Simulation Study of Layer-to-Layer NBTI Degradation Non-Uniformity in GAA Multi-Channel Nanosheet FETs

BACKGROUND: Gate-all-around (GAA) nanosheet pMOSFETs stack multiple channel layers within a single device, creating hydrogen-transport boundary conditions that are absent in planar architectures. Conventional negative bias temperature instability (NBTI) analyses treat all channels as equivalent and may therefore overlook layer-to-layer degradation nonuniformity. OBJECTIVE: This work aims to systematically investigate NBTI-induced layer-to-layer degradation non-uniformity in GAA nanosheet pMOSFETs and to clarify the role of the buried oxide (BOX) in governing the underlying hydrogen-transport mechanism. METHODS: Using the Sentaurus TCAD multi-species reaction-diffusion (R-D/RDD) framework, we simulate NBTI degradation in single-, double-, and triple-channel GAA nanosheet pMOSFETs. Three aspects are examined: the dependence of overall degradation on channelstack number, the spatial distribution of trapped interface charge across individual channel layers, and the hydrogen-transport mechanism. A control experiment that closes the BOX hydrogen-diffusion pathway is designed to quantitatively isolate the BOX proximity effect. RESULTS: Device-level NBTI degradation intensifies with increasing channel-stack number. In the triple-channel device, the three layers exhibit a systematic Ch1 < Ch2 < Ch3 degradation gradient whose magnitude grows as the stress voltage decreases. Closing the BOX diffusion pathway partially collapses this non-uniformity, confirming that the BOX acts as a finitecapacity hydrogen reservoir that both shields the bottom channel and amplifies layer-to-layer asymmetry. CONCLUSIONS: The layer-to-layer degradation non-uniformity in GAA nanosheet pMOSFETs is jointly modulated by stress voltage, stress time, and temperature, and all three dependencies can be unified through BOX saturation behaviour. These findings highlight the necessity of layer-resolved NBTI assessment for multi-channel GAA device design and reliability optimization.