Access Official Technical PDF

Share Summary:LinkedInEmail

A curved CMOS sensor can offer optical designers an alternative to correcting a curved focal plane with additional optics. For engineers evaluating imaging hardware, the central question is not only whether a sensor can be formed into a curve, but whether the process changes its electrical and noise characteristics. Characterization of a 20-megapixel prototype provides a practical example of how to assess that question using measurements of gain, dark current, readout noise, dynamic range, full well capacity and pixel response uniformity.

Why use a curved CMOS sensor?

Many optical systems produce a curved focal plane, while conventional image sensors are flat. A system designed around a flat detector may therefore need additional optical elements to flatten the image. Those elements can add complexity and affect optical throughput and overall performance.

A detector shaped to follow the focal plane may reduce the need for such correction optics. Curved detectors have also been considered as a way to support more compact and simpler imaging-system designs. The practical value depends on the full optical and detector design; sensor characterization is necessary to check that forming the device does not undermine its operation.

The prototype discussed here was a concave CMOS image sensor with 20 megapixels, based on the CMOSIS CMV20000. It had 5,120 by 3,840 pixels, each 6.4 micrometres in size, and was curved to a spherical radius of 150 millimetres across a 24 by 32 millimetre area. The package was kept consistent with the original commercial component, including its electrical connections, so the curved device could be handled as a plug-and-play component.

How the sensor was curved and tested

The fabrication approach involved two main steps. First, the silicon sensor die was thinned by grinding to make it more mechanically flexible. It was then bonded to a curved substrate, whose shape determined the final curvature. The die was wire-bonded to the package, with the original packaging arrangement retained.

To evaluate the effects of this process, the curved device was compared with a flat version of the same sensor using the same general measurement approach. Tests included uniformly illuminated flat fields and exposures taken in darkness. For each exposure time, 30 frames were acquired. Exposure times were varied to examine signal growth, saturation and dark signal; measurements were conducted at camera gain setting 1.

Uniform illumination was produced using an integrating sphere. Exposure times were sampled in an alternating, near-random sequence to limit the influence of illumination drift and small temperature fluctuations. The camera temperature during the measurements was approximately 35°C for both devices, although the reported average operating temperatures differed slightly.

What the characterization measurements reveal

Dark current and readout noise

Dark current is the signal accumulated when a detector is not illuminated. Thermal activity in the semiconductor can release electrons that are collected in pixel wells, where they are indistinguishable from charge created by light. In the test, the mean dark signal was measured at different exposure times. Its approximately linear increase at longer exposures was used to estimate dark current.

After converting the measurements from digital numbers to electrons using the measured gain, the dark current was about 403 electrons per second for the curved sensor and 431 electrons per second for the flat sensor, at approximately 35°C. The readout noise was 10 electrons for the curved device and 11 electrons for the flat one. These are results from the tested samples, not evidence that curvature itself caused the differences.

Gain, saturation and dynamic range

Gain describes the relationship between collected charge and the digital value recorded for a pixel. It was estimated from flat-field measurements by relating the variance in temporal noise to the mean signal across exposure levels. The measured gain was 0.200 digital numbers per electron for the curved sensor and 0.220 for the flat version. The difference was around 10%, but the characterization could not establish whether it resulted from curving or from variation between individual sensor dies.

Both sensors reached a reported saturation level of 4,095 digital numbers, corresponding to the 12-bit analogue-to-digital conversion limit. The estimated full well capacity was approximately 19,871 electrons for the curved device and 18,018 electrons for the flat device. Calculated dynamic range was 66.26 dB and 64.74 dB, respectively. These figures depend on the measurement definitions and the readout noise used in the calculation.

Pixel response uniformity

Pixel response non-uniformity, or PRNU, describes differences in how pixels respond to uniform illumination. The measured PRNU factor was 2.0% for the curved sample and 1.2% for the flat sample. The authors treated this difference as not significant, while noting that further investigation was needed. In particular, a stronger two-dimensional pattern was observed at higher illumination levels and warranted additional study.

Interpreting the results

Overall, the measured characteristics were broadly similar for the curved and flat sensors. Bias levels matched within measurement errors, while the gain, dark current, readout noise and PRNU showed some differences. Because the comparison involved individual samples, those differences cannot be confidently attributed to the curving process. The work therefore supports a measured conclusion: the tested curved prototype remained functional and did not show a clear deterioration across the principal characteristics examined.

For engineers assessing curved image sensors, the broader lesson is to compare like with like. Matching acquisition methods and monitoring temperature help make results more meaningful, while measuring several devices would help separate process effects from normal sample-to-sample variation. A useful evaluation should consider the intended optical design alongside detector metrics, rather than treating curvature as a standalone performance benefit.

Conclusion

The 20-megapixel curved CMOS prototype demonstrated that a thinned, substrate-supported detector could retain broadly comparable electrical and noise characteristics to its flat counterpart. The results do not prove that curving improves sensor performance, nor do they isolate the cause of every measured difference. They do show how systematic characterization can inform decisions about integrating curved detectors into imaging systems: measure the key detector properties, account for test conditions and seek a larger sample set before drawing firm conclusions.

Access Official Technical PDF

About The Author