An Efficient NoC-based ANN Framework for Epileptic Seizure Recognition

Artificial Neural Networks (ANN) which mimic the biological neural networks can process various cognitive functions of the brain. With the advancement in the intelligent systems, the processing power and efficiency of the systems has been challenged due to the increasing size and complexity of data. Hence it is necessary to design an ANN accelerator that would solve the complex computations efficiently. Since, most of the systems are based on conventional bus-based architectures, processing such complex datasets reduces the efficiency and scalability of the system. On the other hand, Network-on-Chip (NoC) architecture has proven to be efficient for computations in heterogeneous multi-core systems. In this paper, we propose a NoC based neural network accelerator for neuromorphic computing. We consider an ANN for real-time identification of epileptic seizure based on continuously monitored EEG signal of the patients. The trained ANN is mapped onto the Network-on-Chip architecture for efficient and real-time computation. The use of NoC benefits the proposed architecture by increasing the throughput, parallelism, power efficiency and scalability of the system. In the proposed model, the neurons are hierarchically interconnected. The proposed architecture is explored in terms of throughput, energy, latency and identification accuracy of an epileptic seizure.