Energy efficiency analysis of Spiking Neural Networks for space applications

TL;DR

The paper addresses the challenge of onboard AI for space missions under tight power budgets by evaluating Spiking Neural Networks (SNN) against traditional ANNs on the EuroSAT scene classification task. It introduces a hardware-agnostic energy metric, EMAC, to compare architectures across neuron models and coding schemes, and validates the approach with both numerical experiments and a hardware demonstration on BrainChip Akida AKD1000. Results show convolutional SNNs can match ANN accuracy while reducing energy per inference by 50–80%, though scaling to deeper networks remains challenging due to training memory and latency constraints. The work provides a practical framework for designing energy-efficient onboard AI and highlights directions for improved training methods and hardware integration to realize the full potential of SNNs in space systems.

Abstract

While the exponential growth of the space sector and new operative concepts ask for higher spacecraft autonomy, the development of AI-assisted space systems was so far hindered by the low availability of power and energy typical of space applications. In this context, Spiking Neural Networks (SNN) are highly attractive due to their theoretically superior energy efficiency due to their inherently sparse activity induced by neurons communicating by means of binary spikes. Nevertheless, the ability of SNN to reach such efficiency on real world tasks is still to be demonstrated in practice. To evaluate the feasibility of utilizing SNN onboard spacecraft, this work presents a numerical analysis and comparison of different SNN techniques applied to scene classification for the EuroSAT dataset. Such tasks are of primary importance for space applications and constitute a valuable test case given the abundance of competitive methods available to establish a benchmark. Particular emphasis is placed on models based on temporal coding, where crucial information is encoded in the timing of neuron spikes. These models promise even greater efficiency of resulting networks, as they maximize the sparsity properties inherent in SNN. A reliable metric capable of comparing different architectures in a hardware-agnostic way is developed to establish a clear theoretical dependence between architecture parameters and the energy consumption that can be expected onboard the spacecraft. The potential of this novel method and his flexibility to describe specific hardware platforms is demonstrated by its application to predicting the energy consumption of a BrainChip Akida AKD1000 neuromorphic processor.

Figures (13)

• Figure 1: Current-based Leaky Integrate and Fire (LIF) neuron model. The neuron takes as input the pre-synaptic binary spikes $S_j$. The internal dynamic governs the trend of the current $i$ and voltage $v$. Whenever the voltage hits the threshold $\theta$, an output spike $S_o$ is emitted, and the voltage is reset to zero.
• Figure 2: Accuracy vs. dimensionless energy (EMAC).
• Figure 3: Dimensionless energy (EMAC per inference) w.r.t. total emitted spikes per inference, average values over the test set.
• Figure 4: EMAC per inference for selected test cases: breakdown across network tasks, average value over the test set. Standard deviation value is indicated by black error bars.
• Figure 5: Total emitted spikes per inference, average value over the test set. The standard deviation value is indicated by black error bars.