Energy-Efficient Adaptive Neural Stimulator with Waveform Prediction by Sub-Threshold Interrogation of the Electrode-Tissue Interface

This paper presents an implantable low-power neural stimulator that generates electrical stimulation pulses based on subject-specific edge-learning of electrode-tissue voltage profiles. The system deploys a low-magnitude constant-current stimulation pulse to create a training dataset, which is subsequently utilized to predict the desired electrode voltage waveforms for higher magnitudes of constant-current stimulation. The predicted waveform dataset has been used to control a custom switched-capacitor output stage, thereby avoiding V_{driver_transistor} · I_stimulation power loss as in the conventional neural stimulator drivers. The proposed system incorporates on-chip learning and prediction implemented within an ultra-low-power microcontroller, which has been optimized for memory- and power-constrained implantable environments. The stimulator output stage reduces power loss by up to 20% as compared to dynamic power supply scaling method, and consumes up to 3.63× lower as compared to conventional constant-current output stages. The intelligent neural interface system has been powered by a wireless inductive energy transfer link and is remotely controlled through a WiFi-based internet network. A custom-developed application interface, compatible with both mobile devices and personal computers, facilitates secure remote adjustments of stimulation parameters. The proposed system has been validated through a combination of in vivo rat peripheral nerve stimulation, in vitro saline tests, and benchtop experiments. These results collectively demonstrate the potential to advance future neural implant technologies by enabling intelligence, safety, energy efficiency, and remotely controllable neural organ modulation.