Traditional cryptography relies on a fundamental assumption: a secret cryptographic key can be securely stored in digital memory, such as non-volatile memory (NVM) or eFuses. However, this assumption is increasingly vulnerable. If an adversary gains physical possession of a device, they can deploy sophisticated side-channel attacks, physical probing, or memory-dumping techniques to extract these static keys.
Physical Unclonable Functions (PUFs) fundamentally alter this paradigm. Instead of storing a digital secret, a PUF treats the physical hardware itself as the key. As highlighted by Chip Hong Chang et al. in The ASHES 2020 special issue at JCEN, PUFs exploit the unavoidable, microscopic manufacturing variations inherent in semiconductor fabrication to create unique, tamper-evident digital fingerprints. These variations are stochastically determined during manufacturing and are completely impossible to replicate—even by the original manufacturer using the same equipment.
The Mechanics of Physical Randomness
To understand why PUFs work, we must look at how sub-micron physical chaos is translated into reliable digital bits. When a chip is manufactured, variations in lithography, oxide thickness, and chemical doping create slight differences in the electrical behavior of individual components.
Silicon Primitives
One classic approach leverages Static Random Access Memory (SRAM). An SRAM cell consists of cross-coupled inverters that form a bistable circuit. When the device powers up, the cell is forced into a race condition to settle into either a logical 0 or 1. Due to microscopic differences in the threshold voltages of the transistors, each cell naturally tilts toward one state.
Ashwija Reddy Korenda et al. demonstrate in A Proof of Concept SRAM-based Physically Unclonable Function (PUF) Key Generation Mechanism for IoT Devices how these unique power-up states provide a high-entropy source for generating cryptographic keys on-demand. Because the key is derived dynamically at power-up and exists only in volatile memory, it completely vanishes when the device is turned off. This eliminates the risk of physical key extraction from non-volatile storage, as noted by Yang Su in SecuCode: Intrinsic PUF Entangled Secure Wireless Code Dissemination for Computational RFID Devices and Muhammad Hammad et al. in Lightweight and Robust Key Agreement for Securing IIoT-Driven Flexible Manufacturing Systems.
Another silicon approach relies on timing variations. In Lightweight and Energy-Efficient Implementation of an Unclonable WSN Nodes on PSoC, Ke Xue et al. describe a mechanism using ring oscillators. By setting up parallel inverter chains, slight differences in wire and gate delays cause each circuit to oscillate at a slightly different frequency. Comparing these frequencies across different hardware instances yields a highly stable, device-specific bitstream.
Beyond-CMOS Primitives
Modern research has expanded into emerging material technologies to find richer, more complex sources of entropy. Bertrand Cambou et al. in A ReRAM Physically Unclonable Function (ReRAM PUF)-Based Approach to Enhance Authentication Security in Software Defined Wireless Networks and Chenchen Ding et al. in Transforming memristor noises into computational innovations leverage the stochastic nature of conductive filament formation in resistive RAM (ReRAM) and memristors. The nanoscale variations in how these filaments form create unique threshold voltages and resistance states that are virtually impossible to clone or physically inspect.
The Challenge-Response Paradigm and the Machine Learning Threat
PUFs are generally categorized based on their interface and the size of their input space:
- Weak PUFs: Have a highly limited number of inputs (challenges) and are primarily used for internal key generation (e.g., SRAM PUFs).
- Strong PUFs: Support an astronomically large number of unique Challenge-Response Pairs (CRPs). This allows them to be used directly for authentication protocols without storing keys, as explored by Damith Ranasinghe et al. in Memristive crypto primitive for building highly secure physical unclonable functions.
However, Strong PUFs face a severe vulnerability: machine learning (ML) modeling attacks. If an adversary intercepts a subset of CRPs over an open digital channel, they can train an ML model to learn the underlying physical parameters of the circuit (such as individual gate delays). Once trained, the model can accurately predict the hardware's response to unseen challenges, effectively creating a digital clone of the physical device.
To counter this, researchers introduce structural and physical non-linearity to break the mathematical relationship between inputs and outputs. Hongxiang Gu et al. in Efficient and Secure Group Key Management in IoT using Multistage Interconnected PUF proposed the Multistage Interconnected PUF (MIPUF). This architecture routes signals through sequential stages of PUFs connected by reconfigurable switching networks, creating a highly complex, non-linear mapping that thwarts ML modeling.
Challenge ───► [ PUF Stage 1 ] ───► [ Reconfigurable Switch ] ───► [ PUF Stage 2 ] ───► Response
▲
Control Signals
Alternatively, researchers look to the quantum and atomic physics of electron transport. Zezhong Tu et al. in A Probability-Based Strong Physical Unclonable Function With Strong Machine Learning Immunity developed a probability-based strong PUF (Prob-PUF) that exploits the stochastic nature of electron trap emissions in transistor oxide defects. Instead of outputting a fixed binary bit, it measures the random timing of charge releases to derive a probabilistic signature, ensuring that the output cannot be modeled by deterministic machine learning algorithms.
Reconfigurability and Erasure
A major limitation of traditional silicon PUFs is that their physical structure is fixed at manufacturing. If a set of CRPs is intercepted or compromised, the hardware primitive becomes permanently insecure. To solve this, researchers have introduced reconfigurable and erasable PUFs.
Rohit Abraham John in Halide perovskite memristors as flexible and reconfigurable physical unclonable functions demonstrated that halide perovskite memristors can achieve physical reconfigurability. By resetting the memristive elements to a high-resistance state, the old state is completely erased. The inherent cycle-to-cycle variability of the material then allows the device to generate an entirely new, uncorrelated set of identifiers over its lifecycle. Similarly, Kyung Seok Woo et al. in Tunable stochastic memristors for energy-efficient encryption and computing utilize weak conductive filaments to hide cryptographic keys via partial resets, revealing them only when specific challenge voltages are applied.
When physical modification of the material is difficult, logical erasure can substitute. Chenglu Jin et al. in PLayPUF: Programmable Logically Erasable PUFs for Forward and Backward Secure Key Management developed a stateful interface that maintains a usage counter for specific challenge-response pairs. Once a predefined usage threshold is met, the interface blocks further access to that response. This software-defined access control simulates physical destruction, preventing adversaries from exploiting intercepted hardware states and guaranteeing forward and backward security (Yansong Gao et al. in Efficient Erasable PUFs from Programmable Logic and Memristors).
Expanding the Physical Substrate
The concept of hardware-intrinsic security has expanded far beyond traditional silicon microchips, finding roots in diverse physical mediums:
- Photonic PUFs: Fabio Pavanello et al. in EUROPULS: NEUROmorphic energy-efficient secure accelerators based on Phase change materials aUgmented siLicon photonicS exploit the microscopic manufacturing variations in photonic integrated circuits to create optical hardware fingerprints, bypassing electronic vulnerabilities entirely.
- Printed Electronics and Inks: Ahmet Turan Erozan et al. in Image PUF: A Physical Unclonable Function for Printed Electronics based on Optical Variation of Printed Inks introduced iPUF, which uses the random microscopic patterns of printed inks from inkjet printing to authenticate ultra-low-cost printed electronics without adding any transistors.
- Material and 3D Printing: Yiming Qiao et al. in MRS Communications embedded high-grade randomness directly into 3D-printed objects to form non-copyable physical security keys. Similarly, Tongtong Zhang et al. in Multimodal dynamic and unclonable anti-counterfeiting using robust diamond microparticles on heterogeneous substrate grew nanodiamond microparticles with unique geometric facets via chemical vapor deposition, creating indestructible, multimodal anti-counterfeiting labels.
System-Level Binding and Applications
To be useful, these physical fingerprints must be cryptographically bound to digital systems, software, and network protocols.
Intellectual Property Protection
Jiliang Zhang et al. in A PUF-FSM Binding Scheme for FPGA IP Protection and Pay-Per-Device Licensing integrated device-specific hardware fingerprints into the finite state machine (FSM) control logic of intellectual property (IP) cores on FPGAs. By adding non-functional locking states, the core remains completely disabled unless triggered by a response that matches the host device's unique PUF. Similarly, Huming Qiu et al. in RBNN: Memory-Efficient Reconfigurable Deep Binary Neural Network with IP Protection for Internet of Things bound deep binary neural network weights to specific hardware using SRAM power-up patterns, ensuring the model remains scrambled and non-functional if copied to an unauthorized chip.
Data-Device Attestation
PUFs can also bind digital data to the physical device that generated it. Yue Zheng et al. in A PUF-Based Data-Device Hash for Tampered Image Detection and Source Camera Identification fused digital image features with a hardware fingerprint (derived from CMOS camera sensor fixed pattern noise) via random projection. This creates a compact hash that is valid only for a specific combination of image content and the unique physical sensor that captured it, preventing attackers from reusing valid hashes on altered content or different devices.
Cloud and Network Security
In multi-tenant cloud FPGAs, Muhammed Kawser Ahmed et al. developed a framework called TrustToken in Trusted IP solution in multi-tenant cloud FPGA platform. It uses PUFs to generate volatile identity tokens at runtime, eliminating the need for persistent on-chip storage and preventing data leakage between successive tenants sharing the same hardware.
Finally, for critical communications in zero-trust environments, M. Gharib et al. in SCC5G: A PQC-based Architecture for Highly Secure Critical Communication over Cellular Network in Zero-Trust Environment bound post-quantum lattice-based public keys to hardware PUFs. This physical-digital link ensures that digital credentials cannot be cloned or utilized on unauthorized devices, providing an unforgeable root of trust grounded in the physical reality of the hardware.
Key Foundational Work
Controlled physical random functions and applications (2008) — contributed 6 concept elements to this direction.
Bootstrapping Trust in a "Trusted" Platform (2008) — contributed 3 concept elements to this direction.