DPMemory - Jianqing Liu

Collaborative Research: CNS Core: Small: Privacy by Memory Design

Project Synopsis

Differential privacy (DP) has been widely accepted as the de facto technique for protecting data privacy. Despite the decade-long research efforts on DP, there still exists a critical research problem that has been largely overlooked, that is all existing DP studies are grounded on the hypothesis that software can easily and faithfully sample and add noises from a probability distribution. However, this hypothesis is being constantly challenged by recent findings about its privacy violation and by the growing demand of privacy protection in low-end devices that may lack high-level software libraries. Hence, this project's innovative research angle is to realize DP mechanisms directly on embedded memories, which are ubiquitous in modern electronic devices.

The fundation of this project is based on the physical property of embedded memeories - data bits stored in memory cells will be corrupted when the supply voltage of the memory is reduced below the nominal voltage. As shown in Figure 1, the cell failure rate is dependent on transistor sizes, supply voltages and several other parameters. This project aims to design and optimize memory architecture and peripheral circuits to control cell failures to let stored bits be corrupted in a way that conforms to the DP notions. Towards this goal, this project proposes three research thrusts:

(1): Design Memory-based (Central and Local) DP Noises and Processing Algorithms;

(2): Design and Optimize Memories for Flexible and Robust (Central and Local) DP;

(3): Chip Tape-out, System Implementation, Evaluation, and Case Studies.

In our prior studies [TIFS'22], we developed a prototype to demonstrate the realization of Local DP (LDP) in SRAM memory and the achievement of significant power savings on memory read/write processes. Our idea was based on a key observation that memory cells exhibit failures at sub-nominal supply voltages, which perturbs data bits (e.g., 0 to 1) stored therein. To utilize this characteristic towards the LDP realization, we designed a new memory architecture with a hybrid 6T-8T cell structure that achieves heterogeneous cell failures when operating at sub-nominal supply voltages. With such design, when data is read out from the memory, some bit positions can be controlled to flip their bit values while others can have their bit values remain unaltered. The cell failures (or bit flipping) are properly controlled via runtime power control to realize the randomized response on some specific binary bits of the stored data, thereby ensuring LDP on the original data. In the meantime, we purposefully rendered cell failures to the least significant bits to minimize the utility loss.

Personnel and Collaborators

PhD students

Hui Sun (Fall 2023 - Fall 2024 )
MS., Computer Science, University of Florida
Email: hsun26 at ncsu dot edu
Role: Differential privacy denoising

Jiarui Deng (Fall 2025 - )
MS., Computer Science, Jinan (暨南) University
Email: TBD
Role: Differential privacy design for AI

Collaborators

Dr. Na Gong (site-PI)
W. Nicholson Associate Professor
Department of Electrical and Computer Engineering
University of South Alabama
Email: nagong at southalabama dot edu

Research Progress

Publications

[5]

Privacy By Memory Design: Visions and Open Problems

J. Liu and N. Gong

IEEE MICRO

(MICRO)

Vol. 44, no. 1, pp. 49-58, 2024.

[4]

Two Birds with One Stone: Differential Privacy by Low-power SRAM Memory

J. Liu, N. Gong, and H. Das

IEEE Transactions on Dependable and Secure Computing

(IEEE TDSC)

Vol. 21, no. 6, pp. 5706-5719, 2024.

[3]

Low-power and Computing-free Privacy Design for IoT Systems

H. Sun, K. Mooney, M. Pinon, T. Ji, H. Das, N. Gong, and J. Liu

IEEE Computer Society Annual Symposium on VLSI

(IEEE ISVLSI),

Knoxville, USA, 2024.

[2]

Query Integrity Meets Blockchain: A Privacy-Preserving Verification Framework for Outsourced Encrypted Data

S. Jiang, and J. Liu, Y. Liu, J. Chen, L. Wang, and Y. Zhou

IEEE Transactions on Services Computing

(IEEE TSC),

Vol. 16, no. 3, pp. 2100-2113, 2023.

[1]

Towards Anonymous yet Accountable Authentication for Public Wi-Fi Hotspot Access with Permissionless Blockchains

Y. Niu, L. Wei, J. Liu, and Y. Fang

IEEE Transactions on Vehicular Technology

(IEEE TVT),

Vol. 72, no. 3, pp. 3904-3913, 2023.

Patents

[1]

Local Differential Privacy Using Static Random-Access Memory

N. Gong and J. Liu

US Patent, under review, filed 2024.

Prototype

A custom 6T SRAM_DP chip using Efabless 130nm technology

Education and Outreach Activities

under development

Disclaimer

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.