Optimal Mapping of Spiking Neural Network to Neuromorphic Hardware for Edge-AI

Abstract

Neuromorphic hardware, the new generation of non-von Neumann computing system, implements spiking neurons and synapses to spiking neural network (SNN)-based applications. The energy-efficient property makes the neuromorphic hardware suitable for power-constrained environments where sensors and edge nodes of the internet of things (IoT) work. The mapping of SNNs onto neuromorphic hardware is challenging because a non-optimized mapping may result in a high network-on-chip (NoC) latency and energy consumption. In this paper, we propose NeuMap, a simple and fast toolchain, to map SNNs onto the multicore neuromorphic hardware. NeuMap first obtains the communication patterns of an SNN by calculation that simplifies the mapping process. Then, NeuMap exploits localized connections, divides the adjacent layers into a sub-network, and partitions each sub-network into multiple clusters while meeting the hardware resource constraints. Finally, we employ a meta-heuristics algorithm to search for the best cluster-to-core mapping scheme in the reduced searching space. We conduct experiments using six realistic SNN-based applications to evaluate NeuMap and two prior works (SpiNeMap and SNEAP). The experimental results show that, compared to SpiNeMap and SNEAP, NeuMap reduces the average energy consumption by 84% and 17% and has 55% and 12% lower spike latency, respectively.

HeterGenMap: An Evolutionary Mapping Framework for Heterogeneous NoC-Based Neuromorphic Systems

“…NEUTRAMS [19] uses the Kernighan-Lin partitioning strategy to split the neurons into groups. In [27], the authors divide the adjacent layers into a so-called sub-network. Then, the sub-network will be partitioned to reduce the complexity of the algorithm.…”

“…In this section, we compare the average number of hops as the main result of the mapping approach. Here, we extract the result from NeuMap [27] where it is compared to SNEAP [10] and SpiNeMap [18]. Here, we evaluate the average number of hops in connection with the mapping results for six different networks (CNN-CIFAR10, CNN-Fashion-MNIST, LetNet5-CIFAR10, LetNet-5-MNIST, MLP-Fashion-MNIST, and MLP-MNIST) as can be seen in the Table 4.…”

“…Here, we evaluate the average number of hops in connection with the mapping results for six different networks (CNN-CIFAR10, CNN-Fashion-MNIST, LetNet5-CIFAR10, LetNet-5-MNIST, MLP-Fashion-MNIST, and MLP-MNIST) as can be seen in the Table 4. The results for NeuMap [27], SNEAP [10], and [18] are for 8 × 8 2D-NoC and 256 neurons/core. In our HeterGenMap evaluations, we use 8×8 and 4×4×4 NoC and 256 neurons/core.…”

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

“…For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ With this assertion, the cell re-grow process thus keeps the brain functioning for an extended period until it can no longer re-grow cells resulting in the brain's eventual crash. Although neuromorphic applications that map the brain's functioning principles to digital hardware have been studied in the last 20 years [15], most of these existing application mapping strategies explored their performance characteristics notably in [16], [17], [18], and [19]. However, the performance attribute of this hardware is a measure of its robustness capabilities in the event of single or multiple fault occurrence [20], [21].…”

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

“…SNNs have been mapped to neuromorphic hardware using several approaches and methodologies. Existing techniques, notably in [14] [15] [16] [17], emphasized hardware performance at the expense of resiliency and robustness, which are critical factors for maintaining reliable computation output for optimal performance. Consequently, it is imperative to find the most efficient way to map SNN applications to neuromorphic hardware which would have significant implications for the performance and reliability of the application.…”