Implanted functional electrical stimulation (FES) has been successfully used in a large set of applications linked to organic deficiencies and sensory disabilities. More recent attempts have been made to use implanted FES for movements or functions restoration in para- and quadriplegic patients. Unfortunately, standing and walking still remain unsatisfactory at the moment. Although not optimal, FES systems remain the only products on offer for movement restoration in a daily use context. The main drawbacks of the technique are well known and include insufficient reliability, the complexity of the surgery, limited stim- ulation selectivity and efficiency, non-physiological recruitment of motor units and the complexity of muscle control. To improve this selectivity, both electrode geometry and current shape have to be explored. A programmable multidimensional stimulus waveform gener- ator provides the opportunity to conduct research on artificial- to-natural interfaces in order to achieve efficient and minimally aggressive activation. Thus, our team has developed a new archi- tecture for advanced implanted FES: we designed and prototyped the basic elements of a network of distributed stimulation and measurement units. We designed the architecture taking into account the a priori constraints and requirements in terms of performance, safety, stimulation sites, as well as diversity in the stimulation profiles to address selectivity issues in nerve fiber recruitment. We choose a power-efficient microstimulator design where the stimulation current generated by a single source is then distributed on the outputs according to programmable ratios. The prototype circuit based on current mirrors guarantees constant ratios between channels all over the dynamic range. It uses a high-voltage technology (0.35 µm AMS HV) and can deal with adaptive power supply up to 20 V.