A Power and Area Efficient CMOS Stochastic Neuron for Neural Networks Employing Resistive Crossbar Array.

A power and area efficient CMOS stochastic neuron for resistive computing device-based neural networks is presented. The stochastic neuron performs both quantization and activation function simultaneously by using a single dynamic comparator and allows power-hungry analog to digital and digital to analog converters to be removed at the cost of the increased computation time. A network learning method utilizing a noisy sigmoid function is also presented to minimize the computation time with little accuracy degradation. A prototype neuron chip fabricated in 0.18µm CMOS process successfully demonstrates the neuron's performance and the learning method is verified through network simulations.

Unsupervised Learning for Depth, Ego-Motion, and Optical Flow Estimation Using Coupled Consistency Conditions.

Herein, we propose an unsupervised learning architecture under coupled consistency conditions to estimate the depth, ego-motion, and optical flow. Previously invented learning techniques in computer vision adopted a large amount of the ground truth dataset for network training. A ground truth dataset, including depth and optical flow collected from the real world, requires tremendous effort in pre-processing due to the exposure to noise artifacts. In this paper, we propose a framework that trains networks while using a different type of data with combined losses that are derived from a coupled consistency structure. The core concept is composed of two parts. First, we compare the optical flows, which are estimated from both the depth plus ego-motion and flow estimation network. Subsequently, to prevent the effects of the artifacts of the occluded regions in the estimated optical flow, we compute flow local consistency along the forward-backward directions. Second, synthesis consistency enables the exploration of the geometric correlation between the spatial and temporal domains in a stereo video. We perform extensive experiments on the depth, ego-motion, and optical flow estimation on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset. We verify that the flow local consistency loss improves the optical flow accuracy in terms of the occluded regions. Furthermore, we also show that the view-synthesis-based photometric loss enhances the depth and ego-motion accuracy via scene projection. The experimental results exhibit the competitive performance of the estimated depth and the optical flow; moreover, the induced ego-motion is comparable to that obtained from other unsupervised methods.

Jitter Elimination in Shape Recovery by using Adaptive Neural Network Filter.

Three-dimensional (3D) cameras are expensive because they employ additional charged coupled device sensors and optical elements, e.g., lasers or complicated scanning mirror systems. One passive optical method, shape from focus (SFF), provides an efficient low cost solution for 3D cameras. However, mechanical vibration of the SFF imaging system causes jitter noise along the optical axis, which makes it difficult to obtain accurate shape information of objects. In traditional methods, this error cannot be removed and increases as the estimation of the shape recovery progresses. Therefore, the final 3D shape may be inaccurate. We introduce an accurate depth estimation method using an adaptive neural network (ANN) filter to remove the jitter noise effects. Jitter noise is modeled by both Gaussian distribution and non-Gaussian distribution. Then, focus curves are modeled by quadratic functions. The ANN filter is designed as an optimal estimator restoring the original position of each frame of the input image sequence in the modeled jitter noise, as a pre-processing step before the initial depth map is obtained. The proposed method was evaluated using image sequences of both synthetic and real objects. Experimental results demonstrate that it is reasonably efficient and that its accuracy is comparable with that of existing systems.

A High Full Well Capacity CMOS Image Sensor for Space Applications.

This paper presents a high full well capacity (FWC) CMOS image sensor (CIS) for space applications. The proposed pixel design effectively increases the FWC without inducing overflow of photo-generated charge in a limited pixel area. An MOS capacitor is integrated in a pixel and accumulated charges in a photodiode are transferred to the in-pixel capacitor multiple times depending on the maximum incident light intensity. In addition, the modulation transfer function (MTF) and radiation damage effect on the pixel, which are especially important for space applications, are studied and analyzed through fabrication of the CIS. The CIS was fabricated using a 0.11 µm 1-poly 4-metal CIS process to demonstrate the proposed techniques and pixel design. A measured FWC of 103,448 electrons and MTF improvement of 300% are achieved with 6.5 µm pixel pitch.

A 3D-Printed Sensor for Monitoring Biosignals in Small Animals.

Although additive manufacturing technologies, also known as 3D printing, were first introduced in the 1980s, they have recently gained remarkable popularity owing to decreased costs. 3D printing has already emerged as a viable technology in many industries; in particular, it is a good replacement for microfabrication technology. Microfabrication technology usually requires expensive clean room equipment and skilled engineers; however, 3D printing can reduce both cost and time dramatically. Although 3D printing technology has started to emerge into microfabrication manufacturing and medical applications, it is typically limited to creating mechanical structures such as hip prosthesis or dental implants. There have been increased interests in wearable devices and the critical part of such wearable devices is the sensing part to detect biosignals noninvasively. In this paper, we have built a 3D-printed sensor that can measure electroencephalogram and electrocardiogram from zebrafish. Despite measuring biosignals noninvasively from zebrafish has been known to be difficult due to that it is an underwater creature, we were able to successfully obtain electrophysiological information using the 3D-printed sensor. This 3D printing technique can accelerate the development of simple noninvasive sensors using affordable equipment and provide an economical solution to physiologists who are unfamiliar with complicated microfabrication techniques.

Author Correction: Zebrafish as an animal model in epilepsy studies with multichannel EEG recordings.

A correction to this article has been published and is linked from the HTML version of this paper. The error has been fixed in the paper.

Low photon count based digital holography for quadratic phase cryptography.

Recently, the vulnerability of the linear canonical transform-based double random phase encryption system to attack has been demonstrated. To alleviate this, we present for the first time, to the best of our knowledge, a method for securing a two-dimensional scene using a quadratic phase encoding system operating in the photon-counted imaging (PCI) regime. Position-phase-shifting digital holography is applied to record the photon-limited encrypted complex samples. The reconstruction of the complex wavefront involves four sparse (undersampled) dataset intensity measurements (interferograms) at two different positions. Computer simulations validate that the photon-limited sparse-encrypted data has adequate information to authenticate the original data set. Finally, security analysis, employing iterative phase retrieval attacks, has been performed.

Zebrafish as an animal model in epilepsy studies with multichannel EEG recordings.

Despite recent interest in using zebrafish in human disease studies, sparked by their economics, fecundity, easy handling, and homologies to humans, the electrophysiological tools or methods for zebrafish are still inaccessible. Although zebrafish exhibit more significant larval-adult duality than any other animal, most electrophysiological studies using zebrafish are biased by using larvae these days. The results of larval studies not only differ from those conducted with adults but also are unable to delicately manage electroencephalographic montages due to their small size. Hence, we enabled non-invasive long-term multichannel electroencephalographic recording on adult zebrafish using custom-designed electrodes and perfusion system. First, we exploited demonstration of long-term recording on pentylenetetrazole-induced seizure models, and the results were quantified. Second, we studied skin-electrode impedance, which is crucial to the quality of signals. Then, seizure propagations and gender differences in adult zebrafish were exhibited for the first time. Our results provide a new pathway for future neuroscience research using zebrafish by overcoming the challenges for aquatic organisms such as precision, serviceability, and continuous water seepage.

Sub-0.5 V Highly Stable Aqueous Salt Gated Metal Oxide Electronics.

Recently, growing interest in implantable bionics and biochemical sensors spurred the research for developing non-conventional electronics with excellent device characteristics at low operation voltages and prolonged device stability under physiological conditions. Herein, we report high-performance aqueous electrolyte-gated thin-film transistors using a sol-gel amorphous metal oxide semiconductor and aqueous electrolyte dielectrics based on small ionic salts. The proper selection of channel material (i.e., indium-gallium-zinc-oxide) and precautious passivation of non-channel areas enabled the development of simple but highly stable metal oxide transistors manifested by low operation voltages within 0.5 V, high transconductance of ~1.0 mS, large current on-off ratios over 10(7), and fast inverter responses up to several hundred hertz without device degradation even in physiologically-relevant ionic solutions. In conjunction with excellent transistor characteristics, investigation of the electrochemical nature of the metal oxide-electrolyte interface may contribute to the development of a viable bio-electronic platform directly interfacing with biological entities in vivo.

Interferometry based multispectral photon-limited 2D and 3D integral image encryption employing the Hartley transform.

We present a method of securing multispectral 3D photon-counted integral imaging (PCII) using classical Hartley Transform (HT) based encryption by employing optical interferometry. This method has the simultaneous advantages of minimizing complexity by eliminating the need for holography recording and addresses the phase sensitivity problem encountered when using digital cameras. These together with single-channel multispectral 3D data compactness, the inherent properties of the classical photon counting detection model, i.e. sparse sensing and the capability for nonlinear transformation, permits better authentication of the retrieved 3D scene at various depth cues. Furthermore, the proposed technique works for both spatially and temporally incoherent illumination. To validate the proposed technique simulations were carried out for both the 2D and 3D cases. Experimental data is processed and the results support the feasibility of the encryption method.

Electronic system with memristive synapses for pattern recognition.

Memristive synapses, the most promising passive devices for synaptic interconnections in artificial neural networks, are the driving force behind recent research on hardware neural networks. Despite significant efforts to utilize memristive synapses, progress to date has only shown the possibility of building a neural network system that can classify simple image patterns. In this article, we report a high-density cross-point memristive synapse array with improved synaptic characteristics. The proposed PCMO-based memristive synapse exhibits the necessary gradual and symmetrical conductance changes, and has been successfully adapted to a neural network system. The system learns, and later recognizes, the human thought pattern corresponding to three vowels, i.e. /a /, /i /, and /u/, using electroencephalography signals generated while a subject imagines speaking vowels. Our successful demonstration of a neural network system for EEG pattern recognition is likely to intrigue many researchers and stimulate a new research direction.

Implementation of a portable device for real-time ECG signal analysis.

BACKGROUND: Cardiac disease is one of the main causes of catastrophic mortality. Therefore, detecting the symptoms of cardiac disease as early as possible is important for increasing the patient's survival. In this study, a compact and effective architecture for detecting atrial fibrillation (AFib) and myocardial ischemia is proposed. We developed a portable device using this architecture, which allows real-time electrocardiogram (ECG) signal acquisition and analysis for cardiac diseases. METHODS: A noisy ECG signal was preprocessed by an analog front-end consisting of analog filters and amplifiers before it was converted into digital data. The analog front-end was minimized to reduce the size of the device and power consumption by implementing some of its functions with digital filters realized in software. With the ECG data, we detected QRS complexes based on wavelet analysis and feature extraction for morphological shape and regularity using an ARM processor. A classifier for cardiac disease was constructed based on features extracted from a training dataset using support vector machines. The classifier then categorized the ECG data into normal beats, AFib, and myocardial ischemia. RESULTS: A portable ECG device was implemented, and successfully acquired and processed ECG signals. The performance of this device was also verified by comparing the processed ECG data with high-quality ECG data from a public cardiac database. Because of reduced computational complexity, the ARM processor was able to process up to a thousand samples per second, and this allowed real-time acquisition and diagnosis of heart disease. Experimental results for detection of heart disease showed that the device classified AFib and ischemia with a sensitivity of 95.1% and a specificity of 95.9%. CONCLUSIONS: Current home care and telemedicine systems have a separate device and diagnostic service system, which results in additional time and cost. Our proposed portable ECG device provides captured ECG data and suspected waveform to identify sporadic and chronic events of heart diseases. This device has been built and evaluated for high quality of signals, low computational complexity, and accurate detection.

Nanoscale RRAM-based synaptic electronics: toward a neuromorphic computing device.

Efforts to develop scalable learning algorithms for implementation of networks of spiking neurons in silicon have been hindered by the considerable footprints of learning circuits, which grow as the number of synapses increases. Recent developments in nanotechnologies provide an extremely compact device with low-power consumption.In particular, nanoscale resistive switching devices (resistive random-access memory (RRAM)) are regarded as a promising solution for implementation of biological synapses due to their nanoscale dimensions, capacity to store multiple bits and the low energy required to operate distinct states. In this paper, we report the fabrication, modeling and implementation of nanoscale RRAM with multi-level storage capability for an electronic synapse device. In addition, we first experimentally demonstrate the learning capabilities and predictable performance by a neuromorphic circuit composed of a nanoscale 1 kbit RRAM cross-point array of synapses and complementary metal-oxide-semiconductor neuron circuits. These developments open up possibilities for the development of ubiquitous ultra-dense, ultra-low-power cognitive computers.