TSMC patent application US20120098975 proposes to have color crosstalk calibration patterns alongside the array, so that any color filter, microlens or metal misalignment can be calibrated on per die or per reticle basis:
 |
| Example of calibration pixels location |
 |
| Example of crosstalk test patterns |
I used to work for a major medical equipment manufacturer (now gone). Every couple of years, they would have a big meeting with their competitors about patent cross-licensing becuase no one could make anything without these cross-licenses. To determine who would pay whom, they stacked their relevant patents on the table. The one with the biggest stack got the most money.
ReplyDeleteThis patent closely resembles the ones my former employer filed only to maximize their stack height.
Hi Dave, Maximizing your stack height in imaging by means of cross-talk patents ? In imaging you should lower your stack height to limit the cross-talk ;-)
DeleteHi, Albert,
DeleteGood one. Rock on, Neil.
Regards, Dave
As 1.1um pixels become more common and the leaders move toward 0.9um pixel, alignment variations become a factor. And crosstalk goes not only to the closest neighbors, but also to many pixels away from them. Having the specially patterned calibration pixels is certainly helpful - no doubt about that.
ReplyDeleteNow, the question is where to place them - either next to the array, on the chip periphery or in the scribe line. Or, may be, even devote one one die in the reticle to such test pixels. Having them next to the array has an advantage of sharing column readout and row driver circuits with the ones of the main array.
If TSMC patent is granted, our placement choice would be narrowed.
This arrangement neglects the len's crosstalk, blur spot, etc. Crosstalk is also 2D in nature, so a simple two-color pattern gives only part of the answer. Yes, the data would be interesting. But there is so much more required to create a decent model that the value of this data alone seems rather limited.
ReplyDeleteSure, the data is limited. But it's not characterization data. It's only supposed to take care of the sensor variations. And it's not giving you all-automatic adaptation to whatever lens is attached to the module. Also, the application does not limit the test pixels by the simple two color patterns, it's just one of many examples in the document.
ReplyDeleteOn the other hand, one can get a lot of useful process variation data from these and other "simple" test pattern. This info can stored in OTP memory at the tester floor, for example. Then one can use this data to correct colors in ISP.
Excuse me, I have a question.
DeleteIs OTP "must be" or "must to have" in CIS?
Thank you very much.
JEnny
No, it's not a must. But it's a very nice to have. Many modern sensors have it.
DeleteSomething similar has been used in the past for dark current calibration. It is also very common to locate electronic and optical calibration test structures in-between bond pads or in scribe lines.
ReplyDeleteDark current calibration is still done in this way : adding black reference lines and/or adding black reference columns.
DeleteI think that this patent protects nothing, the scribe lines will not be included in the final products, so no patent issue in this case. Do you agree?
ReplyDelete