Epilepsy is a common chronic neurological disorder that causes recurring seizures and affects more than 50 million people worldwide. Implantable medical devices (IMDs) are regarded as effective tools to cure patients who suffer from refractory epilepsy. Seizures of patients with intractable epilepsy can be controlled by neither currently available anti-seizure drugs nor resection surgery. In these cases, epilepsy control implants can reduce the frequency and severity of seizures. Furthermore, patients can use IMDs in conjunction with a reduced dosage of medication to prevent severe side effects. A closed-loop epilepsy control implant applies electrical stimulation to specific brain tissues according to signal processing and seizure detection tasks conducted by an automatic seizure detector. Hardware implementation of signal processing algorithms is a challenging task to fulfill the stringent constraints of the implants in terms of power consumption, area occupation, hardware complexity and real-time operation. System-level analysis of closed-loop seizure control implants is performed using the models of different components in Simulink software. The main properties of different components of the implant are modeled to simulate the operation of the implant. In addition, the temperature elevation of the implant, which highly impacts the safety and performance of the system in the long run, was investigated. The temperature elevation is controlled by the adjustment of stimulation parameters in order to maintain the temperature rise of the implant lower than 1 degree Celsius. A novel two-stage feature extraction method is developed in this thesis to offer accurate energy-efficient biomedical signal processing in the implant. The features available in the feature pool are divided into two groups, and extracted in the monitoring and detection stages. Extraction of the majority of the features is disabled during long inter-ictal periods which enables low-power consumption of the implant. Intracranial electroencephalography (iEEG) signals are used to monitor and detect seizure events. iEEG signals are recorded by microelectrode arrays that consist of tens of electrodes. Signal processing applied on a large dimension of EEG signals is a crucial impediment for hardware-friendly and low-power seizure detection. A framework that contains a mutual electrode channel and feature dimension reduction technique is proposed to significantly reduce the computation complexity of signal processing on hardware. Moreover, seizure detection based on machine-learning techniques is studied in this thesis. Various types of machine learning techniques are reported in the literature while their hardware compatibility is not taken into account. Thus, the architecture of a RF classifier is optimized in terms of classification accuracy and hardware complexity. The RF classifier with optimized hyperparameters is implemented on an FPGA to perform the seizure detection task. In addition, the novel concept of programmable seizure detection is introduced for the first time in the literature. Programmable seizure detection enables a continuous interaction between users and implants. As a consequence, neurologists can adjust device's properties and the patients can play an important role in improving the effectiveness of their treatment. A 32-bit RISC processor compatible with implants is designed and implemented on an FPGA to realize the programmability feature.