Due to their low-cost fabrication process and high efficiency, silicon-based Complementary Metal Oxide Semiconductor (CMOS) image sensors are the reference in term of detection in the visible range. However, their optical performances are toughly degraded in the Near Infrared (NIR). For such wavelengths, Silicon has a small absorption coefficient, leading to a very poor Quantum Efficiency (QE). A solution to improve it is to implement structures like pyramids that are etched in the Silicon layer. This will lead to diffraction inside the photodiode, enhancing the light path and therefore the absorption. Using Finite Difference Time Domain (FDTD) simulations, we demonstrated a huge QE enhancement at 940nm on real pixels, by implementing this kind of diffractive structures and we finally confirmed these results by characterizations. We obtained QE values up to 47% at 940nm for our 3.2μm pixel, corresponding to a gain of 2 comparing to a pixel without any diffractive structures. We also measured the Modulation Transfer Function (MTF), to evaluate how this figure of merit is impacted by the addition of these structures. As expected, the MTF was degraded when we added these diffractive patterns but were still high looking at the values. We indeed demonstrated MTF values going up to 0.55 at Nyquist/2 frequency and 0.35 at Nyquist frequency. Looking not only at QE values but also at MTF ones, these are very promising results that could be used in many different NIR applications like face recognition, Light Detection and Ranging (LIDAR) or AR/VR.