Story perspectives

Georgia Tech Boosts AI Efficiency with 3D Integration

2/8/2025

25 4

Summary

Story summary

Georgia Tech researchers have unveiled groundbreaking advancements in 3D heterogeneous integration for RRAM-based compute-in-memory accelerators. Their innovative approach boasts a remarkable 39% boost in energy efficiency and an astonishing 8× increase in operational density over traditional 2D designs. By tackling power supply noise and refining interconnects, they significantly improve AI model classification accuracy, including ResNet-50 and ResNet-152.

1 of 1

Story summary

Georgia Tech researchers have unveiled groundbreaking advancements in 3D heterogeneous integration for RRAM-based compute-in-memory accelerators. Their innovative approach boasts a remarkable 39% boost in energy efficiency and an astonishing 8× increase in operational density over traditional 2D designs. By tackling power supply noise and refining interconnects, they significantly improve AI model classification accuracy, including ResNet-50 and ResNet-152.