A 185 TOPS/W/mm2 Bayesian Inference Engine with 640 aJ Write-Free FeFET GRNG for Uncertainty-Aware Aerial Search and Rescue

2026-06-09

AI Summary

This work addresses the challenge of costly false alarms in aerial search-and-rescue missions caused by overconfident predictions from conventional deterministic deep learning models. To mitigate this, the authors propose a ferroelectric field-effect transistor (FeFET)-based Bayesian inference engine featuring a novel write-free Gaussian random number generator grounded in the Central Limit Theorem (CLT-GRNG). By stochastically selecting and summing currents from pre-programmed FeFET subsets, the CLT-GRNG enables highly efficient Gaussian sampling without any write operations during inference. This design dramatically enhances energy efficiency and device endurance: each CLT-GRNG sample consumes only 640 aJ, yielding a 560× improvement over existing Bayesian accelerators. The resulting compute-in-memory (CIM) chip achieves 185 TOPS/W/mm² and demonstrates superior uncertainty calibration and environmental robustness in search-and-rescue scenarios, substantially reducing false alarm overhead.

Abstract

Aerial search and rescue missions require fast and reliable victim detection under uncertain and rapidly changing environments. Deterministic deep learning models can produce overconfident false positives, forcing unmanned aircraft systems to perform costly verification maneuvers that reduce search coverage and increase rescue delay. Bayesian neural networks provide uncertainty-aware detection, but their sampling overhead is challenging for battery-constrained edge platforms. This work presents a FeFET-based Bayesian inference engine with a write-free central limit theorem Gaussian random number generator embedded in a compute-in-memory macro. By summing currents from a randomly selected subset of minimum-sized, programmed-once FeFETs, the proposed architecture eliminates energy- and endurance-intensive write operations during inference while maintaining scalable Gaussian sampling. The CLT-GRNG consumes 640 aJ per sample, providing a 560x energy-efficiency improvement over prior BNN accelerators, while the CIM tile achieves 185 TOPS/W/mm2. Evaluated on aerial search and rescue detection, the Bayesian model improves uncertainty calibration and robustness under environmental corruption, reducing risk and enabling low-confidence detections to be filtered before costly verification. These results demonstrate an energy-efficient and uncertainty-aware edge AI engine for autonomous search and rescue systems.

Problem

• Bayesian neural networks
• uncertainty-aware detection
• aerial search and rescue
• edge AI
• energy efficiency

Innovation

• FeFET
• Bayesian inference
• write-free GRNG
• compute-in-memory
• energy-efficient AI