Richard Crisp published his paper about Residual Bulk Image (RBI) in CCDs on his web site. The paper was presented at recent IS&T/SPIE's Electronic Imaging Symposium, San Jose Ca. January 21, 2009. One needs to click on picture to download the pdf file.
While I don't entirely agree with Richard's explanations, the RBI phenomenon itself is quite interesting and the paper is certainly worth reading. By the way, Richard's site has a lot of other imaging-related information.
Update: Richard Crisp adds a bit of explanation to my post:
... my intent behind the paper was to really focus on the characterization and management of RBI, not so much to explain the phenomenon. That has been done many times by others. Janesick discusses it in detail in his book on Scientific CCDs. I only wanted to lightly touch on the root cause and then discuss how to quantify it and to assess the impact of its management on noise of ccds.
An analysis of activation energy(energies) is what is missing, this would provide insight into the mechanism/species and then provide an strategy in terms of gettering at depth.
ReplyDeleteI don't think that he has conclusively proven that it is traps past the denuded zone. It could still very well be surface state traps. This would be solved in 2 parts. 1) again actvation energy analysis and seeing if the traps correspond to known contaminants. 2) a plot of leakage wrt to wavelength over many wavelengths. He only samples 2 wavelengths, without knowing the transition probability of captures you cannot conclude it is at depth. I nice correlation would dispel that.
MR
the sensor operates in MPP mode. It is not surface trapping. It is bulk trapping. The intent of the paper wasn't so much to explain the cause of RBI: Jim Janesick does a fine job of that in his "Scientific Charge Coupled Devices". Instead the intent was to discuss how to characterize it, how to manage it and the impact of the management on the noise of the image.
ReplyDeleteAn additional aspect was the tie-in to the amplifier luminescence: that apparently is a single shot event that occurs after power-up and doesn't recur: the luminescence signature decays as does RBI.
here's Jim Janesick's explanation of the RBI phenomenon in "Scientific Charge Coupled Devices"
ReplyDeletehttp://www.narrowbandimaging.com/images/janesick_rbi.pdf
again, my intent was not to so much discuss the phenomenon as it was to discuss a method to characterize it and to assess the impact of management of RBI on noise.
The phenomenon is quite fascinating for sure though so I understand the interest in it, particularly if it is new to you. I've been seeing it for years in my astronomical images taken using KAF series sensors.
-Richard Crisp
MR, on slide 33 Richard shows temperature slope with leakage doubling every 10C or so. So, the activation energy is not far from mid-bandgap.
ReplyDeleteOne thing that puzzles me is that if there are indeed epi-to-substrate defects, we are supposed to see them in CMOS sensors as well. But I have never heard anybody reporting it, considering the wide variety of CMOS and DRAM processes in use. The effect is quite big, I doubt CMOS vendors would miss it. So, what makes CCDs so different?
I am not sure CCDs are so different from CMOS in this regard. But, RBI is sort of a phenomenological thing and shows up when pushing devices to an extreme. When CMOS sensors are used scientifically it may show up again.
ReplyDeleteI don't subscribe to the explanations given in the past for the observed phenomena. First, if there is a trap density at the epi interface it is going to be very low and spatially confined to within a few angstroms of the epi interface.
I would guess the source of traps might be from impurities and oxygen-related defects gettered during the process from the active layer deeper into the silicon bulk. But again, this is a guess. Gettering is highly dependent on temperature cycling profiles so perhaps these older CCDs had a significantly different temperature/time profile than more modern CMOS devices.
I would expect someone at or from Kodak or philips (DALSA) might know something more about this.
-EF
EF- Thanks for more cleanly describing that. I guess I should be clearer. Here goes.
ReplyDeleteIn general, I'd say COPS might be a prime candidate amongst others. There is no real reason to think that an EPI interface should cause trapping per se, the only mechanism is a doping gradient acting in a gettering fashion, which is some what tenuous as source of traps. That interface could just as easily push junk into the epi area.
To the other commenter:
I've reread Janesicks book (rather the section), thanks for pointing that out.
Janesick does make several assumptions, and there isn't any supporting papers from other authors. For example he assumes that as the interface bulk states fill that the QE increases. Why? IF the polarity of trap is different it could also shield and thus decrease QE. So is the mechanism the trap being ionized? or is it being quenched? It's not clear, it's just assumed and NOT studied.
I did drop a hint before, re: the denuded zone.
Image-sensor, if this is the case then this is the most likely explanation for the difference between CCD and CMOS sensors. Janesick states that it doesn't appear in bulk wafers, or in sensors with vertical A/B or in back-side illuminated sensors which is entirely consistent with both proposals (epi interface, Denuded zone interface) but we know that the denuded zone is put in place precisely to eliminate COPS and to sweep up all sorts of nasties. Why not pick the obvious answer?
Prime wafer specifications have been getting more and more strict generationally, it would be interesting to see if the techniques used in modern prime wafers have been back ported to the high resistivity substrates that the CCD vendors use or if they are just ordering the same wafers. If they haven't change the recipes then, yes, I could see the RBI phenomenon.
This all of course suggests that they should get an implant down there to shield the collection areas above. OR better yet, grow it in place.
Richard, thanks for writing the paper and replying here. I guess I should have looked more closely at what the intent of your paper was. I simply saw the assumptions (and had forgot where they had arisen from - from Jim) and focused on that. And in your defense (and here I'm making assumptions -correct me) you're not active in the design of the sensors.
I guess the core of what I am saying is that there looks to be something interesting to be done here. I think that questioning the logic that it's the epi/bulk interface will yield some good results. Carefully reread what Janesick has written and test the assumptions. I see him having jumped to a conclusion and then explaining the effects (and stretching to make it fit in some cases).
- MR
Yes, you are right that I am not actively involved in the design of CCDs or CMOS image sensors. I use them a lot and have designed lots of other chips, but they are really different (High performance CPUs and DRAMs).
ReplyDeleteIn speaking with another author at the SPIE conference (Rich Westhoff of MIT's Lincoln Labs), Rich mentioned to me that the preclean prior to epi growth is implicated in the traps. There can be significant metallic ion contamination in the baths (depending on when it was last changed) and the logic guys don't care about the deep traps etc that remain according to what Rich said to me.
he also mentioned that there are some operations that claim to have "ccd quality" epi processes. I don't know about that either way but Rich is an expert on Epi Growth (we worked together at a materials company in the past so I know his abilities so I tend to believe him).
What I do know is that the KAF09000 I evaluated has the strongest case of RBI I have seen. I have seen it in KAF3200ME and KAF6303E sensors too, and of my group of 3 I have evaluated, the newer sensors have more RBI than the older ones: the KAF3200 is my oldest sensor and shows the least RBI, The 6303 is mid aged and definitely shows more RBI than does the 3200. The 09000 is in a different class altogether... Really bad RBI
I understand the KAF09000 was designed for medical XRAY so who cares about RBI in that application: the exposures are short and use a scintillator so the light is in the green....
-Richard Crisp
MR, let me try to defend Janesick, at least in part. You say:
ReplyDelete"Janesick does make several assumptions, and there isn't any supporting papers from other authors. For example he assumes that as the interface bulk states fill that the QE increases. Why? IF the polarity of trap is different it could also shield and thus decrease QE. So is the mechanism the trap being ionized? or is it being quenched? It's not clear, it's just assumed and NOT studied."
If we talk about epi-to-substrate interface traps, the trap polarity does not matter, assuming we do not have depletion reaching the interface (correct me if I'm wrong, but Richard's paper and Janesick's picture in the cited chapter support it). If a trap is ionized, it's effectively shielded by majority carriers, so it does not affect minority photocarrier's transport (unless we go to extremely low temperatures, well below those mentioned in Richard's paper).
What does matter for QE is the minority carriers lifetime, and filling the traps by minority carrier's "flooding" should increase their lifetime, no matter the traps are ionized or quenched or something else happens to them. So, in that part I agree with Janesick that filling the traps should increase QE.
I also agree with you that the traps might be located either beyond the denuded zone or even inside the zone. All facts considered, I find it hard to believe that surface traps are involved in this effect, even though it's not 100% proven by the experiments we talk about.
Richard: Rich is right, along with the starting quality of the substrate and the ramp up of the reacting species in the epi chamber, you need to first etch to clear off the surface oxide. But we digress. One would hope that they request special starting material. I presume that the old SiTE CCD's had this issue. Do you know if the Dalsa/Philips units show this?
ReplyDeleteImage-Sensor: Let me re-explain, it's obvious I need the practice! Lets assume that these are mid-band traps (which in the discussion of activation energy above you mentioned that they would be mid-band - so hopefully I'm allowed that ;) ) These can both trap holes or electrons and it's not always clear or easy to assign a polarity. here's the thing, if the trap can grab electrons or holes then one would expect an additive or subtractive signal respectively after the delayed readout. If he assumes trap filling increases QE then he is assuming that the trap is populated (during a previous image with decreased QE then?) and not part of that image and then the QE bounces back even further when that puppy comes out (detraps) in the RBI image. BTW, what is the QE of a RBI image? That's a nonsense question, but it should lead to the observation of conservation of charge over all time scales is what is important. After all if RBI is a positive image, which means electrons here, they had to come from somewhere.
But it's not clear to me that it HAS to always be that way (i.e. a positive RBI image, given the mid-band nature) since we aren't even sure what is trapping. Which leads me back to the observation that there is some room for investigation here for the designers. Which is pretty cool considering how long CCD's have been around.
Thanks for the discussion
- MR
Oh and BTW, I don't think Janesick needs defending. He found something, worked out some of it and moved on like all busy people. I don't think that it should be used uncritically however.
ReplyDelete-MR
all the examples of RBI that I have seen in KAF series sensors (the only ones for which I have used frontside illuminated full frame ccds) have been positive images.
ReplyDeleteAggravating factors that I have personally observed in these uses (all astronomical): unfiltered light, NIR longpass filters, red and or Halpha filters (656nm).
with unfiltered light imaging galaxies (rich in NIR), unsaturated images will initiate RBI in subsequent darks:
http://www.narrowbandimaging.com/rbi_page.htm
I also show RBI resulting from saturated regions as well on other images linked to that web page.
the examples shown may be of interest to the discussion. those were from a KAF3200ME sensor manufactured in 2004, the evaluated KAF09000 was manufactured in early 2008
-Richard Crisp
EF is asking for a reaction from Philips/DALSA. Here it is : we have never seen these effects, because the Philips/DALSA CCDs are using vertical anti-blooming by means of a vertical n(buried channel)-p(well)-n(substrate). This structure has a electrostatic potential minimum around 2.5 um away from the Si-SiO2 interface. Everything generated deeper than this 2.5 um is drained to the n-substrate, only the generated electrons in the top 2.5 um are collected. This structure results in an extremely low dark current, very uniform, low cross-talk (normally due to red photons), superb anti-blooming (up to 100,000 x). The price you have to pay : lower red sensitivity and almost no near-IR sensitivity.
ReplyDeleteBased on the high red sensitivity of the Kodak sensors (KAF3200 and KAF09000) and the lateral overflow drain of these devices, I guess (!!!) that Kodak is not using a vertical npn structure. So this may result in getting these RBI's coming from traps deeper in the bulk.
A.T. ex-Philips/ex-DALSA
here is a page prepared by a Kodak Apps Engineer located in Europe:
ReplyDeletehttp://astrosurf.com/cavadore/CCD/ResidualImage_KAF9000/
--R Crisp