Capturing Deep Shadows and Bright Highlights in Tiny Camera Pixels

High-dynamic-range camera pixels must detect faint light and hold the much larger charge produced by bright light. Many designs switch between sensitivity settings, with a noise penalty at each handover that can leave visible bands in mid-tones. In retinal cells, a light-driven fall in calcium activates several feedback mechanisms that adjust sensitivity smoothly. We use the same principle in silicon: as light-generated charge accumulates, the pixel’s internal voltage falls. Its built-in charge storage changes slightly, then two transistors progressively connect a small and a large capacitor, each an extra charge reservoir. Their responses overlap, so storage grows and sensitivity falls without an abrupt switch. The added parts sit beneath the light-collecting layer, which stays unchanged. One paired reset-and-signal measurement captures the full range, and a lookup table converts the smoothly compressed signal back into brightness values, avoiding separate gain readings and their noisy handover.

A camera pixel must measure faint shadows and bright highlights through the same limited electrical range. Many high-dynamic-range sensors use separate sensitivity settings and combine their readings, with a sudden drop in image quality where the settings meet. This design borrows telephone companding: preserve fine steps for weak signals, compress strong signals, then reverse the compression digitally. First, capacitors along the pixel’s isolation-trench walls hold overflow electrons and release them back through the light-sensitive region for one combined transfer to the readout node. Comparing that signal with the node’s reset level cancels reset noise across the full range. Next, a voltage-variable capacitor on the underlying logic chip increases its capacitance smoothly as the signal grows. Small signals produce large voltage changes; bright signals take progressively less voltage per electron. A 14-bit converter measures the compressed signal, and a calibrated lookup table restores a brightness-proportional output. Together, overflow return and smooth compression replace separate gain readings and their stitching, preserving shadow detail without a gain-switching noise cliff.

Camera sensors must capture deep shadows and bright highlights in the same exposure. Some extend their range by storing excess light-generated charge on a large capacitor, but reading that extra store introduces a noise dip where the sensor switches gain. We replace that overflow store with charge metering, borrowed from coulometry, where chemists measure a substance by totaling the electrical charge passed through a reaction. A threshold detector beneath each pixel registers roughly 3,200 overflow electrons, increments a digital counter, then resets the charge-sensing node and refreshes its reference. Brighter pixels simply count faster. At exposure’s end, the processor adds the counted charge, the leftover charge and the charge still in the photodiode. The distinctive addition is calibration every frame: a test ramp passes through each pixel’s own detector to measure its actual packet size. This corrects pixel-to-pixel threshold differences that would otherwise leave a fixed pattern in the image. Dim-light readout stays unchanged, while bright-light capacity grows without a gain switch or a large overflow capacitor.