Functional electrical stimulation (FES) is a palliative solution to improve severe motor and sensory deficiencies; it has been successfully used in numerous applications such as pacemakers, deep brain stimulation, pain control, hearing restoration or vagus nerve stimulation for epilepsy or cardiac neuromodulation. Implanted electrical stimulation is also studied for sensory feedback in amputees or vision for blind people. Implanted FES for movement rehabilitation such as foot droop for hemiplegic’s patients [1] or even more complex movements like ambulation [2] and grasping [3] are used but not so disseminated. One of the reasons is that these applications need for multisite and selective stimulation. When not limited to one site of stimulation, solutions for implanted FES are actually mainly based on centralized architectures, except the Bion implant [4]. They lead to complex surgery, high risk of failure during and after surgery, and global infection problems involving the whole explantation of the device [5]. Moreover, they are not really adequate to face selectivity issues that require multi-polar electrodes and thus an important number of wires or the generation of complex waveforms. Selectivity imposes to use multipolar electrodes in order to be able to control the nerve fiber recruitment [6]. Nerve fiber recruitment impacts muscle fibers recruitment and thus the functional response [7-8]. Moreover, such neural-based stimulation selectivity may need complex multi-phasic stimulation patterns; these patterns must be very accurate from a time point of view (anodal blocking requires for example few micro-seconds synchronization). It won't be possible to manage such real-time constraints simultaneously on many sites from a central implant. Besides, such complex technologies further increase the difficulty of their use in a clinical context. Tuning for adaptation to individuals is often suboptimal and in the worst case, not efficient at all. Very complex hardware designs based on Application Specific Integrated Circuits (ASIC), advanced and proved FPGA based digital architectures and networked global architecture are then proposed by laboratories. The challenge remains to transfer to wide spread clinical use through efficient industrial transfer. But safety, low power consumption, complexity of the design and thus the maintenance should then be taken into account. Software engineering is the last issue to be considered as most of these systems include a huge number of parameters that cannot be manually tuned by a practitioner for each patient. A tradeoff should be found between a high level of flexibility, generic approach, efficiency and usability. We will first present the context of FES, in particular issues to be faced to reach a selective and efficient neural-based stimulation in terms of functional rehabilitation. Then, the use in a clinical context will be discussed in particular in the context of selective sensory feedback stimulation and multipolar stimulation for selective efferent stimulation.