Tech-On: Rohm exhibited a thin film multilayer image sensor at CEATEC Japan 2007. The major difference between Rohm's new image sensor and the existing CMOS sensors is that the former has a structure technically identical to that of a Cu-In-Ga-Se (CIGS) solar cell, instead of employing the existing Si photodiode. The CIGS photodiodes formed directly on an LSI chip.
The sensor has 100K pixels, 10um each. Its QE for visible light is said to be approximately twice of that of crystalline Si photodiode. Assuming that good Si photodiode in large pixel approaches to 60-70% QE, I wonder how Rohm doubles that.
The major remaining issue is how to miniaturize the sensor. The pixel in the latest prototype is a 10um. In the technology used for the latest development, grooves between the adjacent elements are made after the formation of CIGS layer. "We need to develop a more sophisticated technique to miniaturize the picture element blocks," Rohm said.
There are probably 3 problems with this material.
ReplyDeleteDark current, dark current and dark current.
If they use it in photovoltaic mode, like solar cell, there should be no dark current at all. Now, if they solve PRNU problem as well, it might have a chance, a slim chance probably.
ReplyDeleteSaid all that, personally I do not believe this technology ever hits the market.
So, you mean it is a logarithmic sensor? In that case is the improvement in QE all that important, esp. now that the gap is probably narrower?
ReplyDeleteActually, with added reset photovoltaic log sensor can become lin-log and one can use it just in linear region.
ReplyDeleteYou cannot have it both ways. If you are using photovoltaic mode, then there is no current and dark current is not so important but it is a log sensor and QE is not as important. If you reset the diode, and use the linear mode, then dark current matters a lot.
ReplyDeleteOK, let me clarify the pixel operation in lin-log mode.
ReplyDeleteFirst, the photodiode is reset to zero voltage. By zero voltage I mean real zero, that is both photodiode sides are at the same potential, zero for a standard CMOS photodiode.
Second, the integration process starts. If there is no light, the photodiode remains at zero voltage, as this is its thermal equilibrium state. Under the light the photodiode slowly starts to build negative voltage, I mean negative for a regular CMOS sensor, I'm not sure about Rohm's thin film photodiode polarity.
After some time the photodiode voltage saturates when its diffusion current equates photogenerated current.
Now, this saturation voltage logarithmically depends on the light intensity. So, sensor measuring the saturation voltage is innately logarithmic.
However, it takes time to reach this equilibrium state. If one sets the integration time shorter, well before the photodiode voltage reaches the saturation, the measured voltage would be linear. It's only linear if integration time is not too long, for a given light intensity.
In such a linear mode the dark current remains essentially zero. However, some other problems arise. You can read about them on Caeleste web site:
Caeleste web site
ha. Ok. Sure, everything is linearized in small signal analysis. Even Bart points out that this linearity in small signal is also dependent on "constant" capacitance. A PN junction capacitance also gets large and non-linear under the forward bias condition you describe with the integration of light signal. Also, in this case, dark current is remanifested as excessive recombination current (traps are traps are traps).
ReplyDeleteAnyway, we agree that the likelihood of this material going anywhere in consumer land is slim.
Yep, we are in agreement now. Traps are always a problem, in one way or another. Other problems could be solved, but too many traps kill the idea.
ReplyDelete