True random numbers play a critical role in secure cryptography. The generation relies on a stable and readily extractable entropy source. Here, from solution-processed structurally metastable 1T' MoTe2, we prove stable output of featureless, stochastic, and yet stable conductance noise at a broad temperature (down to 15 K) with minimal power consumption (down to 0.05 micro-W). Our characterizations and statistical analysis of the characteristics of the conductance noise suggest that the noise arises from the volatility of the stochastic polarization of the underlying ferroelectric dipoles in the 1T' MoTe2. Further, as proved in our experiments and indicated by our Monte Carlo simulation, the ferroelectric dipole polarization is a reliable entropy source with the stochastic polarization persistent and stable over time. Exploiting the conductance noise, we achieve the generation of true random numbers and demonstrate their use in common cryptographic applications, for example, password generation and data encryption. Besides, particularly, we show a privacy safeguarding approach to sensitive data that can be critical for the cryptography of neural networks. We believe our work will bring insights into the understanding of the metastable 1T' MoTe2 and, more importantly, underpin its great potential in secure cryptography.

翻译：真随机数在安全密码学中扮演着关键角色，其生成依赖于稳定且易于提取的熵源。本文中，我们利用溶液处理的亚稳态1T' MoTe₂，在宽温度范围（低至15 K）和极低功耗（低至0.05微瓦）下，证明了无特征、随机且稳定的电导噪声的稳定输出。通过对电导噪声特征的表征与统计分析，我们提出该噪声源于1T' MoTe₂中铁电偶极子随机极化的波动性。进一步地，正如实验验证及蒙特卡洛模拟所示，铁电偶极子极化是一种可靠的熵源，其随机极化具有持续性和时间稳定性。利用该电导噪声，我们实现了真随机数的生成，并展示了其在常见密码学应用（如密码生成和数据加密）中的使用。此外，我们特别提出了一种对神经网络安全至关重要的敏感数据隐私保护方法。我们相信，本工作将深化对亚稳态1T' MoTe₂的理解，更重要的是，奠定其在安全密码学领域的巨大应用潜力。