User-Programmable Neuromorphic Infrared Focal Plane for Multi-spectral Imaging Applications

The proposed effort represents some of the most powerful work to be done to develop improved sensing systems for multispectral applications that exhibit a higher level of performance than today's available sensors. This work is worthy of support under the stated criteria for developing new concepts for sensor architecture and pre-processing multispectral sensor information because it is truly innovative. Biological sensors provide technically rich examples for use to use in the developmen tof devices which are more efficient with respect to sensitivity, data throughout and post-detection processing of recovered image data. This project proposes an effort which will accomplish a preliminary design for a sensor which extends the previous work that we have done in the development of the "neuromorphic" infrared focal plane device. It will show how this technology may be combined with hyperspectral imaging technology that we have also developed to produce an overall system which performs multispectral image computations directly on-focal plane with a spectral resolution of better than 0.01 um over the desired spectral range. The resulting body of work will provide a solid technical basis for the development of an actual device in a Phase II effort. An extension of the Phase I effort is also proposed in an interim task between Phase I and Phase II that will help further refine design issues in preparation for the hardware development. Devices which mimic biological counterparts are more efficient than conventional sensors in many ways. To a large degree, they process information directly on-plane, reducing the computational load of off-plane processing resources. The ever-increasing pixel count of currently-available devices generates a tremendous quantity of data which ust be handled by off-plane processors. Biological systems do pre-processing of the data to reduce the amount of transmitted data to a level which may be handled by reasonably-sized processors. The data that they deliver has been pre-filtered with respect to spatial and temporal requirements of follow-on devices. They operate to a large extent in the real-time analog domain, so their power dissipati