Active error correction of topological quantum codes - in particular the toric code - remains one of the most viable routes to large scale quantum information processing. In this work, we introduce the concept of a dynamical decoder as a promising route for achieving fault-tolerant quantum memories. We analyze a specific dynamical decoder based on a discrete time cellular automaton decoder and provide evidence of a threshold above 0.05% with measurement errors. Without measurement errors, the threshold increases by a factor of roughly 1.5. We stress that (asynchronous) dynamical decoding gives rise to a Markovian dissipative process, hence equating cellular automaton decoding to a fully dissipative topological quantum memory, which removes errors continuously. Finally, we analyze the required resources, and speculate about an ideal constant resource dynamical decoder.