Search

In deep neural network (DNN) accelerators, it is expensive to transfer model parameters from the main memory to the processing elements. Data movement accounts for a large number of the inference latency and energy consumption. In this paper, we present three position-based techniques to compress the DNN model parameters. The techniques could lead to significant energy and performance improvement. The three presented compression techniques are lossless. The first technique takes into consideration the regularly repeat property of the DNN weights to compress them. The second technique saves the relative distance between weights instead of the weights to compress the model. The third technique applies Huffman coding on the relative distance based on the second technique. The proposed techniques are assessed on several DNNs. The results show that, the first technique could decrease 38% of latency and 36% energy, respectively. The second technique could decrease 41% of latency and 39% energy, respectively. The third technique could decrease 45% of latency and 43% energy, respectively. Applying Huffman code could achieve additional 7% reduction in both latency and energy based on the second technique. [ABSTRACT FROM AUTHOR]

The original article can be found online at https://doi.org/10.1007/s11227-023-05339-4.Correction to: The Journal of Supercomputinghttps://doi.org/10.1007/s11227-023-05339-4In this article, the author name Minghua Tang was incorrectly written as Minging Tang.The original article has been corrected.Publisher's NoteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.By Minghua Tang; Enrico Russo and Maurizio PalesiReported by Author; Author; Author [Extracted from the article]

A transparent, flexible graphene field effect transistor (GFET) based on ferroelectric gate is demonstrated. In this device, the single layer graphene was fabricated by chemical vapor deposition method, and transferred to the polyethylene terephthalate substrate. Then the poly(vinylidene fluoride-trifluoroethylene) (P(VDF-TrFE)) ferroelectric polymer layer film was used as the gate dielectric for the graphene FET. Based on the P(VDF-TrFE)/graphene heterojunction FET, Hall Bar structure was fabricated. The transport properties of the graphene channel at low temperature and retention characteristics at different temperature are investigated in detail. These special properties indicated that the GFET might be useful for many particular applications, such as a non-volatile memories, flexible electronic devices and phototransistors. [ABSTRACT FROM AUTHOR]

Although health benefits of the Dietary Approaches to Stop Hypertension (DASH) diet are established, it is not understood which food compounds result in these benefits. We used metabolomics to identify unique compounds from individual foods of a DASH-style diet and determined if these Food-Specific Compounds (FSC) are detectable in urine from participants in a DASH-style dietary study. We also examined relationships between urinary compounds and blood pressure (BP). Nineteen subjects were randomized into 6-week controlled DASH-style diet interventions. Mass spectrometry-based metabolomics was performed on 24-hour urine samples collected before and after each intervention and on 12 representative DASH-style foods. Between 66–969 compounds were catalogued as FSC; for example, 4-hydroxydiphenylamine was found to be unique to apple. Overall, 13–190 of these FSC were detected in urine, demonstrating that these unmetabolized food compounds can be discovered in urine using metabolomics. Although linear mixed effects models showed no FSC from the 12 profiled foods were significantly associated with BP, other endogenous and food-related compounds were associated with BP (N = 16) and changes in BP over time (N = 6). Overall, this proof of principle study demonstrates that metabolomics can be used to catalog FSC, which can be detected in participant urine following a dietary intervention. [ABSTRACT FROM AUTHOR]

Lanthanum-doped ZnO (ZnLaO) polycrystalline thin films were deposited on Pt/Ti/SiO/Si substrates by a chemical solution deposition method. Metal/La-doped ZnO/Pt sandwich structures were constructed by depositing different top electrodes (Ag and Pt). Unipolar switching and bipolar switching characteristics were investigated in Pt/La-doped ZnO/Pt and Ag/La-doped ZnO/Pt structures, respectively. Compared with the undoped devices (Pt/ZnO/Pt and Ag/ZnO/Pt), the La-doped devices exhibits superior resistive switching performances, such as narrow distribution of the resistive switching properties ( R, R, V, and V), higher R/ R ratio and sharp switching transition. [ABSTRACT FROM AUTHOR]