengya Zhang,et al.  A Sparse Coding Neural Network ASIC With On-Chip Learning for Feature Extraction and Encoding , 2015, IEEE Journal of Solid-State Circuits.

[2]  Joel Zylberberg,et al.  Inhibitory Interneurons Decorrelate Excitatory Cells to Drive Sparse Code Formation in a Spiking Model of V1 , 2013, The Journal of Neuroscience.

[3]  Hong Wang,et al.  Loihi: A Neuromorphic Manycore Processor with On-Chip Learning , 2018, IEEE Micro.

[4]  Gregory K. Chen,et al.  A 4096-Neuron 1M-Synapse 3.8PJ/SOP Spiking Neural Network with On-Chip STDP Learning and Sparse Weights in 10NM FinFET CMOS , 2018, 2018 IEEE Symposium on VLSI Circuits.

[5]  Fabien Clermidy,et al.  Asynchronous Circuit Designs for the Internet of Everything: A Methodology for Ultralow-Power Circuits with GALS Architecture , 2016, IEEE Solid-State Circuits Magazine.

[6]  A. van Ooyen,et al.  Independently Outgrowing Neurons and Geometry-Based Synapse Formation Produce Networks with Realistic Synaptic Connectivity , 2014, PloS one.

[7]  Andrew S. Cassidy,et al.  A million spiking-neuron integrated circuit with a scalable communication network and interface , 2014, Science.

[8]  Michael P. Flynn,et al.  A 3.43TOPS/W 48.9pJ/pixel 50.1nJ/classification 512 analog neuron sparse coding neural network with on-chip learning and classification in 40nm CMOS , 2017, 2017 Symposium on VLSI Circuits.

[9]  Christian Bernard,et al.  8.1 a 4x4x2 homogeneous scalable 3d network-on-chip circuit with 326mflit/s 0.66pj/b robust and fault-tolerant asynchronous 3d links , 2016, 2016 IEEE International Solid-State Circuits Conference (ISSCC).

[10]  David J. Field,et al.  Emergence of simple-cell receptive field properties by learning a sparse code for natural images , 1996, Nature.