Modeling of Statistical Low-Frequency Noise of Deep-Submicrometer MOSFETs

Abstract

The low-frequency noise (LF-noise) of deep-submicrometer MOSFETs is experimentally studied with special emphasis on yield relevant parameter scattering. A novel modeling approach is developed which includes detailed consideration of statistical effects. The model is based on device physics parameters which cause statistical variations in LF-noise behavior of individual devices. Discrete quantities are used and analytical results for the statistical parameters are derived. Analytical equations for average value and standard deviation of noise power are provided. The model is compatible with standard compact models used for circuit simulation.

Statistical Model for the Circuit Bandwidth Dependence of Low-Frequency Noise in Deep-Submicrometer MOSFETs

“…The resulting LF noise power spectrum may 0018-9383/$25.00 © 2007 IEEE be written as the summation of the contribution of each one of the N tr traps found in the device [2], [11], i.e.,…”

“…For deep-submicrometer devices, the number of traps with energy within a few kT close to the surface Fermi level is small [8]. As both the number of traps and their position over the channel are random variables, noise performance may strongly vary between different devices on one chip and, moreover, even between different operation points of a single device [2], [8].…”

“…In a previous work by the same authors, an analytical modeling approach, which is based on microscopic instead of distributed quantities, was presented [2]. This paper advances statistical modeling by including Monte Carlo simulation, experimental data, and detailed modeling of noise behavior dependence on circuit bandwidth.…”

“…This paper is organized as follows: In Section II, the analytical statistical modeling approach presented in [2] is extended to include the frequency dependence of statistical noise behavior. In Section III, Monte Carlo simulations that corroborate the results of the analytical analysis are presented, and the model is compared to experimental data from three different technology nodes.…”

Accurate Prediction of Random Telegraph Noise Effects in SRAMs and DRAMs

“…For today's small devices, however, this assumption is invalid [2], [10], [11]. Detailed trap profile models (corroborated by measured data) suggest that, in deeply scaled technologies, only about one to two traps are active at any given bias [2], [10], [12], [13], causing significant inter-device variation.…”