Using Preformed Resistive Random Access Memory to Create a Strong Physically Unclonable Function

Basic Information

• Paper ID: 2510.02643
• Title: Using Preformed Resistive Random Access Memory to Create a Strong Physically Unclonable Function
• Authors: Jack Garrard, John F. Hardy II, Carlo daCunha, Mayank Bakshi (Northern Arizona University)
• Classification: cs.CR (Cryptography and Security)
• Publication Date: October 6, 2025 (IEEE ACCESS)
• Paper Link: https://arxiv.org/abs/2510.02643

Abstract

Physically Unclonable Functions (PUFs) represent a promising solution for authentication and asymmetric encryption. This paper proposes a novel protocol based on Resistive Random Access Memory (ReRAM) PUF for creating physical ReRAM PUFs with large challenge spaces. The protocol employs differential reading of unformed ReRAM as the response generation method. The paper also provides experimental hardware demonstration of the protocol on physical ReRAM devices, demonstrating excellent performance characteristics as a PUF.

Research Background and Motivation

Problem Definition

Modern secure communication requires reliable authentication and encryption mechanisms. Traditional public-key cryptographic systems depend on secure distribution and storage of private keys; however, digital keys are susceptible to copying and leakage, and are difficult to reissue once compromised.

Research Motivation

1. Vulnerability of Digital Keys: Traditional keys are in digital form, can be copied, and are susceptible to leakage threats
2. Limitations of SRAM PUF:
   • Challenge space scales linearly with limited scalability
   • Directly exposes secret information in raw responses
   • Requires storage of complete response space; cannot reduce registration size through modeling
3. Post-Quantum Cryptography Requirements: Need for quantum-resistant random data sources

Limitations of Existing Methods

SRAM PUF, as the current mainstream solution, has the following problems:

• Linear relationship between challenge space and memory cell count, poor extensibility
• Raw responses directly expose underlying randomness
• Registration process must cover complete response space
• Susceptible to differential attacks

Core Contributions

1. Proposes Novel ReRAM PUF Protocol: New protocol using unformed ReRAM cells as PUF elements
2. Implements Hardware Experimental Verification: Complete hardware demonstration on physical ReRAM devices
3. Significantly Expands Challenge Space: Challenge space scales quadratically with cell count
4. Achieves Excellent Performance Metrics: Realizes extremely low bit error rate (<0.03%) and excellent PUF characteristics
5. Requires No Custom Manufacturing: Can be implemented using existing manufacturing processes by skipping the forming step

Methodology Details

Task Definition

Design a strong ReRAM-based PUF system where input is a challenge byte stream and output is the corresponding unique response byte stream, requiring large challenge space, low bit error rate, and strong security.

ReRAM Technology Fundamentals

ReRAM Structure and Principles

ReRAM employs metal-insulator-metal thin film stack structure:

• Changes resistance state through formation and rupture of conductive filaments (CF)
• Forming process: Apply positive bias to form oxygen vacancy paths
• Reset/set process: Switch between high resistance state (HRS) and low resistance state (LRS)

Unformed ReRAM Characteristics

This research uses unformed ReRAM cells:

• Resistance state is extremely stable and easy to measure
• Avoids time-dependent relaxation caused by oxygen diffusion
• Manufacturing variability provides natural randomness source

Core Protocol Architecture

1. Challenge Generation Protocol

Challenge Generation Flow:
1. Generate initial data using random bytes and cryptographic hash via SHA256
2. Parse into address pair pool (approximately 1.5× target key size)
3. Filter high standard deviation cells
4. Verify cell pair differences are sufficiently large to produce repeatable responses
5. Generate stability mask marking unstable cells

2. Response Generation Mechanism

Response Generation Flow:
1. Receive random bytes and stability mask
2. Regenerate address pairs and apply mask filtering
3. Perform differential reading on dual ReRAM chips
4. Condition signal through analog circuit and provide offset
5. Analog comparator produces single-bit response
6. Concatenate all bits and hash before returning

Technical Innovations

1. Differential Reading Mechanism

• Uses two ReRAM chips for comparison rather than absolute measurement
• Generates responses based on resistance comparison rather than direct resistance values
• Effectively isolates underlying analog measurement information

2. Quadratic Challenge Space Expansion

Challenge space calculation: 4096 × 4096 × 8 × (1-0.33) × (1-0.12) ≈ 80,000,000 cell pairs Final CRP count: Approximately (8×10^7)^256

3. Adaptive Filtering Strategy

• Cell-level filtering: Standard deviation threshold of 30 ADC counts
• Pairing-level filtering: Difference greater than 2× sum of standard deviations
• Dynamic stability mask generation

Experimental Setup

Hardware Platform

• ReRAM Chips: CrossBar Al/Al2O3/W 1R1T configuration
• Array Specifications: 1kb×4 test array, 32 word lines, 128 bits
• Current Range: 105-793 nA (8 preselected values)
• Voltage Limit: 1.5V maximum to prevent accidental forming
• Custom PCB: Signal conditioning and analog comparison

Evaluation Metrics

1. Bit Error Rate (BER): Hamming distance between registered and actual responses
2. Reliability: Consistency of repeated responses to the same challenge
3. Uniqueness: Randomness of responses across different PUFs (ideal value 50%)
4. Diffusivity: Randomness across different challenges from the same PUF
5. Uniformity: Distribution balance of 0s and 1s in responses

Experimental Parameters

• Registration Samples: 50 samples per cell
• Test Scale: 7 chips, 8 current values, 1000 CRPs
• Filtering Parameters: Cell standard deviation threshold 30, pairing multiplier threshold 2

Experimental Results

Primary Performance Metrics

Current Dependency Analysis

BER performance at different currents:

• 105 nA: 2.563×10^-4
• 793 nA: 4.055×10^-4
• Trend: BER slightly increases with current but remains at extremely low levels

Power Consumption Analysis

• Power per challenge-response generation: Microwatt level
• Single voltage read: <40 nW
• Overall power consumption: <1 mW, suitable for client devices

Comparison with Other PUF Technologies

Related Work

PUF Technology Development

1. SRAM PUF: Current mainstream, but challenge space grows linearly
2. ReRAM PUF Variants:
   • Formed ReRAM PUF: Reliability and diffusivity issues
   • Reference cell protocol: Linear challenge space, low chip utilization
3. Other Memory PUFs: Emerging technologies such as STT-MRAM

Advantages of This Work

1. Challenge Space: Quadratic growth vs. linear growth
2. Manufacturing Compatibility: No custom process required
3. Security: Indirect measurement prevents information leakage
4. Energy Efficiency: Ultra-low power consumption suitable for IoT applications

Conclusions and Discussion

Main Conclusions

1. Successfully implemented a strong PUF protocol based on unformed ReRAM
2. Achieved excellent performance metrics: BER < 0.03%, all PUF characteristics close to ideal values
3. Realized quadratic challenge space expansion, significantly superior to existing SRAM PUF
4. Requires no custom manufacturing process, demonstrating practical potential

Limitations

1. Hardware Dependency: Requires specialized ReRAM chips and analog circuits
2. Temperature Sensitivity: Although unformed ReRAM is relatively stable, further verification is needed
3. Scale Validation: Current experiments based on relatively small array scales
4. Long-term Stability: Requires longer-duration stability testing

Future Directions

1. True Random Number Generator: Utilize high standard deviation cells for continuous random bit generation
2. Improved Modeling: Develop precise models of ReRAM cell and current relationships
3. Miniaturized Design: Create tokenized devices for practical applications
4. Neural Network Integration: Combine with ReRAM-based hardware neural networks

In-Depth Evaluation

Strengths

1. Strong Technical Innovation: First systematic use of unformed ReRAM to implement strong PUF
2. Comprehensive Experimental Verification: Complete hardware implementation and comprehensive performance evaluation
3. High Practical Value: Addresses key limitations of SRAM PUF
4. Excellent Security Design: Differential reading mechanism effectively prevents information leakage

Weaknesses

1. Insufficient Cost Consideration: ReRAM chip costs may exceed SRAM
2. Environmental Adaptability: Lacks stability testing under different environmental conditions
3. Attack Model Analysis: Insufficient analysis of resistance to advanced attack methods
4. Standardization Level: Lacks compatibility analysis with existing PUF standards

Impact Assessment

1. Academic Contribution: Provides new technical pathway for PUF field
2. Industrial Value: Likely to promote ReRAM application in security chips
3. Reproducibility: Detailed experimental setup facilitates reproduction by other researchers
4. Application Prospects: Particularly suitable for IoT devices and hardware neural network applications

Applicable Scenarios

1. IoT Device Authentication: Low-power characteristics suitable for resource-constrained devices
2. Hardware Security Modules: Can be integrated into security chips
3. Edge Computing: Collaborative operation with ReRAM neural networks
4. Supply Chain Security: Device fingerprinting and anti-counterfeiting

References

This paper cites 28 relevant references covering multiple domains including PUF technology, ReRAM devices, and cryptographic protocols, providing a solid theoretical foundation for the research.

Overall Assessment: This is a high-quality hardware security research paper that proposes an innovative ReRAM PUF solution and validates the method's effectiveness through complete hardware experiments. The paper demonstrates excellent performance in technical innovation, experimental verification, and practical value, making important contributions to PUF technology development.