No more FF35 Scarlet?

[David Rasberry]

The big questions that still remain are about the specifics of the Red picture format. I believe that it has been confirmed that the stills mode uses .R3D. What we do not know is if you can take single frames or burst frames like a traditional camera, or if you shoot video and extract stills like you do with the Red One. I would love it if Red would elaborate but the stills mode is still one of the mysteries of the DSMC system, along with the low light performance of the 2/3 chip.

At least from the last feature list posted for the 2/3" Scarlet there is a separate still frame mode that supports flash/strobe sync with the side handle module. There is a switch on the body to select stills or motion mode.
If there is, as has been repeatedly stated, no difference in image quality whether one shoots still or motion the mode one uses would depend on what one is shooting.

We discussed the differences in another thread sometime ago. If one is shooting motion frame rates to extract stills, say for sports action photos, one would use higher shutter speeds and different exposure settings to get clean stills than one would use to maintain a film like aesthetic in shooting motion.

I am very much looking forward to experimenting with these features. I think new hybrid shooting styles will develop that can't be done with any other kind of camera.

 

[toke lahti]

Nope not interested. It'd mean another 3 years for RED to change their vision and further delaying their current plans. So, no...not interested.

If they can replace the old sensor in Red1 with new Red1-sized MX-sensor, they already have sensor for "Scarlet One".
How hard can it be?

 

[David Rasberry]

If they can replace the old sensor in Red1 with new Red1-sized MX-sensor, they already have sensor for "Scarlet One".
How hard can it be?

Red One M is APS-C/S35 sized sensor. The slightly larger new M-X version will be in the Scarlet S35 anyway. Where would the cost savings come from?

The 2/3" version will handily exceed the resolution of S16mm film anyway and there are lots of lenses available for the format.

 

[nick and]

1Ds4 size chip would be a nice compromise

The S35 sensor is in the range of the 1D -- if not bigger -- not APS-C.

 

[toke lahti]

Red One M is APS-C/S35 sized sensor. The slightly larger new M-X version will be in the Scarlet S35 anyway. Where would the cost savings come from?

Where does the cost savings come from comparing 2/3" to S35?
Sensor area, I guess, and mass production.
2/3" = 10.1 x 5.35 mm = 54 mm^2
(S16 = 12.52 x 7.41 mm = 93 mm^2)
R1 = 24.4 x 13.7 mm = 334 mm^2
S35 = 30 x 15 mm = 450 mm^2
FF35 = 36 x 24 mm = 864 mm^2


If sensors have constant price per area, S35 will cost 35% more than R1.
Furthermore S16 would cost 72% more than 2/3".

The 2/3" version will handily exceed the resolution of S16mm film anyway and there are lots of lenses available for the format.

There is also lots of s16 lenses out there and they would have wider angle and shallower DOF with s16 or windowed R1 sensor.
S16 sensor would also have 2/3 stops more sensivity than 2/3" sensor.

Looking at Scarlet prices ($2750, $7000, $12000), if the cost of camera is otherwise pretty same, the sensor will cost about $12 per mm^2.
Cost of other parts than sensor seems to be little less than $2k.
Then s16 would be only $500 more than 2/3".
It would be also very nice to use it with 2/3" lenses, when you would have some "off target" area in the picture.
R1 size would be around $6000.

If you can't afford S35, I guess that nobody will really use Scarlet as real DSMC, because 2/3" is so big jump from S35. On the other hand R1 or even s16 would be better for stills, so if Red wants to market Scarlet as DSMC, they should offer these other sizes.

This world is full of lenses that Red should adapt to, not the other way around. It just doesn't feel very economical if you are not going to use Scarlet for stills and use it with s16 lens, you'll have to buy S35 version, from which cost, half is something that you will never benefit.

 

[David Rasberry]

S35 and 2/3" are the most common standards. S16 is 12.4mm wide compared with 9.6 for 2/3". In terms of total production output, 2/3" format is the most commonly used pro format in the world, film or digital.
I have advocated moving to a 4/3 format as the middle step between 2/3" and S35 with a 17mm sensor width, but it is a very new format.

 

[toke lahti]

S35 and 2/3" are the most common standards. S16 is 12.4mm wide compared with 9.6 for 2/3". In terms of total production output, 2/3" format is the most commonly used pro format in the world, film or digital.
I have advocated moving to a 4/3 format as the middle step between 2/3" and S35 with a 17mm sensor width, but it is a very new format.

Although I haven't heard anybody were shooting with 2/3" with film (except some wildlife documentary cranked material to 16mm), 4/3" would be pretty nice if Red would make some lenses for it. I haven't found a single 4/3" lens with manual & mechanical focus, zoom & aperture.

4/3" = 18 x 12 mm = 216 mm^2 = $4500

2 stops more sensivity than 2/3"!
This could be also pretty nice with s16 lenses?

 

[Rick Burnett]

Scarlet S35: 30 x 15 mm sensor
Canon 7D APS-C: 22.3 x 14.9 mm sensor
Canon 1D APS-H: 27.9 x 18.6mm sensor

Interesting, 7D matches closer in height whereas 1D matches closer in width in sensor size. Will be interesting to use my 7D lens on the S35. To really understand how to compare the sizes, I'd need to look at the different aspect ratios on the red / and running in cropped mode to understand what i might see.

Pertaining to larger sensors, I can tell you this, the bigger the die you make, the more process variation comes into play. When designing an integrated circuit, you have to account for variance when you build it, resistance and capacitance are NOT constant across the surface of a die. For digital circuits, this is much easier to deal with, but for analog, it's a much bigger problem. I wouldn't be surprised if when they make these sensors, a lot of them get tossed for not working within the range they need (called speed bining in the CPU world, but this might be for sensitivity range or something).

Just a thought.

 

[Jeff Kilgroe]

Rick, the sensors are already being installed in cameras -- the S35 Mysterium-X going into the RED One upgrades is the same sensor that will be in the upcoming Scarlet S35, EPIC-X and EPIC-S35. You are correct though, not all samples will meet the requirements to be used in the cameras. Most likely those sensors that don't pass QC will probably be ground up as there really isn't any other product out there that would be using them and RED probably doesn't want any such product on the market anyway.

Pertaining to sensor sizes... The S35 Mysterium-X chip is essentially identical in width to APS-H or the Canon 1D series. Actual recordable area for 5K should be about 27.6mm and not the full 30mm width. The pixels are 5.4 microns across, which is the same as the original Mysterium. 5K (5120) x 0.0054mm = 27.648mm. So lens FOV/AOV will be essentially the same as on the Canon 1Dx. Of course, the Canon SLR sensors have a 3:2 aspect ratio, which gives them more vertical image area as opposed to the 2:1 aspect of the S35 Mysterium-X.

 

[David Rasberry]

Although I haven't heard anybody were shooting with 2/3" with film (except some wildlife documentary cranked material to 16mm), 4/3" would be pretty nice if Red would make some lenses for it. I haven't found a single 4/3" lens with manual & mechanical focus, zoom & aperture.

4/3" = 18 x 12 mm = 216 mm^2 = $4500

2 stops more sensivity than 2/3"!
This could be also pretty nice with s16 lenses?

Estimating sensitivity based on sensor size alone is not physically valid. Sony is making 2/3" cameras now with native 1600 iso ratings.

 

[Rick Burnett]

You are correct though, not all samples will meet the requirements to be used in the cameras. Most likely those sensors that don't pass QC will probably be ground up as there really isn't any other product out there that would be using them and RED probably doesn't want any such product on the market anyway.

Some probably won't even work completely depending on the tightness of the digital timing given the size of them, (I am a chip designer). It's interesting that many companies don't even share their yield numbers (number of working parts versus failing parts on a wafer) and especially not with these higher technologies.

I am just curious, does anyone know what feature size is typically used for these sensors (for instance, 45nm is pretty common now for high speed digital)?

 

[Gavin Greenwalt]

Estimating sensitivity based on sensor size alone is not physically valid. Sony is making 2/3" cameras now with native 1600 iso ratings.

True but Sony is also making 3 sensor cameras. So the total surface area of their photo receptors is larger than 2/3".

 

[Rick Burnett]

True but Sony is also making 3 sensor cameras. So the total surface area of their photo receptors is larger than 2/3".

That's actually not true, because the area you are measuring per active pixel size is the same. Just because you use three doesn't mean they are more sensitive, you are just using that sensitivity differently. For instance, if they are separating out RGB and not using a debayer pattern you will get more information, but each of those bits are just as sensitive.

The sensitivity of the active pixel has more to do with the construction of the active pixel and the circuitry supporting it. Like the charge conversion, amplifiers, noise correction, etc.

For once, size doesn't matter

 

[toke lahti]

Estimating sensitivity based on sensor size alone is not physically valid. Sony is making 2/3" cameras now with native 1600 iso ratings.

I guess comparing Red's sensors to each other is pretty valid.

 

[Rick Burnett]

I guess comparing Red's sensors to each other is pretty valid.

Not necessarily, sensor size has nothing to do with how Red designed the active pixel sites or circuitry in each sensor. Even though it too does not tell the whole story, pixel density is the first factor to start looking at. And even beyond that, within the pixel density, how much of that area is active and how much is supportive circuitry. If they used the same exact circuitry in all their sensors, they could be all exactly sensitive, but I'd argue that the larger the sensor, the more issues with voltage drop and on chip variation so you could lose some sensitivity across the entire array if you were stringent enough in your yield selection. Which is why it makes sense to me that the larger sensors are so much more expensive.

 

[David Rasberry]

I guess comparing Red's sensors to each other is pretty valid.

I expect the inherent sensitivity of the 2/3" sensor to be the about the same as the S35 with perhaps 1 stop or so less dynamic range.

 

[Gavin Greenwalt]

That's actually not true, because the area you are measuring per active pixel size is the same.

Yeah but there are 3x as many of them then. And you can average them out giving you dramatically less noise and more sensitivity.

A 3chip camera has 3x the capture surface area. This also lets you have larger photosites. A 3 chip camera (all else being equal) should have comparable sensitivity to a Bayer pattern camera with a sensor 3x as large.

 

[David Rasberry]

Yeah but there are 3x as many of them then. And you can average them out giving you dramatically less noise and more sensitivity.

A 3chip camera has 3x the capture surface area. This also lets you have larger photosites. A 3 chip camera (all else being equal) should have comparable sensitivity to a Bayer pattern camera with a sensor 3x as large.

Luminance is only sampled in the green channel though and you have the optical losses through the prism to contend with too. There are more luminance samples with Scarlet's 2/3" sensor, not including the R and B filtered pixels, than on a 1080p camera. On Scarlet based on the 3.3micron pixel spec, out of a 4 pixel Bayer block, you have two luminance samples with a total area of about 21.8 sq. microns, vs a single luminance sample of about 25 sq. microns with a 3 chip 1080p camera. Little more than 10% difference in area. I think the underlying chip fab process and the inherent noise levels will have more impact than the minor difference in sample area. At low sample levels where read noise or quantization error predominates, the two samples of the Bayer chip may well be superior to the single sample 1080p chip.

 

[Rick Burnett]

Yeah but there are 3x as many of them then. And you can average them out giving you dramatically less noise and more sensitivity.

A 3chip camera has 3x the capture surface area. This also lets you have larger photosites. A 3 chip camera (all else being equal) should have comparable sensitivity to a Bayer pattern camera with a sensor 3x as large.

But aren't those sensors picking up red, green and blue independently? if so, how could you average them since they are not measuring the same thing?

I would also argue that it might not be dramatically less noise as it depends on the distribution of noise and the source of noise. For the most part, if the sensors were used identically, I agree, but I've not seen them used that way. Not to mention that splitting the light into three different sensors if passing the full spectrum to each is going to reduce the light to each of them to 1/3 which means now you are recording a weaker source and pushing yourself down into the noise further. I don't see the benefit in that sense.

With respect to split colors, which I believe is happening, the active pixel window is no different. a red pixel in a bayer pattern will be the same size as a red pixel in the sensor with all red pixels using the same sensors with different optical filters. The difference being that there will be more pixels of red so you just basically remove the disadvantages of using a debayer process. You aren't adding sensitivity to the pixel, but you are adding detail because you don't have to interpolate between pixels like you would on a bayer sensor. That's not a gain in sensitivity. This is of course assuming all the same sensors used in both single or triple sensor design.

 

[toke lahti]

Pertaining to larger sensors, I can tell you this, the bigger the die you make, the more process variation comes into play.

Yes, of course. Silly me, I've been working in chip manufacturing company when I was young. So, let's say that chance of defect is constant per given area of sensor.
This of course makes bigger sensors much more expensive.
So how should the price estimation go?

If we then assume that price of
Scarlet - sensor = about $2.5k, then:

Lets assume that there is 0.09% chance that area of 1mm^2 is defective.
Then lets assume that 1mm^2 of sensor costs $5 divided by the chance it's ok.

Then price of sensors would be:
2/3" = 10.1 x 5.35 mm = 54 mm^2 ($2750): 5/0.9991^54 = $5.25/mm^2 = $280
S16 = 12.52 x 7.41 mm = 93 mm^2 ~ $510
4/3" = 18 x 12 mm = 216 mm^2 ~ $1300
APS-C = 22.3 x 14.9 mm = 332 mm^2 ~ $2200
R1 = 24.4 x 13.7 mm = 334 mm^2 ~ $2300
S35 = 30 x 15 mm = 450 mm^2 ($7000): $7.50/mm2 = $3400
APS-H = 27.9 x 18.6mm = 519 mm^2 ~ $4100
FF35 = 36 x 24 mm = 864 mm^2 ($12000): $10.89/mm^2 = $9400

Second scenario: more expensive cmos process: $10/mm^2.
To fit 2/3" & FF to this scenario, we'll have to assume that chance of defect in 1mm^2 has to be 0.02%:
2/3" = 10.1 x 5.35 mm = 54 mm^2 ($2750): 10/0.9998^54 = $10.11/mm^2 = $550
S16 = 12.52 x 7.41 mm = 93 mm^2 ~ $950
4/3" = 18 x 12 mm = 216 mm^2 ~ $2300
APS-C = 22.3 x 14.9 mm = 332 mm^2 ~ $3600
R1 = 24.4 x 13.7 mm = 334 mm^2 ~ $3600
S35 = 30 x 15 mm = 450 mm^2 = $7000 = $10.94/mm2 = $4900
APS-H = 27.9 x 18.6mm = 519 mm^2 ~ $5800
FF35 = 36 x 24 mm = 864 mm^2 ($12000): $11.89/mm^2 = $10000

Looking at Scarlet prices it looks like the second scenario is closer, which means that their cmos factory has damn good yield. 91% of S35 sensors are good.
Or FF will be strongly subsidized by smaller versions.

And of course you need to add something like $100 to all sensor prices to get a whole module.

But after all, I still think that there should be something between 2/3" and S35!
For example "widescreen 4/3" (18 x 9 mm) would be less than $5k and very nice with about 2kg s16 zoom lenses for motion and APS-C or 4/3" lenses for stills.