Business Wire, EETimes: Cypress 25MP VITA 25K 53fps sensor claims to offer the market's highest throughput for a device with a pipelined and triggered global shutter. The sensor has 32 10-bit LVDS outputs, each runs at a 620 Mbps, which results in a high frame rate of 53fps at full 25MP resolution. The sensor is intended for high-end machine vision applications, such as inspection machines; biometric inspection, such as next-generation palm print readers; and intelligent traffic systems.
The VITA 25K sensor has 5120 x 5120 active pixels with 4.5um pixel size. The sensor's datasheet gives few more details: The pixel fill factor is 80% and FF x QE is 52% at 550nm. FPN is 0.5%rms of the full scale. PRNU is 2%. Dark current is 133e/s at 25C. The power dissipation is 2.5W out of a triple supply (1.8V, 3.3V and 4.5V). Its ADC has 10 bit resolution.
The sensor offers a 35-mm optical format with monochrome or color digital output. Samples of the VITA 25K image sensor are currently available, with production devices expected in the first half of 2011.
What could be the price of a such sensor ?
ReplyDeleteThanks!
a testicle and your left eye
ReplyDeleteWhile the announcement didn't generate many ripples here, I think this type of imager is the wave of the future.
ReplyDelete53 fps is a very common request. That frame rate will probably be standard soon. 100 e-/sec dark current is really not something to be proud of. But if you are running at 53 fps, I guess its OK.
ReplyDeleteAs a consumer, I would definitely prefer to have a single high-quality image sensor for all (or most of) my consumer needs rather than a hodge-podge of lower-quality devices that are essentially disposable. There are definitely some obstacles to this particular sensor meeting that need, but the potential is there.
ReplyDeleteNever mind ripples and waves. There was a sea change over the weekend. On Saturday, Foxconn was proposing a 30% raise for line workers in Shenzen, up from 20%, and a Honda plant in China was in a similar situation. I've just read over at EETimes that Foxconn has increased the raise to 122%.
Out of this I see turblent times ahead all across the global economy, until people can sort out what prices, practices, and business models will work. In image sensors specifically, per-chip costs will likely increase a lot, and suppliers and customers will have to reconsider where and when to use image sensors. I don't know whether the market will grow or contract as a whole, but conditions will probably favor quality over cheapness in the longer term.
There is no miracle: people develop sellable products with minimum production cost. So if they find a reasonable cost device, they use it. Such stuff is definitely not low cost one. But the problem is for a lot of applications, you don't have so many provider. That is why the Cypress sensors, most of time low imaging quality one, have a market place.
ReplyDeleteCan you find a Cypress sensor based camera without image correction ? Very few, but the question is that you can buy a FPGA everywhere, but you don't such kind sensors everywhere. That is why the quality metric can not be applied all the time.
Интересный сенсор. Очень.
ReplyDeleteInteresting indeed, droog. 5120x5120 seems to be an array size that has popped up in certain circles, starting with DALSA CCDs I think.
ReplyDeleteBut it seems to me that anyone who has built a 32 Mpixel UDTV sensor for 60 fps could scale it to a larger snapshot pixel, assuming there was a customer or market for such a sensor to justify the cost. Anyway, in my opinion, respectable but really not a "first" in a breakthru way.
Probably Cypress had a specific request from an old FillFactory customer but was also allowed to offer it as a niche market product.
Nevertheless, glad to see the Cypress team is still breathing. Hopefully this is not their last breath.
Dear Eric,
ReplyDeleteWhat happened in Cypress ? Why do you say that this could the last breath ??
Thanks !
Quick question ... per the spec-sheet, does a dark-current of 133 e-/sec indicate that at 53fps we're only looking at around 2 to 3 e-/frame? And if so, with a full-well capacity of 25,000 electrons, would that mean a SNR somewhere between to 8000:1 and 10,000:1 (at least for dark-current to signal)? That seems awfully high for a chip that only has a 10-bit ADC, so I'm wondering if I'm missing something ...
ReplyDeleteJason,
ReplyDeleteFirst, 53fps is the sensor speed, rather than exposure time. Exposure can be shorter than the frame rate, for example 1ms.
Second, one needs to know the dark current non-uniformity to calculate the fixed pattern noise coming from this component. This is rarely given in datasheets.
Third, there are other noise components in the datasheet, such as FPN (fixed pattern noise) of 0.5% of full scale. This limits dynamic range by 200:1, well below 10b ADC resolution.
The dark current might be the sum of photodetector and storage site. In this case, the full period of 1/53 s applies.
ReplyDeleteSNR is limited by shot noise, so for 25,000 e-, sqrt(25000) = 158 not 10,000:1. Probably the read noise is at least 25 e- rms so 10b ADC is probably reasonable.
FPN can probably be corrected off chip so it has minimal impact on dynamic range.
Wondering if this sensor does really have a FWC of 25k, when a similar Cypress product - vita1300 - only reaches 17k e- FWC in a slightly larger pixel... I guess 25k was the design goal.
ReplyDeleteIf the dark current figure is for both acquisition and storage, there would seem to be a large innate dark current non-uniformity, as queue time range from very short for the first row that gets processed to very long for the last row that gets processed.
ReplyDeleteI bet there are some interesting nuts-and-bolts details underlying the public system specs.
Arggh! "ranges from"
ReplyDeleteCDM - You are right that the integration time is not the same for all pixels but don't you think the read noise along with photon shot noise will be larger than dark signal non-uniformity, given the 133e-/sec avg. dark current?
ReplyDeleteFor fun I will hedge by observing that the available statistics are necessary but not sufficient.
ReplyDeleteAlso, in the bigger picture, the Foxconn effect will probably trump everything else.
It would still be cool to get more than the marketing snapshot in these announcements.
Thanks for the thorough explanations ... that definitely makes more sense
ReplyDeleteSomeone at Cypress seems to be reading this post.
ReplyDeleteOn June 16, a new flyer was posted on http://www.cypress.com/?docID=22563 stating the sensor has a FWC of 15.5k e-