Nikkei says that Sony has developed a sensor for automotive applications, which is "10 times as sensitive to light as existing chips, allowing for easier detection of objects in dark environments. The sensor can provide a color display of the surrounding area even in moonlight, enabling it to detect obstacles more readily than a driver can." Sony plans to start mass production of the sensor and begin sales to automotive equipment manufacturers in the latter half of 2015. The new sensor is expected to appear in car-mounted devices in 2016.
Aptina and OmniVision are leading the automotive image sensor market now. "But Sony aims to eventually pick up 50% of the market." Annual production of smart cars equipped with many cameras is expected to reach 100 million vehicles by 2025.
10x sensitive to light as existing chips....sounds like quite the breakthrough. Wonder what the rest of us missed?
ReplyDeleteYou have definitely problem with SONY :)-. You attack them on each of their annonce. What is wrong between you and SONY please?
ReplyDeleteAs I have stated before, but maybe you missed it, my problem is with Sony Marketing, not the technical guys or their technology.
DeleteYou aren't bothered by a claim of 10x sensitivity improvement compared to existing chips (not cameras)?
And how do you propose they achieved the 10x improvement in sensitivity ... overnight?
DeleteNice. 10x sensitivity! I presume that means either 10x bigger pixels, or 800% QE!
ReplyDeleteOh, or marketing BS.
Or curved sensor enabling F1.0 lens.
DeleteOr 10x relates to IR sensitivity at 900-950nm.
Increased IR sensitivity wouldn't help with color imaging. Larger aperture would have to compare to around F#3 to get a 10x increase. That's a typical cell phone aperture, CCTV or automotive wouldn't bother with anything above F#2 on a wide angle lens.
DeleteMy cynical guess: since Sony still measures and publishes sensitivity ratings in mV/lux (even for IR optimized sensors without any analogoue output) they just need to make the FWC smaller or introduce some horrible gain stage in there somewhere.
On the other hand, Sony has released some great sensors and technologies recently, maybe it is just a combination of all the best ingredients? Curved Starvis Pregius sensor with Panasonic photonic color splitters and Invisage coating, now that would make me drool...
JH.
Or just an extra 10x gain stage. Noise goes up, too.
ReplyDeleteFirst of all, sensitivity can be increased arbitrarily by using gain. However this increment in sensivitivity doesn't improve the image quality.
ReplyDeleteSeccond: the information is in the number of captured photons, the only way to improve really the image quality is by acquiring more photons and this is only possible if highers pixels are used or if the FFQE is improved. But the FFQE for the modern sensor cannot be improved significantly. Then the only real option is increasing the pixel size.
Any other strategy are marketing techniques.
Sony's existing QE at 900nm is too high already to get a 10x improvement. I thought about that one.
ReplyDeleteCurved silicon enables faster lenses (or at least makes it easier to build them with reasonable performance and size), but that doesn't count as a 10x increase in sensitivity sensor. Anyway, there are already f/1.5 and 1.6 automotive lenses on the market, so f/1.0 would only be a 2x increase in CAMERA sensitivity.
Can any one tell me what low sensitivity numbers SONY existing chips have?
ReplyDeleteMaybe they refer to curved rgbc.
ReplyDeleteThe curved sensor haven't more sensitivity that the flat sensors.
DeleteThe only improvement of curved sensors is that allows build more simple optical system. But this will not justify an improvement x10
Take the new sensors to a real test like EMVA1288 to find out exactly how good it is. Standard measurements won't lie; marketing documents do.
ReplyDeleteThree layer organic with no color filter array?
ReplyDeleteOr maybe not...
I'd guess marketing to be the case.
If no color filters are used, how is possible that Sony say that a colour photograph will be performed using the moon light.
Delete3 organic layers. Like Foveon without the issues. Though no CDS?
DeleteBut almost certainly just marketing nonsense.
This technique doesn't improve the FFQE significantly. The main advantage is that allows removing the demosaic process. This technique will increase the number of sensed photons by a factor up to x3.
DeleteSensitivity is a very broad term, used frivolously, and in general is subjective. This is complete nonsense. Unless Sony has discovered to have 0 e- of read noise and 0 e- of mean dark current I do not see how this is possible. Also, with zero cross-talk.
ReplyDeleteIs their QE spectrum a delta function? That's impressive. Fermi & Dirac would both be honored.
It's ppossible obtain FFQE values significantly greater than one, for example using avalanche photodiodes. However this is a form of gain, but the information associated to the number of photons per pixel is not be increased. This method is only useful to reduce the influence of readout noise. But today there's readout channels with a noise bellow one electron.
DeleteIn summary, I doubt that Sony can improve the effective sensitivity in a x10 factor
If I remember well, Bart Dierickx once gave an overview of possible definitions of sensitivity. There were many, maybe 15 of so. I couldn't find them anymore on the web. Maybe Bart can repost them ???
ReplyDeleteFor me, the fundamental sensitivity is the number of sensed photons per exposure unit.
DeleteThis is the only definition that is directly related with the image quality.
It was in an old FillFactory powerpoint. I may still have it somewhere but it is not on his publications page: http://users.telenet.be/sbme/bartdierickx/publicaties/
DeleteLet's ask.
I only remember that in FillFactory datasheet, they stated the sensitivity in terms of V/Lux*s with a very good value on the first page and on the last page, they indicated that the measurement has been done at 650nm. I don't know if this falls in one of the 15 definition :)-
ReplyDeleteThe sensitivity in V/(lux•sec) is very common, but is an ambiguous definition. If in the readout circuitry an analog gain is applied, the sensitivity will increase by this gain. Then the sensitivity in V/(lux•sec) is not good if no other information is supplied.
ReplyDeleteThe key point is that the information and then the image quality is in the number of electrons/holes photo generated per pixel for an exposure unit. And even this is not the more fundamental if the sensor have capability to amplify the charge, as the used by the avalanche photodiodes. In this last case, it is needed divide by the charge gain to obtain the intrinsic sensitivity.
To summarize various comments, camera sensitivity increase possible from various designs elements are as follows:
ReplyDeleteReplace bayer with one of the following:
RGBC color filter with half the pixels clear: ~2x increase. Issues with color moire or resolution loss.
Panasonic’s color splitter: 3x theoretically (paper says 1.85x achieved). Possible issues with resolution loss, moire due to mixing signals between pixels.
3 layer sensors like Foveon: 3x? Issues with color separation, noise
Curved sensor: ~2x from making it easier to make high aperture lenses
QE increase: ~1.25x possible (Invisage being a possibility)
Max possible total increase: 3x*2x*1.25x = 7.5x
Note that we’re talking about camera sensitivity whereas the article talks about sensor sensitivity.
RGBC - this is already in existing chips, so unfair to count this. Of course, Sony does tend to ignore things done elsewhere so perhaps they only are talking about their own existing chips.
DeletePanasonic color splitter - this is theoretical, and probably only works (if at all) for high F-numbers where incoming rays are mostly vertical. This should not be counted at all for low light, low F-no systems.
Curved sensor - maybe since automotive can support more expensive sensors. Then again, one is trading sensor cost against optics cost.
QE increase - ok, sure.
Max possible increase reduced to 2.5x for camera, 1.25x for chip itself.
I fully agree
DeleteThis improvement should be possible if we place the sensor at high temperature. For example, if we place the sensor at 100°C and the PPD, with 20K FWC, generates 10K electrons dark signal. In this case, the noise is dominated by DC to 100e. If they can keep the DC within 100e, then the sensitivity improvement could be 10x in this case.
ReplyDelete-yang ni