Deep Learning Hardware Design Competition 2022

AI accelerator design competition for undergraduate students

Date
Feb 2022 – Jul 2022
Tags
  • Hardware Accelerator
  • Computer Vision
  • FPGA

Project Description

This competition is organized by Polaris, South Korea’s next-generation semiconductor convergence university program, and is open to college students nationwide. Teams run the YOLOv3 model on an FPGA board, and the team that designs the fastest and most energy-efficient hardware wins.

Details

Our team designed a high-performance, power-efficient FPGA implementation for CNN inference. It includes an adder-tree-based computational unit tailored to the Tiny-YOLOv3 model and a datapath that minimizes buffer usage while maximizing computational parallelism, optimizing performance within the constraints of limited on-chip memory.

Remarks

Our team presented at a special session of IEEE AICAS 2022 and won 2nd place out of 111 teams, receiving a $2,000 prize.

Images

AI accelerator architecture
AI accelerator architecture
Competition result
Competition result