The growing number of control models based on combinations of neural networks, fuzzy systems and evolutionary algorithms shows that they represent a flexible and powerful approach. However, most of these models assume that there is enough CPU power for the evolutionary and learning algorithms, which in a large number of cases is an unrealistic assumption. It is usual that the control tasks are performed by small microcontrollers, which are very near to or embedded in the plant, with low power, low cost and dedicated to a single task. This work proposes an architecture for evolution and learning in adaptive control, specifically designed to operate in microcontrollers based environments. An evaluation on a simulated temperature control environment is provided, together with details on the current hardware implementation.