ST patent application US20110073976 aims to solve red QE problem by making a reflective STI pattern 41 as shown on the figure:
A prior art achieving reflection from metals on the front side was rejected due to "in microelectronics, all metals are normally processed and deposited to be non-reflective. Introducing a reflective metal layer may require significant modifications of the manufacturing conditions," and "insertion of an additional metal element in the structure has the disadvantage of causing stray capacitive couplings."
"Unevennesses 41 form a periodic pattern formed of pads surrounded with trenches. These pads may have a square base, but will preferably have a circular or hexagonal basis to avoid privileging a specific type of light biasing. Period Λ of the pattern, height d of the pads, as well as the ratio between the width of the pads and the width of the interval separating them are selected so that the periodic unevennesses altogether form a two-dimensional optic network capable of reflecting wavelengths corresponding to the color red of the visible spectrum."
"For a photodiode associated with a red filter, unevennesses 51 [typo, meant to be 41 - ISW] will, for example, have the following characteristics: period Λ: on the order of from 350 to 400 nm; height d of the pads: on the order of from 50 to 300 nm; ratio between the pad width and the distance between pads close to 1."
"An advantage of such a structure is that the reflection effect is better with a two-dimensional network than with a reflective layer. Another advantage is that this result is obtained without introducing any new material with respect to known structures."