Spiking neural networks with uncertainty model of stochastic sampling for circuit yield enhancement

In semiconductor manufacturing, yield analysis plays a critical role in optimizing production processes, but traditional methods, such as Monte Carlo simulations, often rely on idealized models and require extensive computational resources. These approaches struggle to account for the inherent uncertainties of real-world manufacturing, limiting their practical applicability. Spiking Neural Networks (SNNs), inspired by biological neural processes, offer a promising solution by efficiently handling large-scale data while maintaining low power consumption and real-time processing capabilities. This paper introduces an uncertainty-aware spiking learning model that reduces the impact of non-ideal simulation results by incorporating input uncertainties through stochastic sampling, where neuron firing states are influenced by both input noise and neuronal characteristics. To further improve yield, the model leverages reinforcement learning to optimize process parameters iteratively. Extensive experiments on two circuit yield simulation datasets demonstrate that the proposed method outperforms traditional approaches in handling uncertainties and provides more reliable and accurate yield predictions, offering a robust and efficient alternative for semiconductor process optimization.