Atomic-scale insights on hydrogen trapping and exclusion at incoherent interfaces of nanoprecipitates in martensitic steels

Hydrogen is well known to embrittle high-strength steels and impair their corrosion resistance. One of the most attractive methods to mitigate hydrogen embrittlement employs nanoprecipitates, which are widely used for strengthening, to trap and diffuse hydrogen from enriching at vulnerable locations within the materials. However, the atomic origin of hydrogen-trapping remains elusive, especially in incoherent nanoprecipitates. Here, by combining in-situ scanning Kelvin probe force microscopy and aberration-corrected transmission electron microscopy, we unveil distinct scenarios of hydrogen-precipitate interaction in a high-strength low-alloyed martensitic steel. It is found that not all incoherent interfaces are trapping hydrogen; some may even exclude hydrogen. Atomic-scale structural and chemical features of the very interfaces suggest that carbon/sulfur vacancies on the precipitate surface and tensile strain fields in the nearby matrix likely determine the hydrogen-trapping characteristics of the interface. These findings provide fundamental insights that may lead to a better coupling of precipitation-strengthening strategy with hydrogen-insensitive designs.


Supplementary Method 1
A COMSOL software was used to simulate the hydrogen diffusion in the plate-shaped sample (with a 490×1000um rectangle shape cross-section) by using the physical mode of "Transport of Diluted Species". Following the experimental procedure, we simulated the following two processes: (1) electrochemical charging of hydrogen at a constant current density for 25mins on the sample back surface; (2) the subsequent spontaneous diffusion of hydrogen within the sample.
In the first process, the hydrogen concentration on sample back surface is set as a constant ( 0 ), under a constant current density ( ) in the electrochemical charging; and the two parameters can be related by 1 , where is the sample plate thickness, is the Faraday constant, is the density of the sample (~7.88 g cm -3 for steel) and is the hydrogen diffusion coefficient (~3.3 × 10 −8 cm 2 s -1 for our sample). Therefore, the 0 is calculated to be 2.96 ppm.
By the end of the hydrogen charging process, the hydrogen distribution in the sample can be estimated by 2 : ( , t 0 ) = C 0 (1 − ( √4Dt 0 )) (2) where is the distance along the sample thickness direction (0 ≤ ≤ ), 0 is the charging time (i.e. 25 mins).
In the second process, the pre-charged hydrogen will spontaneous diffuse within the sample, driven by concentration gradients. The hydrogen concentration ( ′) can be described by: The initial condition is defined as the final state of the first process, i.e., ′( , t 0 ) = ( , t 0 ).

17
The desorption rate of hydrogen atoms is set to be proportional to the square of hydrogen where is the recombination rate of hydrogen atoms which is set as 5 × 10 −9 . Note that, in reality, hydrogen atoms desorb by both recombination and oxidation on the sample surface, and the recombination rate of hydrogen atoms is very sensitive to the surface condition and likely changes as the surface oxide film changes with H infusion. Therefore, the above formulation can only give a qualitative estimation on the trend of hydrogen concentration evolution.
By setting = d, temporal evolution of hydrogen concentration at the sample front surface is obtained ( Supplementary Fig. 2a), which incidentally shows the same trend as the measured potential evolution ( Supplementary Fig. 2e), albeit quantitative correlation between the hydrogen concentration and the potential is yet to be explored.

Supplementary Note 1
The precipitates are dispersed in a large surface area; and therefore, it takes time to switch the SKPFM scan from one local region to another. Therefore, at a certain moment, we can only work on one potential mapping of a single precipitate. Actually, we focused on precipitate #1 and #3 in our in-situ SKPFM experiment, and recorded their potential evolution after hydrogen charging (see the results in Fig. 2). We only probed the precipitate #2 at 240hrs after the hydrogen charging which showed a hydrogen-trapping characteristic similar to that of precipitate #1 ( Supplementary   Fig. 4c and Fig. 2d). Unfortunately, precipitate #1 was destroyed in the subsequent FIB process, since the precipitate was embedded in the Pt protection layer and had a risk of being sectioned during the FIB thinning process. Therefore, we chose precipitate #2 which showed a similar hydrogen-trapping characteristic as precipitate #1 for the TEM analysis, while used the in situ SKPFM result of precipitate #1 in Figure 2 of the manuscript.

Supplementary Discussion 1
Per the SKPFM results in Figure 2 (1) Based on our DFT calculations, the lowest solution energy of H atoms within the Ti2CS lattice is ~1.04eV which is much higher than that in α-Fe (~0.23eV). As such, the total amount of H that could enter the precipitate is limited; additionally, many of the H atoms that do enter the precipitates may desorb through recombination into H2 or reaction with O2 in air on the surface.
In other words, the amount of H atoms that flow from the precipitate into the interface should be very limited. (

Supplementary Discussion 2
Due to the following reasons, the oxide films in the vicinity of the interface and on the matrix should qualitatively be the same, and respond to H uptake in the same manner.
1) The steel sample is low-alloyed and hence the oxide film should mainly be composed of iron oxides, hydroxides, and oxyhydroxides.
2) The EDS data shows trivial chemical segregation at the interface (Figs. 3h&i) and hence the matrix nearby the interface and far away from the interface should have the same composition and surface oxides.
3) Referring to the HRTEM images of the surface oxide film near the border and relatively far away from the border, the film thickness and lattice spacings of the crystalline species within the oxide film are generally the same (Supplementary Fig. 8).
In addition, defects within the oxide film (such as vacancies and voids) may also affect the potential response to H 8 . In our study, since all potential responses are captured on the same sample and the oxide films are formed under the same condition (refer to the sample preparation process in the above response), the types and content of defects in oxide films are reasonably the same. Moreover, the potential differences between the interface and the matrix are clearly much larger than the potential fluctuations (Figs. 2e&j), demonstrating that the observed feature cannot be ascribed to the variation of defects in the oxide film.
In fact, the oxide film can change during the SKPFM experiment, due to following reasons. (1) At 26℃ and 38%r.h. humidity, thickness of the oxide film increases very slowly due to continuous oxidation.
(2) H can be irreversibly trapped by defects (such as vacancies and boundaries) within the oxide film, which cannot be completely released at room temperature. However, as shown in Supplementary Fig. 4h, when the sample was charged with H for the second time, the potential response as detected on the sample surface is similar to what has been observed in the prior test; this implies that the above-mentioned changes do not significantly alter the response of oxide film to H activity.

23
The initial slightly higher potential at the border of precipitate #3 (Fig. 2f) may be attributed to the preexisting H in the oxide film which may have been introduced during the mechanical polishing process wherein water was used as a coolant. As such, the sample was not literarily H-free before the H-charging step; a small amount of H may have led to the "bright ring" feature in the initial SKPFM scan. Similarly, a faintly low potential can be seen on the border of precipitate #1 in the initial SKPFM scan (Fig. 2a).

Supplementary Discussion 3
If the interface of the precipitate #3 as shown in Figs. 2f-i traps hydrogen, as the diffusible and shallowly-trapped hydrogen gradually effuse from the matrix, the potential at the border area would eventually become lower than the matrix afar due to the slow release of hydrogen from the trapping site. On the contrary, the relatively high potential ("bright" ring feature) at the interface persists for more than 10 days. Indeed, given sufficiently strong trapping of H by the interface, the release of H at the very local will be trivial when most H has escaped from the sample, rendering no observable potential drop at the border of the precipitate. However, based on our TEM characterization, a type of feature that affect H distribution is the elastic strain nearby the border region, which is not strong H traps 9 . Consistently, the border of precipitate #1 shows a lower potential than the matrix at 4940mins, as H atoms within the tensile strain field are gradually released into the oxide film overtime.

25
In addition, what we saw on the 10 th day ( Supplementary Fig. 4c) was not a "frozen-in" state. As shown in the potential mappings in Supplementary Figs. 4c&g, the potential at the precipitate/matrix interface was still continuously evolving from the 10 th day to the 18 th day. The "dark ring" feature in the 10 th -day scan clearly fades away in the 18 th -day scan, which is consistent with the inference that H at the interface has escaped.