Probabilistic Memory for Trustworthy Edge Intelligence

2026-07-02

AI Summary

This work addresses the throughput gap and high instruction overhead between Gaussian random number generation and computation that hinder efficient deployment of probabilistic computing in edge intelligence. The paper proposes probabilistic memory (p-MEM), which for the first time integrates probabilistic sampling with memory access by storing distribution parameters directly in memory and performing in-situ sampling, treating deterministic data as a zero-variance special case. This design delivers a high-throughput, low-latency, and energy-efficient hardware primitive for probabilistic computation, enabling scalable and trustworthy probabilistic AI. Experimental results demonstrate that p-MEM achieves a Gaussian random number generation throughput of 1000 GSa/s/mm², reducing instruction counts by 2.19× and 4.37×, sampling latency by 562× and 3.45×, and energy consumption by 295.5× and 3.53× compared to conventional CPU and GPU implementations, respectively.

Abstract

Probabilistic computation plays an important role in trustworthy edge intelligence to quantify uncertainty, enhance robustness, reconstruct data, and protect privacy, but its adoption is limited by the orders-of-magnitude data throughput gap between Gaussian random number generation (GRNG) and computation, as well as instruction overhead. This paper introduces probabilistic memory (p-MEM), a unified memory primitive that stores distribution parameters, such as mean and standard deviation, and samples directly at the native memory bandwidth, where deterministic data becomes the zero-variance special case. Using a layout-validated p-MEM simulator, we comprehensively explore device choices, memory specifications, and technology nodes, showing that p-MEM can achieve more than 1000 GSa/s/mm^2 GRNG throughput, including memory-array access. Integrated into CPU/GPU systems, p-MEM reduces instruction count by up to 2.19x/4.37x, sampling latency by 562x/3.45x, and energy by 295.5x/3.53x for Bayesian neural network workloads, providing a scalable hardware substrate for trustworthy probabilistic AI.

Problem

• probabilistic computation
• Gaussian random number generation
• trustworthy edge intelligence
• memory bandwidth
• instruction overhead

Innovation

• probabilistic memory
• Gaussian random number generation
• trustworthy edge intelligence
• Bayesian neural networks
• hardware acceleration