MigSpike: A Migration Based Algorithms and Architecture for Scalable Robust Neuromorphic Systems

Dang, Khanh N.; Doan, Nguyen Anh Vu; Abdallah, Abderazek Ben

doi:10.1109/tetc.2021.3136028

Cited by 12 publications

(35 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent hardware fault injection experiments have shown that they can be highly vulnerable, especially when those are happening after training [33]. However, to understand the impacts of faults on SNNs, we randomly inserted two hardware-level faults as in [34] into 2 different post-trained SNN models.…”

Section: A Background and Motivationmentioning

confidence: 99%

“…Furthermore, with the increase in the size of the system, when a fault occurs, the result of computation from neurons in the cores can yield faulty output. Motivated by these challenges, our work in [34] proposed migration methods named MigSpike using max-flow min-cut flow and genetic algorithms. This strategy addresses the drawbacks of the prior proposal by introducing a fault-tolerant framework during neuron mapping.…”

Section: A Background and Motivationmentioning

confidence: 99%

“…(ii) migrating to another chip would be expensive and thus would lead to many reconfigurations. Motivated by the limitation of the proposal in [34] and the disadvantages of the alternative solutions, we explored the possibility of dropping some neurons during mapping to resolve this bottleneck. We proposed a mechanism based on ranking and selection (RSM) to drop neurons according to their contribution levels.…”

Section: A Background and Motivationmentioning

confidence: 99%

“…However, we traded off improving performance over robustness during mapping. The MigSpike architecture we proposed in [34] is implemented on a 3D network-on-chip-based neuromorphic system. Unlike conventional mapping techniques which are based on clustering and partitioning, the MigSpike supports migrating the task of faulty neurons to spare ones in a system.…”

Section: Handling Faults During Neural Computationmentioning

confidence: 99%

“…For instance, in L 2 , the total number of spare neurons available for repairs is 135 while the total number of faulty ones is 237. Clearly, the mapping method in [34] will fail to remap.…”

Section: Handling Faults During Neural Computationmentioning

confidence: 99%

See 4 more Smart Citations

Fault-Tolerant Spiking Neural Network Mapping Algorithm and Architecture to 3D-NoC-Based Neuromorphic Systems

Yerima,

Ikechukwu,

Dang

et al. 2023

IEEE Access

Self Cite

View full text Add to dashboard Cite

Neuromorphic computing uses spiking neuron network models to solve machine learning problems in a more energy-efficient way when compared to conventional artificial neural networks. However, mapping the various network components to the neuromorphic hardware is not trivial to realize the desired model for an actual simulation. Moreover, neurons and synapses could be affected by noise due to external interference or random actions of other components (i.e., neurons), which eventually lead to unreliable results. This work proposes a fault-tolerant spiking neural network mapping algorithm and architecture to a 3D network-on-chip (NoC)-based neuromorphic system (R-NASH-II) based on a rank and selection mapping mechanism (RSM). The RSM allows the ranking and rapid selection of neurons for fault-tolerant mapping. Evaluation results show that with our proposed mechanism, we could maintain a mapping efficiency of 100% with 20% spare rate and a fault rate (40%) more than in the previous mapping framework. The Monte Carlo simulation evaluation of reliability shows that the RSM mechanism has increased the mean time to failure (MTTF) of the previous mapping technique by 43% on average. Furthermore, the operational availability of the RSM for mapping to a 4 × 4 × 4 (smallest) and 6 × 6 × 6 (largest) NoC is 88% and 67% respectively.

show abstract

Section: A Background and Motivationmentioning

confidence: 99%

Section: A Background and Motivationmentioning

confidence: 99%

Section: A Background and Motivationmentioning

confidence: 99%

Section: Handling Faults During Neural Computationmentioning

confidence: 99%

“…For instance, in L 2 , the total number of spare neurons available for repairs is 135 while the total number of faulty ones is 237. Clearly, the mapping method in [34] will fail to remap.…”

Section: Handling Faults During Neural Computationmentioning

confidence: 99%

See 3 more Smart Citations

Fault-Tolerant Spiking Neural Network Mapping Algorithm and Architecture to 3D-NoC-Based Neuromorphic Systems

Yerima,

Ikechukwu,

Dang

et al. 2023

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

An In-Situ Dynamic Quantization With 3D Stacking Synaptic Memory for Power-Aware Neuromorphic Architecture

et al. 2023

View full text Add to dashboard Cite

R-MaS3N: Robust Mapping of Spiking Neural Networks to 3D-NoC-Based Neuromorphic Systems for Enhanced Reliability

Yerima,

Dang,

Abdallah

2023

IEEE Access

Self Cite

View full text Add to dashboard Cite

Neuromorphic computing utilizes spiking neural networks (SNNs) to offer power/energyefficient solutions for complex machine-learning problems in hardware. However, neural circuits are prone to faults caused by variability in the manufacturing flow, process variations, and manufacturing defects. This work proposes a mapping approach, R-MaS3N, that leverages the reuse of existing neurons for robust mapping of SNNs to a 3D-NoC-based neuromorphic system (NR-NASH). A heuristic-based partitioning technique is employed to partition neurons in the layers of an SNN application using neuron firing patterns. Moreover, a neuronal partitioning approach cluster mapped neurons in the neuromorphic neural circuits based on connectivity patterns and spiking activities. Evaluation results show that the proposed fault-tolerant mapping method maintains a remapping efficiency of 100% with a fault rate of 40% in the 3D NoC-based neuromorphic system. With a system configuration of 4 × 4 × 4 and 256 neurons per cluster, our approach has a remapping time of 71× less than the previous approach with the same configuration parameters. In addition, the MTTF of the mapping method for system configuration 5 × 5 × 5 network size at a 40% fault rate surpasses the previous method at 20% fault rate by 16% for 4 × 4 × 4 network size.

show abstract

MigSpike: A Migration Based Algorithms and Architecture for Scalable Robust Neuromorphic Systems

Cited by 12 publications

References 40 publications

Fault-Tolerant Spiking Neural Network Mapping Algorithm and Architecture to 3D-NoC-Based Neuromorphic Systems

Fault-Tolerant Spiking Neural Network Mapping Algorithm and Architecture to 3D-NoC-Based Neuromorphic Systems

An In-Situ Dynamic Quantization With 3D Stacking Synaptic Memory for Power-Aware Neuromorphic Architecture

R-MaS3N: Robust Mapping of Spiking Neural Networks to 3D-NoC-Based Neuromorphic Systems for Enhanced Reliability

Contact Info

Product

Resources

About