Skip to content

MIT, UMass Amherst Develop Robust Security for AI's Memristive Arrays

Protecting AI's intellectual property in memristive arrays. New security mechanisms ensure data safety without compromising performance.

This picture shows few cross symbols and few papers and key chains on the glass table.
This picture shows few cross symbols and few papers and key chains on the glass table.

MIT, UMass Amherst Develop Robust Security for AI's Memristive Arrays

Researchers at MIT and the University of Massachusetts Amherst have developed robust security mechanisms to protect valuable intellectual property in memristive crossbar arrays used for machine learning and artificial intelligence computations. Their approach safeguards critical data without significant redesign of existing architectures.

The team, led by Muhammad Faheemur Rahman and Wayne Burleson, addressed vulnerabilities in in-memory computing systems. Memristive crossbar arrays, promising significant advantages for machine learning and AI, store learned information as weights. These weights, valuable intellectual property, are vulnerable to unauthorized extraction when hardware is compromised.

The researchers developed integrated security mechanisms to protect these weights and establish verifiable ownership. Their approach incorporates Keyed Permutation and Watermark Protection Columns, safeguarding critical data without requiring substantial redesign of existing architectures. Simulations demonstrate robust protection with minimal impact on area, delay, and power consumption. Tests across 45nm, 22nm, and 7nm CMOS nodes, using realistic interconnect models and a large radio frequency dataset, showed under 10% overhead in these metrics.

This research marks a crucial step towards building secure and efficient in-memory computing systems, addressing a critical need in the rapidly evolving field of artificial intelligence and machine learning. The approach integrates efficiently with existing memristive crossbar architectures, providing a solution without significant design modifications.

Read also:

Latest