Access Official Technical PDF

Share Summary:LinkedInEmail

In industrial machine vision, an HDR image sensor can help capture detail across a scene containing both bright and dim areas. But dynamic range alone is not enough: LED lighting can flicker, and elevated operating temperatures can increase image noise or alter color response. A CMOS sensor developed for automotive use illustrates how pixel design and process optimization can address these challenges, with considerations relevant to factory automation and other demanding imaging applications.

Why HDR and LED flicker mitigation matter

High dynamic range (HDR) imaging combines signals captured under different exposure conditions to represent a wider range of scene brightness. This is useful when a vision system must retain information in bright and shadowed parts of the same image. In LED-lit environments, however, the light source may pulse during image integration. Depending on the relationship between pulse timing and exposure, the captured signal can fluctuate or fail to represent the illuminated object consistently.

The sensor described in the study uses a backside-illuminated, 3.0 µm pixel architecture with lateral overflow. When the photodiode fills, additional photoelectrons overflow into other pixel storage regions. The readout combines the photodiode signal with the overflow signal to extend the usable range. A subsequent, much shorter exposure adds highlight information. The reported implementation provides 97 dB of dynamic range from a single exposure sequence and 120 dB when the additional exposure is included.

For LED flicker mitigation (LFM), exposure timing is central. The study’s simulation found 100% LED pulse detection probability when the longer integration period exceeded the LED pulse period. This result is specific to the modeled conditions, not a universal guarantee for every lighting frequency, duty cycle, or camera configuration. The analysis also showed that signal fluctuation can increase when the shorter exposure begins contributing to the combined signal.

Managing noise at high temperature

HDR pixel designs must maintain image quality where exposure ranges meet. In this sensor, the transition between the photodiode signal and the overflow signal was particularly sensitive to dark signal non-uniformity (DSNU) in the floating diffusion (FD). DSNU is fixed-pattern variation in dark output across pixels; unlike temporal noise, it can create a persistent spatial pattern. At elevated temperatures, the FD contribution could become large enough to limit total signal-to-noise ratio (SNR) in the transition region.

The engineering team used technology computer-aided design (TCAD) to investigate the electric fields around the FD and guide process and layout changes. Measurements indicated exponential relationships between FD junction electric field and both dark current and DSNU. Reducing the relevant electric fields through process modifications lowered FD DSNU by approximately 28 times under the reported test conditions: 80 °C and a 33 ms integration time.

The result was not identical across fabrication sites. Measurements also pointed to factors such as shallow trench isolation interfaces and contact placement as possible limits on DSNU reduction. This is an important design consideration: reducing one noise source may reveal another component of the pixel or fabrication process as the next constraint.

For the reported sensor, minimum total SNR at the exposure transition was measured at 32 dB at 25 °C, 29 dB at 80 °C, and 25 dB at 100 °C, using a 16.6 ms longer integration time. These results show why HDR performance should be evaluated across signal transitions and operating temperature—not only by quoting a maximum dynamic-range figure.

Temperature changes sensitivity and color response

Temperature affects more than dark noise. The study measured changes in quantum efficiency (QE), source-follower gain, and pixel transfer-factor gain. Source-follower gain decreased as temperature increased, consistent with the measured temperature behavior of the pixel’s output signal path. These gain changes need to be considered when interpreting signal levels across an operating range.

Measured sensitivity changed only by a few relative percent in the visible range when comparing room temperature with a 120 °C junction temperature. In the 940 nm band, sensitivity increased by as much as 1.46 times. The size and direction of the change therefore depend on wavelength; a calibration established for visible imaging should not automatically be assumed to describe near-infrared response.

Visible color bands also changed by different amounts: blue sensitivity shifted least, red most, and green fell between them. The measurements showed a modest effect on color accuracy at high temperature. Macbeth chart color error increased from dE 10.2 at room temperature to dE 11.2 at 120 °C, while the resulting tint was reported as difficult to detect visually. The study notes that automatic white balance and temperature-adjusted color correction matrices can address this type of color shift.

What system designers should evaluate

For engineers integrating cameras into inspection, guidance, or monitoring systems, the findings point to several practical checks:

  • Test exposure transitions: Assess SNR where HDR signal ranges are combined, since this region may be more demanding than bright or dark regions alone.
  • Match exposure timing to lighting: Consider LED pulse period and exposure duration, and verify performance under the actual lighting conditions.
  • Validate over temperature: Measure dark-pattern noise, sensitivity, and color response across the intended operating range.
  • Separate visible and NIR behavior: If the application uses near-infrared illumination, evaluate it independently from visible-light performance.
  • Plan calibration for thermal shifts: Account for temperature-related gain and color changes in the image-processing pipeline where color accuracy matters.

Conclusion

Reliable HDR imaging depends on coordinated pixel architecture, readout timing, fabrication controls, and temperature-aware image processing. The reported sensor demonstrates how reducing floating-diffusion dark current and DSNU can support SNR through HDR transitions, while timed exposures can help detect pulsed LED illumination. For industrial automation teams, the practical lesson is to qualify the complete imaging system under representative lighting, temperature, and signal conditions rather than relying on dynamic range as a standalone measure of performance.

Access Official Technical PDF

About The Author