The IMX224MQV sensor has 2,350mV (standard F5.6) sensitivity, a level of sensitivity that is around twice that of older Sony IMX238LQJ sensor with 3.75um pixels and the industry's highest. This was achieved through the adoption of photo diodes that improve the conversion efficiency of light into electrons and through the addition of circuits that improve the conversion efficiency of those electrons into voltage. Also, a PGA with a gain up to 72 dB further contributes to the ultimate achievement of a minimum object illuminance of 0.005 lux. With this design, it now becomes possible to capture high-resolution color images even on a dark road at night with no stars to illuminate the surroundings.
This image sensor also supports a WDR that allows for extended exposure time, which, when compared to conventional multi-exposure WDR systems, makes possible improved image quality in low light conditions when combined with a compatible ISP (Mass production of compatible ISP from Sony is planned on June 2016). Furthermore, the sensor's pixel has an improved NIR sensitivity for better object recognition when used in with NIR LED illumination.
 |
| IMX224MQV image captured at 0.005 lux (F1.4, 16.7ms exposure, 72 dB gain) |
Going forward, Sony intends to expand its image sensor reach beyond the realm of digital imaging devices and into the arena of "sensing" as well. Building off products such as the "IMX224MQV," Sony will pioneer and tap into a new market for image sensors that are capable of acquiring and discerning information from captured images such as objects' distance, movement, shape, and color, for use in vehicles and elsewhere.
72dB gain = 4000. It's crzay. But with such gain, the sensor will saturate rapidly, since 0.25mV will become 1V which is almost the full ADC range.
ReplyDeleteIs someone else than me out there that asks for the opposite - lower overall "gain"?
ReplyDeleteIn our machine vision environment we would rather need attenuation than gain (ok - we could use grey filters but - just to post a question...)
we have both enough light (the desired light has to be bright compared to environment light) and the demand for a exposure time of at least a few (5-10) ms (e.g. to "filter" mechanical vibrations).
so our image sensors usualy are configured at the lower limit of the possible gain range - using no "amplification" by ADC but trying to use as much as possible of the linear range of the pixel.
if I compare the values in the datasheet of one of the sensors we use (CMV4000), I'd say that for the kind of machine vision applications we use the ideal would be <1-2V/lux.s - so less than 1/4 of what it is now... (ideal would be about 4000x4000 3µm global shutter pixels - but thats another story ;-)
72db gain = aprox. iso 400,000
ReplyDeleteSorry, I'm just beginning in this world : what is the relation between ISO and gain or e sensibility / noise ?
Delete