If the wavelengths are long enough, there is a statistical probability that the generated photoelectron will be trapped in one of these interface traps rather than being collected in the well. Once trapped, the charge carriers will escape from these trapping site due to random thermal motion. There's a time constant that turns out to be exponentially related to temperature. Note that it is not necessary to have a saturated sensor to exhibit RBI.
The longer the subsequent exposure taken after trapping photoelectrons in these interface traps, the more charge will leak out into the following frame. This is the source of RBI (Residual Bulk Image).
Since the desired photoelectrons are winding up in trapping sites instead of the pixel's well; the QE is adversely affected. But as these interface traps fill, more of that charge winds up in the pixel's wells. This results in a variation of QE which will affect the linearity of the camera. It is bad for photometry. The effect is called Quantum Efficiency Hysteresis.
If you run the sensor warmer to shorten the time constant of the trap holding time, then you are not avoiding the trapping, you simply are reducing its impact at the expense of reducing your maximum exposure time (limit reached when dark current shot noise exceeds read noise). Additionally there are hot pixels in the typical array that are thermally activated and at warmer temperatures there are more of them.
The approach used in the Galileo and Cassini cameras was to fill the traps prior to exposures and run the cameras cold to lengthen the leakout rate. They want to run them cold anyway because of the need for minimal dark current for their applications. The point is that you aren't limited in exposure time when you run colder; the traps start out filled, remain filled through out the exposure and then are "topped off" again before the next exposure. They don't get QEH, they don't get RBI and they can take arbitrarily long exposures.
The traps are filled by flooding the sensor with NIR light: about 100x over full well is sufficient. Next the array needs to be flushed. Then the exposure can be started.
So the procedure is Flood, Flush, Integrate.
The procedure needs to be done for any sort of frame: Lights, Darks, Biases
Back illuminated devices generally do not exhibit RBI: the bulk substrate is etched away
A vertical antiblooming gate also doesn't exhibit RBI: the signal goes into the substrate. But a standard front-side illuminated full frame CCD built on Epitaxial wafers will exhibit it under the right conditions.