A 665μW silicon photomultiplier-based NIRS/EEG/EIT monitoring asic for wearable functional brain imaging

摘要

Functional brain imaging is considered a powerful and practical solution for understanding the brain and neurological diseases. While EEG is an established method for non-invasive electrical activity, electrical-impedance tomography (EIT) and near-infrared spectroscopy (NIRS) can additionally measure impedance changes and hemodynamic processes. To facilitate long-term multi-channel brain imaging in a wearable form factor without cabling overhead, there is a need for low-power local amplifiers [1] to support all these modalities. The main principle of optical hemodynamic measurements is to send light pulses into the tissue and measure the reflected light, which is modulated by the oxygen levels in the blood (Fig. 17.8.1). State-of-the-art NIRS ICs typically consume a few mW, primarily for the LEDs to meet the required light sensitivity at the photodiodes (PDs). Silicon photomultipliers (SiPMs) are promising alternatives because they have excellent low-light detection capabilities, speed of response and higher detection efficiency in both visible and near infrared range [2]. Hence, SiPMs allow deeper brain sensing depth and the possibility to sample consistent cerebral regions with larger inter-optode distance. This benefit would significantly reduce the number of NIRS channels and the associated power for a wearable NIRS device. Although SiPMs require a higher bias voltage (~30V) than PDs, they achieve similar NIRS responses with a few hundred times less LED current. This results in a low-power NIRS ASIC and an overall power-efficient system. Existing optical sensing ICs are not suitable for a SiPM because of its large and variable output current. Trimming-based calibration methods [3] suffer from drift over time. Auto-zeroing by swapping an integrator capacitor [4][5] compensates ambient light at the cost of the integrator's headroom. Apart from ambient light, the dynamic range (DR) of the amplifier is also limited by a large NIRS signal, leading to a power-hungry readout.