μRNG: A 300–950mV 323Gbps/W all-digital full-entropy true random number generator in 14nm FinFET CMOS

An all-digital full-entropy True Random Number Generator (TRNG) with measured 1.3GHz operation and total power consumption of 1.5mW at 0.75V, 25^oC is fabricated in 14nm FinFET CMOS. Three independent self-calibrating entropy sources, coupled with pre-extraction correlation suppressors and a real-time BIW extractor enable ultra-low energy consumption of 3pJ/bit, while generating cryptographic-quality keys with measured Shannon entropy up to 0.99999999995 and lower-bound min-entropy >0.99. The 100% digital design enables a compact layout occupying 1088μm², with scalable operation down to 300mV, while passing all NIST statistical randomness tests.

I. Introduction

Full-entropy True Random Number Generators (TRNG) serve as the bedrock of secure platforms, providing high-entropy keys for data encryption, secure wireless communications and media content protection [1]. While a variety of non-deterministic physical phenomena like phase jitter [2], telegraph noise [3], and oxide soft-breakdown [4] are excellent sources of randomness, the technology-invariance and frequency-independence of thermal noise have made it a popular choice in recent TRNG implementations [5]–[9]. Although raw bitstreams produced by these sources display high levels of entropy and may pass NIST randomness tests, they may still contain nonuniformities and serial-correlations due to random process variations, intermittent high-frequency supply/coupling noise and feedback control loop artifacts that render them sub-optimal for direct use as full-entropy cryptographic key generators in high-volume manufacturing environments. Conventional TRNGs overcome such nonidealities by using keyed functions like HMAC, CMAC or CBC-MAC to post-process the output of a single nonuniform entropy source [10]. While block-ciphers like AES-128 in CBC-MAC mode are ideal extractors well suited for desktop and server applications, the large area (>30K gates) and high energy consumption render them unsuitable for use in ultra-low energy internet-of-things (IoT) and wearable platforms [11]. These battery-constrained systems require energy-efficient low-area solutions for generating cryptographic-quality keys.