Debayering a DSLR's Bayer matrix.

[Gina]

Here are some photos of the 1100D sensors I've already succeeded in destroying

1. Glass broken and removing the CFA by abrasion with car scratch remover (fine abrasive) - from a while back
   
2. Ditto with more removed
   
3. Current state with the remaining glass removed
   
4. The one I attacked with Dremmel and grinding wheel
   

The abrasive method left an uneven result.

Edited August 5, 2013 by Gina

[Gina]

Very interesting images Luis

[Rottweiler]

I've been following this thread for a bit, now I recovered the cost of many sensors we broke in the process (I could have changed my car for this amount!) I have decided to share our method that we use at JTW Astronomy. It can be combined with the methods here and also we have some safety tips and even improvements for the sensor. We have tried many different ways and have found different methods for different sensors. We accept that some people will do this themselves and if we can use our experience to save one sensor, then we are happy.

First of all, older sensors are MUCH easier to debayer. Not only is the coverglass easier to remove but also the bayer mask is less resiliant. I think this is a result of research into making the mask fade resistent. At least the older cameras have lovely big pixels. A debayered 1100D is FAR more sensitive than a debayered 600D. I have some photometry data that I have been meaning to process. 250 sets... But we are always short on time here. The coverglass off with a knife and try the solvent method. The solvent for this is Dichloroethane, which is a SUSPECTED CARCINGEN (I am told). We actually used basic respirators at first with this and I can confirm it will give you a banging headache, a suitable mask and ventilation or even a fume cupboard are essential. This stuff will lift bayer masks off old sensors and CCDs quite easily.

For newer sensors, the method of the wooden scraper will suffice. We used scrapers made of soft metals that we lapped to a perfectly flat edge. Any mechanical process even with a softer material than silicon can leave micro scratches on the sensor surface or start to remove the coating (reducing sensivity in a local spot). We have tried 24ct gold, silver, selenium and heat treated copper, which is interesting as wood never even crossed our mind, but quite possibly just as good as a lapped gold scraper. With a scraping method regular removal of the dust is essential, especially with soft metallic scrapers which can shed particles which can scratch if tracked across the surface. A small vacuum is fine. A diamond lap to keep the face flat is essential and rubbing is not the best solution. A slight positive cutting edge, which you get under the bayer and then in a forward motion peel it off like a plane peels a door bottom. The width if too thin will cut too deep easily and make the scratches. Too wide and you need too much pressure to get the cut started.

Mechanical methods are unavoidable for newer sensors. You will get traces of bayer mask or micro scratches. To remove this you need to use fine glass polish, such as windscreen repair. You can dilute it with some non-water based sovent such as the dichloroethane. We use 99.999% pure isopropanol. Which also can be used to clean the glass polish away and leave a streak free finish, DCE does NOT leave a streak free finish. You can apply it using a clean cotton bud. It is best to also wash the cotton buds in isopropanol a day before and keep them in a box with holes drilled in the side so they can dry out without dust getting in. The problem with using abrasives is that they do not stop once you are through the bayer like a solvent. You really need to do just a few strokes, clean, check and repeat. You can go through the coating in under a minute once through the bayer.

The major risks in this process are breaking the coverglass, breaking the gold wires or going too deep into the sensor (avoid the Dremel method!). To get the coverglass off the newer sensors you need to use heat. Localised heat like from a butane pen torch. Patience is essential here, heat up one spot where the glass is bonded and the differential thermal expansion will cause the bond to break. Allow it to cool completely and then repeat the process. The broken bond will appear white and you need to 'chase' this break all the way around the coverglass. Which will then simply lift off.

To protect the gold wires, use epoxy resin. Be VERY careful with this, one wrong move and you have epoxy on your sensor. Mix it up and using a fine tip coat the wires in it. When it is hard, you have armoured cables. I cannot guarantee a particular brand of epoxy will not cause weird effects due to electronic properties. We use a type from Bison with no ill effect. An improvement to the sensor is to fit an AR coverglass. We make these ourselves using 0.5% reflectivity AR coated windows. You can get these from Edmund Scientific. If you dont have a fine grinder use a glass cutter with an oiled wheel. The edges will be tatty but you wont see that.

Reassembly of the sensor is also very important, make sure everything is spotless! All traces of the old bonding agent need to be removed with a razor blade. We use epoxy to reattach, only a thin line is needed to get a full width bond, and use a slow setting epoxy so you dont get any potentially harder part causing the coverglass to sit on an angle. Sounds weird but we did experience this.

Good luck with your surgeries, I will add anything else to this as we discover it.

[Luis Campos]

Cheers Patrik,

Many thanks for all that precious info, it is worth a lot trust me

I will try a 450D one of these days, I have my 350D TEC cooling almost finished, with this I can go for very long subs even with narrowband to get that xtra signal

One question please Patrik, do you think hot glue will work to protect the golden wires or is it too risky?

Cheers,

[Rottweiler]

Who is Patrik?

I have not tried hot glue. But people do use it to insulate the driver boards of sensors with a cold finger type cooler. The only concern is that hot glue tends to pour out and you need to keep the height of the glue below the level of the mating face for the coverglass. It also does not set hard and I think if you tried to trim it, it would move and tear the gold wires.

[Rottweiler]

BTW, here are our results with an uncooled monochrome 600D -

[Luis Campos]

Ooopppsss...sorry, my mistake

Yeah, I guess with the epoxy glue you can be much more precise and aply it with a toothpick on the wires steadilly and slowlly, that's a good idea.

When insulating my 450D sensor and board I will use hot glue for sure, I will also use this on my 350D sensor, I broke the cover glass when taking it off, so I'm using just the Baader on the filter holder and aplyed some silicone around just to seal the unit, but obviously some air was trapped inside and when cooling the sensor this air condenses over the sensor surface, so I will disassemble the camera again, clean a bit better the CFA leftovers and then seal the filter hoder and sensor with a thin coat of hot glue all around, and then make two small 1mm holes and purge the thing with Nitrogen, then quickly cover the exiting hole and then the "filling hole" next, I hope to solve my dewing problem like that

Cheers,

Edited August 5, 2013 by Luis Campos

[Luis Campos]

BTW, here are our results with an uncooled monochrome 600D - https://www.facebook...&type=3

Sweet result there...and this was uncooled

Congrats on a fine image,

[Luis Campos]

Very interesting images Luis

Thank you Gina,

Blending the two datas works good

Next will be a Ha RGB test

Cheers,

[Gina]

Thank you "Rottweiler" - that's very useful Few would have revealed their "trade secrets" like that - much appreciated

[pixueto]

Thank you JTW. I Very much appreciate that.

When you mention older models, which cameras are you referring to? are 1000D, a 450D old models?

If I use Dichloroethane with a 1000D or 450D sensor will it remove the CFA without scraping? How do i stop it going beyond the bayer array?

Many, many thanks

[Gina]

I feel a bit wary of applying epoxy resin to the gold wires - I think I'd be happier staying well away from them.

[JamesF]

I feel a bit wary of applying epoxy resin to the gold wires - I think I'd be happier staying well away from them.

You could perhaps mix up some epoxy and apply it by dribbling it onto the wires with a syringe thereby avoiding touching them? No idea if it's just too viscous for that, but Araldite is supplied in syringes so it may work.

James

[Rottweiler]

Dichloroethane will not touch anything later than a 400D unfortunately. It seems to be much better at dissolving CCD bayers, on DSLRs it can take a while. It does not attack silicon though, so it will enable 100% safe removal assuming you dont gas yourself with it

James is correct with the epoxy, you need to dribble it on somehow, we use a pistol which is great for control but these arent cheap. You can chop the end off a cotton bud and allow it to drip on. Maybe draw some targets on a piece of paper and practice your aim? Touching the wires is definitely not recommended.

One point I probably didn't stress enough is how careful to be using the heat method to remove the window. It takes just a second to break the bond, the sensors are fine with this method so long as they don't have the ribbon cable type attachment such as the 5D MkII or 40D. If you rush this, you will probably break the window. It happened to us a few times, worst of it is, it leaves little bits of glass on your sensor, so more cleaning to do.

[Rottweiler]

It is also better to scrape the microlenses off before applying DCE, it seems to be faster. You will know when you are past the microlenses as the surface will be duller and when you swab the sensor with DCE the tip of the cotton bud will be green.

[pixueto]

Dichloroethane will not touch anything later than a 400D unfortunately. It seems to be much better at dissolving CCD bayers, on DSLRs it can take a while. It does not attack silicon though, so it will enable 100% safe removal assuming you dont gas yourself with it

James is correct with the epoxy, you need to dribble it on somehow, we use a pistol which is great for control but these arent cheap. You can chop the end off a cotton bud and allow it to drip on. Maybe draw some targets on a piece of paper and practice your aim? Touching the wires is definitely not recommended.

One point I probably didn't stress enough is how careful to be using the heat method to remove the window. It takes just a second to break the bond, the sensors are fine with this method so long as they don't have the ribbon cable type attachment such as the 5D MkII or 40D. If you rush this, you will probably break the window. It happened to us a few times, worst of it is, it leaves little bits of glass on your sensor, so more cleaning to do.

Thank you JTW. Can the Dichloroethane be applied on the bayer array leftovers on a 100D or 450D to make the scrapping removal process easier? Will it weaken at least the CFA array in a 1000D? I'm having a lot of trouble removing those CFA leftovers even under the microscope!

Many thanks

Edited August 5, 2013 by pixueto

[Rottweiler]

Thank you JTW. Can the Dichloroethane be applied on the bayer array leftovers on a 100D or 450D to make the scrapping removal process easier? Will it weaken at least the CFA array in a 1000D? I'm having a lot of trouble removing those CFA leftovers even under the microscope!

Many thanks

The 100D possibly, the 450D probably not. Take the gold scraper and use the longer side. You can buy 1 grain gold ingots on ebay, they need a bit of modification and the lapping needs to be perfectly flat. Make a holder for it. If you use the short side it will scrape better, the longer side can be used for more selective scraping of remnants. Possibly the most important factor with this is to know when to admit defeat. 100% removal is perfection, perfection is hard to acheive in reality. By chasing that stubborn spot you could scrap a sensor or make a sratch that is far more noticeable. Flat field correction becomes more important than even your dark frames with a mono DSLR. Sub micron defects will be visible, power tools are a bad idea and you won't get that precision with your hands. I have probably debayered 20+ sensors now and I still would not say it is possible to be perfect. I can only hope our hobby grows to the point where we are offered mono DSLR sensors with microlenses from sensor manufacturers. I did enquire about this, sensor prices with 4 zeros on the end put a swift end to that. Production quantities are too low...

It is nice to see so many thank you's too It makes all the mocking from the sensors (they can be cruel) all worthwhile! Perhaps if this mod becomes popular we could take our reward in it being called the JTW method? Similar to the SC1 mods of Mr Chambers We started as a hobby that grew out of control, so to see something we did be potentially helpful to many people really makes us proud!

[pixueto]

Thank you again JTW. Sorry about asking you so many questions but not everyday one gets the chance to talk to someone who could shed light on all the problems I've been encountering with this mod that has now become a kind of personal obsession for me.

So, if I understood correctly, even though it won't work with the 1000D or the 450D, the dichloroethane can actually remove the bayer array in a 350D, right? Whithout any scraping involved, thus removing the posibility of leaving scraches on the silicone bed under the CFA?

And talking about scratches, there is something that puzzles me and I was hopping you could help me understand. After scrapping my 1000D sensor, I can see I've created a good few scratches. However, when I took some photos, there is no evidence that those scratches are there. I even took a flat frame at F36 and still couldn't see any of them. I run a few tests for dead pixels and the sensor seems to be fine. Now, surely those scratches must have an effect, right? How come they don't show up?

Many thanks again for your help.

[Rottweiler]

If a scratch is deep enough you could hit something critical and get things like dead columns or traps appearing (can be interpolated out though). If they are invisible at F36 then I would say whats the problem? I wouldn't even attempt to polish these out. you will only take the layer down to the level of the scratch anyway. Maybe some accurate lab testing would find a discrepancy in count or something, but for astrophotography? Kind of reminds me of a colleague I had who complained his car rattled at 100mph... stick to the speed limit! Hahaha!

The good thing is that a flat field pretty much fixes remaining spots of bayer (not too big) and also the scratches to an extent. I have never tried a 350D. DCE is not so expensive, try it, you may have a pleasant surprise waiting for you. When you modify anything you are always going to be compromising. If the compomise is so small it takes a lab to find it, victory!

The other thing is you need to convert to TIFF before performing any processing, flat fielding a CR2 from a mono modded DSLR can yield some gnarly results, because the computer is still treating it is a colour image and doing some preprocessing based on 2x2s, I think, at least. Image processing is not my forte, but this seems to work, the reasons are pure speculation.

[alistairsam]

Thanks so much Rottweiler, great to see you sharing your professional experience.

i had a few questions. when you say the 1100D is "FAR" more sensitive than the 600D, is it really discernible? would you have any tests with a debayered 1100D?

I really like the 600D for its swivel screen so was hoping to debayer that rather than the 1100D.

with using heat to remove the cover glass, can a soldering iron's tip be used? ever tried that? or would a hot air rework station work as well? The soldering iron tip, if it can get hot enough, would be a lot safer than hot air.

how difficult is it to cool a 600D compared to a 1100D?

what other models have to successfully debayered? that info would help a lot for us in deciding which model to try next.

Thanks so much for the info.

Alistair

[Rottweiler]

Thanks so much Rottweiler, great to see you sharing your professional experience.

i had a few questions. when you say the 1100D is "FAR" more sensitive than the 600D, is it really discernible? would you have any tests with a debayered 1100D?

I really like the 600D for its swivel screen so was hoping to debayer that rather than the 1100D.

with using heat to remove the cover glass, can a soldering iron's tip be used? ever tried that? or would a hot air rework station work as well? The soldering iron tip, if it can get hot enough, would be a lot safer than hot air.

how difficult is it to cool a 600D compared to a 1100D?

what other models have to successfully debayered? that info would help a lot for us in deciding which model to try next.

Thanks so much for the info.

Alistair

The 1100D definitely is more sensitive, how much we don't know. It is in our photometry data somewhere. We were getting saturation from Vega with exposures of 1/500 of a second if I remember rightly (80mm WO 580mm f/l). Surface area of the 1100D pixel is roughly 45% more than the 600D.

Cameras we have successfully debayered - 450D, 600D, 1000D, 1100D and the 5D MkII (took forever, horrible, also new sensors are crazy expensive). I don't think a soldering iron will work, but perhaps a hot air rework station will, it may even be safer than a pen torch. Localised heat still, but not as hot.

As for cooling, the 600D type sensors. which I think are also found in the 550D, 60D and 650D have only a small gap behind the sensor, around 0.6mm. Whereas the 1100D and older models like the 450D, etc have a gap of 1.6mm, this extra mm gives an extra 10-15 degrees of cooling in our cameras. Partly because the extra mass means the cooling is not overpowered by heat production but also this extra thickness means you can cut a channel in the back of the cold finger to reduce losses from unneccesary cooling of the driverboard. The temperature sensor is also on the driver board so this also introduces an inaccuracy in the EXIF that is not in favour of the 1100D either. This channel can cut with a milling maching, dremel or by hand, it needs to be at least 1mm deep. Another thing is that the 1100D has to have this channel, there are small SMT components on the back of the sensor, without this channel they will be ripped off and the sensor will have horrific diagonal noise bands. You cannot get access to reattach them unless you fancy desoldering the sensor and the knowledge of what goes where, realigning the sensor is all but impossibly too. The best you can get from this is a sensor to practice debayering on...

So I would say it is easier to cool a 600D, but the stock perfomance of the cooler will be slightly lower, you can compensate for the lower efficiency of cooling a 600D with a slightly stronger power supply and possibly set point cooling. With the same peltier, power feed and thermal compounds, last winter we had an Ultimate 1100D at -35 while an Ultimate 600D was reporting -20. I don't have any hard data for cold finger type installations with a single stage peltier, but the difference is still there, but less pronounced.

[Gina]

Another thing is that the 1100D has to have this channel, there are small SMT components on the back of the sensor

That's odd - I have added cold finger cooling to three 1100D or Rebel T3 cameras plus a couple of others that I only succeeded in destroying. None of these sensors have/had SMDs on the back of the image processing PCB. The back wasn't clear though - there was a row of chip pins sticking through and soldered on the back from a small square chip on the front. I used a thin sheet of hard plastic between the cold finger and PCB to provide insulation. With this and the height of the pins I was limited to 0.7mm copper sheet for the cold fingers. This was quite adequate - I obtained 28 degrees C cooling below ambient using a Peltier TEC, in turn cooled with a PC CPU cooler. In practice I often obtained -15C as shown by EXIF T so I would think the actual sensor temperature was even lower. Edited August 6, 2013 by Gina

[alistairsam]

The 1100D definitely is more sensitive, how much we don't know. It is in our photometry data somewhere. We were getting saturation from Vega with exposures of 1/500 of a second if I remember rightly (80mm WO 580mm f/l). Surface area of the 1100D pixel is roughly 45% more than the 600D.

hi

would this mean the 350d is potentially more sensitive with its 6.4micron pixels? the 12bit would be a slight disadvantage but not too much. i think the 40d is another good candidate.

Would you have any images taken with the cooled 1100d?

so with removing the cover glass, do we just heat one spot, let it cool and repeat till it turns white and then continue on around the edges?

have you noticed any difference in image quality with debayered ccd vs CMOS sensors?

the ccd's have a larger coverage area for each pixel. I guess nikon's are the only contenders with ccd's. I wonder if their cover glass is similarly glued on.

sorry about the questions, pretty exciting stuff.

thanks

Alistair

[Gina]

Here is a photo of the sensor side of the 1100D imaging PCB. This was after I removed the cover glass remnants with a hot air gun and desoldered the sensor in the process.

[Rottweiler]

Gina, that is what I thought when I first saw them. If you take one of your busted sensors and rip the driver board off, you will see they are SMT components and not pins. At 0.7mm you will have just cleared them. We like to keep the channel as we then get more thermal mass behind the sensor, about 2x as much material. Your method insulates better though. I wonder which is better, more mass/less insulation or less mass/more insulation? As far as accuracy of the EXIF goes, we found it to be just a few degrees at 0C, and anywhere between 10-30 degrees at -30C. You really can't trust it. When you deep cooled a 5D MkII, some of them report +110C! Which is just bizarre

Alistair, if you only consider pixel size yes, but you also have factors like the microlenses, throughput of the bayer, readout noise etc. Also the 350D is 12bit as you say. The heat removal of the cover glass, you heat one spot, cool and then heat the edge of the spot, this makes the white spot longer and you repeat until you take the white spot all the way around the sensor. Obviously heating a sensor is not great and you really have to be quick with the heat and patient with the cooling. As for CCD vs DSLR, a 5um CCD pixel is almost pure active surface area, I think the DSLR sensor pixel size is not the actual size of the photosensitive part. CCDs are superior and here to stay for the foreseeable future, modding a DSLR is a compromise of performance and price. Modding CCDs can also be done, curious why it is so rare. Budget level cameras all really benefit from cooling upgrades.

I don't have any sample images with the 1100D sent in yet. The only ones we made so far are for stock purposes, bar 2 which we have only recently delivered, and clear nights in Holland rarely coincide with a free morning to sleep it off. I don't have the luxury of a garden or balcony, so I have to supervise my equipment all night....