Optimizing Network Routing Optics for use in an “OLED Smart Pixel” Prototype

Abstract

Augmented reality has become an increasingly popular technology in society and often uses AI processing to handle the lengthy computation necessary for it and to reduce latency as much as possible. Universal Display Corporation (UDC) is working in conjunction with the Lightwave Laboratory to develop a prototype of an “OLED smart pixel” that aims to do on chip optical AI processing for this application. The design consists of an Organic LED as a light source that routs through some sort of routing optics that applies the kernel weights used in convolution neural networks to detectors that sum these weights and then surrounding CMOS thin film electronics perform futher computation. Within this design the routing optics have posed a challenge and this project proposes the use of the 3D highly multimode waveguide optical splitters. This project aims to develop a reliable network geometry for the largest possible kernel size of these splitters to route the light from the OLED source to the detector summing these kernel weights. Kernel size correlates to the number of output channels from the optical splitter, and a large kernel reduces the need for further steps of kernel decomposition in the overall computation for the image processing. However, these optical splitter arrays must prove feasible by extensive ray tracing in Zemax to determine loss of light through the network and characterize any stray light that occurs in the system. The ideal end product is a perfect 1/N split of the light at the detectors, where N is the size of the output port array.