Sophoton Inc., established in April 2021 withheadquarters in Suzhou and Shanghai, is a startup company specializing inhigh-performance 3D sensing (dToF) technology. Sophoton designs andmanufactures SPAD-SoC for smart hardware such as automotive LiDAR, smartphoneLiDAR, and 3D modules for robots, aiming to empower all smart devices with 3Dsensing capabilities and advance the 3D era of machine vision.
LiDAR is widely used in robotics applicationsthat require environmental perception. Although LiDAR has advantages such ashigh resolution and long range in environmental detection, there are stillissues that need to be addressed. This project focuses on a problem calledhigh-reflection blooming. The high-reflection blooming effect usually refers tothe point cloud outline of high-reflectivity markers expanding to thesurroundings, forming a point cloud shape larger than the actual size of theobject, with additional point clouds having lower reflection intensity. Thisposes a significant safety hazard in highly intelligent fields such asautonomous driving.
The goal of this project is to design anin-memory computing method to increase data bandwidth and accelerate theprocessing of the high-reflection blooming effect. During the design process,circuit area and time efficiency should be considered simultaneously to ensurethat the processor can quickly and accurately process high-reflectivity markerdata.
We will design the overall architecture ofthe processor, determine the location and number of in-memory computing units,optimize memory access efficiency, and use Verilog for RTL-level circuitdesign. The in-memory computing units and acceleration modules will be designedto efficiently process data related to the high-reflection blooming effect. Byrandomly selecting 100 test sets for testing, we achieved 100% accuracy,verifying the effectiveness of the processor. We will also compare theefficiency with traditional methods to evaluate the performance improvement ofthe in-memory computing accelerated processor in handling the high-reflectionblooming effect, and conduct experimental verification to test the actualworking effect and performance indicators of the processor. The successfulimplementation of this project will significantly improve the performance ofLiDAR in high-precision applications such as autonomous driving and robotics,address the safety hazards caused by the high-reflection blooming effect,provide a reliable data preprocessing solution for future LiDAR chips, andpromote the technological advancement and commercial application of relatedindustries.
