Bidirectional Neural Interface Circuits with On-Chip Stimulation Artifact Reduction Schemes

Abstract

Bidirectional neural interfaces are tools designed to “communicate” with the brain via recording and modulation of neuronal activity. The bidirectional interface systems have been adopted for many applications. Neuroscientists employ them to map neuronal circuits through precise stimulation and recording. Medical doctors deploy them as adaptable medical devices which control therapeutic stimulation parameters based on monitoring real-time neural activity. Brain-machine-interface (BMI) researchers use neural interfaces to bypass the nervous system and directly control neuroprosthetics or brain-computer-interface (BCI) spellers. In bidirectional interfaces, the implantable transducers as well as the corresponding electronic circuits and systems face several challenges. A high channel count, low power consumption, and reduced system size are desirable for potential chronic deployment and wider applicability. Moreover, a neural interface designed for robust closed-loop operation requires the mitigation of stimulation artifacts which corrupt the recorded signals. This dissertation introduces several techniques targeting low power consumption, small size, and reduction of stimulation artifacts. These techniques are implemented for extracellular electrophysiological recording and two stimulation modalities: direct current stimulation for closed-loop control of seizure detection/quench and optical stimulation for optogenetic studies. While the two modalities differ in their mechanisms, hardware implementation, and applications, they share many crucial system-level challenges. The first method aims at solving the critical issue of stimulation artifacts saturating the preamplifier in the recording front-end. To prevent saturation, a novel mixed-signal stimulation artifact cancellation circuit is devised to subtract the artifact before amplification and maintain the standard input range of a power-hungry preamplifier. Additional novel techniques have been also implemented to lower the noise and power consumption. A common average referencing (CAR) front-end circuit eliminates the cross-channel common mode noise by averaging and subtracting it in analog domain. A range-adapting SAR ADC saves additional power by eliminating unnecessary conversion cycles when the input signal is small. Measurements of an integrated circuit (IC) prototype demonstrate the attenuation of stimulation artifacts by up to 42 dB and cross-channel noise suppression by up to 39.8 dB. The power consumption per channel is maintained at 330 nW, while the area per channel is only 0.17 mm2. The second system implements a compact headstage for closed-loop optogenetic stimulation and electrophysiological recording. This design targets a miniaturized form factor, high channel count, and high-precision stimulation control suitable for rodent in-vivo optogenetic studies. Monolithically integrated optoelectrodes (which include 12 µLEDs for optical stimulation and 12 electrical recording sites) are combined with an off-the-shelf recording IC and a custom-designed high-precision LED driver. 32 recording and 12 stimulation channels can be individually accessed and controlled on a small headstage with dimensions of 2.16 x 2.38 x 0.35 cm and mass of 1.9 g. A third system prototype improves the optogenetic headstage prototype by furthering system integration and improving power efficiency facilitating wireless operation. The custom application-specific integrated circuit (ASIC) combines recording and stimulation channels with a power management unit, allowing the system to be powered by an ultra-light Li-ion battery. Additionally, the µLED drivers include a high-resolution arbitrary waveform generation mode for shaping of µLED current pulses to preemptively reduce artifacts. A prototype IC occupies 7.66 mm2, consumes 3.04 mW under typical operating conditions, and the optical pulse shaping scheme can attenuate stimulation artifacts by up to 3x with a Gaussian-rise pulse rise time under 1 ms.