New HDPU moves neural network inference to the sensor

Researchers have developed a Hardwired Distributed Processing Unit (HDPU) capable of performing neural network inference directly on a sensor, freeing up the host system. This HDPU, implemented on an Artix-7 FPGA, can process camera frames for tasks like zebra-crossing detection in under 37.49 ms. It achieves high accuracy, with a 95.7% success rate on street photographs and a significant improvement to 78.2% on dashcam frames after a hardware contrast stage. By offloading inference and reducing data transmission, the HDPU drastically cuts down on processing time and data volume sent to the host. AI

IMPACT Enables more efficient and localized AI processing, potentially reducing latency and power consumption in edge devices.

RANK_REASON The cluster contains a research paper detailing a novel hardware architecture for AI inference. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Artix-7 FPGA
• arXiv
• CNN
• Hardwired Distributed Processing Units
• HDPU
• Python
• ResNet-18

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →