Design and Fabrication of CMOS Microstructures to Locally Synthesize Carbon Nanotubes for Gas Sensing.

Carbon nanotubes (CNTs) can be grown locally on custom-designed CMOS microstructures to use them as a sensing material for manufacturing low-cost gas sensors, where CMOS readout circuits are directly integrated. Such a local CNT synthesis process using thermal chemical vapor deposition (CVD) requires temperatures near 900 °C, which is destructive for CMOS circuits. Therefore, it is necessary to ensure a high thermal gradient around the CNT growth structures to maintain CMOS-compatible temperature (below 300 °C) on the bulk part of the chip, where readout circuits are placed. This paper presents several promising designs of CNT growth microstructures and their thermomechanical analyses (by ANSYS Multiphysics software) to check the feasibility of local CNT synthesis in CMOS. Standard CMOS processes have several conductive interconnecting metal and polysilicon layers, both being suitable to serve as microheaters for local resistive heating to achieve the CNT growth temperature. Most of these microheaters need to be partially or fully suspended to produce the required thermal isolation for CMOS compatibility. Necessary CMOS post-processing steps to realize CNT growth structures are discussed. Layout designs of the microstructures, along with some of the microstructures fabricated in a standard AMS 350 nm CMOS process, are also presented in this paper.

Combining biophysical modeling and deep learning for multielectrode array neuron localization and classification.

Neural circuits typically consist of many different types of neurons, and one faces a challenge in disentangling their individual contributions in measured neural activity. Classification of cells into inhibitory and excitatory neurons and localization of neurons on the basis of extracellular recordings are frequently employed procedures. Current approaches, however, need a lot of human intervention, which makes them slow, biased, and unreliable. In light of recent advances in deep learning techniques and exploiting the availability of neuron models with quasi-realistic three-dimensional morphology and physiological properties, we present a framework for automatized and objective classification and localization of cells based on the spatiotemporal profiles of the extracellular action potentials recorded by multielectrode arrays. We train convolutional neural networks on simulated signals from a large set of cell models and show that our framework can predict the position of neurons with high accuracy, more precisely than current state-of-the-art methods. Our method is also able to classify whether a neuron is excitatory or inhibitory with very high accuracy, substantially improving on commonly used clustering techniques. Furthermore, our new method seems to have the potential to separate certain subtypes of excitatory and inhibitory neurons. The possibility of automatically localizing and classifying all neurons recorded with large high-density extracellular electrodes contributes to a more accurate and more reliable mapping of neural circuits. NEW & NOTEWORTHY We propose a novel approach to localize and classify neurons from their extracellularly recorded action potentials with a combination of biophysically detailed neuron models and deep learning techniques. Applied to simulated data, this new combination of forward modeling and machine learning yields higher performance compared with state-of-the-art localization and classification methods.

A CMOS Multi-Electrode Array for Four-Electrode Bioimpedance Measurements.

This work demonstrates how a multi-electrode array (MEA) dedicated to four-electrode bioimpedance measurements can be implemented on a complementary metal-oxide-semiconductor (CMOS) chip. As a proof of concept, an 8 × 8 pixel array along with dedicated amplifiers was designed and fabricated in the TSMC 180 nm process. Each pixel in the array contains a circular current carrying (CC) electrode that can act as a current source or sink. In order to measure a differential voltage between the pixels, each CC electrode is surrounded by a ring shaped pick up (PU) electrode. The differential voltages can be measured by an on-board instrumentation amplifier, while the currents can be measured with an on-bard transimpedance amplifier. Openings in the passivation layer exposed the aluminum top metal layer, and a metal stack of zinc, nickel and gold was deposited in an electroless plating process. The chips were then wire bonded to a ceramic package and prepared for wet experiments by encapsulating the bonding wires and pads in the photoresist SU-8. Measurements in liquids with different conductivities were performed to demonstrate the functionality of the chip.

Independent Component Analysis for Fully Automated Multi-Electrode Array Spike Sorting.

In neural electrophysiology, spike sorting allows to separate different neurons from extracellularly measured recordings. It is an essential processing step in order to understand neural activity and it is an unsupervised problem in nature, since no ground truth information is available. There are several available spike sorting packages, but many of them require a manual intervention to curate the results, which makes the process time consuming and hard to reproduce. Here, we focus on high-density Multi-Electrode Array (MEA) recordings and we present a fully automated pipeline based on Independent Component Analysis (ICA). While ICA has been previously investigated for spike sorting, it has never been compared with fully automated state-of-the-art algorithms. We use realistic simulated datasets to compare the spike sorting performance in terms of complexity, signal-to-noise ratio, and recording duration. We show that an ICA-based fully automated spike sorting approach can be a viable alternative approach due to its precision and robustness, but it needs to be optimized for time constraints and requires sufficient density of electrodes to cover active neurons in the proximity of the MEA.

A Deep Learning Approach for the Classification of Neuronal Cell Types.

Classification of neurons from extracellular recordings is mainly limited to putatively excitatory or inhibitory units based on the spike shape and firing patterns. Narrow waveforms are considered to be fast spiking inhibitory neurons and broad waveforms excitatory neurons. The aim of this work is twofold. First, we intend to use the rich spatial information from high-density Multi-Electrode Arrays (MEAs) to make classification more robust; second, we hope to be able to classify sub-types of excitatory and inhibitory neurons. We first built, in simulation, a large dataset of action potentials from detailed neural models. Then, we extracted spike features from the simulated recordings on a high-density Multi-Electrode Array model. Finally, we used a Convolutional Neural Networks (CNN), to classify the different cell types. Compared with the ground truth data from the simulated dataset, the results show that this forward modelling/machine learning approach is very robust in recognizing excitatory and inhibitory spikes (accuracy $\ge 92.15$%). Additionally, the approach can, to a certain extent, correctly classify different cell sub-types. As the detail and fidelity of neural models increase and high-density recordings become available, this approach could become a viable and robust alternative for classification of neural cell types from in-vivo extracellular recordings.

Open source modules for tracking animal behavior and closed-loop stimulation based on Open Ephys and Bonsai.

OBJECTIVE: A major goal in systems neuroscience is to determine the causal relationship between neural activity and behavior. To this end, methods that combine monitoring neural activity, behavioral tracking, and targeted manipulation of neurons in closed-loop are powerful tools. However, commercial systems that allow these types of experiments are usually expensive and rely on non-standardized data formats and proprietary software which may hinder user-modifications for specific needs. In order to promote reproducibility and data-sharing in science, transparent software and standardized data formats are an advantage. Here, we present an open source, low-cost, adaptable, and easy to set-up system for combined behavioral tracking, electrophysiology, and closed-loop stimulation. APPROACH: Based on the Open Ephys system (www.open-ephys.org) we developed multiple modules to include real-time tracking and behavior-based closed-loop stimulation. We describe the equipment and provide a step-by-step guide to set up the system. Combining the open source software Bonsai (bonsai-rx.org) for analyzing camera images in real time with the newly developed modules in Open Ephys, we acquire position information, visualize tracking, and perform tracking-based closed-loop stimulation experiments. To analyze the acquired data we provide an open source file reading package in Python. MAIN RESULTS: The system robustly visualizes real-time tracking and reliably recovers tracking information recorded from a range of sampling frequencies (30-1000 Hz). We combined electrophysiology with the newly-developed tracking modules in Open Ephys to record place cell and grid cell activity in the hippocampus and in the medial entorhinal cortex, respectively. Moreover, we present a case in which we used the system for closed-loop optogenetic stimulation of entorhinal grid cells. SIGNIFICANCE: Expanding the Open Ephys system to include animal tracking and behavior-based closed-loop stimulation extends the availability of high-quality, low-cost experimental setup within standardized data formats serving the neuroscience community.

Adaptive WTA with an analog VLSI neuromorphic learning chip.

In this paper, we demonstrate how a particular spike-based learning rule (where exact temporal relations between input and output spikes of a spiking model neuron determine the changes of the synaptic weights) can be tuned to express rate-based classical Hebbian learning behavior (where the average input and output spike rates are sufficient to describe the synaptic changes). This shift in behavior is controlled by the input statistic and by a single time constant. The learning rule has been implemented in a neuromorphic very large scale integration (VLSI) chip as part of a neurally inspired spike signal image processing system. The latter is the result of the European Union research project Convolution AER Vision Architecture for Real-Time (CAVIAR). Since it is implemented as a spike-based learning rule (which is most convenient in the overall spike-based system), even if it is tuned to show rate behavior, no explicit long-term average signals are computed on the chip. We show the rule's rate-based Hebbian learning ability in a classification task in both simulation and chip experiment, first with artificial stimuli and then with sensor input from the CAVIAR system.

Localizing neuronal somata from Multi-Electrode Array in-vivo recordings using deep learning.

With the latest development in the design and fabrication of high-density Multi-Electrode Arrays (MEA) for in-vivo neural recordings, the spatiotemporal information in the recorded signals allows for refined estimation of a neuron's location around the probe. In parallel, advances in computational models for neural activity enables simulation of recordings from neurons with detailed morphology. Our approach uses deep learning algorithms on a large set of such simulation data to extract the 3D position of the neuronal somata. Multi-compartment models from 13 different neural morphologies in layer 5 (L5) of the rat's neocortex are placed at random locations and with different alignments with respect to the MEA. The sodium trough and repolarisation peak images on the MEA serve as input features for a Convolutional Neural Network (CNN), which predicts the neural location robustly and with low error rates. The forward modeling/machine learning approach yields very accurate results for the different morphologies and is able to cope with different neuron alignments.

A Bio-Inspired AER Temporal Tri-Color Differentiator Pixel Array.

This article investigates the potential of a bio-inspired vision sensor with pixels that detect transients between three primary colors. The in-pixel color processing is inspired by the retinal color opponency that are found in mammalian retinas. Color transitions in a pixel are represented by voltage spikes, which are akin to a neuron's action potential. These spikes are conveyed off-chip by the Address Event Representation (AER) protocol. To achieve sensitivity to three different color spectra within the visual spectrum, each pixel has three stacked photodiodes at different depths in the silicon substrate. The sensor has been fabricated in the standard TSMC 90 nm CMOS technology. A post-processing method to decode events into color transitions has been proposed and implemented as a custom interface to display real-time color changes in the visual scene. Experimental results are provided. Color transitions can be detected at high speed (up to 2.7 kHz). The sensor has a dynamic range of 58 dB and a power consumption of 22.5 mW. This type of sensor can be of use in industrial, robotics, automotive and other applications where essential information is contained in transient emissions shifts within the visual spectrum.

Integrated electronic system for implantable sensory NFC tag.

We have designed the complete electronic system for an implanted sensory NFC-A tag (type 1) that monitors a physiological parameter, e.g. blood glucose, dehydration, bladder pressure, to name some of the target applications that we pursue. The tag is meant to be implanted under the skin and is powered by an NFC reader held close to it, such as a smart phone or a smart watch. The electronic system consists of a sensor front-end, ADC, NFC-A transceiver and NFC power harvester. In its present status, the physical layer of the communication and the power harvester have been implemented on one ASIC, and the sensor front-end and ADC on another, while the digital circuits realizing the higher level NFC protocol have been implemented on an FPGA. Simulations and a few preliminary test results are presented in this paper. The ultimate goal after thorough testing of this first prototype is to integrate all of these modules on a single ASIC.

Bio-Inspired Asynchronous Pixel Event Tricolor Vision Sensor.

This article investigates the potential of the first ever prototype of a vision sensor that combines tricolor stacked photo diodes with the bio-inspired asynchronous pixel event communication protocol known as Address Event Representation (AER). The stacked photo diodes are implemented in a 22 × 22 pixel array in a standard STM 90 nm CMOS process. Dynamic range is larger than 60 dB and pixels fill factor is 28%. The pixels employ either simple pulse frequency modulation (PFM) or a Time-to-First-Spike (TFS) mode. A heuristic linear combination of the chip's inherent pseudo colors serves to approximate RGB color representation. Furthermore, the sensor outputs can be processed to represent the radiation in the near infrared (NIR) band without employing external filters, and to color-encode direction of motion due to an asymmetry in the update rates of the different diode layers.

Design and Characterization of an Osmotic Sensor for the Detection of Events Associated with Dehydration and Overhydration.

The level of hydration in the human body is carefully adjusted to control the electrolyte balance that governs the biochemical processes that sustain life. An electrolyte deficiency caused by de- or overhydration will not only limit human performance, but can also lead to serious health problems and death if left untreated. Because humans can withstand a change in hydration of only [Formula: see text], frequent monitoring should be performed in risk groups. This paper presents an osmotic hydration sensor that can record the level of hydration as a function of osmotic pressure in phosphate buffered saline or sodium-chloride solutions that simulate the interstitial fluid in the body. The osmotic pressure is recorded with the aid of an ion-exchange membrane that facilitates the migration of water and cations, in favor of reverse osmosis or gas separation membranes. The hydration sensor is designed to be coupled to an inductively powered readout circuit designed for integration in a micro-implant that has previously been shown to consume only 76 [Formula: see text] of power. The dynamic range spans a state of serious overhydration (220 [Formula: see text]) to a serious state of dehydration (340 [Formula: see text]) with a response time of [Formula: see text] (for a variation of hydration of 20%).

Neuromorphic silicon neuron circuits.

Hardware implementations of spiking neurons can be extremely useful for a large variety of applications, ranging from high-speed modeling of large-scale neural systems to real-time behaving systems, to bidirectional brain-machine interfaces. The specific circuit solutions used to implement silicon neurons depend on the application requirements. In this paper we describe the most common building blocks and techniques used to implement these circuits, and present an overview of a wide range of neuromorphic silicon neurons, which implement different computational models, ranging from biophysically realistic and conductance-based Hodgkin-Huxley models to bi-dimensional generalized adaptive integrate and fire models. We compare the different design methodologies used for each silicon neuron design described, and demonstrate their features with experimental results, measured from a wide range of fabricated VLSI chips.

CAVIAR: a 45k neuron, 5M synapse, 12G connects/s AER hardware sensory-processing- learning-actuating system for high-speed visual object recognition and tracking.

This paper describes CAVIAR, a massively parallel hardware implementation of a spike-based sensing-processing-learning-actuating system inspired by the physiology of the nervous system. CAVIAR uses the asychronous address-event representation (AER) communication framework and was developed in the context of a European Union funded project. It has four custom mixed-signal AER chips, five custom digital AER interface components, 45k neurons (spiking cells), up to 5M synapses, performs 12G synaptic operations per second, and achieves millisecond object recognition and tracking latencies.