Key Issues and Solutions for Characterizing Hot Carrier Aging of Nanometer Scale nMOSFETs

Abstract

Silicon bandgap limits the reduction of operation voltage when downscaling device sizes. This increases the electrical field within-a-device and hot carrier aging (HCA) is becoming an important reliability issue again for some CMOS technologies. For nanodevices, there are a number of challenges for characterizing their HCA: the random charge-discharge of traps in gate dielectric causes "within-a-device-fluctuation (WDF)," making the parameter shift uncertain after a given HCA. This can introduce errors when extracting HCA time exponents and it will be shown that the lower envelope of the WDF must be used. Nanodevices also have substantial device-to-device variation (DDV) and multiple tests are needed for evaluating their standard deviation (σ) and mean value (μ). Repeating the timeconsuming HCA tests is costly and a voltage-step-stress method is applied to reduce the number of tests by 80%. For a given number of devices under tests (DUTs), there is a little information on the accuracy of the extracted σ and μ. We will develop a method to provide this information, based on the defect-centric model. For 40 DUTs with an average of ten traps per device, the extracted μ and σ has an accuracy of ±14% and ±24%, respectively, with a 95% confidence.

A Versatile CMOS Transistor Array IC for the Statistical Characterization of Time-Zero Variability, RTN, BTI, and HCI

“…Fig. 15(a) shows the mean V th as a function of the ST showing that device degradation is strongly dependent on the V GS and V DS stress conditions [5]. As an illustrative example, Fig.…”

“…This is due to imperfections during the fabrication process, causing effects that worsen with technology scaling such as random dopant fluctuations, line edge roughness, or gradient effects [3], [4]. TDV, on the other hand, includes transient effects, like random telegraph noise (RTN), and aging effects, like hot-carrier injection (HCI) [5], [6] and both types of bias temperature instability (BTI) [7], [8]: negative BTI (NBTI) and positive BTI (PBTI).…”

Assessing the accuracy of statistical properties extracted from a limited number of device under test for time dependent variations

“…In test 2, engineer 2 also used X DUTs, but will obtain different µ and σ, because a different set of devices were used [10]. Fig.…”

“…6 The impact of the average number of traps, Nt, per DUT on the µ (a) and σ (b) extracted for DUTs=100 when the tests were repeated 1000 times (M=1000 in Fig. 4) [10]. Fig.…”

“…8 The dependence of the accuracy of standard deviation, σ, on the number of DUTs used in a test for Nt=10 (a) and Nt=40 (b). The accuracy with a 95% confidence is marked out for 40 devices [10].…”