Quantization occurs when an analog signal is converted to a digital representation. Quantization adds a small amount of noise into a measurement; for example, if a sample value is 7.3, it will be represented as 7. That difference is referred to as quantization noise.
Typical Analog-to-Digital converters can produce numbers with 8 to 16 bits resolution. An 8-bit converter can represent values from 0 to 255, whereas a 16-bit converter can represent values between 0 and 65,535.
In order to adequately represent image sensor data, the converter must have enough bits of resolution to record both the brightest value possible (saturation) while simultaneously recording the noise background. In order to avoid adding extra quantization noise into the image, the noise should typically be sampled by 2-3 bits of resolution.
The number of bits required to adequately record a sensor's image depends on the well depth, or maximum number of photoelectrons it can record. The well depth is larger for physically larger pixels, but also depends on the exact details of the particular sensor technology used.
For typical sensors, 8-bit resolution is insufficient for high quality images; whereas 16-bit converters are overkill in most cases, except for large pixel sensors.