This paper presents design, implementation and evaluation of an efficient embedded hardware for accurate automated detection of epileptic seizures. Three hardware configurations are proposed and evaluated in terms of accuracy of detection, utilization of hardware resources, and power consumption. The results show that a solution based on combination of the statistical function of variance (for feature extraction) and an artificial neural network (ANN) classifier allows to achieve high detection accuracy (99.18%) with moderate hardware footprint (around 44% of the FPGA resources). Furthermore, use of algorithmic and architectural optimization techniques (reduction in precision of the fixed-point number representation and reuse of hardware components) allows reducing hardware footprint by a factor of 4.4 and power consumption by a factor of 2.7 as compared with an un-optimized hardware configuration. High accuracy, real-time detection, simplicity, power efficiency and small hardware footprint make our approach a good candidate for embedded epileptic seizure detection implementation.