Talk:Active pixel sensor
|WikiProject Electronics||(Rated C-class, High-importance)|
|WikiProject Photography||(Rated C-class)|
I will be adding much more to this in the coming days. I recently completed a masters thesis on this matter and would like to contribute. --MattCohen 23:23, 8 May 2006 (UTC)
I slightly disagree with paragraph 2 as it implies that the APS concept was "validated" by JPL in the mid 1990's. In reality the APS concept was in full-scale production as early as 1978 for use in Infrared Focal Plane Arrays (IRFPA). Although these devices were CMOS / InGaAs or HgCdTe hybrids, the only superficial difference with the JPL CMOS devices was that the photodiodes were not integrated into the readout substrate.
It is often spun that CMOS APS devices have relatively new and novel detection and readout architectures but the IR FPA designers had been there long before.
For more information read:
"The Infrared and Electro-Optical Systems Handbook: Volume 3 - Electro-Optical Components - Chapter 5 -Readout electronics for infrared sensors" John L. Vampola (1993) - A great text, one of the first books on CMOS imager array design
"Infrared readout electronics: a historical perspective" M J. Hewitt, J L. Vampola (1994) SPIE Vol.2226
"General noise processes in hybrid infrared focal plane arrays" M D. Nelson, J F. Johnson, T S. Lomheim (1991)
- I've added this to the page, though someone with deeper subject knowledge needs to review my changes! - David Gerard 12:17, 21 September 2006 (UTC)
Out of date?
This article is out of date and misleading. Even Nikon has finally switched over to CMOS because of better performance, particularly in terms of noise. —Preceding unsigned comment added by 184.108.40.206 (talk) 03:37, 1 September 2007 (UTC)
- Give us a reference so we can know what to change. Dicklyon 04:55, 1 September 2007 (UTC)
- This link has recent press releases (too many to list individually!) announcing new DSLRs by both Canon (which has long used CMOS sensors in its EOS line), and Nikon (which is using CMOS in all of its newest professional and semi-professional cameras): http://www.dpreview.com/news/0708/archive.asp. The relevant models are the Canon 1Ds III and 40D, and the Nikon D3 and D300. It also has a press release about a new Sony CMOS DSLR sensor (which are widely used by most non-Canon DSLR manufacturers). CMOS on the high end has clearly gained the upper hand. Test results on DPR also clearly show CMOS's general superiority to CCD at high ISO (amplification), indicative of their low noise performance. The latest Nikon goes up to ISO 25600! (Though it remains to be seen how successfully.) While some recent Nikon CCD-based cameras are competitive in terms of noise, this is generally at the cost of image detail--that is, noise reduction is being used to blur away noise, rather than the sensor itself being intrinsically lower noise (and Canon CMOS-based cameras have generally still won at the highest ISOs). DPR is a very reputable site; it was recently purchased by Amazon.com, has been referred to in mainstream magazines like Popular Photography, etc. --220.127.116.11 16:53, 3 September 2007 (UTC)
- I agree it's a reliable site, but you haven't pointed out precisely enough what they've said that you want to add to the article. Feel free to add anything that's sourced there; here is their new search for active pixel (not all of which are used in the sense of this article). Dicklyon 17:07, 3 September 2007 (UTC)
- Now that's odd... What I'd heard so far from other sources and had expected to see here in the "comparison to CCD" section was that CMOS sensors had lower overall light sensitivity, and therefore would create a comparitively grainier picture with poorer colour reproduction in low light conditions - as seen with typical cheap webcams and most types of phone camera when used indoors of an evening. Almost looking as if you're using a higher ISO even when the camera itself is set to the same apparent one (and exposure), or silently using high ISO settings in those cases where such things are not exposed to or adjustable by the user (and going up to 25600? ouch. that's got to be either an insanely sensitive chip or Nikon allowing compromise to work around some horrendous shooting conditions). A more expensive webcam that I have bought and used for work purposes happily touts itself as using a "real CCD" for just these purposes, and it certainly does give a better image, though whether that's down to the claimed reasons or just better optics (wider aperture, larger sensor, more precisely ground lens...) and post-processing circuitry, I cannot tell. Can we have hard, reliable sources to back things up one way or the other please, Mr CCD Thesis Man? 18.104.22.168 (talk) 01:20, 9 December 2009 (UTC)
Active Pixel Sensor Concept Origin
The term "active pixel sensor" was coined by Tsutomu Nakamura at Olympus in reference to their Charge Modulation Device (CMD) image sensor - a 1T pixel - that proved (thus far) to be too fussy to produce in volume and has the disadvantage of putting a dead spot almost dead center in the pixel.
Our work at JPL was inspired by this and other attempts at an active pixel image sensor elsewhere. While I wanted to do more interesting devices, we were stuck by budget constraints to just using CMOS as available thru MOSIS.
The essence of the JPL contribution, presently embodied in numerous Caltech-owned patents, is to use a single stage of charge transfer within the pixel to allow kTC noise reduction. We had to use a MOS photogate but would have liked to use a JFET photogate (and we did when we collaborated with Kodak in the following year - now known as the 4T pixel). This combined with on-chip fixed pattern noise removal and on-chip timing circuits allowed a high performance "camera-on-a-chip" for the first time.
Any argument that this was all done before by the IR FPA community in the previous decade is misleading, at best. In fact, I worked at Hughes Missile Systems Group in the early 1980's on IR FPAs with both CCD and CMOS readouts as part of my Howard Hughes Doctoral Fellowship (with Les Kozlowski as my mentor - thanks Les) The CMOS circuits used (source-follower per detector, for example) are more similar to 3T pixels as originally described by Peter Noble in the late 1960's. I heard much later that it was well known that the readout circuits themselves could create a poor quality visible image but no one thought about using them for that, probably because of the horrendous fixed pattern noise, and relatively high readout noise for tactical readout circuits. In fact, years later, I invited Les to my lab at JPL to show him our results with the CMOS APS devices. He was excited by what he saw (it seemed to me) and before long Rockwell was off doing related visible sensor work. (In the meantime I had invited Les to join us in founding Photobit but he thought it was not a good time for him to leave Rockwell for a risky start up).
Back to JPL. It also turned out that 3T devices gave pretty good performance when coupled with the FPN reduction circuits and compared to MOS photogate QE, PD QE was much better, so we also wrote a lot of papers on 3T devices but one should not get confused on what the JPL contribution was. It was, and still is, the basis of all high performance CMOS APS devices - intra pixel charge transfer.
Someday I will write a fuller account of what happened in those days, but we worked hard to transfer the technology to US industry including AT&T Bell Labs (Lucent), Schick Technologies (dental x rays), Kodak, National Semiconductor (Dick Merrill got so inspired he invented the 3-layer 3T derivative and left National for Foveon, and Kevin Brehmer, our other collaborator at National founded his own company as well). The tech transfer was going so slowly (at first) that we decided to form Photobit Corp in 1995 and commercialize our own technology. At the time, VLSI Vision was using passive pixel sensors, and Omnivision just decloaked, also with passive pixel sensors. Within a couple of years they converted to active pixel sensors and Omnivision licensed the technology from Photobit/Caltech.
Thinking that it would be difficult to compete on a fabless basis, we sold Photobit to Micron Technology in 2001.
-Eric R. Fossum (see www.ericfossum.com for list of publications and patents) 18 April 2008
p.s. by the way, Dick, Photobit worked on DSC (DSLRs) with Olympus. The project was killed by the Micron acquisition.
- Eric, thanks for your perspective. I was aware of your main contribution being intra-pixel charge transfer in an active pixel sensor with CDS for kTC noise cancellation. I wasn't aware of Nakamura coining the term. But the origin of the APS concept (not the name) is much older. Not just Noble, but also Weckler, Chamberlain, Weimer, List, and maybe others, published designs in the late 1960s using active amplifiers per pixel. By the 70s, as someone pointed out, they were in use in IR focal plane arrays. I had a variant in an integrated imager I made in 1980 (in nMOS, pre-MOSIS). So it would be good to have all this stuff in the history; but we shouldn't do a lot of historical synthesis and commentary, unless we find such published. Dicklyon (talk) 08:28, 19 April 2008 (UTC)
- Dick, I believe Noble was the first published work on an amplifier in the pixel based on a literature search I did many years ago and it has not been contradicted thus far. Strangely, I think I coined the term "passive pixel" to describe the devices that were more DRAM like, despite being known for evangelizing the "active pixel". I don't think Peter actually made one but he suggested a source-follower per photosite AND a buried photodiode (no intra pixel charge transfer tho). I am well aware of Gene Weckler's work as well as that of Savvas' but I am pretty sure both contributions are predated by Noble. You probably know that these early efforts were pretty much a disaster due to threshold variations and Vt instability inherent in MOS technology at the time. That was why CCDs were quickly adopted since charge domain devices were not affected by Vt issues as much. Also, I was not aware of IRFPAs using MOS readouts until the very early 80's. Are you sure they were around in the 70's? -EF 21 April 2008
- The stuff on IR FPAs was from someone else; I don't know much about that; but if you have a source from the early 80s, we should cite that, and drop the bit about the 70s until someone finds a source. We could use a source on the new assertion about your 1992 invention, too; I can't find a patent filing that early; is it covered in the dinosaurs paper? I can't find my copy of that. As to Peter Noble and the others, I don't have specific info of who was first, just that there were a flurry of papers in 1967/68 by all those guys, each with active readout transistors in the pixels. As you say, CCDs won out at the time; my impression was that was more about density, since MOS couldn't compete on that until it got to be submicron. MOS devices, such as my nMOS 16-pixel binary imager in 1980 and Mead's CMOS silicon retinas were limited to low resolutions until we got through a lot of years of Moore's law. Dicklyon (talk) 16:13, 21 April 2008 (UTC)
- There is no open literature on IRFPAs in that era - it was restricted at the ITAR level or more so. As far as I have heard, density had little, if anything to do with CCDs advantage. One technology worked, one didnt. It was about that simple. The JPL invention was in 1992 and the first paper presented in Feb-ish 93 - the dinosaur paper. To publish that paper, a JPL New Technology Report was filed (not a citeable document) and within a year, the patent was filed by Caltech, and it did not issue for a few more years (5,471,515). By that time the work was well documented. Anyway, the Dinosaur paper was the first published paper, followed by a paper later in the Spring at the IEEE Workshop on CCDs that Savvas hosted at Waterloo. We had a very public debate about the merits of CMOS image sensors vs. CCDs. He was emphatic that he knew that the CMOS (MOS) image sensors would never work well due to his personal experience, as was about 70% of the audience. It was an interesting event. —Preceding unsigned comment added by 22.214.171.124 (talk) 21:48, 21 April 2008 (UTC)