Memristor crossbar-based neural networks perform parallel operations in the analog domain. In ex-situ training approach, predetermined resistance values need to be programed into the memristor crossbar. Due to the stochastic nature of memristor devices, programming a memristor needs to read the device resistance value iteratively. However, reading an individual memristor in a crossbar, especially without an isolation transistor, is challenging because of sneak path currents. Programming a memristor to either the RON or ROFF state is relatively straight-forward. Neural networks that use higher-precision weights typically achieve better classification accuracy than Ternary Neural Networks (TNNs). This paper presents a memristor-based neural network implementation that uses only the two resistance states (RON, ROFF). We have examined the impact of the device RON/ROFF ratio and driver size on the scalability of memristor-based neural network circuits. A large neural network is implemented by integrating multiple smaller 3D-stacked crossbar arrays. Additionally, we have proposed novel neuron circuits to support higher weight precision. Experimental results show that the proposed high-precision synapses are easy to program and offer improved classification accuracy compared to a TNN.

The proposed work demonstrates a 3D memristor crossbar-based pattern classification system. Analogue signals from a 2D sensor array are directly applied to the memristor crossbar circuit using TSVs for computation. The proposed near-sensor computing system reduces the wiring complexity between the sensor unit and the compute unit significantly. The 3D memristor crossbar occupies a very small area compared to an equivalent 2D crossbar. We have leveraged ex-situ training for the memristor-based neural network using the two extreme resistance levels of the device (RON, ROFF) which enables simple training circuit. To process a 100 x 100 sensor array, the proposed approach requires 5.6x less wiring between the sensor array and the memristor crossbar circuit compared to a non-stacked approach.

Memristor crossbar-based neural network systems offer high throughput with low energy consumption. A key advantage of on-chip training in these systems is their ability to mitigate the effects of device variability and faults. This paper presents an efficient on-chip training circuit for memristor crossbar-based multi-layer neural networks. We propose a novel method for storing the product of two analog signals directly in a memristor device, eliminating the need for ADC and DAC converters. Experimental results show that the proposed system is approximately twice as energy efficient and 1.5 times faster than existing memristor-based systems for training multi-layer neural networks.

An ultralow-power, high-performance online-learning and anomaly-detection system has been developed for edge security applications. Designed to support personalized learning without relying on cloud data processing, the system employs sample-wise learning, eliminating the need for storing entire datasets for training. Built using memristor-based analog neuromorphic and in-memory computing techniques, the system integrates two unsupervised autoencoder neural networks—one utilizing optimized crossbar weights and the other performing real-time learning to detect novel intrusions. Threshold optimization and anomaly detection are achieved through a fully analog Euclidean Distance (ED) computation circuit, eliminating the need for floating-point processing units. The system demonstrates 87% anomaly-detection accuracy; achieves a performance of 16.1 GOPS—774× faster than the ASUS Tinker Board edge processor; and delivers an energy efficiency of 783 GOPS/W, consuming only 20.5 mW during anomaly detection.

This paper proposes a low power consuming system for monitoring elderly people’s activities and their health conditions. The proposed system has two activity recognition modules: smartphone sensor-based wearable module; infrared grid sensor-based remote module. The two activity recognition modules work in a coordinated way. The fraction of the time the person is detected by the infrared sensor, the smartphone remains idle. As a result, energy consumption in the smartphone is reduced significantly, and hence the battery lifetime is increased. In the smartphone, a Feed-forward Neural Network (FNN) based activity recognition algorithm is implemented using fixed-point computation to further reduce energy consumption. A Convolutional Neural Network is used in the infrared sensor-based activity recognition module. The proposed system also has real-time health monitoring capability, which is based on ECG signal classification. A FNN leveraging fixed-point operation is used for ECG signal classification on an embedded ARM processor. Proposed fixed-point implementations of the FNNs are faster than floating-point implementation and require 50% less memory to store the neural network model parameters without loss of classification accuracy.

IoT devices can enable low cost and interactive health care services. In this paper we have proposed an affordable telemedicine system to bring healthcare services within the reach of the rural people of Bangladesh. Proposed system enables transmission of patient’s body parameters in real-time to a remote doctor. The proposed system also has real-time patient monitoring capability which is based on ECG signal classification. A feed-forward neural network is used for ECG signal classification on an embedded ARM processor. For low power operation, we have utilized fixed-point (integer) arithmetic instead of floating-point arithmetic for the ECG signal classification task. Proposed fixed-point implementation is 1.06x faster than floating-point implementation and requires 50% less memory to store the neural network model parameters without loss in the classification accuracy.

This paper presents an efficient string matching circuit based on a memristor crossbar. It uses a reduced number of memristors compared to the previous string matching circuits. The circuit leverages the computation in memory feature and hence avoids data movement for computation. The proposed circuit is about 1.97x area and energy efficient compared to the previous work. Tuning the conductance of the memristors in the crossbar, a single crossbar circuit can be used for matching strings of different lengths.

Abstract Memristor crossbar-based neural networks perform parallel operation in the analog domain. Ex-situ training approach needs to program the predetermined resistance values to the memristor crossbar. Because of the stochasticity of the memristor devices, programming a memristor needs to read the device resistance value iteratively. Reading a single memristor in a crossbar (without isolation transistor) is challenging due to the sneak path current. Programming a memristor in a crossbar to either RON or ROFF state is relatively straightforward. A neural network implemented using higher precision weights provides higher classification accuracy compared to a Ternary Neural Network (TNN). This paper demonstrates the implementation of memristor-based neural networks using only the two resistance values (RON, ROFF). We have considered the crossbar scaling limits and proposed a novel technique to implement a large neural network using multiple smaller crossbar arrays. We have proposed novel neuron circuits to achieve higher weight precision. Our experimental result shows that the proposed higher precision synapses are easy to program and provide better classification accuracy compared to a TNN. Proposed technique of implementing a large neural network on memristor crossbar circuits has a slight loss in the classification accuracy compared to the software implementation. But the memristor-based implementation uses only 51.7% of the synapses used in the software implementation.

This article proposes a fully integrated single-channel bistatic frequency-modulated continuous-wave (FMCW) radar transceiver (TRX) that operates at a center frequency of 256 GHz. The main focus of this work is to realize a wideband and efficient radar TRX that offers high resolution of target detection in the short-range FMCW radar sensing application. The radar TRX chip is designed and manufactured using the 130 nm silicon–germanium (SiGe) bipolar complementary metal-oxide-semiconductor (BiCMOS) technology which offers heterojunction bipolar transistors (HBTs) with $f_{\mathbf {T}}/f_{\mathbf {MAX}}$ of 300/500 GHz. The transmitter (TX) of the radar TRX is based on a fundamentally operated multiplier-by-8 chain architecture that offers a 3-dB bandwidth of around 65 GHz with a saturated output power of −5.4 dBm. On the other hand, the receiver (RX) is based on a subharmonic architecture that provides a conversion gain (CG) of 10.4 dB with an average noise figure (NF) of 23.5 dB. This TRX is realized with two integrated on-chip folded dipole antennas. The antenna offers high antenna gain and radiation efficiency due to the use of the selective localized backside etching (LBE) technique. This chip consumes 305 mW of power from a 3.3-V supply and occupies a silicon area of 3.3 mm2. The radar range measurement is performed in an anechoic chamber, and it shows the maximum dynamic range (DR) of around 34 dB at 1-m range of the target.

Memristor crossbar-based neural networks perform parallel operation in the analog domain. Ex-situ training approach needs to program the predetermined resistance values in the memristor crossbar. Because of the stochasticity of the memristor devices, programming a memristor needs to read the device resistance value iteratively. Reading a single memristor in a crossbar (without isolation transistor) is challenging due to the sneak path current. Programming a memristor in a crossbar to either RON or ROFF state is relatively straightforward. A neural network implemented using higher precision weights provides higher classification accuracy compared to a Ternary Neural Network (TNN). This paper demonstrates the implementation of memristor-based neural networks using only the two resistance values (RON, ROFF). At the same time, it achieves higher weight precision. The experimental result shows that the proposed higher precision synapses are easy to program and provide better classification accuracy compared to a TNN.

This paper proposes a compact and efficient frequency quadrupler (FQ) intended for a sub-harmonic transceiver. The circuit is based on stacked double bootstrapped Gilbert cell (GC) and works at the frequency band of 110-150GHz. It consists of two stacked mixing stages. The GC stages individually perform as a doubler for the input frequency, and the resultant quadrupled signal appears at the output of the second GC stage. The FQ achieves the peak conversion gain of 6.6dB and the maximum output power of -1dBm. The 3-dB bandwidth of this FQ is 40GHz(110-150GHz). Due to the compact design without compromising the performance, the stand-alone FQ Chip occupies only $0.056\text{mm}^{2}$ chip area and consumes only 13.4mA of current from the 3.3V supply. It achieves the maximum drain efficiency of 1.81%. The presented FQ is suitable for the frequency generation circuit in various system design applications above 100GHz.

We describe practical improvements for parallel BWT-based lossless compressors frequently utilized in modern day big data applications. We propose a clustering-based data permutation approach for improving compression ratio for data with significant alphabet variation along with a faster string sorting approach based on the application of the [Formula: see text] complexity counting sort with permutation reindexing.

A number of selected fungicides were evaluated to determine their efficacy for controlling collar rot disease of soybean plants caused by Sclerotium rolfsii. The experiment was conducted under the controlled condition at the Plant Pathology Laboratory and Field laboratory of BINA, Bangladesh Agricultural University campus from November 2018 to August 2019. In-vitro research was done for the observation of radial mycelial growth of S. rolfsii on potato dextrose agar (PDA), treated with five fungicides viz. Antracol 70 WP (T1), Ridomil Gold MZ 68 WP (T2), Secure 600 WG (T3), Bavistin DF (T4), Dithane M-45 (T5), and one non-treated (T0) treatment. The highest percentage of mycelial growth inhibition of S. rolfsii in PDA medium was recorded in treatment T5 (Dithane M-45) 100% and lowest in treatment T3 (Secure 600 WG) 37.33% at 6 days after inoculation. Then the selected five fungicides were again applied to pot under controlled conditions to observe the best effect of selected fungicides against collar rot pathogen of soybean plants. The inoculation was done on a variety of BINA soybean 4 in pot condition. The highest mortality percent for the collar rot disease was found in treatment T0 (controlled) 100% soybean plants conversely, the lowest mortality percent was found in treatment T5 (Dithane M-45) 27.28% besides 38.92% in T2 (Ridomil Gold MZ 68 WP), 43.42% in T1 (Antracol 70 WP), 46.18% in T3 (Secure 600 WG) and 50.00% in treatment T4 (Bavistin DF) respectively. Thus, Dithane M-45 was found superior in controlling collar rot pathogen S. rolfsii of Soybean over all other fungicides tested in both in vitro and in vivo.

The dream towards fully autonomous vehicles brings a lot of challenges in terms of reliability, high performance computing and sensing capabilities in a vehicle. Nowadays, autonomous vehicles are equipped with several sensors, which allow the vehicle to sense everything on the road and to collect the information needed to drive safely. Altogether, these sensors generate a lot of data, roughly 4TB in a single day [1]. In order to process this abundance of sensory data faster and reliably, there is a need for high performance microcontroller units (MCUs) and fault tolerant data processing respectively. For this reason, we have developed a smart reconfigurable sensor (SRS) platform, which aims to solve the challenges of high performance computing and safety. An in-house fabricated entire silicon-based millimeter-wave transceiver, using IHP's 130 nm SiGe BiCMOS technology, is used as SRS front-end and software (SW) based triple modular redundancy (TMR) system is implemented for fault tolerant radar data processing. Our highly adaptive system supports fault tolerant modes (i.e. fail operational, fail safe), low power mode and distributed execution of tasks among different cores, all while meeting the strict automotive standards.

We designed and implemented a deep learning based RF signal classifier on the Field Programmable Gate Array (FPGA) of an embedded software-defined radio platform, DeepRadio™, that classifies the signals received through the RF front end to different modulation types in real time and with low power. This classifier implementation successfully captures complex characteristics of wireless signals to serve critical applications in wireless security and communications systems such as identifying spoofing signals in signal authentication systems, detecting target emitters and jammers in electronic warfare (EW) applications, discriminating primary and secondary users in cognitive radio networks, interference hunting, and adaptive modulation. Empowered by low-power and low-latency embedded computing, the deep neural network runs directly on the FPGA fabric of DeepRadio™, while maintaining classifier accuracy close to the software performance. We evaluated the performance when another SDR (USRP) transmits signals with different modulation types at different power levels and DeepRadio™receives the signals and classifies them in real time on its FPGA. A smartphone with a mobile app is connected to DeepRadio™to initiate the experiment and visualize the classification results. With real radio transmissions over the air, we show that the classifier implemented on DeepRadio™achieves high accuracy with low latency (microsecond per sample) and low energy consumption (microJoule per sample), and this performance is not matched by other embedded platforms such as embedded graphics processing unit (GPU).

This paper describes a differential microstrip patch antenna array and a rectangular waveguide to coupled differential microstrip line transition operating at 122 GHz. The antenna array is realized in series-fed topology/architecture and is very suitable for MIMO radar applications. This on-board antenna provides a high antenna gain and radiation efficiency at millimeter-wave frequencies. The rectangular waveguide transition with low insertion loss offers the facilities to characterize the on-board antenna and can also be utilized in a radar system in combination with a horn antenna. The measured bandwidth of the antenna array is 7 GHz with a maximum gain of 12.98 dBi at 122 GHz. The waveguide transition has a bandwidth of 20 GHz at 10-dB return loss. Radar measurements were performed using radar sensors that were equipped with the developed antenna array as well as the waveguide transition in combination with horn antenna for comparison purposes. The radar measurement results with on-board antenna array show a good performance for detecting the range of the target.

Memristor crossbar arrays carry out multiply–add operations in parallel in the analog domain, and can enable neuromorphic systems with high throughput at low energy and area consumption. Neural networks need to be trained prior to use. This work considers ex-situ training where the weights pre-trained by a software implementation are then programmed into the hardware. Existing ex-situ training approaches for memristor crossbars do not consider sneak path currents, and they may work only for neural networks implemented using small crossbars. Ex-situ training in large crossbars, without considering sneak paths, reduces the application recognition accuracy significantly due to the increased number of sneak current paths. This paper proposes ex-situ training approaches for both 0T1M and 1T1M crossbars that account for crossbar sneak paths and the stochasticity inherent in memristor switching. To carry out the simulation of these training approaches, a framework for fast and accurate simulation of large memristor crossbars was developed. The results in this work show that 0T1M crossbar based systems can be 17–83% smaller in area than 1T1M crossbar based systems.

Custom low power hardware for real-time network security and anomaly detection are in great demand, as these would allow for efficient security in battery-powered network devices. This paper presents a memristor based system for real-time intrusion detection, as well as an anomaly detection based on autoencoders. Intrusion detection is based on a single autoencoder, and the overall detection accuracy of this system is 92.91% with a malicious packet detection accuracy of 98.89%. The system described in this paper is also capable of using two autoencoders to perform anomaly detection using real-time online learning. Using this system, we show that anomalous data is flagged by the system, but over time the system stops flagging a particular datatype if its presence is abundant. Utilizing memristors in these designs allows us to present extreme low power systems for intrusion and anomaly detection, while sacrificing little accuracy.