Early Access Note:
Early Access articles are new content made available in advance of the final electronic or print versions and result from IEEE’s Preprint or Rapid Post processes. Preprint articles are peer-reviewed but not fully edited. Rapid Post articles are peer-reviewed and edited but not paginated. Both these types of Early Access articles are fully citable from the moment they appear in IEEE Xplore.
This paper presents results of using a simple bit-serial architecture as a method of designing an extremely low-power and low-cost neural network processor for epilepsy seizure prediction. The proposed concept is based on a novel bit-serial data processing unit (DPU) which implements the functionality of a complete neuron and uses bit-serial arithmetic. Arrays of DPUs are controlled by simple finite state machines. We show that epilepsy detection through such dedicated neural hardware is feasible and may facilitate development of wearable, low-cost and low-energy personalized seizure prediction equipment. The proposed processor extracts epileptic seizure characteristics from electroencephalogram (EEG) waveforms. In order to facilitate the classification of EEG waveforms we develop a dedicated feature extraction hardware that provides inputs to the neural network. This approach has been tested using various network configurations and has been compared with related work. A complete system which can predict epileptic seizures with high accuracy has been implemented on an ALTERA Cyclone V FPGA using 3931 ALMs which constitutes about 7% of the Cyclone V A7 capacity. The design has a prediction accuracy of 90%.READ FULL ARTICLE ON IEEE XPLORE