The distributed version of the Ramadge and Wonham control problem asks for the existence of a correct controller of a plant modeled as an asynchronous transition system. The controllable plant is distributed on several finite-state processes which interact asynchronously using shared actions. The controller is distributed as well, and decomposes as one local controller on every process. Each local controller can choose to block some of the actions, called controllable actions, but it cannot block the uncontrollable actions from the environment. A controller is correct if it guarantees that every possible execution of the plant satisfies some specification. The controller synthesis problem is the decision problem which, given a plant as input, asks whether the system admits a correct controller. In case such a controller exists, the algorithm should compute one as well. The difficulty of the controller synthesis depends on several factors, including the architecture of the information flow between processes (pipeline, ring, ...), the information available to the controllers, and the specification. The choices of the local controllers can be based on several sources of information. A minima, a local controller monitors the sequence of states and actions of the local process. This information is called the *local view* of the controller. Pnueli and Rosner proved that the distributed controller synthesis problem is undecidable when local controllers are restricted to use their local view for taking decisions, even for very simple architectures. Gastin, Lerman and Zeitoun have proposed in 2004 a variant of the distributed controller synthesis problem where local controllers are allowed to base their decisions on their *causal past*. When a shared action is played by several processes then all the controllers of these processes are allowed to exchange as much information as they want, and in particular together they can compute their mutual view of the global execution: their causal past. In this setting, Gastin et al. established the decidability of the synthesis for series-parallel architectures, and left open the decidability in the general case. In this talk, we provide a negative answer to their open question: the distributed controller synthesis problem with causal memory is undecidable. We also review several architectures which where proved decidable or for which the decidability is still open.