Charge Trapping and Variability in CMOS Technologiesat Cryogenic Temperatures

7 Charge Noise Characterization

Charge noise is inevitable during device operation and thus can also be seen in electrical characterization. It occurs in different forms on both large area devices and scaled devices. On scaled devices, there are discrete steps in the measured drain-source current (math image), which correspond to charge capture and emission events of single defects. On large area devices, thousands of such charge capture and emission events take place simultaneously. Therefore in large devices discrete steps are not visible anymore, however, the superposition of the defects can be observed as 1/f noise.

In the following first the experimental characterization and the theoretical models describing charge noise in the frequency domain are introduced. Afterwards, algorithms for detecting and extracting the step heights and capture and emission time distributions are explained before measured RTN at cryogenic temperatures will be analyzed for different technologies.

7.1 Experimental Characterization

Experimentally, charge noise characterization on a MOSFET is straight-forward: A constant voltage is applied at the gate of the DUT, and the drain-source current is recorded for a defined period using an equidistant sampling scheme. Using an initially measured (math image)((math image)) curve, the current signal can be mapped to an equivalent (math image) signal. Active charge traps can capture and emit charges which can be seen as discrete steps in the recorded current and the corresponding (math image) curve. This is shown on four exemplary traces in Fig. 7.1 (left).


Figure 7.1: Four exemplary RTN signals show different mean charge capture and emission times and step heights. From these RTN signals the power spectral density (represented by the same color as the corresponding RTN signal) can be obtained, which follows a Lorentzian distribution. The corner frequency of the Lorentzian corresponds to the mean capture/emission time \( \tau    \) of the RTN signal. The superposition of uniformly distributed Lorentzian signals follows a 1/f behavior, which is well known from large area devices.

Such traces can be recorded for various gate voltages and temperature conditions. This allows to determine the dependence of different parameters such as mean charge capture and emission times or mean step heights on (math image) and \( T \). By varying the sampling frequency from high to low frequencies, defects with high and low capture and emission rates can be accessed.