Thermal Modeling of 2.5D Integrated Package of CMOS Image Sensor and FPGA for Autonomous Driving

Abstract

Deep learning algorithms for autonomous driving require significant data movement between the camera and the processor. We propose using 2.5D integration of a CMOS image sensor (CIS) and FPGA on silicon interposer to reduce the latency and energy consumption due to data movement. Thermal simulations of the full system show an increase in CIS temperature due to 2.5D integration. The reduction in energy consumption due to data movement is $48\times$ and in latency is $24\times$.