Bi₂O₂Se-Based True Random Number Generator for Security Applications

Abstract

The fast development of the Internet of things (IoT) promises to deliver convenience to human life. However, a huge amount of the data is constantly generated, transmitted, processed, and stored, posing significant security challenges. The currently available security protocols and encryption techniques are mostly based on software algorithms and pseudorandom number generators that are vulnerable to attacks. A true random number generator (TRNG) based on devices using stochastically physical phenomena has been proposed for auditory data encryption and trusted communication. In the current study, a Bi2O2Se-based memristive TRNG is demonstrated for security applications. Compared with traditional metal–insulator–metal based memristors, or other two-dimensional material-based memristors, the Bi2O2Se layer as electrode with non-van der Waals interface, high carrier mobility, air stability, extreme low thermal conductivity, as well as vertical surface resistive switching shows intrinsic stochasticity and complexity in a memristive true analogue/digital random number generation. Moreover, those analogue/digital random number generation processes are proved to be resilient for machine learning prediction.

Reliable Memristor Based on Ultrathin Native Silicon Oxide

“…The filament-type memristor (FM) is known as one of the most promising technologies for constructing neuromorphic computing systems, − and its intrinsic randomness can be exploited for security hardware such as true random number generators (TRNGs). − However, most FMs exhibit unreliable resistive switching (RS) with device parameters (i.e., set/reset voltages, high/low resistance states) drifting during cycle-to-cycle (C2C) operations, − which severely decreases the learning accuracy of neural networks and damages the probability distribution as required for a TRNG. In response to the challenge, there is a fast-growing interest in developing FM based on 2D nanomaterials (thickness typically <5 nm). − Such a device can exhibit ultralow energy consumption and improved C2C uniformity, because the conducting filaments (CF) can be strictly constricted in atomic-scale regions. − Unfortunately, because of the poor control of local defects (e.g., holes, cracks, folds, wrinkles, and grain boundaries) of polycrystalline 2D layered materials, those devices suffer from unfavorable leakage and inhomogeneous charge conduction, resulting in low production yields and large device-to-device (D2D) variability. − Furthermore, the general performance of memristors based on 2D nanomaterials is not superior to that of traditional memristors based on thick electrolytes, potentially because of the formation of large amounts of CFs, which would induce significant filament instability. , In addition, these atomically thin materials and associated fabrication processes are not fully compatible with the complementary metal-oxide-semiconductor (CMOS) technology, hindering implementation in practical circuits.…”

Application of 2D Materials in Hardware Security for Internet‐of‐Things: Progress and Perspective

“…Al/AlO x /Bi 2 O 2 Se/Pt memristor [79] C2C variation RTN Stable switching under different temperature (47-107 °C) ---Yes…”

“…Recurrent neural network (RNN) A type of RNN that allows the outputs at last time step to be conjunctively used as coinputs at the next time step Bi 2 O 3 Se-based TRNG [79] Support Vector Machines (SVM)…”

“…Lai et al demonstrated a CVD‐grown 2D Bi 2 O 3 Se‐based memristive TRNG for voice encryption/decryption and TRNG‐based Diffie‐Hellman Key Exchange protocol. [ 79 ] The RTN signals and the C2C variation of the Bi 2 O 3 Se‐based memristor are utilized as entropy sources for the analog mode or digital mode of random bit generation, respectively. The Bi 2 O 3 Se layer was used as an electrode by taking the advantage of high carrier mobility, air stability, extremely low thermal conductivity, and vertical surface resistive switching.…”

“…[ 16 ] Lai et al adapted Long Short‐Term Memory (LSTM)‐type recurrent neural network (RNN) to predict the output of the Bi 2 O 3 Se‐based TRNG. [ 79 ] Their results showed that LSTM could only predict the trend of output but fail to perfectly fit all the features, suggesting resilience against ML modeling attack. 2D material‐based PUFs and TRNGs are promising against ML attacks, but their resilience should be tested more extensively using other ML models.…”