Tuesday, June 24, 2014
Howard Rhodes' Presentation at IS 2014
Yole Developpement reviews Howard Rhodes, Omnivision's CTO, presentation at 2014 Image Sensor Conference in London, UK. The "The Future of CMOS Imaging" presentation shows stacked sensor as the next big thing and discusses the different stacking approaches, among other stuff:
This comment has been removed by the author.
ReplyDeleteIs by "pixel sharing" meant two photodiodes using the same amp and filter perhaps?
DeleteReally new in the area - 4t and 3t don't bring many bells. Would love some link or short explanation or alternative name to the technology.... perhaps....
Thanks in advance.
3T and 4T stand for 3 and 4 transistors (per pixel). The latter allows for significantly lower noise.
DeleteI assume pixel sharing in this context means that pixels share some transistors - this is a technique used especially in the smaller sensors to reduce the amount of silicon not used by the photodiode. You may notice terminology like 1.75T - this is normally a 4 transistor design where several pixels share some transistors, so in practise every pixel only uses 1.75 transistors, using much less silicon than 4 transistors per pixel woul.
Have a look at this tutorial paper from Theuwissen:
Delete"CMOS image sensors: State-Of-the-art and future perspectives"
http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=4430875
It will probably answer some of your questions.
Not to be picky, but didn't the first commercial CMOS sensor from VVL come out in 1995 or thereabouts?
ReplyDeleteVVL was doing passive pixel CMOS image sensors. (I think IVP started before VVL and was doing high speed passive pixel CMOS image sensors for machine vision.) Omnivision and Photobit were both founded in 1995 with Photobit doing APS and Omnivsion doing PPS. According to Peter Denyer, founder of VVL, Omnivision started out as copying VVL passive pixel sensors. After a couple of years, VVL and Omnivision both switched to 3T active pixel CMOS image sensors.
DeleteWell, the development of CMOS imaging devices are much earlier than 1995. During my PhD work (1988), I've designed a 16x16 processing chip for massively parallel computer and I have put a photodiode with an comparator in each processing element in order to be able to provide a binary image directly to the processing elements. Is this APS or PPS ?
Delete-yang ni
do you consider this mass production?
DeleteEric, the slide says CIS MP. I take that to mean the first mass production CMOS sensor of any architecture. That would be the VVL mono, passive pixel sensor. I forget the part number, but I used it around '95.
DeleteFor better or worse, CIS almost always refers to active pixel sensor devices, and in fact, active pixels with intra pixel charge transfer (4T) these days. But this is Omnivision's slide and apparently they are just talking about their contributions to the field. I saw a recent Sony presentation that talked about their invention of the column parallel ADC a few years ago....so go figure.
DeleteAnyway, I think "mass production" is a relative thing. It wasn't long ago (ok, 20 years ago) when Kodak was selling a million CCDs a year and for sure that meant mass production. If you count MOS passive pixel sensors then Reticon was in mass production way before anyone else. Surely VVL was mass producing PPS for Barbie-cams etc. well before Omnivision or Photobit.
Eric, CIS stands for CMOS Image Sensor. Therefore, if it is an image sensor made in a CMOS technology then it is a CIS. There is no active or passive requirement.
DeleteInteresting definition : CCDs are also made in CMOS processes, a CCD is simply a connection of millions of MOS capacitors. So a CCD is a CIS ?
DeleteHmm, many big factories list CIS in their list of available processes. To my knowledge they refer to the modified CMOS process with the extra layers for the pinned Photodiodes (4T pixel)
DeleteAnd,.. for some companies, CIS still means Contact Image Sensor, those line scanners in fax machines, in bill readers and label verifiers, that operate close to the paper.
DeleteAlbert,
DeleteCCDs aren't CMOS, are they? Isn't one of the big problems with CCD processes that both MOS polarities can't (or at least don't need to) be made on the same device? Isn't this the limitation that leads to output stages that are only followers?
Albert, a CCD is not made using a Complimentary-MOS process. It does not contain both n-type and p-type devices, so it is not CMOS.
DeleteMy dear beloved Anonymous friend, we have made for several decades n-channel CCDs in a p-well on an n-type substrate, perfectly suited for n-type and p-type MOS transistors. We have realized CMOS transistors (n-type and p-type) in combination with the CCD, published at IEDM 1984. I do not claim that a CCD is a CMOS image sensor, I only want to point out that you can make CMOS transistors in a CCD proces !!!
DeleteIs it correct that we never had a CCD integrated with ADCs, timing controls, or some image processing digital circuits?
DeleteThat begs the question: why can't "CCD" process allow for ADCs, reference voltage generators,amplifiers, and other circuits in the same substrate? Surely, "CCD" technology is shrinking in minimum feature sizes as well. In fact,we can even introduce transistor per CCD pixel.Then it would be indistinguishable from the "CMOS" APS. I think what Albert is trying to says is that it is the presence of buffer amp per pixel (which has its own advantages/disadvanates) that distinguishes CCD or CMOS, not the technology per-se.
DeleteWell, in a way, I do think it is the process technology that matters. When you start with a CCD process optimized for CCD performance, you may have working NMOS & PMOS devices, but you cannot make them working as well as a CMOS process developed from start without considerations for CCD. Similarly, when you have developed a CMOS process at the beginning, it would be difficult from the cost point of view to add CCD later without an overhaul of the whole thing. Some would simply say the CCD process is not compatible with the modern CMOS process. While for CIS, it is relatively easier to start with a baseline CMOS process, and do some modifications, add-ons, here & there, to have the best of both worlds. For CCD, it is just harder to do, and nobody has done that probably for cost reasons.
DeleteI think this all depends upon what the meaning of the word "IS" is. ;-)
DeletePhilips made a CCD with all timing and driving electronics in full CMOS ! This was presented at IEDM 1984, and was present in products, even in a broadcast camera. If I remember well, Kodak made a CCD with a CDS circuit on-chip. I do not know whether this was CMOS or only n-MOS. It is really relatively simple to make n-type and p-type MOS transistors in a CCD process, but I have to admit, it is not simple to make high performance n-type and p-type transistors in a CCD process. Gate oxides are too thick, junctions are too deep.
DeleteSome foundries offer CCD structures integrated in CMOS process, such as ESPROS and TowerJazz (Kirana project for RAL):
Deletehttp://www.espros.ch/products-and-services
http://www.imagesensors.org/Past%20Workshops/2013%20Workshop/2013%20Papers/Slides/11-02_Turchetta_SLIDES.pdf
Something different, Sony manufactures "Digital Output CCDs" where CCD die and CMOS AFE die are integrated in a single package, such as this one:
http://www.sony.net/Products/SC-HP/new_pro/september_2013/icx804sqb_e.html
The CCD in CMOS is good for certain applications where CTE is not a great concern; I.e.., short chains. It cannot completely replace a "CCD" imager. But this hybrid does have some very interesting application in my view.
DeleteI just recently heard that the hybrid bonding process will likely be available to all customers at TSMC.
ReplyDeleteEric, I heard this rumor last time when I was in Taiwan. Do you have any URL pointing to this information please ? Thanks !
Delete-yang ni