We present results of an in‐situ geochemical study using laser‐ablation inductively coupled plasma–mass spectrometry (LA‐ICP‐MS) analyses along a ~4.3 cm long section across the K‐Pg event bed, ...drilled during IODP Expedition 342 at J Anomaly Ridge south of St. John's, Newfoundland. This section comprises the Maastrichtian with a sharp boundary to the graded, between 1.5 and 1.8 cm thick ejecta layer with totally altered impact glass spherules, which in turn is topped by Danian sediments. The porous and clayey material required elaborate preparation in order to yield reliable data. The ejecta bed shows a highly variable depletion in rare earth elements that even results in strongly subchondritic concentrations. The Ce/Ce* varies strongly (0.81–34), Ni/Cr ranges from 0.38 to 2.79. The maximum platinum group elements (PGE) concentrations are located in one LA‐spot exactly at the basis of the ejecta layer; they amount (in μg g−1) to 0.35 (Rh), 1.64 (Pd), 2.79 (Pt), and 0.86 (Au). The Nb/Ta ratio increases in the Ma from ~10 to 35.9 toward the ejecta horizon, which itself has higher Nb, Ta, Zr, and Hf concentrations than the background sedimentation, combined with low Nb/Ta (~5–10), and low Zr/Hf (~20–30). The overall result is that alteration processes changed totally the original geochemical characteristics of this K‐Pg spherule bed. To explain the exorbitant element mobility at distances of hundreds of μm, we discuss a combination of mostly reducing redox processes and interaction with organic compounds. This study demonstrates the high potential of in‐situ analyses with high spatial resolution at complex geological materials. Moreover, our results indicate that some caution is necessary in determining the projectile type in impactites via PGE ratios.
The presence of water (hydrogen) in nominally anhydrous mantle minerals may have profound effects on their physical properties (e.g., electrical conductivity, diffusivity, rheology), and these ...effects are expected to depend on how water is incorporated in the crystal structure. For olivine, the most abundant upper mantle mineral, despite extensive studies, mostly using vibrational spectroscopy, the interpretations are still not well constrained. To provide better understanding on this issue, we carried out a comprehensive 1H and 29Si NMR study on an Mg2SiO4 forsterite sample containing about 0.5 wt% H2O synthesized at 12 GPa and 1200 °C, complemented by Raman measurement and first-principles calculation of the geometry, stability, and NMR parameters of model structures. The Raman spectra contain relatively sharp O-H stretching bands near 3612, 3579, and 3567 cm-1 and a broader band near 3547 cm-1, similar to previous reports. The 1H static and MAS NMR data revealed that there are two main populations of protons in the hydrous forsterite structure, one experiencing strong 1H-1H homonuclear dipolar couplings and contributing to a broad peak near 2.4 ppm, and another with weaker dipolar couplings and contributing to a narrower peak near 1.2 ppm in the MAS NMR spectrum at 30 kHz. Two-dimensional 1H CRAMPS-MAS NMR measurements confirmed that the two proton components belong to the same phase and the contrast in MAS NMR peak width is largely due to difference in the strength of 1H-1H homonuclear dipolar couplings. In addition, there is also a very weak, narrow 1H MAS NMR peak near 7.3 ppm (contributing to <0.1% of the total intensity) due to protons that are more remote from the two main components. First-principles calculation confirmed that the two main proton components can be attributed to the hydrogarnet-like substitution mechanism of four H ions for one Si (4H)Si in a tetrahedral site of olivine, but unlike hydrogarnet with one of the protons pointing away from the tetrahedral center and located in an adjacent interstitial site, thus experiencing weaker dipolar couplings than those in the vicinity of the vacant tetrahedron; the very weak narrow peak near 7.3 ppm can be attributed to the substitution mechanism of two H ions for one Mg in an M1 site (2H)M1 of forsterite. The 1H-29Si CP-MAS NMR spectra revealed both a broad peak encompassing the position for OH defect-free forsterite (-61.7 ppm) and a narrower peak at higher frequency (-60.9 ppm). First-principles calculation indicates that these peaks are accountable by the same models as for the 1H NMR data. Thus, this study has provided unambiguous evidence supporting that hydrogen is incorporated in forsterite at relatively high-pressure dominantly as (4H)Si defects, with (2H)M1 defects playing only a very minor role. The much larger 1H chemical shift for protons associated with the latter (than the former) is correlated with stronger hydrogen bonding for the latter, which in turn reflects difference in bonding environments of the OH groups (with the latter bonded to a Si, and the former only bonded to Mg). Similar correlation applies to the O-H stretching frequency. The (4H)Si defects are responsible for the observed high-frequency O-H stretching bands (>3450 cm-1), and the (2H)M1 defects give lower frequencies (undetected here due to low abundance, but most likely near 3160-3220 cm-1 as previously reported) in vibrational spectra. These results can serve as a guide for (re-)interpretation of infrared and Raman spectroscopic data on hydrous olivine produced under different pressure and silica activity conditions, and require reconsideration of any models for the effects of water on physical properties of olivine based on different interpretations of such data. This study also demonstrated the usefulness of the combined solid-state NMR and first-principles calculation approach in unraveling the hydrogen incorporation mechanisms in nominally anhydrous minerals.
Deep learning is finding its way into the embedded world with applications such as autonomous driving, smart sensors and aug- mented reality. However, the computation of deep neural networks is ...demanding in energy, compute power and memory. Various approaches have been investigated to reduce the necessary resources, one of which is to leverage the sparsity occurring in deep neural networks due to the high levels of redundancy in the network parameters. It has been shown that sparsity can be promoted specifically and the achieved sparsity can be very high. But in many cases the methods are evaluated on rather small topologies. It is not clear if the results transfer onto deeper topologies. In this paper, the TensorQuant toolbox has been extended to offer a platform to investigate sparsity, especially in deeper models. Several practical relevant topologies for varying classification problem sizes are investigated to show the differences in sparsity for activations, weights and gradients.
Top gate, global back gate and buried gate CNTFET structures with a channel length of 5.9 nm are studied in the scope of the 2026 ITRS requirements. The studies are performed using a numerical device ...simulator. Figures of merit and performance parameters such as the switching speed, the switching energy, I on /I off -ratio, among others, are obtained for each structure and compared with the 2026 ITRS requirements for different application scenarios. Most of the requirements are met with the buried gate CNTFET. The requirement for the I on /I off -ratio is met at the cost of other performance parameters.
Recent research implies that training and inference of deep neural networks (DNN) can be computed with low precision numerical representations of the training/test data, weights and gradients without ...a general loss in accuracy. The benefit of such compact representations is twofold: they allow a significant reduction of the communication bottleneck in distributed DNN training and faster neural network implementations on hardware accelerators like FPGAs. Several quantization methods have been proposed to map the original 32-bit floating point problem to low-bit representations. While most related publications validate the proposed approach on a single DNN topology, it appears to be evident, that the optimal choice of the quantization method and number of coding bits is topology dependent. To this end, there is no general theory available, which would allow users to derive the optimal quantization during the design of a DNN topology. In this paper, we present a quantization tool box for the TensorFlow framework. TensorQuant allows a transparent quantization simulation of existing DNN topologies during training and inference. TensorQuant supports generic quantization methods and allows experimental evaluation of the impact of the quantization on single layers as well as on the full topology. In a first series of experiments with TensorQuant, we show an analysis of fix-point quantizations of popular CNN topologies.
HALF: Holistic Auto Machine Learning for FPGAs Ney, Jonas; Loroch, Dominik; Rybalkin, Vladimir ...
2021 31st International Conference on Field-Programmable Logic and Applications (FPL),
2021-Aug.
Conference Proceeding
Odprti dostop
Deep Neural Networks (DNNs) are capable of solving complex problems in domains related to embedded systems, such as image and natural language processing. To efficiently implement DNNs on a specific ...FPGA platform for a given cost criterion, e.g., energy efficiency, an enormous amount of design parameters must be considered from the topology down to the final hardware implementation. Interdependencies between the different design layers must be taken into account and explored efficiently, making it hardly possible to find optimized solutions manually. An automatic, holistic design approach can improve the quality of DNN implementations on FPGA significantly. To this end, we present a cross-layer design space exploration methodology. It comprises optimizations starting from a hardware-aware topology search for DNNs down to the final optimized implementation for a given FPGA platform. The methodology is implemented in our Holistic Auto machine Learning for FPGAs (HALF) framework, which combines an evolutionary search algorithm, various optimization steps, and a library of parametrizable hardware DNN modules. HALF automates both the exploration process and the implementation of optimized solutions on a target FPGA platform for various applications. We demonstrate the performance of HALF on a medical use case for arrhythmia detection for three different design goals, i.e., low-energy, low-power, and high-throughput. Our FPGA implementation outperforms a TensorRT optimized model on an Nvidia Jetson platform in both throughput and energy consumption.
Deep Neural Networks (DNNs) are capable of solving complex problems in domains related to embedded systems, such as image and natural language processing. To efficiently implement DNNs on a specific ...FPGA platform for a given cost criterion, e.g. energy efficiency, an enormous amount of design parameters has to be considered from the topology down to the final hardware implementation. Interdependencies between the different design layers have to be taken into account and explored efficiently, making it hardly possible to find optimized solutions manually. An automatic, holistic design approach can improve the quality of DNN implementations on FPGA significantly. To this end, we present a cross-layer design space exploration methodology. It comprises optimizations starting from a hardware-aware topology search for DNNs down to the final optimized implementation for a given FPGA platform. The methodology is implemented in our Holistic Auto machine Learning for FPGAs (HALF) framework, which combines an evolutionary search algorithm, various optimization steps and a library of parametrizable hardware DNN modules. HALF automates both the exploration process and the implementation of optimized solutions on a target FPGA platform for various applications. We demonstrate the performance of HALF on a medical use case for arrhythmia detection for three different design goals, i.e. low-energy, low-power and high-throughput respectively. Our FPGA implementation outperforms a TensorRT optimized model on an Nvidia Jetson platform in both throughput and energy consumption.
The SARS-CoV-2/COVID-19 pandemic has inflicted medical and socioeconomic havoc, and despite the current availability of vaccines and broad implementation of vaccination programs, more easily ...accessible and cost-effective acute treatment options preventing morbidity and mortality are urgently needed. Herbal teas have historically and recurrently been applied as self-medication for prophylaxis, therapy, and symptom alleviation in diverse diseases, including those caused by respiratory viruses, and have provided sources of natural products as basis for the development of therapeutic agents. To identify affordable, ubiquitously available, and effective treatments, we tested herbs consumed worldwide as herbal teas regarding their antiviral activity against SARS-CoV-2. Aqueous infusions prepared by boiling leaves of the Lamiaceae perilla and sage elicit potent and sustained antiviral activity against SARS-CoV-2 when applied after infection as well as prior to infection of cells. The herbal infusions exerted in vitro antiviral effects comparable to interferon-beta and remdesivir but outperformed convalescent sera and interferon-alpha2 upon short-term treatment early after infection. Based on protein fractionation analyses, we identified caffeic acid, perilla aldehyde, and perillyl alcohol as antiviral compounds. Global mass spectrometry (MS) analyses performed comparatively in two different cell culture infection models revealed changes of the proteome upon treatment with herbal infusions and provided insights into the mode of action. As inferred by the MS data, induction of heme oxygenase 1 (HMOX-1) was confirmed as effector mechanism by the antiviral activity of the HMOX-1-inducing compounds sulforaphane and fraxetin. In conclusion, herbal teas based on perilla and sage exhibit antiviral activity against SARS-CoV-2 including variants of concern such as Alpha, Beta, Delta, and Omicron, and we identified HMOX-1 as potential therapeutic target. Given that perilla and sage have been suggested as treatment options for various diseases, our dataset may constitute a valuable resource also for future research beyond virology.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK