A 3D-VLSI Architecture for Future Automotive Visual Recognition

Abstract

3D-VLSI s are indispensable to achieve future high-performance and low-power systems for automotive application which should be highly dependable. An image sensor system module with visual recognition function is required for future automatic driving vehicles. On the basis of the research being done by the authors, a conceptual 3D-VLSI image sensor system module for future automatic driving vehicle has been developed that uses heterogeneous 3D integration technology and 2.5D technology where a pair of 3D-stacked image sensors and 3D-stacked multicore processor s are integrated on a silicon interposer. A 3D-stacked image sensor with four layers of CMOS image sensor layer, analog circuit layer, ADC layer, and interface circuit layer was designed and an extremely high frame rate of 10,000 flames/s was achieved by introducing a block-parallel architecture. A high-performance 3D-stacked multicore processor was designed for visual recognition and the possibility of achieving the high-performance of 5 TFlops and the low-power consumption of 5 W was evaluated using the fabricated prototype 3D-stacked multicore processor. In addition, it was also evaluated whether the high dependability of 80 Fit was achieved by introducing the self-test and self-repair function in the 3D-stacked multicore processor. A prototype multicore processor was built to demonstrate the basic feasibility of 3D integration.