The current trend towards embedding more and more light sensors in portable and wearable devices is calling for higher integration and reuse of the sensor interface electronics. In many applications, the light needs to be generated locally, which becomes the dominant source of power consumption. Power can hence be saved by making the sensor more sensitive to lower light. At low light the noise is totally dominated by the noise coming from the electronic readout chain. This paper shows how this noise can be minimized all along the readout chain, from the pixel to the ADC. It also shows that many low-light applications actually share the same readout architecture. The latter is made of the pixel including a source follower, a shared amplifier and a correlated double sampling (CDS) or correlated multiple sampling (CMS) stage that is key for eliminating the low-frequency noise and reducing the white noise for CMS. It is shown how each of these building blocks can be optimized for reaching a minimum input-referred noise. The design methodology is then illustrated by three applications, including a 0.5 electon(rms) CMOS VGA imager, a 2.6 mu W PPG sensor and a 10 mu W 1D time-of-flight distance ranging device.