Jun
30
Paper Accepted at DAIES 2026
June 30, 2026 | | Comments Off on Paper Accepted at DAIES 2026
The paper entitled “Dependability Analysis of a Spiking Convolutional Neural Network (S-CNN)”, written by Joaquín Gracia-Morán, J. Carlos Baraza-Calvo, Juan-Carlos Ruiz, David de Andrés, Daniel Gil-Tomás, and Luis-J. Saiz-Adalid, has been accepted in the 1st Int. Workshop on Dependable AI in Embedded Systems (DAIES 2026).
The paper investigates the reliability of Spiking Convolutional Neural Networks (S-CNNs) when exposed to memory faults, an increasingly important topic as neuromorphic computing gains traction in safety- and mission-critical embedded applications. Through an extensive software-implemented fault injection campaign, we evaluate the effects of adjacent multi-bit memory faults on both convolutional and spiking neuron layers.
Our results show that the proposed S-CNN exhibits a high degree of intrinsic robustness, maintaining correct classifications in most fault scenarios. The study also identifies the LIF neuron parameters as the most critical components from a dependability perspective, providing valuable insights for the design of future fault-tolerant neuromorphic systems.
Comments