Resource-Constrained Acquisition Circuits for Next Generation Neural Interfaces

Abstract

The development of neural interfaces allowing the acquisition of signals from the cortex of the brain has seen an increasing amount of interest both in academic research as well as in the commercial space due to their ability to aid people with various medical conditions, such as spinal cord injuries, as well as their potential to allow more seamless interactions between people and machines. While it has already been demonstrated that neural implants can allow tetraplegic patients to control robotic arms, thus to an extent returning some motoric function, the current state of the art often involves the use of heavy table-top instruments connected by wires passing through the patient’s skull, thus making the applications impractical and chronically infeasible. Those limitations are leading to the development of the next generation of neural interfaces that will overcome those issues by being minimal in size and completely wireless, thus paving a way to the possibility of their chronic application. Their development however faces several challenges in numerous aspects of engineering due to constraints presented by their minimal size, amount of power available as well as the materials that can be utilised. The aim of this work is to explore some of those challenges and investigate novel circuit techniques that would allow the implementation of acquisition analogue front-ends under the presented constraints. This is facilitated by first giving an overview of the problematic of recording electrodes and their electrical characterisation in terms of their impedance profile and added noise that can be used to guide the design of analogue front-ends. Continuous time (CT) acquisition is then investigated as a promising signal digitisation technique alternative to more conventional methods in terms of its suitability. This is complemented by a description of practical implementations of a CT analogue-to-digital converter (ADC) including a novel technique of clockless stochastic chopping aimed at the suppression of flicker noise that commonly affects the acquisition of low-frequency signals. A compact design is presented, implementing a 450 nW, 5.5 bit ENOB CT ADC, occupying an area of 0.0288 mm2 in a 0.18 μm CMOS technology, making this the smallest presented design in literature to the best of our knowledge. As completely wireless neural implants rely on power delivered through wireless links, their supply voltage is often subject to large high frequency variations as well voltage uncertainty making it necessary to design reference circuits and voltage regulators providing stable reference voltage and supply in the constrained space afforded to them. This results in numerous challenges that are explored and a design of a practical implementation of a reference circuit and voltage regulator is presented. Two designs in a 0.35 μm CMOS technology are presented, showing respectively a measured PSRR of ≈60 dB and ≈53 dB at DC and a worst-case PSRR of ≈42 dB and ≈33 dB with a less than 1% standard deviation in the output reference voltage of 1.2 V while consuming a power of ≈7 μW. Finally, ΣΔ modulators are investigated for their suitability in neural signal acquisition chains, their properties explained and a practical implementation of a ΣΔ DC-coupled neural acquisition circuit presented. This implements a 10-kHz, 40 dB SNDR ΣΔ analogue front-end implemented in a 0.18 μm CMOS technology occupying a compact area of 0.044 μm2 per channel while consuming 31.1 μW per channel.