Direct detection and characterization of a planet around a star by nulling interferometry, must be efficient in a large wavelength domain in order to detect simultaneously the infrared bio-tracers CO _II, O ₃ and H _IIO. This condition requires that an achromatic phase shift of pi be implemented, with an accuracy sufficient for achieving a deep nulling at all considered wavelengths. Several solutions have been presented. We present here a new concept for designing such an achromatic phase shifter. It is based on two cellular mirrors (alternatively, transparent plates can be used) where cells have thickness which are respectively odd and even multiples of a quarter of the central wavelength. Each cell introduces then a phase shift of (2k 1)pi or of 2kpi, on the fraction of the wave it reflects. Each mirror is introduced in the collimated beam issued from one or the other telescopes. Because of the odd/even distribution, a destructive interference is obviously produced on axis for the central wavelength when recombining the two beams. The trick to obtain a quasi-achromatisation is to distribute the thickness of the cells, so that the nulling is also efficient for a wavelength not too far from the central wavelength. We show that if the thicknesses are distributed according to the Pascal triangle, a fair quasi-achromatism is reached. This effect is the more efficient that the number of cells is large. For instance, with 256 × 256 cells, where phase shift range is between -6pi and 6pi one shows that the nulling reaches 10 ^-6 on the wavelength range [0.7lambda ₀, 1.3lambda ₀] which corresponds roughly to the DARWIN specification. In a second step, we study the optimum way to distribute the cells in the plane of the pupil. The most important criterion is the isolation of the planet image from the residual image of the star. Several efficient configurations are presented. Finally we consider some practical aspects on a device belonging to the real world and on the bench we are developing. The major interest of this solution is that it allows a compact, simple and fully symmetric design, with essentially no ajustable sub-systems ; extension to multi-telescopes interferometers with phase shift other than pi can also be envisioned.