Search

Computationally intensive Inference tasks of Deep neural networks have enforced revolution of new accelerator architecture to reduce power consumption as well as latency. The key figure of merit in hardware inference accelerators is the number of multiply-and-accumulation operations per watt (MACs/W), where, the state-of-the-arts MACs/W remains several hundreds Giga-MACs/W. We propose a Tera-MACS/W neural hardware inference Accelerator (TMA) with 8-bit activations and scalable integer weights less than 1-byte. The architectures main feature is configurable neural processing element for matrix-vector operations. The proposed neural processing element has Multiplier-less Massive Parallel Processor to work without any multiplications, which makes it attractive for energy efficient high-performance neural network applications. We benchmark our systems latency, power, and performance using Alexnet trained on ImageNet. Finally, we compared our accelerators throughput and power consumption to the prior works. The proposed accelerator outperforms the state of the art in terms of energy and area achieving 2.3 TMACS/W@1.0 V, 65 nm CMOS technology.

For newborns suspected having childhood interstitial lung disease (ChILD), the sequencing of genes encoding surfactant proteins is recommended. However, it is still difficult to interpret the clinical significance of those variants found. We report a full-term born female infant who presented with respiratory distress and failure to thrive at 2 months of age and both imaging and lung biopsy were consistent with ChILD. Her genetic test was initially reported as a variant of unknown significance in surfactant protein C (c.202G > T, p.V68F), which was modified later as likely pathogenic after reviewing a report of the same variant as causing ChILD. The infant was placed on noninvasive ventilation and treated with IV Methylprednisolone, Hydroxychloroquine, and Azithromycin but did not show significant clinical and radiological improvement underwent tracheostomy and is awaiting lung transplantation at 8 months of age. The challenges interpreting the genetic results are discussed.

For newborns suspected having childhood interstitial lung disease (ChILD), the sequencing of genes encoding surfactant proteins is recommended. However, it is still difficult to interpret the clinical significance of those variants found. We report a full-term born female infant who presented with respiratory distress and failure to thrive at 2 months of age and both imaging and lung biopsy were consistent with ChILD. Her genetic test was initially reported as a variant of unknown significance in surfactant protein C (c.202G T, p.V68F), which was modified later as likely pathogenic after reviewing a report of the same variant as causing ChILD. The infant was placed on noninvasive ventilation and treated with IV Methylprednisolone, Hydroxychloroquine, and Azithromycin but did not show significant clinical and radiological improvement underwent tracheostomy and is awaiting lung transplantation at 8 months of age. The challenges interpreting the genetic results are discussed.

Study Objectives Obstructive sleep apnea (OSA) is characterized by recurrent airway collapse during sleep, resulting in intermittent hypoxia and sleep fragmentation that may contribute to alternations in brain structure and function. We hypothesized that OSA in children reorganizes and alters cortical structure, which can cause changes in cortical thickness correlation between brain regions across subjects. Methods We constructed cortical structure networks based on cortical thickness measurements from 41 controls (age 15.54 ± 1.66 years, male 19) and 50 children with OSA (age 15.32 ± 1.65 years, male 29). The global (clustering coefficient [CC], path length, and small-worldness) and regional (nodal betweenness centrality, NBC) network properties and hub region distributions were examined between groups. Results We found increased CCs in OSA compared to controls across a wide range of network densities (p-value Conclusions Our findings suggest that children with OSA exhibit altered global and regional network characteristics compared to healthy controls. Our approach to the investigation of cortical structure in children with OSA could prove useful in understanding the etiology of OSA-related brain functional disorders.

Recent progress in parallel computing machines, deep neural networks, and training techniques have contributed to the significant advances in artificial intelligence (AI) with respect to tasks such as object classification, speech recognition, and natural language processing. The development of such deep learning-based techniques has enabled AI-based networks to outperform humans in the recognition of objects in images. The graphics processing unit (GPU) has been the primary component used for parallel computing during the inference and training phases of deep neural networks. In this study, we perform training using a desktop or a server with one or more GPUs and inference using hardware accelerators on embedded devices. Performance, power consumption, and requirements of embedded system present major hindrances to the application of deep neural network-based systems using embedded controllers such as drones, AI speakers, and autonomous vehicles. In particular, power consumption of a commercial GPU commonly surpasses the power budget of a stand-alone embedded system. To reduce the power consumption of hardware accelerators, reductions in the precision of input data and hardware weight have become popular topics of research in this field. However, precision and accuracy share a trade-off relationship. Therefore, it is essential to optimize precision in a manner that does not degrade the accuracy of the inference process. In this context, the primary issues faced by hardware accelerators are loss of accuracy and high power consumption.

Advances in the architecture of inference accelerators and quantization techniques of neural networks allow effective on-device inference in embedded devices. Privacy issues for user data, as well as increasing needs of user-specific services, have led to a need for on-device training. The dot product operation required in backpropagation can be computed efficiently by multiplier–accumulators (MACs) in the inference accelerator if forward and backward propagation of the neural network have the same precision. This chapter introduces a quantization technique to enable computation by the digital neuron inference accelerator with the same precision as that using the forward path. Updating the 5-bit weights with gradients of higher precision is challenging. To address this issue, this chapter also introduces a probabilistic weight update. It also describes the hardware implementation of the probabilistic weight-update scheme. The proposed training technique achieves 98.15% recognition accuracy on the MNIST dataset.

We propose a single inductor multiple output (SIMO) auto-buck-boost DC&ndash, DC converter with error-driven randomized control (EDRC). The conventional controls in a SIMO DC&ndash, DC converter supply power to outputs that have been selected in a sequential order. Furthermore, they control the inductor current levels at either edge of a switching period in a steady state to be at the same level to alleviate cross-regulation. However, this limits the flexibility of the converter to respond to changes in load requirements. A sequential selection of light loads results in these loads being selected more often than a load demand, degrading the efficiency for light loads. In addition, limited flexibility leads to delayed responses. This paper introduces an auto-buck-boost topology that selects outputs based on output errors, and instantaneously adjusts the inductor current level. Moreover, we propose a technique for allowing any output to avoid selection when all outputs are fully supplied. The proposed EDRC scheme achieves improvements in efficiency in regards to light loads, cross-regulation, and output driving capability.

We have presented both modeling and a method for extracting parasitic thermal conductance as well as intrinsic device parameters of a thermoelectric module based on information readily available in vendor datasheets. An equivalent circuit model that is compatible with circuit simulators is derived, followed by a methodology for extracting both intrinsic and parasitic model parameters. For the first time, the effective thermal resistance of the ceramic and copper interconnect layers of the thermoelectric module is extracted using only parameters listed in vendor datasheets. In the experimental condition, including under condition of varying electric current, the parameters extracted from the model accurately reproduce the performance of commercial thermoelectric modules.

We propose a way of achieving maximum power and power-transfer efficiency from thermoelectric generators by optimized selection of maximum-power-point-tracking (MPPT) circuits composed of a boost-cascaded-with-buck converter. We investigated the effect of switch resistance on the MPPT performance of thermoelectric generators. The on-resistances of the switches affect the decrease in the conversion gain and reduce the maximum output power obtainable. Although the incremental values of the switch resistances are small, the resulting difference in the maximum duty ratio between the input and output powers is significant. For an MPPT controller composed of a boost converter with a practical nonideal switch, we need to monitor the output power instead of the input power to track the maximum power point of the thermoelectric generator. We provide a design strategy for MPPT controllers by considering the compromise in which a decrease in switch resistance causes an increase in the parasitic capacitance of the switch.

We propose an output error-driven random controlled Single Inductor Multiple Output (SIMO) buck/boost DC-DC converter operated in the Continuous Conduction Mode (CCM). Random selection of output switches based on maximum instantaneous output error enables us a novel control scheme which leads to a fast transient response with mitigating the cross-regulation problem of buck-boost converters. A freewheeling period for CCM control is not necessary in the proposed output error-driven control scheme thanks to the instantaneous control of output switches, hence, we can achieve high power conversion efficiency as well as stability of operation. Both simulated and experimental results show the effectiveness of the proposed control scheme. The measured power conversion efficiency of the converter with 3 buck/boost output nodes is approximately 91.0%.

The present paper demonstrates a feasible real-time path planning algorithm and tracking method for autonomous driving vehicles based on a 5th-order Bezier curve. It is demonstrated that a 5th-degree Bezier curve is required to generate a feasible path to achieve lane change maneuvers in a dynamic on-road environment. The steering rate control input is calculated based on the curvature extracted from the generated trajectory. Matlab simulations were conducted both for path generation and instantaneous planning of driving path for single and double lane change maneuvers. The proposed method was evaluated in an experiment using a miniature-sized unmanned ground vehicle.

Evodiamine (evo) has been shown to exert anti-inflammatory, antinociceptive and anticancer effects. In this study, we investigated the effects of evo alone and in combination with rosiglitazone (rosi) on in vitro adipocyte differentiation and in vivo obesity related to diabetes.Adipocyte differentiation was investigated in vitro using 3T3-L1 and C3H10T1/2 cells. To determine the degree of differentiation, Oil Red O staining and reverse transcription-PCR were carried out. Four groups of db/db mice were treated intraperitoneally once per day with vehicle, evo, rosi and evo+rosi. The mice were killed after 14 days and the blood, liver and adipose tissue were analyzed.The presence of evo or evo combined with rosi during adipogenic induction has been shown to inhibit adipocyte differentiation to a significant degree, particularly at the commitment and early induction stages. The evo and evo+rosi groups of db/db mice evidenced significant reductions in body weight gain. The ratio of epididymal white adipocyte tissue weight to body weight of the evo group was also significantly reduced. It is important to note that in the evo+rosi treatment, blood glucose levels were reduced to a degree similar to that of the rosi group, and plasma insulin levels were reduced significantly better than that of rosi group. Furthermore, hepatic lesions associated with fat and glycogen deposition were morphologically improved in the evo and evo+rosi groups.The results of this study showed that evo exerts an inhibitory effect on in vitro adipocyte differentiation and in vivo obesity, and also an improvement effect on insulin resistance. These desirable effects of evo were noted even in the presence of rosi. These results indicate that evo improves the undesirable effects of rosi, including adipogenesis, body weight gain and hepatotoxicity, while preserving its desirable blood-glucose-lowering effect.

openaire: EC/H2020/825496/EU//5G-MOBIX National roaming, multi-SIM and edge computing constitute key 5G technologies for the cooperative perception and remote driving of L4 (automated) vehicles. To that end, this article reports our progress to trial these technologies at the multi-PLMN experimental 5G SA testbed of Aalto University, Finland. Overall, the objective is to qualify 5G as a core connectivity for connected, cooperative and automated mobility.