Akshay Ananthakrishnan

The brain’s small-world network utilizes its short-range and long-range synaptic connections to process information in a complex and energy-efficient manner. To emulate the former, neuromorphic hardware typically leverages the conductance switching properties of thin-film dielectrics and semiconductors. Because these materials offer low ion mobilities, long-range connections built from thicker dielectrics require impractically-large forming voltages. To overcome this intrinsic shortcoming of solid-state active media, we present in this paper a simple Ag–H ₂ O–Au cell that takes advantage of the relatively high ion mobility offered by deionized water to enable programmable connectivity switches between neurons separated by large gaps (∼40 µ m). We introduce dual voltage programming schemes that allow the switch conductance to be modulated in analog and digital steps. When operating in the analog mode, the switch conductance could be potentiated and depressed over a relatively large (3.5×) range. In the digital mode, the Ag–H ₂ O–Au switch delivered a high ON/OFF current ratio of ∼600 and sustained this margin over 200 switching cycles. Additionally, both switch states could be maintained for at least 3 h without external power. We show that unlike their solid-state counterparts, the water-gap in the Ag–H ₂ O–Au cell can be easily refreshed without compromising the switching functionality. These attributes of Ag–H ₂ O–Au switches in addition to their biocompatibility and simple design make them attractive for neuromorphic wetware implementations.

Bottom-up-fabricated crossbars promise superior circuit density and 3-D integrability compared with the traditional CMOS-based implementations. However, their inherent stochasticity presents difficulties in building complex circuits from components that demand precise patterning and high registration accuracies. With fewer terminals than active devices, passive components offer higher device densities and registration tolerances, making them amenable to bottom-up synthesized nanocrossbars. Motivated by this preference for passivity, we explore, in this article, neuromorphic classifiers based on passive neurons and passive synapses. We demonstrate via SPICE simulations how a shallow network of the diode-resistor-based passive rectifier neurons and resistive voltage summers, despite its inherent inability to buffer, amplify, and negate signals, can recognize MNIST digits with 95.4% accuracy. We introduce weight-to-conductance mappings that enable negative weights to be implemented in hardware without excessive memory overheads. The influences of soft and hard defects on the classification performance are evaluated, and the methods to boost fault-tolerance are proposed. The first-order evaluation of the area, speed, and power consumption of the passive multilayer perceptron classifiers is undertaken, and the results are compared with a benchmark study in neuromorphic hardware.

Inertial harvesters utilize the motion of a translating or rotating mass to scavenge power from ambient vibrations. For translational harvesters, analytical expressions describing the influence of size on the output power of these harvesters have been derived previously by modeling vibrations as sinusoids or white noise. In this paper, we present a general method to estimate the size-dependent power outputs of both translational and rotational harvesters for vibrations with an arbitrary power spectral density. Using the heights and widths of the peaks in the vibrational power spectral density, we derive simple analytical expressions for the maximum power output of harvesters and the minimum harvester size required to achieve it. To illustrate an application of these estimates, we analyze the case of walking-induced vibrations produced at the hip-pocket and upper-arm locations. We find that rotational designs can outperform translational ones only when the harvester is larger than 26 cm and 8 cm for the hip-pocket and arm locations, respectively.

Intracranial electrodes are a vital component of implantable neurodevices, both for acute diagnostics and chronic treatment with open and closed-loop neuromodulation. Their performance is hampered by acute implantation trauma and chronic inflammation in response to implanted materials and mechanical mismatch between stiff synthetic electrodes and pulsating, natural soft host neural tissue. Flexible electronics based on thin polymer films patterned with microscale conductive features can help alleviate the mechanically induced trauma; however, this strategy alone does not mitigate inflammation at the device-tissue interface. In this study, we propose a biomimetic approach that integrates microscale extracellular matrix (ECM) coatings on microfabricated flexible subdural microelectrodes. Taking advantage of a high-throughput process employing micro-transfer molding and excimer laser micromachining, we fabricate multi-channel subdural microelectrodes primarily composed of ECM protein material and demonstrate that the electrochemical and mechanical properties match those of standard, uncoated controls. In vivo ECoG recordings in rodent brain confirm that the ECM microelectrode coatings and the protein interface do not alter signal fidelity. Astrogliotic, foreign body reaction to ECM coated devices is reduced, compared to uncoated controls, at 7 and 30 days, after subdural implantation in rat somatosensory cortex. We propose microfabricated, flexible, biomimetic electrodes as a new strategy to reduce inflammation at the device-tissue interface and improve the long-term stability of implantable subdural electrodes.

Intracortical neural microelectrodes, which can directly interface with local neural microcircuits with high spatial and temporal resolution, are critical for neuroscience research, emerging clinical applications, and brain computer interfaces (BCI). However, clinical applications of these devices remain limited mostly by their inability to mitigate inflammatory reactions and support dense neuronal survival at their interfaces. Herein we report the development of microelectrodes primarily composed of extracellular matrix (ECM) proteins, which act as a bio-compatible and an electrochemical interface between the microelectrodes and physiological solution. These ECM-microelectrodes are batch fabricated using a novel combination of micro-transfer-molding and excimer laser micromachining to exhibit final dimensions comparable to those of commercial silicon-based microelectrodes. These are further integrated with a removable insertion stent which aids in intracortical implantation. Results from electrochemical models and in vivo recordings from the rat’s cortex indicate that ECM encapsulations have no significant effect on the electrochemical impedance characteristics of ECM-microelectrodes at neurologically relevant frequencies. ECM-microelectrodes are found to support a dense layer of neuronal somata and neurites on the electrode surface with high neuronal viability and exhibited markedly diminished neuroinflammation and glial scarring in early chronic experiments in rats.

Maximum output powers of vibration-driven inertial power harvesters reported in literature exhibit sizable variations, even when normalized by the device weight or their maximum linear size. To help establish a common benchmark, we present a power-spectral-density based approach for estimating the maximum power that can be obtained using a resonant inertial power harvester from a random (aperiodic) vibration source with a given power spectral density. In the simplest case of unlimited harvester size, the maximum obtainable power is simply proportional to the maximum value of the power spectral density of vibration acceleration. We describe in detail the underlying theory and the practical method for evaluating these limits. We also present a simple analytical formula to estimate the minimum harvester size required for obtaining the maximum possible power. Specific power limits are derived as function of harvester size for three practical examples of vibration sources: (a) pneumatic power tool, (b) the body of an idling Mazda RX7 sports car, and (c) human walking motion. Characteristic power spectra and optimum design parameters (quality factor and resonant frequency) are presented for both translational and rotational harvesters. Translational harvesters generally outperform rotational ones for realistic harvester sizes, with the power tool vibrations yielding a practical power limit of ~300 mW per gram of inertial mass, followed by walking at ~1mW/g, while the vibrations of a car body yield ~0.1 mW/g or less.