A deadlock-free physical mapping method on the many-core neural network chip

Ma, Cheng; Zhao, Qianchuan; Li, Guoqi; Deng, Lei; Wang, Guanrui

doi:10.1016/j.neucom.2020.03.078

Cited by 9 publications

(5 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Each case is tested with multiple physical mapping methods, including HLA, sequential neighboring placement with and without multicast (Myung et al, 2021 ), and several prior placement methods, including sequential placement (BS) (Wu et al, 2020 ), random search (RS) (Wu et al, 2020 ), simulated annealing (SA), and the RL-based approach (Wu et al, 2020 ). Due to the deadlock issue, we do not give the results of the zigzag physical mapping (Ma et al, 2020 ).…”

Section: Resultsmentioning

confidence: 99%

“…Furthermore, the logical mapping can be divided into two sets of models, which are synchronization and asynchronization. Most of the previous researches adopt the synchronization model based on the 2D mapping system (Shao et al, 2019 ; Ma et al, 2020 ; Wu et al, 2020 ; Myung et al, 2021 ), which only partitions the in-channel and out-channel of the neural network. And these researches focus on the physical mapping based on the 2D mapping system, while the 4D mapping system is a general model that has wider applications.…”

Section: Conclusion and Discussionmentioning

confidence: 99%

“…Let's define E | h | as the energy of transmitting a routing packet through a single hop distance, and define L ij as the communication latency with a routing packet between map ( v i ) and map ( v j ), respectively. With the above definitions, the total communication cost C cost (Myung et al, 2021 ), the average communication latency L (Amin et al, 2020 ), and the average power consumption P (Pei et al, 2019 ; Ma et al, 2020 ) can be calculated by…”

Section: Preliminaries and Related Workmentioning

confidence: 99%

“…There are many heuristic solutions to physical mapping, such as the genetic algorithm (GA) (Lei and Kumar, 2003 ; Zhou et al, 2006 ) and the simulated annealing (SA) (Ma et al, 2020 ) algorithm. Some teams (Davies et al, 2018 ; Shao et al, 2019 ; Zimmer et al, 2020 ) also use the greedy algorithm to optimize the communication latency and energy.…”

Section: Preliminaries and Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Approaching the mapping limit with closed-loop mapping strategy for deploying neural networks on neuromorphic hardware

Wang¹,

Yu²,

Xie³

et al. 2023

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

The decentralized manycore architecture is broadly adopted by neuromorphic chips for its high computing parallelism and memory locality. However, the fragmented memories and decentralized execution make it hard to deploy neural network models onto neuromorphic hardware with high resource utilization and processing efficiency. There are usually two stages during the model deployment: one is the logical mapping that partitions parameters and computations into small slices and allocate each slice into a single core with limited resources; the other is the physical mapping that places each logical core to a physical location in the chip. In this work, we propose the mapping limit concept for the first time that points out the resource saving upper limit in logical and physical mapping. Furthermore, we propose a closed-loop mapping strategy with an asynchronous 4D model partition for logical mapping and a Hamilton loop algorithm (HLA) for physical mapping. We implement the mapping methods on our state-of-the-art neuromorphic chip, TianjicX. Extensive experiments demonstrate the superior performance of our mapping methods, which can not only outperform existing methods but also approach the mapping limit. We believe the mapping limit concept and the closed-loop mapping strategy can help build a general and efficient mapping framework for neuromorphic hardware.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Conclusion and Discussionmentioning

confidence: 99%

Section: Preliminaries and Related Workmentioning

confidence: 99%

Section: Preliminaries and Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Approaching the mapping limit with closed-loop mapping strategy for deploying neural networks on neuromorphic hardware

Wang¹,

Yu²,

Xie³

et al. 2023

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

show abstract

“…The controllable relevance prediction influence elements are obtained through Eq (7), and the dynamic intelligence classification makes the prediction entities have a strict physical mapping relationship [16].…”

Section: Plos Onementioning

confidence: 99%

International trade and finance exploration using network model of computer trade platform

Zhang

Yuan

2021

PLoS ONE

View full text Add to dashboard Cite

International trade becomes increasingly frequent with the deepening of economic globalization. In order to ensure the stable and rapid development of international trade and finance, it is particularly crucial to predict the sales trend of foreign trade goods in advance through the network model of computer trade platform. To optimize the accuracy of sales forecasts for foreign trade goods, under the background of "Internet plus foreign trade", the controllable relevance big data mining of foreign trade goods sales, personalized prediction mechanism, intelligent prediction algorithm, improved distributed quantitative and centralized qualitative calculation are taken as the premise to design dynamic prediction model on export sales based on controllable relevance big data of cross border e-commerce (DPMES). Moreover, after the related experiments and comparative discussions, the forecast error ratios from the first quarter to the fourth quarter are 2.3%, 2.1%, 2.4% and 2.4% respectively, which are also within the acceptable range. The experimental results show that the design combines the advantages of openness and extensibility of Internet plus with dynamic prediction of big data, and achieves the wisdom, quantitative and qualitative prediction of the volume of goods sold under the background of "Internet plus foreign trade", which is controlled by the relevant data of foreign trade. The overall performance of this design is stronger than the previous models, has better dynamic evolution and high practical significance, and is of great significance in the development of international trade and finance.

show abstract

A two-level network-on-chip architecture with multicast support

Yazdanpanah

2023

Journal of Parallel and Distributed Computing

View full text Add to dashboard Cite

A deadlock-free physical mapping method on the many-core neural network chip

Cited by 9 publications

References 30 publications

Approaching the mapping limit with closed-loop mapping strategy for deploying neural networks on neuromorphic hardware

Approaching the mapping limit with closed-loop mapping strategy for deploying neural networks on neuromorphic hardware

International trade and finance exploration using network model of computer trade platform

A two-level network-on-chip architecture with multicast support

Contact Info

Product

Resources

About