紧凑型交错热控制可提高片上网络的吞吐量和可靠性.pdf

1、Compact Interleaved Thermal Control for Improving Throughput and Reliability of Networks-on-ChipJan.22,2025Tong Cheng,Zirui Xu,Xinyi Li,Li Li,Yuxiang FuNanjing UniversityNanjing,Jiangsu,China Background Motivation Interleaved Thermal Control Optimal Control Phase Assignment Experiment Results and An

2、alysis Outline The scope of a particular node The scope of the whole systemEmerging ApplicationsPowerful Computing Arch Pattern Recognition LLM ADAS NoC:higher interconnect bandwidthIncreasing PEs:more computing power BackgroundSevere Thermal/Reliability Problems thermal hotspots thermal gradients t

3、hermal cycling Reliability problems*Zhang,J.,et al.(2022).Hot-Trim:Thermal and Reliability Management for Commercial Multi-core Processors Considering Workload Dependent Hot Spots.IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems:1-1.*MotivationThe NoC system is inherentl

4、y distributiveand achieves excellent scalabilityCurrent distributed DTMs(Dynamic Thermal Management):Deploy identical control mechanisms at each node Initiate control mechanisms synchronouslycentralized thermal management distributed thermal management System Oscillation!MotivationPreliminary experi

5、ment The adjacent nodes exhibit higher traffic correlation and share more similar thermal characteristics Result in significant performance losses during temperature drops and pose high thermal risks during temperature risesIntensified Thermal Cycling Interleaved Thermal Control The scope of a parti

6、cular node Do not make control decisions synchronously!Staggers the control phases of neighboring nodes to evenly divide the control periodGive more chances to adjust:divide a period evenly.Source node throttling Make routing decisions based on each nodes throttling mark Interleaved Thermal Control