Search

Implicit neural representations such as Neural Radiance Field (NeRF) have focused mainly on modeling static objects captured under multi-view settings where real-time rendering can be achieved with smart data structures, e.g., PlenOctree. In this paper, we present a novel Fourier PlenOctree (FPO) technique to tackle efficient neural modeling and real-time rendering of dynamic scenes captured under the free-view video (FVV) setting. The key idea in our FPO is a novel combination of generalized NeRF, PlenOctree representation, volumetric fusion and Fourier transform. To accelerate FPO construction, we present a novel coarse-to-fine fusion scheme that leverages the generalizable NeRF technique to generate the tree via spatial blending. To tackle dynamic scenes, we tailor the implicit network to model the Fourier coefficients of timevarying density and color attributes. Finally, we construct the FPO and train the Fourier coefficients directly on the leaves of a union PlenOctree structure of the dynamic sequence. We show that the resulting FPO enables compact memory overload to handle dynamic objects and supports efficient fine-tuning. Extensive experiments show that the proposed method is 3000 times faster than the original NeRF and achieves over an order of magnitude acceleration over SOTA while preserving high visual quality for the free-viewpoint rendering of unseen dynamic scenes., Project page: https://aoliao12138.github.io/FPO/

Energy Internet is regarded as the development form of smart grid in the future by realizing horizontal multi energy complementary and vertical highly coordinated "generation-grid-load-storage". In order to evaluate the effectiveness of regional energy Internet optimization, firstly, the energy sources and energy routers in the regional energy Internet are modeled. Then, the evaluation index of multi energy collaborative optimization is proposed from three aspects of economy, power loss and environmental benefits. Finally, based on a five-node regional energy Internet, the economic index are optimized and calculated, and the results verify the effectiveness of the proposed indexes.

This paper studies the characteristic of MCUs induced by monoenergetic and quasi-monoenergetic low-energy neutrons in 65-nm SRAM. This work is highlighted with the characterized dependence of the MCU cross section on the neutron energy, whereas the evaluated energy dependence of the SEU cross section was consistent with the literature. Both the MCU cross section and its proportion to the total events increase with higher neutron energy. At the onset energy of 6.0 MeV, the MCU proportion is 51 %, while the maximum value of 66 % is at 70 MeV. On the other hand, even at a lower energy of 4.1 MeV, the MCU proportion keeps around 50 %. The high MCU proportion over the wide range of neutron energy suggests the possibility of large-LET secondary ions being generated even by 4.1 MeV neutrons. Also, based on the measured MCU dependence, we calculated the contribution of the low-energy neutron to the total SER of MCU. The calculated result indicates that the low-energy neutrons below 10 MeV are mostly negligible in the terrestrial environment.

The impacts of hydrided and non-hydrided materials near transistors on neutron-induced single event upsets (SEUs) were investigated by simulating monoenergetic neutron irradiations on 65-nm technology bulk static random access memories. The onset energy of the SEUs induced by H ions depends on the shielding capability, i.e., the material and thickness, of components placed in front of transistors when those components do not contain hydrogen atoms. The shielding capability also influences the initial slope observed in the energy-dependence of SEU cross sections. Taking into account the non-hydrided component attached to memory cells used in the simulation, all experimental data measured at each neutron facility were reproduced well using SEU cross sections obtained by simulation. We also find that the effect of components near transistors on neutron-induced soft error rates is not negligible even for irradiation by white neutrons.

This paper discusses soft errors in digital chips consisting of SRAM, flip-flops, and combinational logic in the terrestrial environment. We review the effectiveness of error-correction coding (ECC) in processor systems and point out the importance of radiation-hardened flip-flops for further error mitigation. The discussion includes the difference between planar and FD-SOI transistors, and the type of secondary cosmic rays including neutron and muon, using irradiation test results. Also, the difficulty in characterizing SER of a commercial GPU chip is exemplified.

We propose an autonomous driving system using programmable SoCs. Our system consists of two FPGA boards of Zybo Z7-20, three cameras, and one motor driver of H-bridge circuits with independent left and right motors. The main part of our system is divided into two blocks: driving control and object detection, which are implemented on each Zybo Z7-20. Within the development framework of programmable SoC of Zynq 7000, we adopt a HW/SW co-design to blance the design period and system performance to satisify the goals of the FPGA design competition.

In this paper, we compare the negative and positive muon-induced SEU event cross sections of 28-nm and 65-nm planar bulk CMOS SRAMs. Our measurement results show a 3.6 X increase in muon-induced SEU event cross section from 65-nm to 28-nm technology, and negative muon-induced SEU event cross section is 3.3 X larger compared to positive muons at 28-nm technology. This result is consistent with the previous works reporting muon-induced SEU event cross section increases with technology scaling. The measured result also suggests the contribution of direct ionization to the total SEU event cross section is 54.1% at 28-nm node with operating voltage of 0.6 V while it is 1.8% at 65-nm node with 0.9 V.

This paper comparatively discusses soft error immunity of silicon on thin BOX (SOTB) SRAM and conventional bulk SRAM presenting neutron- and muon-induced soft error rates (SER) in 65-nm 6T SRAM over a wide range of supply voltages. The results show that the neutron-induced multiple cell upset (MCU) rates of SOTB at 0.4 V and 1.0 V are 0.01 times and 0.003 times lower than those of bulk at 0.4 V and 1.0 V, respectively. In advanced bulk low-voltage SRAM, protons can be dominant secondary particles, which brings drastic SER elevation, but this will not arise in SOTB SRAM due to its thin SOI layer. We also characterized the immunity to positive and negative muons. We observed negative muon induced more upsets due to muon capture process. We also confirmed that SOTB SRAM was more robust to muon than bulk SRAM.

We have proposed a neutron-induced soft error rate (SER) estimation method that incorporates machine learning with Monte Carlo radiation transport simulation. Multiple sensitive volumes based machine learning discriminator makes fast SER estimation possible for a unit circuit (e.g. SRAM cell) consisting of several transistors. The discriminator takes charges deposited by a secondary ion to individual volumes of all the transistors as input and outputs the discrimination result, i.e. upset or non-upset. Supervised learning with the training data obtained by TCAD simulations constructs the discriminator. This paper discusses the discriminator construction for 65-nm ultra-thin-box FD-SOI SRAM with TCAD. We experimentally demonstrate the multiple sensitive volumes assignment is useful for building a precise discriminator. We also discuss the critical volumes and transistors for discriminator performance.

Feature Extraction for sEMG signals is the key technique of the sEMG-based recognition system. More and more different types of features are extracted at present. Aiming at the problem of blind selection and low information utilization rate of features, an improved extraction method based on the multiple features combination and dimension reduction is proposed by this paper to build new features with high information proportion to promote the accuracy of recognition system. The basic sEMG features of time domain and AR model coefficients are used in this paper to be projected on the direction of largest error via PCA method, then the proportion and cumulative of the components on each direction are calculated for selecting the main components that contain most information to rebuild the feature vectors while reducing the dimension of the vectors. In the process of extraction, aiming at the contradiction between the stable principles of the signal in short time window and fluctuation of feature values, the length of the time window is optimized. Experiment shows that the feature extraction method proposed in this paper can promote the accuracy of recognition, which proves the efficiency of the improved method.

With respect to the ill-posed problem when calculating output weights of the ELM (Extreme Learning Machine), an improved ELM algorithm based on TSVD (Truncated Singular Value Decomposition) is proposed in this paper. The degree of ill-condition is severe if the hidden layer output matrix has a large condition number. In such case, the output weights computed by general SVD (Singular Value Decomposition) method will be large and unevenly distributed, which would result in a worsened stability and anti-interference ability. Also, the over-fitting phenomenon presented easily. TSVD is an effective regularization method. It can eliminate the influence caused by small singular values and enhance the generalization ability of the model. As for selecting truncation parameter, it is determined by minimizing the GCV (Generalized Cross-Validation) function with the relationship between TSVD and Tikhnovo Regularization. Simulation results illustrate that TSVD-ELM performs higher prediction accuracy than original ELM on data with noise and increases the model's robustness. Finally, the proposed method is used to build a soft-sensor model to predict the quality of iron ore pellet and gets an acceptable error rate.

We propose a novel solid-state electrical ammonia sensor based on a vertical organic diode with high-density nanopores in the top electrode. The proposed ammonia sensor has a detection limit of 20 ppb, a real-time response, sufficient selectivity, simple structure, high reproducibility, and a low production cost. In addition, we used the proposed sensor to detect the nasal breath of rats directly. After controlling the relative humidity and filtering out the CO 2 , we successfully obtained repeated real-time, reliable responses. This is the first demonstration of a solid-state sensor having reliable responses to the breath ammonia of an animal. The sensing response between normal and acute groups is significantly different (p < 0.05). These results demonstrated that the proposed ammonia sensor can detect liver that undergoes moderate hepatopathy.

The DWDM-PON transmission of coherently injection-locked WRC-FPLDs with ER (7.55 dB), SNR (7.6 dB) and Q factor (6.7) improved by 4.6%, 25.6% and 13.6% are demonstrated by enlarging AR coating reflectance from 0.5% to 1.2%.

This paper presents the design and domination of single-joint driving system of bionic finger based on shape memory alloy (SMA). A pre-shaped SMA wire is employed as the finger skeleton and the joint. To realize bending and stretching the proposed finger flexibly, a couple of thermoelectric devices (TEDs), being as a drive, touch a certain location of the SMA wire which acts as the joint of the finger heating and cooling the wire directly. In order to control angles of the finger joint accurately and rapidly, the model of relationship between the temperature and the corresponding angle of the finger are developed. A control strategy based PI control is discussed to improve the dynamic response time of the system. Several experiments are used to evaluate the dynamic and steady state performances, and the results show that the proposed driving system controls bend and stretch of the finger with satisfied static accuracy and rapid dynamic response.