Binary memristive crossbars have gained huge attention as an energy-efficient deep learning hardware accelerator. Nonetheless, they suffer from various noises due to the analog nature of the crossbars. To overcome such limitations, most previous works train weight parameters with noise data obtained from a crossbar. These methods are, however, ineffective because it is difficult to collect noise data in large-volume manufacturing environment where each crossbar has a large device/circuit level variation. Moreover, we argue that there is still room for improvement even though these methods somewhat improve accuracy. This paper explores a new perspective on mitigating crossbar noise in a more generalized way by manipulating input binary bit encoding rather than training the weight of networks with respect to noise data. We first mathematically show that the noise decreases as the number of binary bit encoding pulses increases when representing the same amount of information. In addition, we propose Gradient-based Bit Encoding Optimization (GBO) which optimizes a different number of pulses at each layer, based on our in-depth analysis that each layer has a different level of noise sensitivity. The proposed heterogeneous layer-wise bit encoding scheme achieves high noise robustness with low computational cost. Our experimental results on public benchmark datasets show that GBO improves the classification accuracy by ∼ 5–40% in severe noise scenarios.

An energy-harvesting interface that incorporates a continuously scalable-conversion-ratio (CSCR) switched-capacitor (SC) dc-dc converter with maximum power point tracking (MPPT) is introduced in this paper. By exploiting unique characteristics of a CSCR SC converter, an MPPT based on the hill climbing algorithm is implemented with a real-time power monitoring scheme with a frequency-mapped sampling capacitor DAC, allowing variation-tolerant MPPT for various types of energy sources. The harvestable voltage range is significantly extended by a step-up/down convertible CSCR SC converter, and the voltage conversion efficiency is further improved with matrix-structured and load-mapped power switches. With the test chip fabricated in 28nm FDSOI technology, the proposed energy harvesting interface shows peak conversion efficiency of 89% and average efficiency of > 80% for 60-2,$794~\mu \text{W}$. MPPT efficiency of > 97 % is confirmed with energy harvesting measurement with a solar cell.

To control smart watches available in the market, both hands should be used. A surface electromyography (SEMG)-based interface will enable controlling the watch with the hand where it is worn. However, fewer studies have been conducted on SEMG-interface for healthy subjects for smart watch applications. This study developed an algorithm to classify 3-5 finger gestures from SEMG signals recorded on the upper wrist. Also, we compared the classification accuracies between intra-subject models and an inter-subject model. We concluded that when there is a low number of gestures, factory calibration for matching SEMGs to gestures is sufficient. However, when there are a high number of gestures, individual calibration is required just after purchasing the watch.

For wearable devices that monitor personal health, secure access to private medical data becomes a crucial feature. Nowadays, device authentication based on biometrics such as fingerprint or iris has become increasingly popular. In this work, we investigate using electrocardiogram (ECG) signals as the biometric modality for device authentication, and we present accurate and low-power ECG-based authentication hardware. Deep neural networks (DNNs) have been employed with a cost function that maximizes inter-individual distance and minimizes intra-individual distance over time. During DNN training, we also introduce joint optimization of low-precision and structured sparsity, so that the real-time authentication hardware can consume minimal energy and area. Experimental results of custom hardware designed in 65nm LP CMOS technology exhibit low power consumption of 59.4 μW for real-time ECG authentication with a low equal error rate of 1.002% for a large 741-subject inhouse ECG database.

Energy efficient bio-feedback stimulation system is implemented for implantable medical device applications. The adaptive parameters optimization (APO) closed loop is proposed to find the most energy efficient values of stimulation parameters including an amplitude (AMP), a pulse-width (PW), a frequency (FREQ) of stimulation currents and a supply of the driver (VDDH). The key mechanism is based on neuron's biologic characteristics; a rheobase current (IRh), a chronaxie time (TCh) and a refractory period (TRefractory). To realize the APO loop, a digital parameters optimizer, an impedance estimator and a programmable current driver is proposed in 180nm HV BCD technology. It forms the closed-loop together with an external neural spike detector and a variable VDDH generator. As a result, the energy consumption can be reduced to less than 1% with the APO closed loop.

This paper presents a fully integrated RF transceiver, capable of small form factor and ultra-low-power performance. The transceiver is achieved by employing LC-oscillator which has RF antenna as inductor components. The proposed receiver converts oscillation start-up time responding to input OOK signal to digital outputs, sensitively. The low complexity transmitter allows to improved implementation size and power efficiency with class-C operation. The transceiver was manufactured in 0.18 μm CMOS. The transceiver achieves a transmitter efficiency of 32.2% at -28.5 dBm EIRP while dissipating 2.2 mW and receiver sensitivity of -85 dBm in 25kbps OOK input. In experiments of implant condition replicated as human body, transceiver communicates to outside base station at 1m wireless and 8 cm implant depth.

This paper presents microwatt end-to-end digital signal processing (DSP) systems for deployment-stage real-time upper-limb movement intent decoding. This brain computer interface (BCI) DSP systems feature intercellular spike detection, sorting, and decoding operations for a 96-channel prosthetic implant. We design the algorithms for those operations to achieve minimal computation complexity while matching or advancing the accuracy of state-of-art BCI sorting and movement decoding. Based on those algorithms, we architect the DSP hardware with the focus on hardware reuse and event-driven operation. The VLSI implementation of the proposed systems in a 65-nm high-VTHshows that it can achieve 4.82 μW at the supply voltage of 300mV in the post-layout simulation. The area is 0.16 mm2.

Near Field Communication (NFC) applied with resonant coupling would lead to an innovative improvements such as power transfer efficiency and robustness, which can be most suitable for medical applications. However, there are several technical issues of the conventional NFC system that makes it difficult to apply resonant coupling. One of the critical issues is that the system performance becomes excessively sensitive to impedance matching and bandwidth of its antenna. In this paper, we propose a novel high-Q NFC system that overcomes the limitations of applying resonant coupling to conventional NFC system. As a result, the high-Q NFC system showed remarkable improvement in energy efficiency. Moreover, the high-Q NFC system showed perfect compatibility with conventional NFC system, while it quadrupled the power transmission efficiency.

Near field channel is fundamentally different from conventional far field channel. Instead of using electromagnetic wave propagation, magnetically coupled resonance is utilized in near field. In order to apply this resonance phenomenon to communication system, Resonance Isolation (RI) system was recently proposed. In the RI system, a resonator stored with initial energy is mutually coupled with another resonator in near field, alternately exchanging its energy with the resonator. We establish two corresponding communication channel models using circuit theory and coupled mode theory, considering three classical network scenarios: two resonators (peer-to-peer), three resonators (equilateral), and three resonators (relay). We analyze the channel models for each scenario by both models and compare the performance with numerical results. We found significant relaying gain in the relaying scenario, and we defined this scenario as the Resonate and Forward technique.

In wireless sensor networks, limited battery capacity in each sensor node restricts the reliability of the network. This restriction can be somewhat removed by recharging the node batteries through wireless energy transmission. In a mobile wireless sensor network, if there is only one energy station that can transmit energy, the sensor nodes have to rely on their mobility to receive energy from an energy station. When a node battery is depleted, the node stops activating and stays idle until it recharges its battery. As the number of such idle nodes increases, the reliability of the wireless sensor network will decrease. If energy can be relayed via mobile nodes, a battery-depleted node can recharge its battery with higher rate, which affects the reliability of the network. In this paper, we show that if the energy transmission efficiency exceeds a certain threshold then the reliability of the network can always be improved.

Technology of using mid-range magnetic resonant coupling has enabled efficient power transmission in near field. The advantage of using resonant coupling as a power transmitting method lies in the high efficiency, omnidirectional, and no electromagnetic radiation on human body. When a continuous wave (AC) is used as the power source, the resonant coupling method is required to match the impedances at both transmitting and receiving ends in order to achieve the high efficiency. Instead of using an AC, the Resonator Isolation (RI) system uses an instantaneous periodic delivery (IPD) method to effectively transfer power and data at the same time. Using IPD method, the RI system does not require impedance matching at both ends. However, the operation between the transmitter and the receiver in the RI system needs to be synchronized. Since the transmission efficiency highly depends on the synchronization performance, the RI system requires an accurate synchronizing method. In this paper, we describe how to synchronize the transmitter and the receiver in the RI system, and present an accurate synchronization can be achieved with simple circuitry and operation.

In this paper, we propose a power autonomous sensor platform with ultra-low power radio frequency (RF) transceiver and wireless charging. The ultra-low power RF transceiver consists of super-regenerative receiver and direct-modulation transmitter with multi-cell power amplifier. Although the proposed transmitter and receiver only consume 1.79 and 0.683 mW, respectively, the wearable sensor platform requires more power efficient transmitter, due to the limited capacity of rechargeable thin film battery, e.g, 50 ~ 250 mWhr [1]. Therefore, we apply wireless charging in order to prolong the lifetime and investigate the impact of wireless power transfer on the power autonomy of sensor platform.

Wireless power transmission using mid-range magnetic resonance has become an important technology in body area sensor systems. Recent researches have used fine-tuned impedance matching for wireless power transmission at fixed distance. However, this method is significantly sensitive to the distance change between two devices since the impedances at both ends must be perfectly matched. Hence, it is a major challenge to achieve efficient wireless power transmission in mobile environment. We propose a new circuit architecture for wireless power transmission in the near field; the Resonator Isolation (RI) structure. Using the RI structure, the wireless power transmission becomes highly robust under mobile environment. In this paper, we present the circuit design of the RI structure and evaluate its performance with the actual systems we constructed.

We consider a cellular network inspired channel model which consists of the bi-directional exchange of information between a base-station and M terminal nodes with the help of a relay. The base-station has a message for each of M terminals, and conversely, each terminal node has a message for the base-station. A single relay assists the bi-directional communication endeavor. We assume an AWGN channel model with direct links (omitted in previous studies) between the base-station, relay, and half-duplex nodes. In this scenario, we derive achievable rate regions for two temporal protocols - needed in half-duplex networks - which indicate which users transmit when. These achievable rate regions are based on a novel lattice encoding and decoding strategy which outperforms previously derived regions using random-coding based decode-and-forward strategies under certain channel conditions. The terminal nodes employ nested lattice codes, and the relay decodes a series of codeword combinations - one of the main novelties of our scheme - from which it deduce the sum-codewords of the base-station to terminal node i, which it then broadcasts. This scheme differs markedly from previously considered successive-decoding based lattice strategies and provides a more general framework for the joint decoding of lattice codewords in a MAC. Numerical evaluations of our lattice-based inner bounds are shown to improve upon previous random-coding based schemes under certain channel conditions, and are compared to half-duplex cut-set outer bounds. We further demonstrate a constant sum-rate gap result using this lattice-based scheme for symmetric channels for one of the two protocols.

In a bi-directional relay channel, two nodes wish to exchange independent messages over a shared wireless half-duplex channel with the help of a relay. In this paper, we derive achievable rate regions for four new half-duplex protocols and compare these to four existing half-duplex protocols and outer bounds. In time, our protocols consist of either two or three phases. In the two phase protocols, both users simultaneously transmit during the first phase and the relay alone transmits during the second phase, while in the three phase protocol the two users sequentially transmit followed by a transmission from the relay. The relay may forward information in one of four manners; we outline existing amplify and forward (AF), decode and forward (DF), lattice based, and compress and forward (CF) relaying schemes and introduce the novel mixed forward scheme. The latter is a combination of CF in one direction and DF in the other. We derive achievable rate regions for the CF and Mixed relaying schemes for the two and three phase protocols. We provide a comprehensive treatment of eight possible half-duplex bi-directional relaying protocols in Gaussian noise, obtaining their relative performance under different SNR and relay geometries.

The multi-pair bi-directional relay network under consideration consists of one base-station, multiple (say m) terminal nodes and one relay, all of which are half-duplex, in which, contrary to prior work, each node has a direct link with every other node. Each of the m terminal nodes exchanges messages with the base-station in a bi-directional fashion, leading to 2m total messages to be communicated with the (possible) help of the relay. Our contributions are: 1) the introduction of three new temporal protocols which fully exploit the two-way nature of the data, over-heard side-information through network coding, random binning, and compress-and-forward terminal node cooperation, 2) derivations of achievable rate regions and 3) cut-set based outer bounds for the multi-pair network, and 4) a numerical evaluation of the derived regions in Gaussian noise which illustrate the performance of the proposed protocols.

In a protocol assisted switched diversity (PASD) system, the time separated repetition blocks are accumulated in order to increase decoding reliability. In this paper, we consider transmit antenna selection (TAS) protocols in a PASD system. To minimize system complexity we assume the feedback information is limited to the SNR of the current pair of antennas. We propose three TAS protocols : the random TAS (R-TAS) protocol which selects antennas randomly, the threshold TAS (T-TAS) protocol which keeps the current selection if the predetermined SNR threshold level requirement is met and the minimum number of transmission TAS (MNT-TAS) protocol which determines the selection of a new antenna set or keeping the current set based on the evaluation of expectation of the number of necessary remaining transmissions. We analyze the average number of transmissions of three proposed protocols in a Rayleigh fading channel and compare numerical evaluations to simulation results under different system conditions.

In a bi-directional relay channel, two nodes wish to exchange independent messages over a shared wireless half-duplex channel with the help of relays. Recent work has considered information theoretic limits of the bi-directional relay channel with a single relay. In this work we consider bi-directional relaying with multiple relays. We derive achievable rate regions and outer bounds for half-duplex protocols with multiple decode and forward relays and compare these to the same protocols with amplify and forward relays in an additive white Gaussian noise channel. We consider three novel classes of half-duplex protocols: the (m, 2) 2 phase protocol with m relays, the (m, 3) 3 phase protocol with m relays, and general (m, t) Multiple Hops and Multiple Relays (MHMR) protocols, where m is the total number of relays and 3 < t ¿ m + 2 is the number of temporal phases in the protocol. Finally, we provide a comprehensive treatment of the MHMR protocols with decode and forward relaying and amplify and forward relaying in Gaussian noise, obtaining their respective achievable rate regions, outer bounds and relative performance at different SNRs. The (m,m+2) DF MHMR protocol achieves the largest rate region under simulated channel conditions.

In coded bidirectional cooperation, two nodes wish to exchange messages over a shared half-duplex channel with the help of a relay. In this correspondence, we derive performance bounds for this problem for each of three decode-and-forward protocols. The first protocol is a two phase protocol where both users simultaneously transmit during the first phase and the relay alone transmits during the second. In this protocol, our bounds are tight. The second protocol considers sequential transmissions from the two users followed by a transmission from the relay while the third protocol is a hybrid of the first two protocols and has four phases. In the latter two protocols the bounds are not identical. Numerical evaluation shows that in some cases of interest our bounds do not differ significantly. Finally, in the Gaussian case with path loss, we derive achievable rates and compare the relative merits of each protocol. This case is of interest in cellular systems. Surprisingly, we find that in some cases, the achievable rate region of the four phase protocol contains points that are outside the outer bounds of the other two protocols.