Normal scenes captured by cameras involve light intensities spanning more than five orders of magnitude, which is often taken care of with auto-exposure/multiple exposures in high dynamic range (HDR) imaging technologies. Such technologies assign one or several uniform exposure levels and dynamic ranges for all pixels in the image, significantly squeezing the data precision for dimmer regions, causing fundamental information loss at the photodetection stage.
Ideally, each detector should independently decide its tradeoff between responsivity and speed. For pixels receiving low-intensity light, they adapt to higher responsivity and data precision at the cost of longer integration time, while pixels with brighter illumination provide higher temporal resolution without wasting an unnecessarily high responsivity. Event camera technologies can use additional circuitry in each pixel to achieve similar goals in the visible band, but at a high hardware budget.
In this project, we adopt material and device innovations to allow each detector pixel to intelligently determine the tradeoff based on local illumination levels, then encode the readout as neuromorphic spikes at a relatively low hardware cost. The detectors respond to a broad spectral range from visible to mid-infrared, with adaptable high peak responsivities. We explore the device’s potential in HDR imaging, IR sensing, and night vision.
Project is currently funded by: Federal