A 62.45 TOPS/W Spike-Based Convolution Neural Network Accelerator with Spatiotemporal Parallel Data Flow and Sparsity Mechanism.
Chen-Han HsuYu-Hsiang ChengZhaofang LiPing-Li HuangKea-Tiong TangPublished in: AICAS (2022)
Keyphrases
- data flow
- neural network
- parallel implementation
- database machine
- systolic array
- object oriented software
- control flow
- artificial neural networks
- object oriented
- high dimensional
- spatial and temporal
- spiking neurons
- data transfer
- digital signal processing
- parallel processing
- shared memory
- sparse representation
- message passing interface
- image processing
- distributed memory
- space time
- back propagation
- feed forward
- workflow model
- moving objects
- database systems
- bottle neck