Answers are grounded in this paper and cite their sources.
In particular, Single Event Upsets and Multiple Bit Upsets created by radiation lead to severe problems for mission critical computing systems such as aerospace, nuclear, as well as safety critical embedded processors. Although QNU remains the most severe error, TMR Latches [3], ECC Latches [4], and DICE [1] are no longer viable because while remaining low power consuming, they incur significant delay overhead to inject QNU. In this paper, the Quadruple Node Upset Self Recovery Latch (LQNTL) is proposed, a fault tolerant latch that employs multi feedback redundancy with machine learning based optimization aimed at soft error resilience improvement. The methodology is evaluated in terms of the fault recovery efficiency through simulation using HSPICE and transistor level, and fault injection using MC; and using AI models based on LSTM, SVM and Random Forests networks. In the second category, Maskov Models are used to probabilistically model transitions between fault state and Reinforcement Learning (RL) algorithms are employed to choose the best self-recovery mechanism through RL in order to make transitions between fault states adaptively in real time. Simulation shows that LQNTL is superior at PVT variations to conventional latches and has 95% recovery success, a much reduced Mean time to Recovery (MTTR) (~10–15ns vs. 20–30ns in conventional latches), and a much safer and more robust latch using state-of-the art processes. LQNTL consumes a factor of 10–20 % more power than LQNTL but has superior fault tolerance and capability to determine errors a la deterministic. This thesis contributes to a thorough literature review of previous fault tolerant latch implementation, the gaps in the current design and the future direction for fault tolerant latch design using AI driven optimizations for next generation of radiation hardened computing architecture. The results indicate that LQNTL could constitute a foundation for future fault tolerant microprocessors, FPGAs and memory architectures in such environments of extreme demand, low power and self-recovery. Adaptive power management, neuromorphic inspired fault prediction, as well as extensions for quantum error correction of latch reliability are all considered for future work.
No takes yet. Share an insight, caveat, or question.
Shah et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: