Despite tremendous progress over the years, current brain-machine interface (BMI) systems are relatively bulky, highly invasive, and limited in their effectiveness except for highly constrained tasks such as moving a cursor on a computer screen. To improve performance of current BMI systems, it is necessary to dramatically increase spatial resolution and coverage across the brain without constraining the mobility of the subject. This calls for innovative approaches to high-density integrated neural recording and stimulation using non-invasive or minimally invasive microelectrode and custom silicon integrated circuits at extreme energy and area efficiency.

In this thesis, I present energy-efficient fully integrated miniaturized implants for electrocortical recording and stimulation, and unobtrusive body-area networks systems for subcutaneous power delivery and data communication, as fundamental building blocks to next generation BMI. First I describe a fully wireless, encapsulated neural interface and acquisition chip (ENIAC) in 180nm silicon-on-insulator (SOI) complementary metal-oxide-semiconductor (CMOS) technology for 16-channel neural recording and stimulation including integrated 4x4 electrode array, coil antenna, and wireless power transfer and data telemetry without any external components, completely contained in less than 3mm3 volume suitable for minimally invasive surgical insertion on the cortical surface. A novel fully integrated wireless power receiver design with an RF-decoupled H-tree signal distribution network delivers 1mW power over 1 cm distance while mitigating RF interference in the sensitive analog front-end and acquisition circuits for recording of electrocorticography (ECoG) signals transmitted through the skull. Second I highlight a 1mm2 16-channel neural recording and acquisition system-on-chip in 65nm CMOS offering 92 dB input dynamic range and <1μVrms input-referred noise covering DC-500Hz signal bandwidth at 0.8μW power consumption per channel, offering a record 2.6 power efficiency factor (PEF). Validation with in vivo recordings in frontal cortex of marmoset primates reveals infra-slow (<0.1 Hz) local-field potentials (LFP) indicative of subject arousal during a visual attention task alternating with periods of rest and feeding. I will further present advances in cognitive radio baseband transceivers for ultra-low power RF wireless non-line-of-sight body-area network communication with spectral tuning and spatial selectivity across 48MHz signal bandwidth at 1.2mW power. These technology advances combine to support further developments towards modular wireless, minimally invasive, whole-brain electrocortical recording at near-cellular resolution.

The importance of health sensing continues to grow in light of an aging global population, the prevalence of chronic illness, a dramatic rise in mental health disorders, and a healthcare system with gaps in equal coverage for everyone. Wearable sensing technologies have greatly expanded health monitoring and human-computer interface applications, bridging the gaps between traditional clinical instrumentation and urgent need for remote, daily healthcare. Noninvasive electrophysiology tools in particular seek to measure signals from the brain and body at conventional sensing locations such as the scalp, but require bulky instrumentation and cumbersome setup procedures, limiting prolonged use in real-world settings. As an appealing alternative, electrophysiological sensing within the ear provides continuous physiological and cognitive state monitoring in an unobtrusive form factor. The proximity of the ear to the central nervous system and the high density of exocrine glands in the ear canal motivate the integration of multiple sensing modalities including electroencephalography (EEG), electrodermal activity (EDA), and metabolite detection into a single in-ear device. The work in this dissertation extends the capabilities and amplifies the impact of in-ear sensing systems addressing several of their limitations including electrode performance, manufacturing scalability, and low-power acquisition hardware. A primary aim of this work is to enhance in-ear sensors to support higher electrode count, offering reliable, long-term wearability, user comfort, and straightforward fabrication methods to benefit investigation of physiological signals in a broader subject pool. Another primary aim of this work is to provide high-performance data acquisition hardware to meet the needs of mobile health monitoring from multimodal sensor arrays. First, new methods for the development of in-ear sensors are presented which deliver user-generic dry-contact electrodes made from standard printed circuit boards (PCBs). Electrodes are characterized using electrochemical impedance spectroscopy (EIS) and a series of in-vivo electrophysiological studies. The second part presents a complete and versatile wireless electrophysiology data acquisition system (weDAQ) that is demonstrated for in-ear EEG and on-body electrophysiology as an open-source resource for the biomedical community. Subsequently, a neural interface system-on-chip (NISoC) integrated circuit (IC) is presented which substantially lowers the power and improves signal integrity through a new correlated double sampling (CDS) technique. Finally, the sensing system is evaluated for use with subjects for auditory steady-state response (ASSR), electrodermal activity (EDA), and attention state classification (vigilance), demonstrating its utility as a versatile, mobile health monitoring platform.

Vision and movement are instrumental components of normal functioning in everyday life. Due to this importance, it is valuable to understand the neurological mechanisms which underlie these processes and how they are employed. Correspondingly, much research is done on deficits in these processes and how people are impacted by such deficits. This dissertation focuses on the development of techniques to better be able to perform investigations into vision, movement, and the interaction between them. In order to study vision it is often important to know where the eyes are looking and how they are moving; this is often done using a method of eye tracking, typically using a camera. As part of this work, eye tracking based upon electroocculography (EOG) was developed as a potentially inexpensive, high temporal resolution eye tracker system, with capabilities of tracking gaze fixation in 3D space. An experimental calibration setup and procedure were created to calibrate and test the capacities of the EOG based eye tracker, including comparison to a standard, camera based eye tracker. Vision is also involved in movement through space. Navigation within an environment requires the utilization of cues, typically visual ones, to determine location, plan a path, and subsequently execute it. In this project a virtual reality system and experimental platform, the Virtual Environment Human Navigation Task (VE-HuNT), was created to assess the navigation skills and strategies utilized within conditions applicable to everyday life. VE-HuNT was employed to investigate the effects of visual impairment due to glaucoma on navigation, and the aspects of the environment that would improve functioning in these individuals. Behavioral metrics for wayfinding ability were assessed and the impact of glaucoma on the ability of subjects to utilize environmental cues for wayfinding was determined. Together the outcomes of these two complementary projects further the ability to perform human cognitive research under conditions representative of typical human activity. As an important tangible benefit of these advances, the improved capabilities for assessment of clinical conditions which impact such activities open up new avenues for neurological understanding of vision and movement.

Quantification of dynamic causal interactions among brain regions constitutes an important component of conducting research and developing applications in experimental and translational neuroscience. Furthermore, cortical networks with dynamic causal connectivity in brain-computer interface (BCI) applications offer a more comprehensive view of brain states implicated in behavior than do individual brain regions. However, models of cortical network dynamics are difficult to generalize across subjects because current EEG signal analysis techniques are limited in their ability to reliably localize sources across subjects. We propose an algorithmic and computational framework for identifying cortical networks across subjects in which dynamic causal connectivity is modeled among user-selected cortical regions of interest (ROIs). We demonstrate the strength of the proposed framework using a “reach/saccade to spatial target” cognitive task performed by ten right-handed individuals. Modeling of causal cortical interactions was accomplished through measurement of cortical activity using electroencephalography (EEG), application of independent component clustering to identify cortical ROIs as network nodes, estimation of cortical current density using cortically constrained low resolution electromagnetic brain tomography (cLORETA), multivariate autoregressive (MVAR) modeling of representative cortical activity signals from each ROI, and quantification of the dynamic causal interaction among the identified ROIs using the Short-time direct Directed Transfer function (SdDTF). The resulting cortical network and the computed causal dynamics among its nodes exhibited physiologically plausible behavior, consistent with past results reported in the literature. This physiological plausibility of the results strengthens the framework’s applicability in reliably capturing complex brain functionality, which is required by applications such as diagnostics and BCI.

Machine learning and related statistical signal processing are expected to endow sensor networks with adaptive machine intelligence and greatly facilitate the Internet of Things (IoT). As such, architectures embedding adaptive and learning algorithms on-chip are oft-ignored by system architects and design engineers, and present a new set of design trade-offs. We focus on topologies efficiently implementing mixed-signal matrix-vector multiplication for applications in spatial filtering for IoT, where substantial processing gain in the analog domain alleviates the need for highly accurate and energy-consuming analog-to-digital conversion. We present a micropower, high-dynamic-range multichannel multiple-input multiple-out (MIMO) mixed-signal linear transform system, with analog signal path and digital coefficient control, composed of an array of 14-bit Nested Thermometer Multiplying DACs (NTMDACs) implementing analog multiplication, and variable gain amplifier (VGA) implementing accumulation. Implemented in 65nm CMOS, the NTMDAC MISO system-on-chip measures 84 dB in interference suppression at 2 pJ of energy per mixed-signal multiply-accumulate. We demonstrate state-of-the art performance on two tasks, spectrally oblivious interference suppression in communication signals and EEG signal separation. We then provide experimental demonstration of the use of a MIMO mixed-signal linear-transform system within a radio-frequency receiver chain. Over-the-air experiments demonstrating signal separation for two broad-band modulated signals further validate the adaptive beamforming capabilities under severe multipath conditions even in the absence of line-of-sight communication path.

In order to mitigate adverse effects of radix errors and capacitive mismatch encountered in compact low-power realizations of high-resolution, high-dimensional MIMO analog processing systems, we introduce Stochastic Successive Approximation, or S2A, as an on-line adaptive optimization algorithm amenable to efficient implementation in massively parallel analog hardware. S2A offers a direct alternative to stochastic gradient descent overcoming several of its shortcomings, such as its sensitivity to analog mismatch model errors, while improving on the rate of convergence for high-dimensional analog computation. The S2A algorithm enables convergence to values closer to the optimal when facing non-convex optimization landscapes induced by mismatch in capacitive multiplying digital-to-analog converter components when applied to adaptive analog signal processing. We experimentally demonstrate, in fewer than 25 iteratations of S2A, 65 dB of processing gain in adaptive beamforming, over-the-air, multipath interferer suppression.

Neuromorphic engineering pursues the design of electronic systems emulating function and structural organization of biological neural systems in silicon integrated circuits that embody similar physical principles.

The work in this dissertation extends neuromorphic engineering to neural interfaces that directly couple biological neurons to their equivalents in silicon integrated circuits, dynamically probing their function through silicon emulation of biophysical chemical and electrical synapses. Our aim in this work is to enable study of hybrid networks of biological and silicon neurons with highly configurable topology and biophysically based properties, providing windows on the inner workings of biological neural circuits from the cellular to the network levels, and hence promoting new synergies between theory in computational neuroscience and experimentation in systems neuroscience.

In the first part, membrane dynamics and ion channel kinetics of biological neurons, obtained from experimental electrophysiological data, were accurately mapped onto equivalent continuous-time analog dynamics in NeuroDyn, a highly reconfigurable neuromorphic silicon microchip. To this end, songbird individual neuron dynamics from intracellular neural recordings were extracted, modeled, and then mapped onto silicon neurons in NeuroDyn by data assimilation to estimate and configure biophysical parameters.

Further, the NeuroDyn framework was extended to serve as a versatile tool for biophysical dynamic clamp electrophysiology, connecting biological and silicon neurons through synthetic virtual chemical synapses. To this end, the response properties of five different types of chemical synapses, including both excitatory (AMPA, NMDA) and inhibitory (GABAA, GABAC, Glycine) ionotropic receptors were reproduced with neuromorphic integrated circuits. In addition, electrical synapses (gap junctions) were emulated in a network of four silicon neurons.

The second part entails the design, implementation and functional validation of high-density multi-channel neural interfaces, establishing bidirectional electrical communication between silicon artificial neurons and biological neurons at very large scale. Our work produced a neural interface system-on-chip (NISoC) with 1,024-channels of simultaneous electrical recording and stimulation at record noise-energy efficiency, with sub-µW power consumption per channel at 6 µVrms input referred voltage noise over 12.5 kHz signal bandwidth. Integrating an array of 32 × 32 electrodes on a 2mm × 2mm chip in 65nm CMOS, the NISoC supports both voltage and current clamping through a programmable interface, ranging 100~dB in voltage, and 120~dB in current, for high-resolution high-throughput electrophysiology.

Further, we demonstrated extended functionality for scalable multichannel in vitro intracellular electrophysiology in a second 256-channel hybridized NiSoC with sharp-tipped Pt nanowire electrodes deposited on the silicon top-metal surface, recording action potentials from rat cortical neurons cultured directly on top of the chip.

These advances combine to enable bidirectional communication between artificial neurons and biological neurons in vitro, with precise probing of neural function and flexible control over synaptic interactions ranging from intracellular dynamics of individual cells to network dynamics comprising potentially thousands of neurons. In addition to applications in closed-loop electrophysiology, in vitro neuromorphic neural interface can be used as testbed for prototyping the next generation of neuroprosthetics.

Recent interest and initiatives in brain research have driven a worldwide effort towards developing implantable neural interface systems with high spatiotemporal resolution and spatial coverage extending to the whole brain. Electrocorticography (ECoG) promises a minimally invasive, chronically implantable neural interface with resolution and spatial coverage capabilities that, when appropriately scaled, meet the needs of recently proposed brain initiatives. Current ECoG technologies, however, typically rely on cm-sized electrodes and wired operation, severely limiting their resolution and long-term use.

The work presented here has advanced micro-electrocorticography (uECoG) technologies for wireless high-density cortical neural interfaces in two main directions: flexible active uECoG arrays; and modular fully integrated uECoG systems. This dissertation presents a systematic design methodology which addresses unique design challenges posed by the extreme densities, form factors and power budgets of these fully implantable neural interface systems, with experimental validation of their performance for neural signal acquisition, stimulation, and wireless powering and data communication. Notable innovations include 1) first demonstration of simultaneous wireless power and data telemetry at 6.78 Mbps data rate over a single 13.56 MHz inductive link; 2) integrated recording from a flexible active electrode ECoG array with 85 dB dynamic range at 7.7 nJ energy per 16-b sample; and 3) the first fully integrated and encapsulated wireless neural-interface-on-chip microsystem for non-contact neural sensing and energy-replenishing adiabatic stimulation delivering 145 uA current at 6 V compliance within 2.25 mm3 volume.

In addition, the work presented here on advancing the resolution and coverage of neural interfaces extends further from the cortex to the retina. Despite considerable advances in retinal prostheses over the last two decades, the resolution of restored vision has remained severely limited, well below the 20/200 acuity threshold of blindness. Towards drastic improvements in spatial resolution, this dissertation presents a scalable architecture for retinal prostheses in which each stimulation electrode is directly activated by incident light and powered by a common voltage pulse transferred over a single wireless inductive link. The hybrid optical addressability and electronic powering scheme provides for separate spatial and temporal control over stimulation, and further provides optoelectronic gain for substantially lower light intensity thresholds than other optically addressed retinal prostheses using passive microphotodiode arrays. The architecture permits the use of high-density electrode arrays with ultra-high photosensitive silicon nanowires, obviating the need for excessive wiring and high-throughput data telemetry. Instead, the single inductive link drives the entire array of electrodes through two wires and

provides external control over waveform parameters for the common voltage stimulation. A complete system comprising inductive telemetry link, stimulation pulse

demodulator, charge-balancing series capacitor, and nanowire-based electrode device is integrated and validated ex vivo on rat retina tissue. Measurements demonstrate control over retinal neural activity both by light and electrical bias, validating the feasibility of the proposed architecture and its system components as an important first step towards a high-resolution optically addressed retinal prosthesis.

Electroencephalography (EEG) has made much progress since the 1920s, when Dr. Hans Berger began his exploration of what he initially thought of as supernatural psychic energy. Now almost a century later, most laypeople are familiar with the concept of EEG; some may even casually say “we’re on the same wavelength” when in agreement with each other. In the scientific community, rigorous data analysis has protected EEG from dismissal as a pseudoscience. Furthermore, the successful application of independent component analysis (ICA) to EEG data and the subsequent release of EEGLAB, a GUI-based Matlab toolbox for processing EEG data, allows for and facilitates source-level analysis. Today, the lowering costs of data acquisition and storage continue to drive demand for tools and methods to process large datasets. Researchers may default to channel-level analysis, but some types of analysis (e.g. brain connectivity) are more sensible at the source-level. Each chapter in this thesis documents a method of analysis for an EEGLAB STUDY (a grouping of EEG datasets, source-resolved by ICA, for comparative analysis). Chapter 2 shows the EEGLAB plug-in std_envotpo, which creates STUDY-level envelope plots of the topography, and its sub-function statPvaf, which allows for non-parametric statistical analysis of the percent variance accounted for. Chapter 3 explores parameter selection in a new data processing pipeline, which utilizes the Source Information Flow Toolbox (SIFT) and Measure Projection Toolbox (MPT) to calculate brain connectivity estimators at the group level.

Advances in neural engineering are enabling targeted neural recording and stimulation towards high-resolution brain computer interfaces. High-throughput bidirectional communication to the brain is paramount to reach translational impact with neural prostheses. Useful vision restoration through microelectronic retinal prosthesis implants continues to be a difficult challenge despite commendable attempts. Achieving practical control of thousands and millions of electrode channels calls for architectural advances to improve the scalability of biopotential recording, power, wireless communications, and biocompatible interconnect. This dissertation presents an interdisciplinary approach towards high-resolution retinal prostheses that overcomes these challenges. The requirements and interactions between biopotential recording circuits and electrodes are formulated in the context of neural recording. Next, a resonant inductive power transfer link for retinal prostheses is designed and validated. The effect of ocular movements on the efficiency of power transfer is demonstrated in a constructed phantom frame. Subsequently, a novel integrated circuit is designed and fabricated to provide highly energy-efficient power delivery and waveform control to drive a nanowire-based microphotodiode subretinal electrode array. Energy efficiency and scalability of the power and communication link is accomplished through external control over global stimulation waveform parameters. Efficiency improvements derive from two main approaches: adiabatic unregulated power delivery to the stimulator, and duty cycling of the external transmitter. Global charge metering and calibration minimize the required data transmission and system operating frequency for bidirectional communication and charge-balanced stimulation. The dissertation concludes with in vivo validation of the nanowire microelectrode array transducer in rabbits using the principle of synchronous detection of cerebral cortex visual evoked biopotential signals. The various components of this dissertation present a full-stack development of a system to remediate blindness and advance the field of neural interfaces.

Neuromorphic computing aims to emulate the highly adaptive and efficient processing of the biological brain in silicon. Field-Programmable Gate Arrays offer a way to create general-purpose reconfigurable neuromorphic systems that when coupled with High-Bandwidth Memory are able to offer massive parallelism and throughput. Using these advances, we aim to build a general-purpose neuromorphic computing platform be broadly available to the general research community. This real-time system is reconfigurable, operates in real-time, and implements highly flexible neural dynamics with up to 160 million neurons and 40 Billion synapses. The first part of the thesis discusses the overall system algorithm and investigates the organization of creating an interface between a potential user and the entire reconfigurable platform. We discuss the creation of an intermediate representation for our reconfigurable system. We show the algorithms used in order to map from an intermediate representation of the network into either the High-Bandwith Memory or the python simulator. Finally, we discuss optimization algorithms that can be used to improve both memory savings and execution speed on the network. The second part of the thesis focuses on an efficient and scalable method for large-scale network partitioning on this neuromorphic system across different layers of hierarchy. The partitioning method results in a balanced assignment of neurons to each core in the network while providing flexibility to any input network structure. We demonstrate results on various input networks and compare different system router configurations in order to show the improved performance of the partitioning algorithm.