Friday, August 14, 2009

Brief Introduction to CMOS Pixels and PTC Measurements

Rebecca Coath from Rutherford Lab, UK published a summary of pixel architectures and PTC, probably extracted from Albert Theuwissen's Advanced Imaging Course. A nice 17-slide presentation clearly and compactly describes pixel architectures from 4T to 7T and how to interpret their performance from PTC measurements. There are pdf and ppt versions of the presentation.

Update: The files have been removed from CERN by Rutherford Lab request, see comments.


  1. Dear Image Sensor World, thanks for referring to my courses, because Rebecca forgot to do this. Indeed, part of the material comes from my courses.

  2. A.T.,

    Actually, if you go to File|Document Properties in Adobe Reader menu, there is a reference to your course. I think the same is in ppt version. The reference is well hidden, though.

  3. Apparently the reference was hidden in the original title of the file, but that is not shown in the downloads. A.T.

  4. Albert: I hope you can clarify some points here, or rebut what I am about to say. I hope the distortions are due to second hand information, dropping of subtle arguements etc.

    To start with I have the utmost respect for your body of work. I have indeed learned a lot about CCD's from your book. Normally, I'd not say anything when I see a posting like the *.pdf above, but the fact that has some genesis with you I find I must reply, for proactive reasons. Simply put, I can easily see my customers saying "Albert says in his course that 4T doesn't give you global shutter" or some such thing. And I will have to spend a lot of time explaining what is actually happening.

    issue 1) The claim is that the contact or s/d is just so bad that it eliminates it's usage in a 4T structure. Or perhaps I've missed the point of the slide, please correct me if I am wrong.

    The simple logical arguement against this statement is: A 3T structure is very much the same as a 4T, you just have an intervening transfer gate. In fact, you cannot get around the fact that you need a contact some where to be able to attach the gate of the output amplifier to the charge being sensed. If contacts are bad in 4T structures then they will be bad in 3T structures and therefore 3T structures cannot exist, to exagerate for emphasis. Since the bulk of sensors shipped today in fact use this structure AND are 3T then the statement must be false. Or perhaps we know how to make acceptable contacts.

    issue 2) The 5T structure CAN have full CDS, but it's a matter of time scales. What is true is that you cannot have simultaneous integration and readout. But you can have CDS if the minimum integration time is sufficient to read out the dark frame + kTC noise (whilst) integrating. And the frame time is sufficiently long enough that you can fit your integration period AND your signal level readout into it. Yes, you do throw out some collected electrons (during signal read out) but there are applications where this is perfectly acceptable. And to be clear, I am talking about doing the CDS in the digital domain.

    This second point is less contentious because there is the implicit assumption that one must have readout and integration happening at the same time. This may have easily been dropped, especially if you were comparing this to ILT CCD's etc. And generally you may not run across this scenario in a general image sensor.

    Reagrds MR

  5. I tried to look at the file but it says the file does not exist. Is the link broken or does it still work for other people?

  6. Yes, the files have disappeared for me too. May be they just exceeded their download limits - just my guess.

  7. Hi M.R. Thanks for your comments.
    1) You are fully right w.r.t. to CDS : 3T, 4T and 5T pixels can be used in combination with digital CDS. It is a bit more complicated, needs more readout cycles, needs external memory and will suffer from more 1/f noise.
    2) You are also right w.r.t. to global shuttering with a 4T pixel, that is possible as well. But in the case of global shuttering with a 4T you have to give up several advantages of the rolling shuttering 4T. One of them is the excellent dark current performance due to the large dark current generation of the floating diffusion (storage node). The latter has a very high dark current in comparison with the pinned photodiode, and the reason for that is the presence of the electric contact at the floating diffusion (see paper of Loukianova et al). 3T pixels always suffer from this FD dark current, 4T in rolling shutter do not suffer from this FD dark current. That is the mean reason for the excellent dark performance of 4T rolling shutter pixels.

  8. You can get quite low dark current even with a metal contact but it is a non-standard contact.

    In the meantime, I would think the posting of someone's course material without permission would be against CERN policy (in particular, material from which Albert earns a living.) Maybe this is why the material disappeared.

    There are many many cases of people lifting figures from my talks and presentations over the past 15+ years on CMOS image sensors. I am always appreciative when they take the time to attribute the original figure. Lifting an entire presentation with proper attribution is just rude and/or lazy.

  9. Albert: Thanks for going "on record".

    Just a slight typo. I think you meant (Capital letters are mine).

    "One of them is the excellent dark current performance due to the LACK of large dark current generation of the ..."

    As Eric helpfully pointed out we can contact without too much problem, but certainly not contacting at all is easier. This all just points to opportunities for innovation.


  10. Hi,
    Yes, you are right. The floating diffusion (contact) is generating most of the dark current, but in the case of a rolling 4T pixel it is not an issue.
    BTW, who is behind the initials M.R. ? A.T.

  11. M.R.,

    The best contacts I've heard of have 30-60e/s leakage at room temperature. Do you imply it can be better than that?

  12. This is the same "quite low" range I have seen and is low enough for 98% or more of consumer applications.

    On the other hand, my understanding is that the white spot or outlier density is significantly higher. It is not the average/median dark current that is an issue, it is the distribution tail.

    My information dates from around 2002 so perhaps there are better results out there by now. I would think the Advasense guys, in particular, would have alot of information on this.

  13. Eric,

    You probably meant Altasens, rather than Advasense to know about the contact dark current. Advasense uses 4T pixels with no contact to the photodiode.

    Vladimir Koifman,

  14. No, I just temporarily misremembered how your pixel works. Sorry about that.

    I was thinking that with feedback you did not need to do complete charge transfer and you could make a contact directly to the photodiode - but I guess you are doing some multiplexing with the 4th T. Anyway, nevermind.

    I thought AltaSens had switched to 4T pixels.
    Maybe high performance 3T is a lost art these days.

  15. Dear all,

    the files were promptly removed from the CERN server as they should have never been there in first instance. We were infuriated when we discovered they had been made public against our explicit request. They were created by one of my guys just as a tool for revisiting what learnt in Albert's course, and most of the drawings had been re-created for this learning purpose.

    You all know Albert very well, and if you have been lucky enough to attend his course, you also know he is an excellent teacher. The people in my group have all enjoyed his course and would warmly re-commended it (not that any of the people in this blog need it, don't get me wrong :-) ).

    Renato Turchetta
    CMOS Sensor Design Group Leader
    Rutherford Appleton Laboratory


All comments are moderated to avoid spam.