Process variation impact on TFET-based leaky integrate-and-fire neurons and a novel lightweight PUF application in SNNs

Abstract

Spiking neural networks (SNNs) often face security and resource constraints, highlighting the need for lightweight hardware solutions. Tunneling field-effect transistors (TFETs) offer low-power operation for leaky integrate-and-fire (LIF) neurons due to their unique current mechanism. This work proposes a physical unclonable function (PUF) architecture based on TFET-LIF neurons for secure edge SNN applications. Leveraging random dopant fluctuation (RDF)-induced randomness in germanium-channel DG-TFETs as an entropy source, we emulate LIF behavior and analyze the impact of RDF, and work function variation (WFV). RDF is identified as the dominant factor influencing firing threshold (V_TH-LIF). A compact circuit is used to extract PUF responses, and every spike consumes 0.27 fJ. The PUF shows high reliability (≥ 97.11%), 49.8% uniqueness, and passes the NIST randomness test, enabling secure key generation and authentication.