Design fast and energy-efficient spiking neural networks
- Genetic algorithm based biological parameter optimization selects the optimal set of parameters that can greatly accelerate training and inference.
- High-order numerical solver improves the precison of neural dynamics and the effciency of information processing.
- Adaptive pruning techniques can significantly reduce memory size and energy consumption while maintaining high network performance.
Parallel online spiking neural network embedded accelerator on FPGA
- Parallel memory allocation and access of synaptic weights for accelerating synaptic operations.
- Time-multiplexed multiple neuron processing cores for energy-efficient and low hardware cost design.
- Online unsupervised local learning core.
- High-speed data communication between the accelerator and external device.
Reconfigurable AER-based neuromorphic system
- Reconfigurable AER routing architecture.