The ever increasing demand for large quantities of bio-engineered pharmaceuticals and the tendency for doses to become larger stimulate the development of more productive and reliable mammalian cell culture technologies. In this context, perfusion bioreactors appear an efficient means to achieve high volumetric productivity and high product concentration. Since large-scale implementation of perfusion is limited by the reliability and scale-up potential of the device used to achieve cell retention, the present work proposes to tackle the problem bottom-up from the fundamentals of physics in order to develop, characterize and model novel particle retention strategies able to overcome the limitations of state-of-the-art systems. The first part of the present thesis focuses on the identification of physical force fields chosen for their ability to drive particles or cells into highly localized regions of a suspension. The ability of ultrasonic, dielectrophoretic and magnetophoretic interactions to make cells levitate over a surface was identified as an innovative way of implementing a Boycott effect (similar to that occurring in inclined sedimentation channels) while maintaining the flow channel in a vertical position. Therefore, inclined settlers were chosen as a reference system and the Ponder-Nakamura-Kuroda theory describing their performance was extended to take into account ultrasonic, dielectrophoretic and magnetic effects. The second part of the work provides deep modeling insight into the mechanisms underlying mammalian cell retention in these new filter implementations. Potentials and limitations are discussed in details. In particular, coupling dielectrophoresis with the Boycott effect proved to be successful in separating polystyrene particles from their moderately electrically conductive suspending medium. However, in highly conductive media, this process was shown to be very sensitive to order-disruptive electrokinetic effects such as ac electroosmosis and electrothermal fluid flow. Heat dissipation is however not specific to dielectrophoresis. This phenomenon is also a known limitation of the ultrasonic or magnetophoretic implementations, in certain conditions. The last part focused its work on the development of magnetophoretically-enhanced sedimentation, which was proposed as a means to partly overcome the heat dissipation problem observed with dielectrophoresis in highly conductive culture media. For such a magnetic approach to work correctly with intrinsically nonmagnetic materials, a paramagnetic suspending medium has to be used. The current implementation proposes to prepare media with fair quantities of gadolinium(III) or dysprosium(III) salts dissolved in them. Using such a paramagnetic buffer, the feasibility of magnetically retaining polystyrene particles could be verified in a vertical separation channel, hence providing first demonstration of the principle. The use of paramagnetic substances in mammalian cell culture media is naturally a matter of concern. The present work addresses this question as well and highlights the feasibility of growing Chinese Hamster Ovary cells in Gd(III)- or Dy(III)-rich environments. In contrast, attempts to develop low electrically conductive culture media failed and replacement of the NaCl content by D-mannitol or glycerol demonstrated severe growth inhibition effects.