Mechanical properties are critical to the quality of hot-rolled steel pipe products.Accurately understanding the relationship between rolling parameters and mechanical properties is crucial for effective prediction an...Mechanical properties are critical to the quality of hot-rolled steel pipe products.Accurately understanding the relationship between rolling parameters and mechanical properties is crucial for effective prediction and control.To address this,an industrial big data platform was developed to collect and process multi-source heterogeneous data from the entire production process,providing a complete dataset for mechanical property prediction.The adaptive bandwidth kernel density estimation(ABKDE)method was proposed to adjust bandwidth dynamically based on data density.Combining long short-term memory neural networks with ABKDE offers robust prediction interval capabilities for mechanical properties.The proposed method was deployed in a large-scale steel plant,which demonstrated superior prediction interval performance compared to lower upper bound estimation,mean variance estimation,and extreme learning machine-adaptive bandwidth kernel density estimation,achieving a prediction interval normalized average width of 0.37,a prediction interval coverage probability of 0.94,and the lowest coverage width-based criterion of 1.35.Notably,shapley additive explanations-based explanations significantly improved the proposed model’s credibility by providing a clear analysis of feature impacts.展开更多
Aiming to enhance the bandwidth in near-memory computing,this paper proposes a SSA-over-array(SSoA)architecture.By relocating the secondary sense amplifier(SSA)from dynamic random access memory(DRAM)to the logic die a...Aiming to enhance the bandwidth in near-memory computing,this paper proposes a SSA-over-array(SSoA)architecture.By relocating the secondary sense amplifier(SSA)from dynamic random access memory(DRAM)to the logic die and repositioning the DRAM-to-logic stacking interface closer to the DRAM core,the SSoA overcomes the layout and area limitations of SSA and master DQ(MDQ),leading to improvements in DRAM data-width density and frequency,significantly enhancing bandwidth density.The quantitative evaluation results show a 70.18 times improvement in bandwidth per unit area over the baseline,with a maximum bandwidth of 168.296 Tbps/Gb.We believe the SSoA is poised to redefine near-memory computing development strategies.展开更多
Optical interconnects based on photonic integrated circuits(PICs)are emerging as a pivotal technology to address the relentless surge in data traffic driven by compute-intensive applications.Combining mode-division mu...Optical interconnects based on photonic integrated circuits(PICs)are emerging as a pivotal technology to address the relentless surge in data traffic driven by compute-intensive applications.Combining mode-division multiplexing(MDM)with wavelength-division multiplexing(WDM)offers a compelling approach to significantly enhance the shoreline density of optical interconnects.However,existing on-chip MDM systems encounter considerable challenges in simultaneously achieving a large optical bandwidth,multi-band operation,and ultra-compactness,thereby limiting scalability as conventional telecom band resources become increasingly constrained.Here we introduce,to our knowledge,the first inverse-designed multi-band mode multiplexer(MUX)utilizing a digital metamaterial structure to support the first three-order TE modes.The proposed device features an ultra-compact footprint of 6μm×4.8μm and exhibits an exceptionally flat spectral response,with numerical simulations confirming spectral variations of less than 0.94 dB across the 1500–2100 nm range.Experimental results further validate its performance,demonstrating insertion losses below 4.3 dB and 4.0 dB,and crosstalk below−11.6 dB and−11.3 dB,within the 1525–1585 nm and 1940–2040 nm bands,respectively.Additionally,system-level optical interconnect experiments using a multi-band MDM circuit successfully achieve single-wavelength transmission rates of 3-modes×180 Gb∕s at the 1.55μm band and record-setting 3-modes×114 Gb∕s in the 2μm band.This work highlights the transformative potential of employing multi-band MDM technology to enhance bandwidth density and scalability,providing a robust foundation for next-generation high-capacity on-chip optical interconnects.展开更多
基金supported by the National Key Research and Development Plan(Grant No.2023YFB3712400)the National Key Research and Development Plan(Grant No.2020YFB1713600).
文摘Mechanical properties are critical to the quality of hot-rolled steel pipe products.Accurately understanding the relationship between rolling parameters and mechanical properties is crucial for effective prediction and control.To address this,an industrial big data platform was developed to collect and process multi-source heterogeneous data from the entire production process,providing a complete dataset for mechanical property prediction.The adaptive bandwidth kernel density estimation(ABKDE)method was proposed to adjust bandwidth dynamically based on data density.Combining long short-term memory neural networks with ABKDE offers robust prediction interval capabilities for mechanical properties.The proposed method was deployed in a large-scale steel plant,which demonstrated superior prediction interval performance compared to lower upper bound estimation,mean variance estimation,and extreme learning machine-adaptive bandwidth kernel density estimation,achieving a prediction interval normalized average width of 0.37,a prediction interval coverage probability of 0.94,and the lowest coverage width-based criterion of 1.35.Notably,shapley additive explanations-based explanations significantly improved the proposed model’s credibility by providing a clear analysis of feature impacts.
基金supported in part by the Strategic Priority Research Program of the Chinese Academy of Sciences under Grant No.XDB44000000。
文摘Aiming to enhance the bandwidth in near-memory computing,this paper proposes a SSA-over-array(SSoA)architecture.By relocating the secondary sense amplifier(SSA)from dynamic random access memory(DRAM)to the logic die and repositioning the DRAM-to-logic stacking interface closer to the DRAM core,the SSoA overcomes the layout and area limitations of SSA and master DQ(MDQ),leading to improvements in DRAM data-width density and frequency,significantly enhancing bandwidth density.The quantitative evaluation results show a 70.18 times improvement in bandwidth per unit area over the baseline,with a maximum bandwidth of 168.296 Tbps/Gb.We believe the SSoA is poised to redefine near-memory computing development strategies.
基金National Key Research and Development Program of China(2023YFB2905700)National Natural Science Foundation of China(62235005,61925104,62171137).
文摘Optical interconnects based on photonic integrated circuits(PICs)are emerging as a pivotal technology to address the relentless surge in data traffic driven by compute-intensive applications.Combining mode-division multiplexing(MDM)with wavelength-division multiplexing(WDM)offers a compelling approach to significantly enhance the shoreline density of optical interconnects.However,existing on-chip MDM systems encounter considerable challenges in simultaneously achieving a large optical bandwidth,multi-band operation,and ultra-compactness,thereby limiting scalability as conventional telecom band resources become increasingly constrained.Here we introduce,to our knowledge,the first inverse-designed multi-band mode multiplexer(MUX)utilizing a digital metamaterial structure to support the first three-order TE modes.The proposed device features an ultra-compact footprint of 6μm×4.8μm and exhibits an exceptionally flat spectral response,with numerical simulations confirming spectral variations of less than 0.94 dB across the 1500–2100 nm range.Experimental results further validate its performance,demonstrating insertion losses below 4.3 dB and 4.0 dB,and crosstalk below−11.6 dB and−11.3 dB,within the 1525–1585 nm and 1940–2040 nm bands,respectively.Additionally,system-level optical interconnect experiments using a multi-band MDM circuit successfully achieve single-wavelength transmission rates of 3-modes×180 Gb∕s at the 1.55μm band and record-setting 3-modes×114 Gb∕s in the 2μm band.This work highlights the transformative potential of employing multi-band MDM technology to enhance bandwidth density and scalability,providing a robust foundation for next-generation high-capacity on-chip optical interconnects.
