Authors: 

Jani Babu Shaik, Sonal Singhal, Nilesh Goel, Atul Ranjan Srivastava, Siona Menezes PicardoJournal

Publication  Name: Integration, Vol. 105, Nov. 2025.

Keywords: Circuit reliability; Neuromorphic system; Leaky integrate-and-fire circuit; SRAM

Abstract: This work presents a reliability-aware neuromorphic system architecture, rel-SLIFMEM. The rel-SLIFMEM integrates a reliability-aware Simplified Leaky Integrate-and-Fire (rel-SLIF) neuron circuit with conventional spiking neural network (SNN). The rel-SLIFMEM SNN architecture aims to mitigate the reliability issues such as Bias Temperature Instability (BTI) and Hot Carrier Injection (HCI). Reliability issues affect critical device parameters such as threshold voltage and current, resulting in performance degradation in neuromorphic circuits similar to traditional CMOS circuits. The present work addresses these challenges by including rel-SLIF neurons, designed to reduce the effects of device degradation, into the SLIFMEM SNN. The designed rel-SLIFMEM SNN is characterized using inter-spike interval (ISI) as a performance metric. Extensive simulation analysis demonstrates the advantages of rel-SLIFMEM in maintaining system performance under reliability stress and energy consumption. Our study compares conventional and reliability-aware neuron circuits and provides insights into mitigating aging-induced performance degradation in future spiking neural networks.