Ionizing radiation influence on 28-nm MOS transistor's low-frequency noise characteristics

In this paper, we explore the transitions of low-frequency noise characteristics in high-k metal-gate bulk CMOS transistors induced by Total Ionizing Dose (TID). Due to the strong bias dependence of the noise characteristics, differentiating between noise shifts caused by the effective biasing change and the contribution of the newly generated traps becomes extremely challenging. In order to better understand the effects of irradiation, transistor noise had to be characterized at several biasing points, both in linear and saturation regions, before and after exposure to 1 Grad (SiO2) of TID. Correlation between shifts in time and frequency domain is presented in this work, along with possible explanations for each variation that occurs. We present examples of irradiation-generated Random Telegraph Noise (RTN) defects as well as various TID effects on noise Power Spectral Density (PSD) curves with pre-existing RTN sources.


Introduction
The origin of Low-Frequency Noise (LFN) in planar MOSFET transistors has been a point of debate for over 50 years, and it remains unresolved to this day [1][2][3].For older technology nodes, it has a Power Spectral Density (PSD) curve that is proportionally dropping with the frequency; hence, it was named 1/f, or as it is commonly referred to, flicker noise.With the rapid decline in transistor dimensions, occurrence of dominant Random Telegraph Noise (RTN) with a Lorentzian PSD shape became more frequent.The explanation for both flicker and random telegraph noise is charge carrier trapping and detrapping between the gate oxide and the semiconductor channel.1/f noise is often modeled as a sum of a large enough number of individual Lorentzians distributed over all the trap locations and energies [4,5].In deep sub-micron technologies, the PSD curve rarely has a straight 1/f shape, but rather contributing traps can often be observed.In electronic devices operating in conditions with constant stressing, such defects become susceptible to change.
High-energy physics experiments are creating a demand for electronics that can reliably operate under exposure to ultra-high doses of ionizing radiation.Therefore, it is important to understand what effects irradiation can have on the noise characteristics of transistors used in novel radiation-robust ASIC designs.The Total Ionizing Dose (TID) of the new High-Luminosity Large Hadron Collider (HL-LHC) project at the Conseil Européen pour la Recherche Nucléaire (CERN) is expected to reach levels of up to 1 Grad (SiO 2 ).For that reason, we are using 1 Grad of TID as a benchmark for our experiments.Previous investigations into radiation influence on bulk MOSFETs show that the primary source of performance degradation in a transistor is the charge buildup in the gate, STI, and spacer oxides above the Lightly Doped Drain (LDD) regions, as well as the gate channel/oxide interface [6][7][8].Initial experiments showed improvement in radiation tolerance, particularly due to a decrease in oxide dimensions.With further decrease in transistor dimensions, Radiation-Induced Short Channel (RISCE) and Narrow Channel (RINCE) Effects were demonstrated [9], revealing a decrease in TID tolerance.Lastly, the sub-100 nm transistors show a reversal of the RISCE effect, with the leading explanation indicating that the halo implantation overlap increases the overall doping in the channel region and, with that, reduces the influence of the trapped oxide charge in the spacer region [10].For these reasons, together with the speed and power consumption improvements, the 28-nm CMOS technology has been identified as a strategic one for the application in future high-energy physics instrumentation [11].
-1 -In this paper, we are exploring how TID influences the LFN characteristics of minimum-sized 28 nm technology MOSFETs.The previous works in this field only show the possible level shift of the noise PSD [12,13], while we cover all the observed scenarios, including the creation and shutting down of dominant RTN defects, changes in PSD shape, and shifts in overall noise level.

Methods
The irradiation campaign was carried out in the Seifert XRD Cabinet at room temperature with a tungsten tube biased at 40 kV and 70 mA.The distance between the tube shutter and the surface of the chip is approximately 30 cm, resulting in a 6.2 Mrad/h dose rate.The transistors were biased in a diode configuration with 0.9/−0.9V at the gate and drain terminals of the NMOS/PMOS transistors, as that is the reported worst-case scenario for performance degradation under TID influence [9,14].To achieve 1 Grad of TID, the irradiation exposure lasted approximately 7 days.
Low-frequency noise was measured before and after the stressing procedure on a custom-built noisemeasurement setup (figure 1), allowing for both time-and frequency-domain noise data acquisition.To avoid any annealing effects due to long measurement times, no intermediate measurement steps were taken during the irradiation process.The drain-current noise power spectral density is obtained by averaging the Fast Fourier Transform (FFT) of 80 1-second time window data sets acquired at a sample rate of 2 GSa/s.The observable range of the frequency domain signal is from 10 Hz to 100 kHz, due to low-pass and high-pass filters inside the measurement system.DC measurements were also performed, primarily for the extraction of threshold voltage and sub-threshold slope shifts.Devices used for these experiments were high-k, Si bulk, commercial 28-nm transistors with minimum-sized dimensions (W/L = 100 nm/30 nm and 100 nm/40 nm).In order to avoid the influence of ESD protection leakage current on the measurement results, the ESD concept implemented is significantly reduced from the commercial standards.As a consequence, many devices were damaged during this study, with a higher survival rate shown in PMOS transistors.The examples demonstrated in this work were extracted from 30 irradiated and measured devices.

Results and discussion
Due to high variability in noise characteristics between transistors with the same geometries and biasing conditions (shown on figure 2a and b), to obtain a precise statistical overview of the irradiation influence, a very large number of experiments is required.In this paper, our primary focus is on providing a qualitative description of the potential effects that TID can have in individual cases.Normalizing the current noise spectral density for the drain DC current is a common tool used in the pre-and post-stressing analysis.This technique proved valid for PSD spectra with 1/f-like curve shapes, where the existing noise models [15,16] suggest that taking the DC current into account negates the difference in noise PSD associated with the TID-inflicted DC parameter degradation.Our investigations have confirmed the findings in [17], demonstrating the difference in bias dependence between an individual RTN defect and the cummulative noise contribution of multiple defects giving a 1/f-like characteristic.Therefore, we will use normalized current PSD to demonstrate TID-induced noise level shifts, and absolute current PSD to demonstrate RTN behavior changes.
The most common effect of TID on noise PSD (shown on figure 3a) is the increase in normalized PSD.However, in a smaller number of cases, the PSD can be non-affected, or even decreased (shown on figure 3b).This can be attributed to the irradiation-induced defect shutdowns.This effect will be further discussed in the next part of this section.
In the majority of cases observed, aside from inducing level shifts, irradiation causes changes in the shape of noise PSD curves.These changes are a consequence of single-trap behavior fluctuations.The most common example is the new dominant RTN defect activation (figure 4(1)).Random telegraph noise is discernible by its characteristic of exhibiting two-or multi-step discrete jump behavior of drain current in the time domain.In the frequency domain, RTN has a Lorentzian shape, flat until it reaches the corner frequency and 1/f 2 -like drop afterwards.The corner frequency of a RTN Lorentzian curve is calculated as in (3.1), with   representing the mean time required for the trap to capture electric charge and   the time to emmit it back to the channel [18].
The opposite effect to the one depicted in figure 4(1) has also been observed, meaning that the RTN center stops capturing charge at the same biasing point after irradiation (figure 4(2)).Possible causes would be one or a combination of the known oxide degradation mechanisms accounted for in bias temperature instability models [19].These include transformations of preexisting oxide defects (either in energy or physical location) by hydrogen relocation, generation and annealing of oxide defects, passivation and activation of dangling bonds.Since transistors undergo TID stress under bias, which also results in self-heating, bias temperature effect acts in combination with the high TID radiation stress.The effects have not been isolated from each other, which is a topic worth further investigation.
To compensate for the net biasing change, previous works used the method of scaling for the DC drain current.Our results show that the single RTN defect may have a different bias dependence than the 1/f-like noise; therefore, by using this method, the results can be distorted.Instead, to improve our understanding of radiation's influence on single RTN defects, better results can be obtained by biasing the transistor with a voltage that compensates for the threshold voltage shift.Biasing compensation after exposure to TID in figure 5b was performed by adjusting the gate voltage to reconcile for the V th shift, and the drain voltage to achieve the same I d as before the stressing.The second adjustment is not a required step and is often impossible due to the TID-induced I ON decrease.The results from figure 5b demonstrate that the trap did not disappear in this instance, as suggested by the measurement at the same biasing point after irradiation, but it rather stopped capturing and emitting charge due to the net biasing change of the transistor.Furthermore, we can see that the rest of the noise spectrum -4 - This method is solely useful for investigating changes in RTN behavior caused by TID-induced biasing shifts, but is limited in uncovering the potential reasons for RTN shutdown.Additionally, for a more precise analysis of irradiation effects on single RTN traps, a clear distinction has to be made between the alterations in noise PSD caused by net biasing changes and those attributable to newly generated defects.

Conclusion
The noise current spectral density characteristics of 28-nm CMOS transistors show large variability before TID stressing.We demonstrate the possible scenarios of how these characteristics further evolve after exposure to ultrahigh doses of ionizing radiation.In our measurements of multiple PMOS and NMOS transistors, we register an increase as well as a decrease in noise PSD when compared under the same bias conditions before and after TID stress.The observed variability is attributed to the high probability of dominant RTN occurrence, their bias dependence, and the complex oxide degradation phenomena due to TID stress.The fact that the 1/f noise PSD and the RTN amplitude don't follow the same bias dependence leads to the conclusion that both absolute noise PSD as well as normalized noise PSD for DC current variation have to be considered in the analysis of low-frequency noise characteristics.Finally, we show that analysis at a fixed biasing point might be misleading due to the irradiation-induced threshold voltage shift combined with the strong bias dependence of RTN.Further investigations require a better understanding of the 1/f mechanisms in very small devices, an insight into the physical model of RTN traps, and the correlation between them.

Figure 4 .
Figure 4. Example of an irradiation-induced dominant RTN center in frequency (a) and time domain (b and c).NMOS transistors with W/L = 100/30 nm, biased at Vgs = 0.5 V, Vds = 0.05 V.