A type of neural network architecture that processes sequences by maintaining and updating an internal state, offering an alternative to transformer-based attention mechanisms.