Abstract
Much research is being done on implantable devices, such as cochlear implants, retinal prostheses, motor prostheses, etc. Acquiring information regarding the stimulated neurons and recording the neural activities in these neural prosthetic devices is essential. It is also crucial to monitor and process the neural action potential signals in real time for closed-loop controlled deep brain stimulation (for example, in epilepsy and Parkinson’s disease). The front end of the implantable device is an array of stimulating/recording electrodes. These electrodes read extracellular neural signals (ENG), which are very small (10μV-200μV) and have a low frequency (0.1Hz-10kHz), requiring a low-noise amplifier (LNA) for signal amplification to acquire the neural signal. Unfortunately, MOSFET process has inherent 1/f noise that dominates at low frequencies, while ENG is in the same frequency band as 1/f noise, resulting in a very poor SNR. When reading the neural signals in an implantable device, the DC offset of the tissue-electrode interface can easily saturate the amplifier. To solve this problem, feedback capacitors and resistors are connected to the preamp to form a high pass function to filter the DC offset. The corner frequency of this HPF has to be lower than 0.1Hz, which requires a very huge capacitor/resister to implement. A MOSFET resistor is used to provide an area-efficient means of creating a large resistance. Behind the HPF, a fully differential OTA is used to implement the preamp. EKV model is used for analysis because the transistors are operated in different inversion levels. The input transistors of the preamp are operated in modereate inversion, while non-input transistors in higher inversion level and long channel, so that the input stage will dominate the thermal noise, 1/f noise and offset contribution. Transistors operated in deep trode region are used as source-degenerated resistors to lower down the trans-conductance of PMOS operated in strong inversion, while the PSD of 1/f noise get higher with inversion level of PMOS in the process we used. Behind the preamp, a gm-C first order filter removes the high-frequency noise carried in the unwanted band. Measurement results shows the input-referred noise of the system is 5.62μVrms from 0.1Hz to 10 kHz, power consumption of preamp is 14.2μW, NEF is 7.45, the gain is 49.5dB, CMRR is 70dB and PSRR is 57dB. The amplifier was fabricated using a TSMC 0.18μm 1P6M CMOS process.