Activation Redistribution Based Hybrid Asymmetric Quantization Method of Neural Networks

The demand for adopting neural networks in resource-constrained embedded devices is continuously increasing.Quantization is one of the most promising solutions to reduce computational cost and memory storage on embedded devices.In order to reduce the complexity and overhead of deploying neural networks on Integeronly hardware,most current quantization methods use a symmetric quantization mapping strategy to quantize a floating-point neural network into an integer network.However,although symmetric quantization has the advantage of easier implementation,it is sub-optimal for cases where the range could be skewed and not symmetric.This often comes at the cost of lower accuracy.This paper proposed an activation redistribution-based hybrid asymmetric quantizationmethod for neural networks.The proposedmethod takes data distribution into consideration and can resolve the contradiction between the quantization accuracy and the ease of implementation,balance the trade-off between clipping range and quantization resolution,and thus improve the accuracy of the quantized neural network.The experimental results indicate that the accuracy of the proposed method is 2.02%and 5.52%higher than the traditional symmetric quantization method for classification and detection tasks,respectively.The proposed method paves the way for computationally intensive neural network models to be deployed on devices with limited computing resources.Codes will be available on https://github.com/ycjcy/Hybrid-Asymmetric-Quantization.

Reinforcement Learning Based Quantization Strategy Optimal Assignment Algorithm for Mixed Precision

The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to databit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determinethe optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantizationcan effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In thispaper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bitwidth is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets theconstraints of hardware resources. In the state-space design, the standard deviation of weights is used to measurethe distribution difference of data, the execution speed feedback of simulated neural network accelerator inferenceis used as the environment to limit the action space of the agent, and the accuracy of the quantization model afterretraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. Theexperimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategyunder the condition that the computational resources are satisfied, and themodel accuracy is effectively improved.The proposed method has strong intelligence and certain universality and has strong application potential in thefield of mixed precision quantization and embedded neural network model deployment.

Oscillation of Dzyaloshinskii–Moriya interaction driven by weak electric fields

Dzyaloshinskii–Moriya interaction(DMI) is under extensive investigation considering its crucial status in chiral magnetic orders, such as Néel-type domain wall(DW) and skyrmions. It has been reported that the interfacial DMI originating from Rashba spin–orbit coupling(SOC) can be linearly tuned with strong external electric fields. In this work, we experimentally demonstrate that the strength of DMI exhibits rapid fluctuations, ranging from 10% to 30% of its original value, as a function of applied electric fields in Pt/Co/MgO heterostructures within the small field regime(< 10-2V/nm). Brillouin light scattering(BLS) experiments have been performed to measure DMI, and first-principles calculations show agreement with this observation, which can be explained by the variation in orbital hybridization at the Co/MgO interface in response to the weak electric fields. Our results on voltage control of DMI(VCDMI) suggest that research related to the voltage control of magnetic anisotropy for spin–orbit torque or the motion control of skyrmions might also have to consider the role of the external electric field on DMI as small voltages are generally used for the magnetoresistance detection.

Improvement of Ga_(2)O_(3)vertical Schottky barrier diode by constructing NiO/Ga_(2)O_(3)heterojunction

The high critical electric field strength of Ga_(2)O_(3)enables higher operating voltages and reduced switching losses in power electronic devices.Suitable Schottky metals and epitaxial films are essential for further enhancing device performance.In this work,the fabrication of vertical Ga_(2)O_(3)barrier diodes with three different barrier metals was carried out on an n--Ga_(2)O_(3)homogeneous epitaxial film deposited on an n+-β-Ga_(2)O_(3)substrate by metal-organic chemical vapor deposition,excluding the use of edge terminals.The ideal factor,barrier height,specific on-resistance,and breakdown voltage characteristics of all devices were investigated at room temperature.In addition,the vertical Ga_(2)O_(3)barrier diodes achieve a higher breakdown volt-age and exhibit a reverse leakage as low as 4.82×10^(-8)A/cm^(2)by constructing a NiO/Ga_(2)O_(3)heterojunction.Therefore,Ga_(2)O_(3)power detailed investigations into Schottky barrier metal and NiO/Ga_(2)O_(3)heterojunction of Ga_(2)O_(3)homogeneous epitaxial films are of great research potential in high-efficiency,high-power,and high-reliability applications.

A Secure and Cost-Effective Training Framework Atop Serverless Computing for Object Detection in Blasting

The data analysis of blasting sites has always been the research goal of relevant researchers.The rise of mobile blasting robots has aroused many researchers’interest in machine learning methods for target detection in the field of blasting.Serverless Computing can provide a variety of computing services for people without hardware foundations and rich software development experience,which has aroused people’s interest in how to use it in the field ofmachine learning.In this paper,we design a distributedmachine learning training application based on the AWS Lambda platform.Based on data parallelism,the data aggregation and training synchronization in Function as a Service(FaaS)are effectively realized.It also encrypts the data set,effectively reducing the risk of data leakage.We rent a cloud server and a Lambda,and then we conduct experiments to evaluate our applications.Our results indicate the effectiveness,rapidity,and economy of distributed training on FaaS.

Heavy ion energy influence on multiple-cell upsets in small sensitive volumes:from standard to high energies

The 28 nm process has a high cost-performance ratio and has gradually become the standard for the field of radiation-hardened devices.However,owing to the minimum physical gate length of only 35 nm,the physical area of a standard 6T SRAM unit is approximately 0.16μm^(2),resulting in a significant enhancement of multi-cell charge-sharing effects.Multiple-cell upsets(MCUs)have become the primary physical mechanism behind single-event upsets(SEUs)in advanced nanometer node devices.The range of ionization track effects increases with higher ion energies,and spacecraft in orbit primarily experience SEUs caused by high-energy ions.However,ground accelerator experiments have mainly obtained low-energy ion irradiation data.Therefore,the impact of ion energy on the SEU cross section,charge collection mechanisms,and MCU patterns and quantities in advanced nanometer devices remains unclear.In this study,based on the experimental platform of the Heavy Ion Research Facility in Lanzhou,low-and high-energy heavy-ion beams were used to study the SEUs of 28 nm SRAM devices.The influence of ion energy on the charge collection processes of small-sensitive-volume devices,MCU patterns,and upset cross sections was obtained,and the applicable range of the inverse cosine law was clarified.The findings of this study are an important guide for the accurate evaluation of SEUs in advanced nanometer devices and for the development of radiation-hardening techniques.

Machine learning method to predict dynamic compressive response of concrete-like material at high strain rates

Machine learning(ML)methods with good applicability to complex and highly nonlinear sequences have been attracting much attention in recent years for predictions of complicated mechanical properties of various materials.As one of the widely known ML methods,back-propagation(BP)neural networks with and without optimization by genetic algorithm(GA)are also established for comparisons of time cost and prediction error.With the aim to further increase the prediction accuracy and efficiency,this paper proposes a long short-term memory(LSTM)networks model to predict the dynamic compressive performance of concrete-like materials at high strain rates.Dynamic explicit analysis is performed in the finite element(FE)software ABAQUS to simulate various waveforms in the split Hopkinson pressure bar(SHPB)experiments by applying different stress waves in the incident bar.The FE simulation accuracy is validated against SHPB experimental results from the viewpoint of dynamic increase factor.In order to cover more extensive loading scenarios,60 sets of FE simulations are conducted in this paper to generate three kinds of waveforms in the incident and transmission bars of SHPB experiments.By training the proposed three networks,the nonlinear mapping relations can be reasonably established between incident,reflect,and transmission waves.Statistical measures are used to quantify the network prediction accuracy,confirming that the predicted stress-strain curves of concrete-like materials at high strain rates by the proposed networks agree sufficiently with those by FE simulations.It is found that compared with BP network,the GA-BP network can effectively stabilize the network structure,indicating that the GA optimization improves the prediction accuracy of the SHPB dynamic responses by performing the crossover and mutation operations of weights and thresholds in the original BP network.By eliminating the long-time dependencies,the proposed LSTM network achieves better results than the BP and GA-BP networks,since smaller mean square error(MSE)and higher correlation coefficient are achieved.More importantly,the proposed LSTM algorithm,after the training process with a limited number of FE simulations,could replace the time-consuming and laborious FE pre-and post-processing and modelling.

Topological magnetotransport and electrical switching of sputtered antiferromagnetic Ir_(20)Mn_(80)

Topological magnetotransport in non-collinear antiferromagnets has attracted extensive attention due to the exotic phenomena such as large anomalous Hall effect(AHE),magnetic spin Hall effect,and chiral anomaly.The materials exhibiting topological antiferromagnetic physics are typically limited in special Mn_3X family such as Mn_3Sn and Mn_3Ge.Exploring the topological magnetotransport in common antiferromagnetic materials widely used in spintronics will not only enrich the platforms for investigating the non-collinear antiferromagnetic physics,but also have great importance for driving the nontrivial topological properties towards practical applications.Here,we report remarkable AHE,anisotropic and negative parallel magnetoresistance in the magnetron-sputtered Ir_(20)Mn_(80)antiferromagnet,which is one of the most widely used antiferromagnetic materials in industrial spintronics.The ab initio calculations suggest that the Ir_4Mn_(16)(IrMn_4)or Mn_3Ir nanocrystals hold nontrivial electronic band structures,which may contribute to the observed intriguing magnetotransport properties in the Ir_(20)Mn_(80).Further,we demonstrate the spin–orbit torque switching of the antiferromagnetic Ir_(20)Mn_(80)by the spin Hall current of Pt.The presented results highlight a great potential of the magnetron-sputtered Ir_(20)Mn_(80)film for exploring the topological antiferromagnet-based physics and spintronics applications.

High repetition granular Co/Pt multilayers with improved perpendicular remanent magnetization for high-density magnetic recording

Thanks to the strong perpendicular magnetic anisotropy(PMA), excellent processing compatibility as well as novel spintronic phenomenon, Co/Pt multilayers have been attracting massive attention and widely used in magnetic storage.However, reversed magnetic domains come into being with the increasing layer repetition ‘N’ to reduce magneto-static energy, resulting in the remarkable diminishment of the remanent magnetization(Mr). As a result, the product of Mrand thickness(i.e., the remanent moment-thickness product, Mrt), a key parameter in magnetic recording for reliable data storing and reading, also decreases dramatically. To overcome this issue, we deposit an ultra-thick granular [Co/Pt]80multilayer with a total thickness of 68 nm on granular SiNxbuffer layer. The Mrt value, Mrto saturation magnetization(Ms) ratio as well as out of plane(OOP) coercivity(Hcoop) are high up to 2.97 memu/cm^(2), 67%, and 1940 Oe(1 Oe = 79.5775 A·m^(-1)),respectively, which is remarkably improved compared with that of continuous [Co/Pt]80multilayers. That is because large amounts of grain boundaries in the granular multilayers can efficiently impede the propagation and expansion of reversed magnetic domains, which is verified by experimental investigations and micromagnetic simulation results. The simulation results also indicate that the value of Mrt, Mr/Msratio, and Hcoopcan be further improved through optimizing the granule size, which can be experimentally realized by manipulating the process parameter of SiNxbuffer layer. This work provides an alternative solution for achieving high Mrt value in ultra-thick Co/Pt multilayers, which is of unneglectable potential in applications of high-density magnetic recording.

Multiple SiGe/Si layers epitaxy and SiGe selective etching for vertically stacked DRAM

Fifteen periods of Si/Si_(0.7)Ge_(0.3)multilayers(MLs)with various Si Ge thicknesses are grown on a 200 mm Si substrate using reduced pressure chemical vapor deposition(RPCVD).Several methods were utilized to characterize and analyze the ML structures.The high resolution transmission electron microscopy(HRTEM)results show that the ML structure with 20 nm Si_(0.7)Ge_(0.3)features the best crystal quality and no defects are observed.Stacked Si_(0.7)Ge_(0.3)ML structures etched by three different methods were carried out and compared,and the results show that they have different selectivities and morphologies.In this work,the fabrication process influences on Si/Si Ge MLs are studied and there are no significant effects on the Si layers,which are the channels in lateral gate all around field effect transistor(L-GAAFET)devices.For vertically-stacked dynamic random access memory(VS-DRAM),it is necessary to consider the dislocation caused by strain accumulation and stress release after the number of stacked layers exceeds the critical thickness.These results pave the way for the manufacture of high-performance multivertical-stacked Si nanowires,nanosheet L-GAAFETs,and DRAM devices.

Proton irradiation-induced dynamic characteristics on high performance GaN/AlGaN/GaN Schottky barrier diodes

Dynamic characteristics of the single-crystal Ga N-passivated lateral AlGaN/GaN Schottky barrier diodes(SBDs)treated with proton irradiation are investigated.Radiation-induced changes including idealized Schottky interface and slightly degraded on-resistance(RON)are observed under 10-Me V proton irradiation at a fluence of 10^(14)cm^(-2).Because of the existing negative polarization charges induced at GaN/AlGaN interface,the dynamic ON-resistance(RON,dyn)shows negligible degradation after a 1000-s-long forward current stress of 50 mA to devices with and without being irradiated by protons.Furthermore,the normalized RON,dynincreases by only 14%that of the initial case after a 100-s-long bias of-600 V has been applied to the irradiated devices.The high-performance lateral AlGaN/GaN SBDs with tungsten as anode metal and in-situ single-crystal GaN as passivation layer show a great potential application in the harsh radiation environment of space.

Trimmable bandgap reference circuit with exponential curvature compensation

This paper proposes an improved exponential curvature-compensated bandgap reference circuit to exploit the exponential relationship between the current gainβof the bipolar junction transistor(BJT)and the temperature as well as reduce the influence of resistance-temperature dependency.Considering the degraded circuit performance caused by the process deviation,the trimmable module of the temperature coefficient(TC)is introduced to improve the circuit stability.The circuit has the advantages of simple structure,high linear stability,high TC accuracy,and trimmable TC.It consumes an area of 0.09 mm^(2)when fabricated by using the 0.25-μm complementary metal-oxide-semiconductor(CMOS)process.The proposed circuit achieves the simulated power supply rejection(PSR)of about-78.7 dB@1 kHz,the measured TC of~4.7 ppm/℃over a wide temperature range from-55℃to 125℃with the 2.5-V single-supply voltage,and the tested line regulation of 0.10 mV/V.Such a high-performance bandgap reference circuit can be widely applied in high-precision and high-reliability electronic systems.

Progress in reentry trajectory planning for hypersonic vehicle

The reentry trajectory planning for hypersonic vehicles is critical and challenging in the presence of numerous nonlinear equations of motion and path constraints, as well as guaranteed satisfaction of accuracy in meeting all the specified boundary conditions. In the last ten years, many researchers have investigated various strategies to generate a feasible or optimal constrained reentry trajectory for hypersonic vehicles. This paper briefly reviews the new research efforts to promote the capability of reentry trajectory planning. The progress of the onboard reentry trajectory planning, reentry trajectory optimization, and landing footprint is summarized. The main challenges of reentry trajectory planning for hypersonic vehicles are analyzed, focusing on the rapid reentry trajectory optimization, complex geographic constraints, and coop- erative strategies.

A CAD oriented quasi-analytical large-signal drain current model for 4H-SiC MESFETs

This paper reports that a 4H-SiC MESFET （Metal Semiconductor Field Effect Transistor） large signal drain current model based on physical expressions has been developed to be used in CAD tools. The form of drain current model is based on semi-empirical MESFET model, and all parameters in this model are determined by physical parameters of 4H-SiC MESFET. The verification of the present model embedded in CAD tools is made, which shows a good agreement with measured data of large signal DC I-V characteristics, PAE （power added efficiency）, output power and gain.

Laser cleaning of steel structure surface for paint removal and repaint adhesion

Paint removal from steel structure is executed for shipyards of marine and offshore engineering.Due to environmental unfriendliness and unhealthy drawbacks of sand blasting technique, laser ablation technique is proposed as a substituting method.By absorbing high energy of the 1064 nm pulsed laser, the paint is vaporized quickly.The ablated debris is then collected by using a suction pump.Initial metal surface of the steel is exposed when laser beam irradiates perpendicularly and scans over it.The cleaned surface fulfills the requirements of surface preparation standards ISO 8501 of SA2.The adhesion is further characterized with pull-off test after carrying out painting with Jotamastic 87 aluminum paint.The repainting can be embedded onto the laser cleaned surface to bond much more tightly.The excellent adhesion strength of 20 MPa between repainted coating and the substrate is achieved, which is higher than what is required by shipyards applications.

A novel analytical thermal model for multilevel nano-scale interconnects considering the via effect

Based on the heat diffusion equation of multilevel interconnects, a novel analytical thermal model for multilevel nano-scale interconnects considering the via effect is presented, which can compute quickly the temperature of multilevel interconnects, with substrate temperature given. Based on the proposed model and the 65 nm complementary metal oxide semiconductor （CMOS） process parameter, the temperature of nano-scale interconnects is computed. The computed results show that the via effect has a great effect on local interconnects, but the reduction of thermal conductivity has little effect on local interconnects. With the reduction of thermal conductivity or the increase of current density, however, the temperature of global interconnects rises greatly, which can result in a great deterioration in their performance. The proposed model can be applied to computer aided design （CAD） of very large-scale integrated circuits （VLSIs） in nano-scale technologies.

Enhanced Thermal Conductivity and Bending Strength of Graphite Flakes/aluminum Composites Via Graphite Surface Modification

The effect of graphite surface modification on the thermal conductivity(TC) and bending strength of graphite flakes/Al composites(Gf/Al) prepared by gas pressure infiltration were investigated. Al3 Ni and Al4C3 phase may form at the interface in Ni-coated Gf/Al and uncoated Gf/Al composites, respectively, while the Al-Cu compound cannot be observed in Cu-coated Gf/Al composites. The Cu and Ni coatings enhance TC and the bending strength of the composites in the meantime. TC of Cu-coated Gf/Al composites reach 515 Wm^-1·K^-1 with 75 vol% Gf, which are higher than that of Ni-coated Gf/Al. Meanwhile, due to Al3 Ni at the interface, the bending strength of Ni-coated Gf/Al composites are far more than those of the uncoated and Cu-coated Gf/Al with the same content of Gf. The results indicate that metal-coated Gf can effectively improve the interfacial bonding between Gf and Al.

Yield estimation of metallic layers in integrated circuits

In the existing models of estimating the yield and critical area, the defect outline is usually assumed to be circular, but the observed real defect outlines are irregular in shape. In this paper, estimation of the yield and critical area is made using the Monte Carlo technique and the relationship between the errors of yield estimated by circular defect and the rectangle degree of the defect is analysed. The rectangular model of a real defect is presented, and the yield model is provided correspondingly. The models take into account an outline similar to that of an original defect, the characteristics of two-dimensional distribution of defects, the feature of a layout routing, and the character of yield estimation. In order to make the models practicable, the critical area computations related to rectangular defect and regular （vertical or horizontal） routing are discussed. The critical areas associated with rectangular defect and non- regular routing are developed also, based on the mathematical morphology. The experimental results show that the new yield model may predict the yield caused by real defects more accurately than the circular model. It is significant that the yield is accurately estimated using the proposed model for IC metals.

Electrical and optical characteristics of vanadium in 4H-SiC

A semi-insulating layer is obtained in n-type 4H-SiC by vanadium-ion implantation. A little higher resistivity is obtained by increasing the annealing temperature from 1450 to 1650 ℃. The resistivity at room temperature is as high as 7.6 ×10^6 Ω. cm. Significant redistribution of vanadium is not observed even after 1650 ℃ annealing. Temperaturedependent resistivity and optical absorption of V-implanted samples are measured. The activation energy of vanadium acceptor level is observed to be at about Ec - 1.1 eV.

Critical area computation for real defects and arbitrary conductor shapes

In current critical area models, it is generally assumed the defect outlines are circular and the conductors to be rectangle or the merger of rectangles. However, real defects and conductors associated with optimal layout design exhibit a great variety of shapes. Based on mathematical morphology, a new critical area model is presented, which can be used to estimate the critical area of short circuit, open circuit and pinhole. Based on the new model, the efficient validity check algorithms are explored to extract critical areas of short circuit, open circuit and pinhole from layouts. The results of experiment on an approximate layout of 4 × 4 shifts register show that the new model predicts the critical areas accurately. These results suggest that the proposed model and algorithm could provide new approaches for yield prediction.