Archive for January, 2011


Saturday, January 29th, 2011

Being primarily an R&D sensor engineer, the most important paper to me on Thursday morning was the one delivered by Toshiba, entitled : “Dark noise in a CMOS imager pixel with negative bias on transfer gate”, by Hirofumi Yamashita et al.

It is well known that a negative bias on the transfer gate (TG) of a 4T PPD pixel will reduce the dark current and especially the dark current non-uniformities coming from the PPD-TG overlap region.  At Delft University we did some research in this field as well.  What was reported in the Toshiba paper is the following : if you bias the TX to a negative voltage, the amount of hot pixels becomes smaller, but apparently the amount of warm pixels becomes larger. (Warm pixels have a smaller amplitude than hot pixels). 

The hot pixels seem to have an activation energy of 0.55 eV, while the warm ones are characterized by an activation energy of 0.20 eV.  So apparently another mechanism is responsible for generating these new and extra warm pixels.  Research has revealed that these warm pixels are depending on the difference between the negative voltage on the TG gate and the voltage on the floating diffusion.  Conclusion : the warm pixels are generated at the TG-FD transition region, and are generated by means of trap-assisted tunneling through the (transfer-) gate oxide.

Testdevices with a thicker gate oxide as well as with a kind of graded n-doped floating diffusion confirmed the abovementioned model.  So the advice is to make sure that the electric fields in the TG-FD region are kept small.  This can/is done by splitting the floating diffusion into two parts : the one closest to the TG gets a doping level of 66 % of the rest of the floating diffusion.

This was a very complete paper : description of the problem, formulating a model, checking the model with simulations and teststructures, and ultimately the implementation of the countermeasures.

To me it is still a surprise that such a simple structure as a 4T pixel can keep us, the imaging community, busy after an incredible amount of person-years of R&D already spent on it.  Apparently the end is still not there.  By itself good news !

Also in the morning session, Morley Blouke gave a nice overview of the early CCD days.  He presented several of the projects he worked on.  For each project he explained the lessons learned.  Nice to get such a look backwards by one of the CCD pioneers.  Perfect idea of the conference committee to invite Morley for this.  Thanks !

This was my third and last message w.r.t. Electronic Imaging 2011.  I reported about just a very limited amount of papers, which does not mean that the other presentations were of less importance, quality or interest.        

Albert, 29-01-2011. 


Wednesday, January 26th, 2011


Today, Wednesday, the conference part on Cameras and Sensors started.  Several interesting papers were presented, but I would like to highlight two of them :

1)            Paper from NHK, presented by Ryohei Funatsu, entitled : “Single-chip color imaging for UHDTV camera with a 33M-pixel CMOS image sensor”.

In bullet form, the following key items were mentioned :

       In het ultra-HD project, NHK went from 4 x 8 Mpixels CCD sensors, to 4 x 8 Mpixels CMOS sensors, to 3 x 33 Mpixel CMOS imagers to now 1 x 33 Mpixel CMOS imager,

       The theoretical MTF of a single chip camera is larger than the MTF of a 4 chips solution (nothing was mentioned about the aliasing),

       A new demosaicing algorithm was illustrated (based on directional correlation in G and correlation with green in B and R), based on a 1D 8-taps filter.  The results shown were quite impressive compared to other demosaicing algorithms,

       Pixels were still 3T, 3.8 x 3.8 mm2,

       Noise figures and speed were mentioned as well, but I missed them in my notes,

       Camera has 16 output channels,

       The coefficients of the CCM were shown : some large negative numbers off-diagonal.

Nice presentation, but I think that there is still room to improve for this impressive camera : switching to a more advanced pixel structure with pinned photodiodes as well as an improvement of the color filters.  Probably some interesting stuff for next year’s conference or for the International Image Sensor Workshop ?

2)            Paper entitled : “Optimizing quantum efficiency in a stacked CMOS sensor”, by Lumiense Photonics, HanVision Co., and Alternative Vision Corp.  Presentation was done by Dave Gilblom of Alternative Vision Corp.

The basic idea of the sensor is not new : the photodiodes are made in a top layer of Si, the in-pixel transistors in a second layer, and the third layer of silicon contains the last part of the circuitry.  What actually surprised me is that 3 small companies took on this very challenging technology, and by the way, are realizing these devices.  Within a couple of days, the first silicon is expected.  If the predictions can be realized, this is something to watch !  The quantum efficiency was boosted by two actions : an optimized AR layer on top of the diodes to increase the transmission of the incoming light and an optimized coating below the diodes to reflect the light (that is not absorbed in the photodiode) as much as possible back into the diode.  Looks all straight forward, but you still have to make it.

Pixels are 7 T (no problem to put a lot of transistors below the actual photo sensitive part), 12 e noise, 180 ke full well, 0.18 mm technology, huge quantum efficiency close to 100 % across a wide spectral range, 5 x 5 mm2 pixels and a dark current at room temperature of less than 1 e/s/pixel.  To be honest, this last number I cannot believe.  At the end of the talk I deliberately asked to the speaker to confirm this number and he did.  But a photodiode lying in the top layer absolutely needs an electric contact to the second layer and this will kill the dark current.  Wait and (literally !) see.

Albert, 26-1-2011.


Tuesday, January 25th, 2011


At the Electronic Imaging conference a great paper was presented this morning (in the session on digital photography), talking about a new type of image sensor.  A high-speed imager was introduced by people from Tohoku University, Japan, paper was entitled : “A prototype high-speed CMOS image sensor with 10,000,000 burst-frame rate and 10,000 continuous-frame rate”.  Presentation was done by Yasuhisa Tochigi.

Basically it is a CMOS imager with a 4T pixel including a LOFIC capacitor for high dynamic range (not surprising to see a LOFIC here, because the co-authors of the paper worked on LOFIC in cooperation with TI Japan).  In total the sensor has 72 H x 32 V pixels.  Remarkable is to find the current source of the source follower in the pixel itself.  This has to do with the speed and voltage losses in resistive column busses.

In the column circuitry, every column has 104 analog memories/pixel.  So every column has 104 x 32 capacitors included.  Capacitors are made as a stacked combination of a MOS transistor and a PIP capacitor.  The 104 analog capacitors per pixel are used to store the information of maximum 52 images in burst mode.  Every pixel signal can be sampled (and stored) twice to measure the reset reference level and the signal level.  So CDS is possible !  Without CDS, the capacitor bank can store 103 images, apparently 1 capacitor is needed to store the offset of the pixel.

To maintain the speed, every pixel has its own column bus, so a complete set of 32 column lines is coming from the pixel array into the column circuitry.  Very impressive !

Some numbers : pixel size : 48 x 48 mm2, fill factor : 35 %, 0.18 mm technology with 2P3M, 60 mV/e, 5 e input-referred noise (speed ?), dissipation : 1W @ 100 kpixels.

The presentation was concluded with some nice demonstrations of the high-speed mode of the sensor, both in burst and in continuous mode.  Great piece of work, very nice presentation.  Congratulations !

Albert, 25-01-2011

Continuation of the blog

Friday, January 14th, 2011


After finishing the long series of PTC stories, I asked for new ideas and suggestions for this blog.  I got several reactions through this blog, through the as well as through private mail communication.  And the result is that for the time being I will continue for 6 or 7 more blogs around the PTC.  As you should be aware of, I developed a software tool to simulate images and to study the influence of different noise sources and specification parameters on the image quality.  I will use this software tool to illustrate the effect on the PTC analysis of the following parameters :

       Temperature at which the camera is running,

       Number of photons,

       Conversion gain,

       QE specification,

       Number of ADC bits,

       Non-linearity of the readout node and the source-follower.

Because I do not want to repeat over and over all the settings of the sensor and the camera, I will list the “standard” settings here that will be use in all further simulations (unless otherwise specified).  Here they are :

       Sensor size : 640 x 480,

       Number of images generated : 25,

       Conversion gain : 40 µV/e,

       Analog gain : 1,

       Temperature : 30oC,

       Exposure time changing from 0 s to 6.5 s,

       Number of ADC bits : 12,

       Dark current : 200 e/s at 22oC,

       Dark random non-uniformity : 30 e/s at 22oC,

       Dark current doubling temperature : 8oC,

       Saturation level : 17,500 e,

       Number of photons : 150,000/s,

       QE : 33 %,

       Saturation level non-uniformity : 5 %,

       DC offset of the output : 125 mV,

       Output amplifier noise : 0.3 mV,

       Temporal row noise : 6 e,

       Fixed pattern row noise : 3e,

       Repetition frequency row FPN : 16 lines,

       Temporal column noise : 8 e,

       Fixed pattern column noise : 12e,

       Temporal pixel noise : 10 e,

       Fixed pattern pixel noise (offset) : 3 e,

       Defects : 100 pixels stuck at “1”, 100 pixels stuck at “0”,

       RTS pixels : 900 pixels, RTS chance : 20 %,

       Non-linearity SF : switched off,

       Non-linearity FD : switched off.

When studying the influence of the aforementioned parameters, the first measure taken is the correction of the defect pixels.  With this knowledge in hand, the first exercise can start : what is the temperature influence on the PTC, or how can the temperature being used to generate a Photon Transfer Curve ?  The answer will follow shortly.


Albert, 14-01-2011.