It is today that the ISSCC officially starts, but yesterday the tutorials, two forums and two evening sessions were organized.  For the imaging community, the “Silicon 3D-Integration, Technology and Systems”  forum was of interest. And especially the talk by Jean-Luc Jaffard of ST Microelectronics, entitled : “Chip Scale Camera Module Using Through Silicon Via Technology”.  The agenda of his presentation looked as follows :  imager market outlines, camera module physical size contributing factors, optical technologies, through silicon via process, camera module assembly and reliability, future evolutions and conclusions.

In general this talk could be seen as a great overview of the state-of-the-art of TSV for imagers.  For the non-imaging experts in the room the talk contained a lot of new information (and basically the forum was addressing a broad spectrum of interested people, not just imaging experts).  For the imaging engineers, the most interesting part was situated in the second half of the talk.

Jean-Luc showed a nice comparison of pros and cons of an injected plastic lens, a molded glass lens and wafer level optics.  Basically wafer level optics go hand-in-hand with TSV technology.  A series of cartoons illustrated the process of the TSVs, as well as the future integration of camera modules : wafer lens, TSV sensor, image processing, memory all stacked on top of each other as a real 3D masterpiece.  Discussing about the wafer level camera roadmap, Jean Luc mentioned that today the modules already make use of wafer level optics, but that the modules are still individually assembled.  Next step will be the combination of wafer level optics and TSV sensors bonded before dicing, while in a couple of years from now  we will see wafer level optics, including auto-focus means, bonded to a TSV sensor before dicing. 

If someone thinks that the development of camera modules soon will come to an end, the answer is simply NO ! 

Albert, 08-02-2010

Wishing your a good mornig from my hotel room in San Francisco.  Basically the International Solid-State Circuit Conference is ready to start !  Over 95 % of the presenters arrived already and had yesterday, Saturday, a long rehearsing day.  Today the plenary speakers will rehearse in the big ballroom, while in parallel the educational activities will start.  On Sunday we will have 9 tutorial sessions (each 1.5 hour) running in two parallel tracks.  That allows those who are interested to attend maximum 2 or 3 tutorials.  Attending a tutorial is a very nice way of getting familiar with certain topics in new fields.  Also on Sunday the first 2 forums are organized.  A forum is a full-day event focusing on one particular topic with for instance about 7 invited top experts in their field.  In one of two today’s forums (”Silicon 3D Integration Technology and Systems” and “Reconfigurable RD and Data Converters” , Jean-Luc Jaffard of ST Microelectronics will talk about TSV for imaging modules.   I do hope that I can attend Jean-Luc’s presentation, I can not follow the complete program due to other obligations.  As there are the plenary rehearsals that I have to chair.

This evening a first evening session is organized (”Beyond CMOS - Emerging Technologies”)  in parallell to the Student Research Preview.  The latter gives MSc and PhD students the opportunity to present their work during a flash presentation and by means of a poster.

Starting from tomorrow, Monday, the paper presentations will start.  I do hope to give you a short update on a daily basis of what is/was important for the imaging community.  I hope you will like it.

Albert, 07-02-2010

Last time we discussed the effects of column noise on the PTC, this time we highlight the row noise.  Row noise can have a temporal character as well as an FPN part, but moreover, this FPN can be repetitive.  For instance, a certain FPN pattern will be repeated every x lines.  The strategy in this study is always the same : set all other noise sources to zero, except the generation of the dark current and the dark-current non-uniformities (DSNU), and check how the row noise components influence the PTC. 

By means of the mathematical model 100+ dark images were generated at various exposure times (between 0 s and 65 s). The result of this exercise in dark can be seen in the following six figures :

-       Figure 1 contains the average dark signal (left axis), and its fixed-pattern noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.)

-       Figure 2 shows the dark fixed-pattern noise versus the dark signal, based on the data shown in Figure 1.

-       Figure 3 shows the row fixed-pattern noise (measured in dark) as a function of the integration time.

-       Figure 4 shows the Fourier transform of the row fixed-pattern noise.

-       Figure 5 contains the average dark signal (left axis), and its temporal noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.) 

-       Figure 6 shows the dark current temporal noise versus the dark signal, based on the data shown in Figure 5.

Figure 1 : Dark current and its FPN component as a function of the exposure time.

As can be seen in Figure 1, the average dark signal is linear with the integration time, at least for the exposure times that do not saturate the pixel.  This indicates that the dark current is responsible for the signal in dark.  The linear relation between the dark signal and the exposure time (t_exp expressed in ms !), shown in Figure 1, holds for the linear part of the curve.  Notice that the expression, as well as the curve, shows the presence of a DC offset.  The curve of the fixed-pattern noise, shown on the right axis, is not influenced by this DC offset.   The slope of the curve representing the FPN as a function of the integration or exposure time is not influenced by the row FPN either.  This is not surprising because any row FPN will be totally independent of the exposure time. 

From the two formulas shown, it can be deduced that the FPN component is 1/6.6 or 15.2 % of the dark signal in the linear region and becomes 4.9 % of the full-well level when the pixels are saturated.  The latter is representing the pixel non-uniformity in saturation.  Notice also the change in offset contained in the noise formula : the offset was (see previous blogs) 0.168 DN, and is now changed in 0.260 DN.  The offset is referring to the FPN at an exposure time equal to 0 s, and the increase in FPN can be attributed for 100 % to the increase in row FPN.  But the curve mentioned in Figure 1 is the FPN on pixel level !

Figure 2 : Dark FPN versus dark signal.

The corresponding “PTC” curve is illustrated in Figure 2 : the FPN versus the (dark) signal is shown.  From this PTC curve several interesting parameters can be deduced :

-       The DSNU can be found to be equal to : 1/10^0.816 = 0.153 or 15.3 % at 30^oC,

-       The saturation non-uniformity : 10^2.138 = 137 DN.

Notice that also Figure 2 shows the data on pixel level.  Next, to find more information with respect to the row behavior, the FPN is calculated on row level.  This can be done by calculating an average value for every row, and then, calculating the standard deviation on these row-average values.  The result of this exercise is shown in Figure 3.

Figure 3 : Row-level FPN.

The graph of row FPN versus the exposure time can be explained as follows :

-       For very low values of the exposure time, the FPN is dominated by the FPN introduced by the row circuitry, and is independent of the exposure time.  It can be found that the row FPN is equal to 10^-0.268 DN = 0.54 DN,

-       For larger values of the exposure time, the FPN proportionally increases with the exposure time.  In this region, the DSNU on pixel level is the dominant source of FPN, even if all pixels within a row are averaged.  The DSNU (on row level !) found is equal to 1/S_d at the moment the noise is equal to 1 DN.  This situation happens at an exposure time equal to 10^3.090 ms = 1.230 s.  From Figure 1, it can be deduced that for an integration time of 1.230 s a signal value equal to 80 DN can be found.  Or the row level FPN is equal to 1/Sd = 1/80 = 0.0125 = 1.25 %, which is a factor of 16/1.25 = 12.8 lower than the same parameter on pixel level presented in Figure 2.

-       The FPN curve reaches a maximum and then the FPN moves to a steady-state value of 10^1.030 DN = 10.72 DN due to the anti-blooming non-uniformities on row level.  This value is a factor 137/10.72 = 12.8 lower than the same parameter on pixel level.

Notice that the reduction of FPN (in the non-saturated region as well as in the saturated region) from pixel level to row level equals to a factor of 12.8.  This value should not be a surprise, because it is the reduction in noise after averaging the noise of 160 pixel values into one row value : sqrt(160) = 12.6 !

Figure 4 : Fourier transform of the row fixed-pattern noise.

Something not yet discussed in the study of the PTC in relation to the various noise sources is the calculation illustrated in Figure 4 : the row-FPN is shown in the frequency domain.  To get this result, the fixed-pattern noise on row level, as shown in Figure 3, is Fourier transformed.  The outcome is shown on the vertical axis of Figure 4, as a function of the sample frequency.  (The sample frequency is defined by the number of lines.)  As can be seen, a very regular frequency pattern is popping up at multiples of 0.0625 times the sample frequency.  In combination with the 512 lines of the imager, this corresponds to a repetitive row fixed-pattern noise with a repetition period of 0.0313 x 512 = 16 lines.  Finding a certain repetitive pattern in the FPN can give very important information to find the root cause of the FPN during the optimization process of a sensor and/or camera design. [Remark : all data shown in this blog is based on an imager size of 160 (H) x 120 (V) pixels, except Fig.4 which is based on an imager with 160 (H) x 512 (V) pixels, all other parameters remained the same for all calculations.]

Figure 5 : Dark current and its temporal noise component as a function of the exposure time.

Figure 5 shows the dark signal and the temporal noise as function of the exposure time.  Not really that much new information can be extracted from these graphs, except the minimum temporal pixel noise being equal to 0.98 DN.

Figure 6 : “PTC” of the sensor.

 Figure 6 shows the real Photon Transfer Curve, in which the temporal noise is shown as a function of the signal.  The curve shows an indication of the part that is (nearly) independent of the dark signal (with a slope of 0) as well as the part that is directly depending on the dark signal (with a slope of 0.5).  The region in the graph showing a collapsing curve indicates the saturation of the pixels. 

From the PTC curve the following parameters can be deduced :

-       The conversion gain, being equal to 1/10^0.724 DN/e^- = 0.188 DN/e^-,

-       The total temporal pixel noise (without any influence of the dark current)  = 10^0.010 DN = 1.02 DN, this minimum noise level is representing the temporal row noise because all other temporal noise sources are set to zero (even the dark current shot noise is zero at an exposure time equal to zero),

-       The onset of anti-blooming = 10^3.32 DN = 2090 DN,   

-       The saturation level of the pixels = 10^3.45 DN = 2818 DN,

Conclusion : by generating images in dark at different exposure times, very important information can be extracted w.r.t. to further optimization of a new sensor/camera or to evaluate an existing sensor/camera.  The methods described are strong tools when it comes down to benchmark existing products available on the market.

Next blog will focus on the noise contribution of the output amplifier (CDS, AGC, …).

Albert 2010-01-21

After discussing the FPN and temporal noise generated in the pixels, the attention will now be shifted to the noise generated in the column circuitry.  This can be FPN as well as temporal noise.  To study these column-wise noise sources, all other noise sources will be set to zero, except the generation of the dark current and the dark-current non-uniformities (DSNU).

By means of the mathematical model 100+ dark images were generated at various exposure times (between 0 s and 65 s). The result of this exercise in dark can be seen in the following five figures : 

-       Figure 1 contains the average dark signal (left axis), and its fixed-pattern noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.)

-       Figure 2 shows the dark fixed-pattern noise versus the dark signal, based on the data shown in Figure 1.

-       Figure 3 shows the column fixed-pattern noise (measured in dark) as a function of the integration time.

-       Figure 4 contains the average dark signal (left axis), and its temporal noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.) 

-       Figure 5 shows the dark current temporal noise versus the dark signal, based on the data shown in Figure 4. 

Figure 1 : Dark current and its FPN component as a function of the exposure time.

As can be seen in Figure 1, the average dark signal is linear with the integration time, at least for the exposure times that do not saturate the pixel.  This indicates that the dark current is responsible for the signal in dark.  The linear relation between the dark signal and the exposure time (t_exp expressed in ms !) shown in Figure 1 holds for the linear part of the curve.  Notice that the expression as well as the curve show the presence of a DC offset.  No effect of the column FPN can be seen.  This is not surprising because the curve shown refers to the FPN on pixel level as a function of the integration time, and any column FPN will be totally independent of the exposure time. 

From the two formulas shown, it can be deduced that the FPN component is 1/6.6 or 15.2 % of the dark signal in the linear region and becomes 4.9 % of the full-well level when the pixels are saturated.  The latter is representing the pixel non-uniformities in saturation.

Also notice the change in offset contained in the noise formula : the offset was (see previous blogs) 0.168 DN, and is now changed in 1.212 DN.  The offset is referring to the FPN at an exposure time equal to 0 s, and the increase can be attributed for 100 % to the increase in column FPN.  But the parameter mentioned in Figure 1 is the FPN on pixel level ! 

Figure 2 : Dark FPN versus dark signal.

The corresponding “PTC” curve is illustrated in Figure 2 : the FPN versus the (dark) signal is shown.   From this PTC curve several interesting parameters can be deduced :

-       The DSNU can be found to be equal to : 1/10^0.782 = 0.165 or 16.5 % at 30^oC,

-       The pixel FPN (without DSNU) : 10^0.278 DN = 1.897 DN (or after finding the conversion gain, this corresponds to 11.9 e^-),

-       The saturation non-uniformity : 10^2.138 = 137 DN (or after finding the conversion gain, this corresponds to 859 e^-).

Notice that also Figure 2 shows the data on pixel level.  To find more information with respect to the column behavior, the FPN is calculated on column level as well.  This can be done by calculating an average value for all pixels in every column, and next, calculating the standard deviation on these column-average values.  The result of this exercise is shown in Figure 3.

Figure 3 : Column-level FPN.

The graph of FPN versus the exposure time can be explained as follows :

-       For very low values of the exposure time, the FPN is dominated by the FPN introduced by the column circuitry.  It can be found that the column FPN is equal to 10^0.283 DN = 1.92 DN,

-       For larger values of the exposure time, the FPN proportionally increases with the exposure time.  In this region, the DSNU on pixel level is the dominant source of FPN, even if all pixels within a column are averaged.  The DSNU (on column level !) found is equal to 1/S_d at the moment the noise is equal to 1 DN.  This situation happens at an exposure time equal to 10^3.003 ms = 1.007 s.  From Figure 1, it can be deduced that for an integration time of 1.007 s a signal value equal to 66 DN can be found.  Or the column level FPN is equal to 1/Sd = 1/66 = 0.015 = 1.5 %, which is a factor of 16/1.5 = 10.7 lower than the same parameter on pixel level,

-       The FPN curve reaches a maximum and the FPN moves to a steady-state value of 10^1.112 DN = 12.94 DN due to the anti-blooming non-uniformities on column level.  This value is a factor 137/12.94 = 10.6 lower than the same parameter on pixel level.

Notice that the reduction of FPN (in the non-saturated region as well as in the saturated region) from pixel level to column level equal is to a factor of 10.7.  This value should not be a surprise, because it is the reduction in noise after averaging the noise of 120 pixel values into one column value : sqrt(120) = 11 !

Figure 4 : Dark current and its temporal noise component as a function of the exposure time.

Figure 4 shows the dark signal and the temporal noise as function of the exposure time.  Not really that much new information can be extracted from these graphs, except the minimum temporal pixel noise being equal to 0.97 DN.

Figure 5 : “PTC” of the sensor.

Figure 5 shows the real Photon Transfer Curve, in which the temporal noise is shown as a function of the signal.  The curve shows a part that is independent of the dark signal (with a slope of 0) and a part that is directly depending on the dark signal (with a slope of 0.5).  The region in the graph showing a collapsing curve indicates the saturation of the pixels. 

From the PTC curve the following parameters can be deduced :

-       The conversion gain, being equal to 1/10^0.753 DN/e^- = 0.176 DN/e^-, (this is a relative large value, due to the fact that the slope of the shot-noise limited part is not really equal to 0.5),

-       The total temporal pixel noise (without any influence of the dark current)  = 10^0.012 DN = 1.02 DN = 6.4 e^-, this minimum noise level is representing the temporal column noise because all other temporal noise sources are set to zero (even the dark current shot noise is zero at an exposure time equal to zero),

-       The onset of anti-blooming = 10^3.32 DN = 2090 DN = 13480 e^-,   

-       The saturation level of the pixels = 10^3.45 DN = 2818 DN = 18180 e^-,   

Conclusion : it is amazing how many parameters can be deduced from the various curves shown in this part of the study.  But it should be remarked that in this discussion the specific column-related parameters can be calculated in a simple way because all other noise sources are set to zero.  It will be shown later that when more noise sources are influencing the output signal, it will be much more difficult to extract the column parameters …

Next time the effect of row noise will be investigated.  As can be expected, it will be very similar to the discussion of the column noise, but with an very interesting extra characteristic : repetitive row noise.  Come and see next time !

Albert 2010-01-06

At the end of the 2009 I would like to take the opportunity to wish all my readers a Merry Christmas and a Happy New Year. 

Everyone is telling that the economy is recovering, let’s hope that this will be the driver for a successful 2010.  I heard from various industrial sources that CCD fabs as well as CMOS fabs are completely filled with image sensor wafers.  Apparently the future for imaging looks bright.

Looking back to 2009, it was a difficult year for many of us in the imaging business.  If I look to my own activities, 2009 started with a drastic reduction in attendees of the ISSCC2009.  If I recall the numbers, the amount of participants was about 66 % of the previous years.  Being the vice chair of the technical program committee, I was reponsible for the evening and educational activities.  Also within the ISSCC organization ”cost-cutting” became a buzz-word.    As far as my own teaching activities were concerned, in the first half of 2009 the number of in-house trainings dropped, and several public courses had to be cancelled because of lack of participants.  Despite all the negative signals, the International Image Sensor Workshop in June was a big success.  Being the general chair of the workshop, I was very much concerned about the financial consequences if we had to face a low number of papers and/or participants.  But the workshop was sold out in just a few days, Johannes Solhusvik helped me putting together a very strong technical program, and we combined the workshop with a one-day symposium on BSI which attracted a lot of attention. 

After the Summer holidays, the teaching and training activities started to recover.  New in-house courses were organized, and in the Fall of 2009 only one public course needed to be cancelled.  The drop in number of courses in the middle of the year gave me the time to work on other items, and the result of that will be a brand new course in 2010.  Although I am still working on it, the course will become available in the second half of 2010.  It will be the first course ever in digital imaging with hands-on evaluations and measurements in the class room.  New hardware is acquired (11 cameras, light boxes, power supplies, laptops, etc.) and at this moment I am preparing the exercises for the course.  The first four experiments are ready : illustrating the improvement in S/N when multiple images are averaged, measuring the fixed-pattern in dark (pixel, column and row FPN), measuring the temporal noise in dark and measuring the pixels with random-telegraph pattern (RTS pixels).  More assignments will follow.  It is a lot of work, but it is also a lot of fun to prepare the set-ups.  I am really looking forward to work with it in a class room.  So 2010 will bring its own challenges, but it so much more motivating to work on these positive challenges to expand and improve my own business.

I wish all of you the very best for 2010, and hope that we will regularly “meet” through this blog.  Thanks for visiting the website of Harvest Imaging !

Albert, 24-12-2009.

During the last blog about the study of the PTC (Photon Transfer Curve) I promised to continue with the gain of the on-chip PGA (Programmable Gain Amplifier).  But in the mean time I changed my mind.  The PGA gain will follow at a later stage.  At this moment I would like to focus on the noise of the pixel : FPN (coming from some residual off-set) and the temporal noise (coming from remaining 1/f noise that is not cancelled by the CDS, Johnson noise in the pixel, and maybe kTC noise if not cancelled by CDS).  The FPN due to dark current non-uniformities (DSNU) is already discussed in an earlier blog, while the FPN from light sensitivity non-uniformities (PRNU) will be addressed in a future one. 

By means of the mathematical model 100+ dark images were generated at various exposure times (between 0 s and 65 s). The result of this exercise in dark can be seen in the following four figures (notice that all noise sources are set to zero, except the dark current shot noise, the DSNU, pixel FPN and pixel temporal noise) :

-       Figure 1 contains the average dark signal (left axis), and its fixed-pattern noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.)

-       Figure 2 shows the dark fixed-pattern noise versus the dark signal, based on the data shown in Figure 1.

-       Figure 3 contains the average dark signal (left axis), and its temporal noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.) 

-       Figure 4 shows the dark current temporal noise versus the dark signal, based on the data shown in Figure 3. 

Figure 1 : Dark current and its FPN as a function of the exposure time.

As can be seen in Figure 1, the average dark signal still is linear with the integration time, at least for these exposure times that do not saturate the pixel.  This indicates that the dark current is responsible for the signal in dark.  The relation between the dark signal and the exposure time (t_exp expressed in ms !) shown in Figure 1 holds for the linear part of the curve.  Notice that the expression as well as the curve show the presence of the DC offset (introduced by the analog circuitry). 

The curve of the fixed-pattern noise, shown on the right axis, is not influenced by this DC offset : for an exposure time of 0 s, the FPN is 0 DN as well.    

From the two formulas shown, it can be calculated that the FPN component is 1/6.6 or 15.2 % of the dark signal in the linear region and becomes 4.9 % of the full-well level when the pixels are saturated.  The latter is representing the pixel non-uniformities in saturation. 

Figure 2 : Dark FPN versus dark signal.

The corresponding “PTC” curve is illustrated in Figure 2 : the FPN versus the (dark) signal is shown.   From this PTC curve several interesting parameters can be deduced :

-       The DSNU can be found to be equal to : 1/10^0.814 = 0.153 or 15.3 % at 30^oC,

-       The pixel FPN (without DSNU) : 10^-0.196 DN = 0.631 DN (after finding the conversion gain, this corresponds to 3.9 e^-),

-       The saturation non-uniformity : 10^2.140 = 138 DN (after finding the conversion gain, this corresponds to 862 e^-).

Figure 3 : Dark current and its temporal noise component as a function of the exposure time.

Figure 3 is showing the signal and the temporal noise as function of the exposure time.  Not really that much new information can be extracted from these graphs, except the minimum temporal pixel noise being equal to 1.63 DN. 

Figure 4 : “PTC” of the sensor.

Figure 4 shows the real Photon Transfer Curve, in which the temporal noise is shown as a function of the signal.  The curve clearly shows the part that is independent of the dark signal (with a slope of 0), the part that is directly depending on the dark signal (with a slope of 0.5) and the part showing a collapsing curve indicating the saturation of the pixels. 

From the PTC curve the following parameters can be deduced :

-       The conversion gain, being equal to 1/10^0.810 DN/e^- = 0.155 DN/e^-,

-       The total temporal pixel noise (without any influence of the dark current)  = 10^0.213 DN = 1.63 DN = 10.5 e^-,   

-       The onset of anti-blooming = 10^3.32 DN = 2090 DN = 13480 e^-,   

-       The saturation level of the pixels = 10^3.45 DN = 2818 DN = 18180 e^-,   

As could be expected : additional pixel noise is popping up in the flat part of the PTC curves.  In these regions where the pixel noise is the dominant noise source (and where it is not overruled by dark-current shot noise or by saturation non-uniformities), its value can be easily deduced by means of the PTC curves.

Next time the effect of column noise will be investigated.

Albert 2009-12-19

In this short blog I would like to report on (what were for me !!) the highlights of the workshop entitled : “CMOS Detectors for High Performance Applications”, organized by Alex Materne (CNES), Oliver Saint-Pe (Astrium) and Christophe Renard (Thales Alenia Space).  Unfortunately I could not attend all presentations, so there might have been more important things presented than just shown over here.  Sorry to the presenters that I missed.

-          P. Robert (ULIS) explained how to increase the dynamic range of infra-red imagers by implementing extra capacitance in the column circuitry,

-          B. Dupont (Caeleste) showed a hybrid photon counting sensor based on SPADs,

-          B. Fowler (Fairchild Imaging) had another talk about their 5.5 Mpixel low-noise sensor.  New to me was the availability of the colour version of this sensor, as well as the back-side thinned version.  The combination of colour and back-side illumination will not be offered.  In the same talk a rad-hardness of 30 Mrad was mentioned.  Seems extremely high to me, but then the discussion popped up how the rad-hardness is defined ?  No data was given about the performance of the device in global shutter mode.  We have to wait till the upcoming SPIE conference in January 2010.

-          T. Baechler (CSEM) presented their work on in-pixel amplification with p-type transistors configured as a regulated cascade in the columns.  Cleaver idea, but at this moment no measurements are available.  An interesting model was presented to show the importance of the dark current if single photon detection is needed.

-          G. Lepage (CMOSIS) talked about the TDI option in CMOS, and how his project evolved from an analogue storage towards a digital storage on-chip for the intermediate data of the TDI.

-          J. Bosiers (DALSA) showed that a front-side illuminated CMOS imager can be light sensitive till 250 nm, at least if the appropriate measures are taken in the technology,

-          J. Pratlong (E2V) presented a large pinned (?) photodiode of 24 um x 24 um with the readout circuitry nicely placed in the middle of the diode.  Unfortunately no real details of the lay-out were shown, it would be of great interest to see how  a large conversion gain (90 uV/e-) can be obtained by such a structure.  Image lag of 5 electrons are reported, this seems a very good value for such a large pixel.

-          B. Cremers (Cypress) reported about an imager with a partially pinned photodiode.  Interesting to notice is the fact that the author talked about a couple of artifacts of the devices.  This is not that often the case that companies report about issues.

-          A. Peizerat (CEA) gave an interesting overview of requirements and options for ADCs in CMOS image sensors (die-level, column-level and pixel-level).

-          I. Djite (ISAE) talked about MTF simulation and showed his simulation results together with measurements.

-          S. Demiguel (SAGEM) focused on back-side illumination of CMOS imagers for low-light level, but the major part of his presentation was on a very nice overview of low-light level imaging options (tubes, EM, intensified, back-side illumination, other materials).

-          P. Jerram (E2V) talked about 2 CMOS imagers that are back-side illuminated.  Very nice talk with a lot of information on the BSI process itself. 

-          J. Bosiers (DALSA) highlighted their large CMOS tile of 77 mm x 145 mm, 3 sides buttable and intended for X-ray applications.  During the talk, the BSI process of DALSA was described as well, based on back-side charging to passivate the back-side of the thinned sensors.

-          P. Cemeli (Soitec) explained the back-side process in the case the starting material is SOI.  Quite a bit of technology information was given and actually this talk together with the one from P. Jerram gave an excellent overview of the ins and outs of BSI on bulk silicon and on SOI.

-          B. Dryer (E2V) presented first results on radiation damage introduced in 0.18um CMOS imagers.  Nice results but more experiments will follow.

-          Round table discussion : a panel of 8 people discussed the issues of getting access to CIS foundries for small volumes and/or for process changes.  It was concluded that process changes are out of the question (except for some small changes in implantation dose) and that the scientific/space community is that much fragmented that higher volumes can never be reached.  The only option left is to search for a common ground to increase the volumes.  This can be done by putting all roadmaps of all agencies on the table and focus on a common interest for all devices.  Otherwise the situation will not change …. and in times that the fabs are filled, the situation can only get worse.

Conclusion : interesting workshop, no registration fee, small group of people (100), good atmosphere, open discussions, no proceedings.  Thanks to the organizers !

Albert, 10-12-2009.

During the last two weeks I was not able to spend enough time preparing a new technical blog about the continuing story of the PTC.  The reason is very simple : a trip to Brazil kept me busy.

I was invited by prof. Jose Gabriel Rodriguez Carneiro Gomes (Universidade Federal do Rio de Janeiro, UFRJ) and prof. Davies William de Lima Monteiro (Universidade Federal de Minas Gerais, UFMG) to give a talk in order to promote solid-state imaging technology in Brazil.  The workshops were organized in close cooperation with the local IEEE Chapter on Circuits & Systems. 

At this moment the overall semiconductor activities in Brazil are continuously growing, but there is still very limited work going on in the field of imagers.  A few start-up companies are doing quite nice work, but they do not (yet) reach the level of the companies present in North America, Europe or the Far East.  So it was a great initiative of prof. Gomes and prof. de Lima Monteiro to promote image sensors in front of an audience composed out of undergraduate students, graduate students, professors and people from industry.  If the R&D work on image sensors gets more attention, it will be easier to get funding from the government for future projects.  And this holds for academic work as well as for industrial activities.

My talk was split into two parts (each part lasted for 2 hours) :

-       “CMOS Image Sensors : Past, Present and Future”.  The content of this session was based a short historical background, a brief overview of the state-of-the-art and focused mainly on future challenges of CMOS imaging.  If pixels get smaller and/or if more pixels are put on a sensor, the overall speed of the sensor will go down (in frames/s), the dynamic range will lower, the light sensitivity will decrease, and ultimately the signal-to-noise will deteriorate.  The talk gave an overview of the few techniques that can be used to increase the speed, to create a wide dynamic range, to increase the light sensitivity (by means of back-side illumination) and to optimize the signal-to-noise ratio (by means of pixel binning),

-       “Colour Processing”.  In a talk of 2 hours it is pretty difficult to give a detailed overview of the complete signal processing present in a digital camera.  For that reason the content of the presentation was limited to the following subjects : short overview of colour imaging, auto-white balance, colour matrixing and demosaicing.

Other speakers in the workshop were Simon Schneiter (talked about 3D Imaging) and Carlos Mendoza (talked about Smart Vision Systems).  Each workshop was attended by about 100 participants.  Based on the reaction after the workshops, the talks were very well received by the audience.  Hopefully the participants were convinced that imaging in general is a great field to work in.  If so, then more students might be attracted by the subject of solid-state imaging, and it can be a great stimulation to submit more projects.  In the end more projects will be granted, and that was the original goal of the bringing all these people together.

My first visit to Brazil was possible thanks to this great initiative taken by prof. Gomes and prof. de Lima Monteiro.  Thanks to them for inviting me this opportunity to promote digital imaging and congratulation for the perfect organization of the workshops !

Albert 01-12-2009.

The bible rewritten ? Apparently the answer to this question is YES, at least, the bible for the solid-state imaging engineer got a new edition.  Recently IEEE has published the latest special issue of Transactions on Electron Devices focusing on Solid-State Image Sensors.  After similar publications in August 1985, May 1991, October 1997 and January 2003, this is already the 5^th edition of the modern bible.  And as can be noticed, every 6 years a new edition is prepared.  In the late ‘70s, IEEE also published a special issue on Charge-Transfer Devices, but this is not considered as a special issue on image sensors.  In the case the young generation is still interested in the older ED special issues, IEEE has put all Transactions on Electron Devices ever published on a single DVD. 

The latest special issue is guest edited by Eric Fossum together with several guest co-editors (Jerry Hynecek, John Tower, Nobukazu Teranishi, Junichi Nakamura, Pierre Magnan and Albert Theuwissen).  The book contains 29 full-length papers spread over 256 pages.   All papers are grouped in the following categories :

-       Visible spectrum image sensors : several techniques to improve resolution, noise, dynamic range, conversion gain and full well are described.  What a super great performing imager could be made if we could combine all these techniques in a single device … ?

-       Modeling and simulation : also in this group the noise characteristics of the imagers get quite a lot of attention.  Not surprising of course, because noise performance is an important parameter in the definition of the dynamic range of an imager, as well as in the determination of the image quality,

-       On-chip signal and image processing : interesting to read that the CMOS world is still trying to “copy” the TDI architecture introduced many years ago in CCD technology,

-       Emerging technologies and applications : the research on alternative colour imaging techniques is still hot, as well as the retinal implants,

-       X-ray and particle image sensors : the main focus is put on medical applications, in combination with radiation damage effects.

As could be expected, most of the papers deal with CMOS image sensors, but nevertheless, a few CCD papers are included in this special issue (ultra low dark current, high-speed imaging, BSI on high-resistivity).  A nice coincidence with the announcement of the Noble Prize.

Although the quality of the published papers is quite high, it is a bit disappointing to see that the big companies active in the field are not present in this special issue.  Apparently they try to hide the information from their competitors …  Of the big ones active in the consumer imaging business, only Aptina, Toshiba, ST Microelectronics and Texas Instruments are publishing a paper in the special issue on image sensors.  Unfortunately the papers from these 4 companies contain information that was already (partly) presented in other papers or at conferences.  And that is a pitty, the R&D work performed in the large companies is of great quality and quantity.  Hopefully the trend of keeping all that information for themselves will not continue … 

Albert 2009-11-17

What is the status of the CMOS pixel model so far ?  In Figure 1 the general overview of the pixel structure, the column circuit together with the chip-level programmable gain amplifier (PGA) and analog-to-digital converter (ADC) is shown. 

Figure 1: Basic configuration as used in the CMOS pixel model.

At this moment the following parameters are included in the model : dark current, dark current non-uniformities, influence of the temperature, anti-blooming capability, anti-blooming or saturation level non-uniformity, and in this blog the DC-offset introduced by the electronic circuitry will be included.  Because this DC-offset is equal for all pixels, it should not add any additional FPN.  Neither will it change the temporal noise, because the DC-offset is constant over time.  But the presence of such an additional DC value can deteriorate the PTC curves.  As will be shown, to construct the PTC curves the DC-offset needs to be subtracted from the obtained pixel signal.

By means of the mathematical model 100+ dark images were generated at various exposure times (between 0 s and 65 s). The result of this exercise in dark can be seen in the following four figures :

-       Figure 2 contains the average dark signal (left axis), and its fixed-pattern noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.)

-       Figure 3 shows the dark fixed-pattern noise versus the dark signal, based on the data shown in Figure 2.

-       Figure 4 contains the average dark signal (left axis), and its temporal noise component (right axis) as a function of the integration time (horizontal axis).  (See previous blogs to learn how the calculation of the fixed-pattern noise is done.) 

-       Figure 5 shows the dark current temporal noise versus the dark signal, based on the data shown in Figure 4. 

 Figure 2 : Dark current and its FPN component as a function of the exposure time.

As can be seen in Figure 2, the average dark signal is still linear with the integration time, indicating that the dark current is responsible for the signal in dark.  The linear relationship between the dark signal and the exposure time (t_exp expressed in ms !) shown in Figure 2, holds for the linear part of the curve.  Notice that the expression as well as the curve show the presence of a DC-offset.  The curve of the fixed-pattern noise, shown on the right axis, is not influenced by this DC-offset. 

From the two formulas shown, it can be deduced that the FPN component is 1/6.6 or 15.2 % of the dark signal in the linear region and becomes 4.9 % of the full-well level when the pixels are saturated.  As could be expected, neither of these values are influenced by the presence of the DC-offset.  

Figure 3 : Dark FPN versus dark signal, with and without the compensation of the DC offset.

The corresponding “PTC” curve is illustrated in Figure 3 : the FPN versus the dark signal is shown.  Notice the irregular behavior of the curve in the case the “measured” data is not compensated for the DC-offset.  The curve is no longer linear and is much steeper than expected.  But when the DC-offset is subtracted from the signal values (on the horizontal axis), again the “perfect” PTC is becoming available.  The value used for the DC-offsetcompensation is 512.04 DN as can be deduced from Figure 2.

 Figure 4 : Dark current and its temporal noise component as a function of the exposure time.

From Figure 4, showing the signal and the temporal noise as function of the exposure time, the same conclusion can be made as from Figure 2 : the presence of any DC-offset has no influence on the temporal noise component, only the absolute value of the signal is changed.

Figure 5 : “PTC” of the sensor, with and without the compensation of the DC offset.

Next to the conversion gain, also the onset of saturation as well as the full-well level of the sensor can be deduced.  Anti-blooming starts at 1800 DN, and the saturation level is equal to 10^3.450 DN = 2818 DN.

Conclusion : on one hand the presence of any DC-offset (being equal for all pixels) has no influence on the fixed-pattern noise, neither on the temporal noise behavior of the sensor.  On the other hand it needs to be compensated in the case PTC curves are to be generated.

Next time the effect of an extra gain (in the programmable gain amplifier) will be explored.

Albert 2009-11-02