Filter-free color pixel sensor using gated PIN photodiodes and machine learning techniques

This work addresses the possibility of light wavelength discrimination, using a specially designed Lateral PIN Gated Photodiode and Machine Learning signal processing techniques. The devices were designed at FEI University using IBM BiCMOS 8HP 0.13 μm technology. The terminal gate of the lateral PIN photodiode controls the vertical depletion layer, in order to modulate the capture of the electron-hole pairs generated by different wavelengths along the intrinsic region. The collected data were organized in optical power classes with the corresponding cathode currents. The experimental characterization shows that the terminal gate voltage can be used to discriminate different incident colors. The simulation results allowed the acquisition of a complete dataset to apply machine learning techniques and classify the photodetector captured color. This technique can be extended to photosensors arrays and eliminate the organic color filters. The best result was obtained using the “Linear and Quadratic Discriminant” algorithms, with the average accuracy of 92.46% and 96.82%, respectively.