The quantum efficiency (QE) of CMOS image sensors does not reach 100% due to a combination of optical and electrical losses, as well as the impact of various physical and technological factors. Here are the primary reasons for the loss in quantum efficiency:

Optical Losses

• Reflection and Absorption: Light incident on the sensor can be reflected away or absorbed before reaching the photodiode, which is the light-sensitive part of the pixel. Antireflective coatings can mitigate this to some extent, but not completely[1].
• Crosstalk: Light intended for one pixel can scatter into adjacent pixels, reducing the QE and also affecting the sensor’s signal-to-noise ratio (SNR). This is particularly significant in front-side illuminated (FSI) sensors where light has to pass through several layers before reaching the photodiode[3].
• Microlens Efficiency: While microlenses are used to focus light more directly onto the photodiodes, their efficiency is not perfect. The design and alignment of these microlenses, especially as pixel sizes shrink, are critical for maximizing QE but cannot eliminate all losses[23].

Electrical Losses

• Charge Collection Efficiency: Not all electron-hole pairs generated by the incoming photons are successfully collected by the photodiode. Some recombine before they can be collected, reducing the QE[12].
• Dark Current and Noise: The presence of dark current (current that flows through the sensor even in the absence of light) and various forms of noise (e.g., read noise, shot noise) can obscure the signal from real photons, effectively reducing QE[17].

Radiation Damage

• Total Ionizing Dose (TID) Effects: Exposure to radiation can increase the dark current, decrease the charge-to-voltage conversion factor, and reduce QE. Radiation can induce defects in the silicon lattice, affecting the sensor’s performance[22].
• Displacement Damage: Radiation can also cause displacement damage in the silicon, altering the doping profiles and depleting volumes in photodiodes, which further decreases QE[2].

Technological Limitations

• Manufacturing Imperfections: Variations in the manufacturing process can lead to defects and non-uniformities in the sensor, affecting QE. This includes variations in the doping profiles, surface passivation quality, and the uniformity of antireflective coatings[10].
• Material Limitations: The intrinsic properties of silicon, such as its bandgap and absorption spectrum, limit its efficiency in converting photons to electron-hole pairs, especially at certain wavelengths[1].

Recent advancements aim to mitigate these losses through innovative approaches such as nanostructured surfaces for enhanced light absorption, atomic layer deposited surface passivation to reduce optical and electrical losses, and the integration of high-energy photon attenuation layers for X-ray detection[3]. Despite these efforts, achieving 100% QE remains challenging due to the fundamental physical and practical limitations described above.

Citations:

• [1] https://www.semanticscholar.org/paper/2926184ba6d6260b8d65edb1468e0b28b966f7ec
• [2] https://www.semanticscholar.org/paper/1000a504c276499a0967c9f49b3068fffe2da978
• [3] https://www.semanticscholar.org/paper/2bc857c569efd091a5aa1d1057eb6d4a561880f2
• [4] https://www.semanticscholar.org/paper/7e1cac36388049515d02eb2816b3983319b5fc28
• [5] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10747789/
• [6] https://www.semanticscholar.org/paper/b4bac65446e19f1286b34fa6913a68def70684ae
• [7] https://www.semanticscholar.org/paper/38d79197f3fe57f027186682a9ddb32e249921c4
• [8] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9860920/
• [9] https://www.semanticscholar.org/paper/8f9609d355fa8264a6402c1ae101300f2dfc42e9
• [10] https://www.semanticscholar.org/paper/06551568d3d3999fa5782da19613a254cdd0c7a7
• [11] https://www.semanticscholar.org/paper/04eea0fb42b8a2e61d611e8e627a54dc3f9eae70
• [12] https://www.semanticscholar.org/paper/9d7b80ce7c3711ee967f8eb32a7308992ad08630
• [13] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7038367/
• [14] https://arxiv.org/abs/2203.13583
• [15] https://www.semanticscholar.org/paper/ca0b2ffaedd33ddf37350d5053d149146eb88dce
• [16] https://www.semanticscholar.org/paper/adf3c33806d8a9cb182fe311db6de7d2007c11ca
• [17] https://www.semanticscholar.org/paper/848585d18503c9d476b583b110cc54b9246d0864
• [18] https://pubmed.ncbi.nlm.nih.gov/33770960/
• [19] https://www.semanticscholar.org/paper/cb7aafa1d53f22a958069df9138f02dccdcab6d6
• [20] https://arxiv.org/abs/2207.13052
• [21] https://www.semanticscholar.org/paper/3d1264b2183376a6bc89b124f25d1f496c631489
• [22] https://www.semanticscholar.org/paper/9e12df935d722d9d902612899bf25d87e5cc4db3
• [23] https://www.semanticscholar.org/paper/29a450107b014930244373faa116a773eb01f1b1