Adimec continues its series of CCD vs CMOS sensor comparisons. The latest part shows that DTI in CMOS sensors, such as Sony IMX174, greatly improves MTF in NIR:
So, Adimec's explanation is incorrect. Then, what makes IMX174's NIR MTF so good?
Adimec says that "In the past, CCD technology had better MTF in NIR because of higher electric fields in the depletion layer, and deep-P diffusions for isolation of the columns in the Interline Transfer (IT) structure."
So, IMX174 is somehow improved over the CMOS sensors of the past, right?
I think that the secret lies in the design of the light shield. IMX174 is using a tungsten (?) light shield that looks like a copy of the lightshield of a CCD-ILT. The lightshield seems oversized in the sense that the opening in the lightshield is smaller than the active area of the PPD. As a consequence the QE goes down a bit, but this is compensated by an efficient micro-lenslets.
I do not exactly understand what you mean : the Nyquist is half the sampling frequency, which is independent of the opening ! But the MTF depends on the opening, but on top of the physical open area, micro-lens(es) are located which artificially grow the size of the opening.
"micro-lens(es) are located which artificially grow the size of the opening" agreed, but does the "artificial" size of the opening/spot-size with a focusing u-lens become as large as the photodiode area, or is it still smaller?
The maximum sampling frequency would not change as Albert says but the susceptibility to aliasing at frequencies above Nyquist would change with a smaller pixel opening provided there is no anti-aliasing filter. In practice this is unlikely even without an antialiasing filter because of diffraction at this pixel pitch size.
"In practice this is unlikely even without an antialiasing filter because of diffraction at this pixel pitch size": In that case the light-shield design or micro-lenses should not have/very minimal effect on MTF. So what am I missing here?! Perhaps it is a new process where the location of the light-shield very close to the silicon that helps to reduce photonic cross-talk?
If PD footprint is shrunken while improved micro-lens can concentrate the same amount of photons onto the PD, optical characteristics should be preserved more or less. What Albert mentioned must be related to electrical crosstalk. That is, increased distance between shrunken PDs can lead to reduced crosstalk.
Pixel size in IMX174 is somewhat larger than in ICX674 (5.86um vs 4.54um). This helps a bit.
Does anyone know what is the epi-thickness of both sensors (or at least QE at 950nm)?
There is an Adimec article about CMOS/CCD sensitivity, although they do not mention which sensors they measured. They show QE/noise. CMOS is better at low wavelengths, due to lower noise, I presume. At 950nm, QE/noise is about the same, meaning that QE is lower. Could this mean that epi-thickness is lower in IMX174 is lower, explaining so the higher MTF?
I wrote : ".... like a copy of the lightshield of a CCD-ILT". Apparently the lightshield is lying on top of the gate oxide just at the edge of, and around the photodiode. The light shield comes extremely close to the Si-SiO2 interface, exactly as it did in the past with the CCD-ILT. For an example see Morimoto 1992 (yes, 1992 this is not a typo).
To my knowledge, the Sony IMX174 does not have DTI, deep trench isolation.
ReplyDeleteSo, Adimec's explanation is incorrect. Then, what makes IMX174's NIR MTF so good?
DeleteAdimec says that "In the past, CCD technology had better MTF in NIR because of higher electric fields in the depletion layer, and deep-P diffusions for isolation of the columns in the Interline Transfer (IT) structure."
So, IMX174 is somehow improved over the CMOS sensors of the past, right?
I think that the secret lies in the design of the light shield. IMX174 is using a tungsten (?) light shield that looks like a copy of the lightshield of a CCD-ILT. The lightshield seems oversized in the sense that the opening in the lightshield is smaller than the active area of the PPD. As a consequence the QE goes down a bit, but this is compensated by an efficient micro-lenslets.
DeleteAlbert,
ReplyDeleteSo the Nyquist frequency should be calculated from the small opening rather than the pixel pitch?
I do not exactly understand what you mean : the Nyquist is half the sampling frequency, which is independent of the opening ! But the MTF depends on the opening, but on top of the physical open area, micro-lens(es) are located which artificially grow the size of the opening.
Delete"micro-lens(es) are located which artificially grow the size of the opening"
Deleteagreed, but does the "artificial" size of the opening/spot-size with a focusing u-lens become as large as the photodiode area, or is it still smaller?
The maximum sampling frequency would not change as Albert says but the susceptibility to aliasing at frequencies above Nyquist would change with a smaller pixel opening provided there is no anti-aliasing filter. In practice this is unlikely even without an antialiasing filter because of diffraction at this pixel pitch size.
Delete"In practice this is unlikely even without an antialiasing filter because of diffraction at this pixel pitch size":
DeleteIn that case the light-shield design or micro-lenses should not have/very minimal effect on MTF. So what am I missing here?! Perhaps it is a new process where the location of the light-shield very close to the silicon that helps to reduce photonic cross-talk?
If PD footprint is shrunken while improved micro-lens can concentrate the same amount of photons onto the PD, optical characteristics should be preserved more or less.
DeleteWhat Albert mentioned must be related to electrical crosstalk. That is, increased distance between shrunken PDs can lead to reduced crosstalk.
Pixel size in IMX174 is somewhat larger than in ICX674 (5.86um vs 4.54um). This helps a bit.
ReplyDeleteDoes anyone know what is the epi-thickness of both sensors (or at least QE at 950nm)?
There is an Adimec article about CMOS/CCD sensitivity, although they do not mention which sensors they measured. They show QE/noise. CMOS is better at low wavelengths, due to lower noise, I presume. At 950nm, QE/noise is about the same, meaning that QE is lower. Could this mean that epi-thickness is lower in IMX174 is lower, explaining so the higher MTF?
I wrote : ".... like a copy of the lightshield of a CCD-ILT".
ReplyDeleteApparently the lightshield is lying on top of the gate oxide just at the edge of, and around the photodiode. The light shield comes extremely close to the Si-SiO2 interface, exactly as it did in the past with the CCD-ILT. For an example see Morimoto 1992 (yes, 1992 this is not a typo).
有老大能有imx的在550nm的MTF曲线么？
ReplyDelete