This paper presents recent improvements introduced in production lines of Mid-Wavelength Infra-Red (MWIR) and Long-Wavelength Infra-Red (LWIR) HgCdTe detectors that increase performances, image quality, and reliability. This was achieved thanks to accurate characterization of RMS noise distributions. Based on many MWIR and LWIR devices RMS distributions, a RMS noise distribution model that accounts for both Background Limited diodes and 1/f noise affected isolated diodes is first proposed. Then, a figure of merit for quantifying the defective pixels is introduced. This figure of merit is shown to be easy to use and robust to statistical variability. Moreover, it does also very well correlate with physics : there is high correlation between the total number of calculated defects and other figures of merit that gauge the material quality or the low frequency noise. The ability to accurately and efficiently quantify RMS noise benefits to Sofradir in its development of highly reliable and performant technologies. Such benefits are illustrated on the latest Sofradir MWIR and LWIR technologies that are demonstrated to be very robust regarding thermal stress and thermal cycling. Finally those technologies are shown to reach high image quality and stability.