Introduction

The behaviour of silicate glasses over geological timescales raises challenging scientific issues. One of the most prominent topics with a high social impact is the safe management of highly radioactive wastes originating from spent nuclear fuel reprocessing that, among others, depends on our capacity to predict the long-term corrosion rate of borosilicate glass1,2,3. Geochemists face similar fundamental problems with basaltic glass when calculating the chemical mass balance of the oceans4 or when attempting to assess potential CO2 sequestration by silicate rocks5. Current rate laws still cannot accurately predict the long-term behaviour of synthetic or natural silicate glasses over geological timescales mostly because of the existence of surface layers between pristine glass and bulk solution that both affect the transport of reactive species and change fluid properties at the glass surface compared with that of the bulk6,7. Surface layers are made of amorphous and crystalline metastable phases, formed by in situ condensation and precipitation reactions from aqueous species8. Glass dissolves following the Ostwald rule of stages, forming a series of alteration products progressively evolving towards thermodynamically stable phases6,9. Depending on the glass composition and on the conditions (for example, temperature, pH, solution composition and flow rate), this transformation into stable compounds can vary from hours to millions of years6,10,11!

In this study, glass corrosion mechanisms were investigated from macroscopic to atomistic scale, through isotopically tagged corrosion experiments performed with international simple glass (ISG)—a six-oxide borosilicate glass used as a reference material by the nuclear glass community1—and the use of complementary analytical techniques to provide evidence of the long-term rate-limiting mechanisms. The main experiment was performed with 16 coupons of ISG glass (with isotopes at the natural abundance) altered for 1 year in a static mode at 90 °C, pH90 °C 7, in a solution initially saturated with amorphous 29SiO2 (Supplementary Fig. 1). Previous experiments showed that when altered in deionized water the leaching solution of ISG glass eventually reaches the equilibrium with amorphous silica (SiO2am) but this requires several years12. This starting solution thus enables the first transient stages of glass corrosion to be bypassed and focus placed on the processes governing the long-term rate. In parallel, a similar experiment was conducted with glass powder until its complete alteration, in order to examine the structural changes within the silicate network. The choice of pH 7 was motivated by two reasons. First, after similar investigations carried out at pH 9 and 11.5, it was found that results at pH 7 were the most convincing because glass corrodes about four times faster at pH 7 than at pH 9, although the fundamental processes are the same13. This led to larger amounts of material available for in-depth characterization. Second, a pH of 7 is typical of many natural waters; for instance, the water in equilibrium with the claystone studied in France for the storage of nuclear wastes is pH 7.3 at room temperature14. Together, these studies improve our understanding of silicate glass corrosion processes, by linking dissolution kinetics and structural modifications following water ingress into the solid. At this pH, the glass is passivated by a self-healing and poorly hydrated silica layer in which water is trapped in subnanometric pores. Conclusions can be extended at least until pH 9.5, leading to general recommendations for the safe management of highly radioactive waste glass.

Results

Glass still corrodes despite silica saturation conditions

Despite the fact that the solution was initially saturated with respect to amorphous silica, the ISG glass still corrodes. This phenomenon occurred under very stable conditions: pH was kept constant (Fig. 1a) and the Si concentration did not change for the entire duration of the test (Fig. 1b). In these conditions, the glass dissolution rate, given by the release time of B, Na or Ca (the three elements behave similarly), dramatically diminished during the experiment (Fig. 1c,d). The rate decrease is ~3.7 orders of magnitude from the first hours (~500 nm d−1, which is close to the maximum rate measured far from equilibrium15), to 363 days (around 0.1 nm d−1). In the meantime, Si, Al and Zr remained almost undissolved (at 363 days, the normalized loss of Al and Zr are 100 and 1,300 times smaller than that of B, respectively) (Supplementary Table 1). In addition, as the 29Si/28Si ratio remained stable over time, it could be concluded that 29Si(aq) did not—or if so, only weakly—interact with Si from the glass during corrosion (Fig. 1b). From this, it seems that the three mobile elements behaved independently from the three low-soluble cations constituting the glassy network. This might seem at odds with the most of accepted models that attributed the drop in the corrosion rate to an increasing activity of the dissolved silica16,17,18,19.

Figure 1: Data from solution analyses.
figure 1

Time evolution of the solution pH (a), the concentration of Si, 29Si/28Si and 29Si/30Si ratios (b), the equivalent thickness of altered glass (ETh) calculated from B release into solution (c) and the glass dissolution rate (d). The pH90 °C was maintained at 7±0.25 by regular additions of diluted HNO3. Boron was used to calculate glass corrosion thickness and rate. This element is known to be a good corrosion tracer as it is very soluble and only slightly retained in the alteration layer (as verified by ToF-SIMS and EFTEM analyses).

Full size image

Mobile species display local gradients

Deeper insights into the glass corrosion mechanisms were obtained from a detailed characterization of the altered glass coupons withdrawn at 7, 209 and 363 days. For the three samples, the alteration layer displays three sublayers (Fig. 2). From the solution to the pristine glass a gradient area can first be noticed, which B, Na and Ca concentrations dropped similarly. As time-of-flight secondary ion mass spectroscopy (ToF-SIMS) is known to potentially broaden chemical profiles especially in the case of rough or tilted interfaces20, B profile was also characterized by transmission electron microscope (TEM) in order to avoid misinterpretation. Energy-filtered TEM (EFTEM) B map** performed on a focused ion beam cross-section prepared with the 209-day sample confirmed the position and the width of the B profile given by ToF-SIMS (Fig. 2a). ToF-SIMS also revealed that B, Na and Ca gradients are well anticorrelated with that of H (Fig. 2b and Supplementary Fig. 2). This pattern is generally attributed to ion exchange between exogenous positively charged species (H+, H3O+ and K+) and Na+, Ca++ from the glass and accompanied by the fast hydrolysis of B–O bonds21,22. Beyond this gradient area, B and Ca concentrations slowly decreased to zero, whereas Na remained at a low but constant concentration, likely to act as a charge compensator along with K for fourfold coordinated Al species and sixfold coordinated Zr species23. This large, central area showed a nearly flat concentration of all the glass constituents. Finally, an external area is visible, mainly characterized by significant changes in Si isotopic ratios (Fig. 2c).

Figure 2: Elemental profiles within the 209-day sample.
figure 2

(a) EFTEM map** and the superimposed profile of B concentration (normalized to that of the pristine glass). (b) The H profile (not-normalized) displaying an anticorrelated steeper gradient than that of the other mobile species of the glass (B, Ca and Na). (c) 29Si/28Si ratio exhibiting a slight enrichment in the first 250 nm of the alteration layer (this depth is possibly overestimated according to the caveat given in Methods) and then a constant value a few percent above the natural abundance (1(±1)% at 7 days, 5(±1)% at 209 days and 4(±1)% at 363 days). This ratio has to be compared with that of the bulk solution (~30). (d,e) Normalized profiles of sparingly soluble glass formers (Si and Al) and soluble elements (Na, Ca and B). Si and Al remain at a near constant concentration in the alteration layer, whereas mobile species (B, Ca and Na) are leached out of the layer except near the reaction front, where a gradient is clearly visible. Note that K is highly present in the alteration layer, but it is not displayed because its signal was saturated.

Full size image

The alteration layer is not a precipitate

In the central area region the 29Si/28Si ratio plateaus at a few percent above the natural abundance (Fig. 2c). This slight enrichment of Si species supplied by the solution increased and stayed constant between 209 and 363 days. At the steady state, it represents one atom of Si supplied by the solution per 600 coming from the glass. This result suggests that the central part of the alteration layer is not a diffusion barrier for Si(aq) as diffusion-limiting process would have led to chemical gradients. Therefore, one can assume that a thermodynamic equilibrium between the pore solution and the silicate network was achieved. The external area, ~250-nm thick (or less, according to the caveat given in the Methods section) showed a significant increase in the 29Si/28Si, especially in the first nanometre (Supplementary Fig. 3) where the isotopic ratio goes up to 1. Moreover, the profiles of the outermost sublayer are similar at 7 and 209 days (Supplementary Fig. 3). The latter observations strongly suggest that the glass surface underwent dissolution and reprecipitation reactions, and that an equilibrium between the external surface of the glass and the bulk solution was achieved within a few days. Overall, these findings are consistent with those obtained in a far more diluted medium, favouring the hydrolysis of Si–O–M bonds (M=Si, Al and Zr), where the whole alteration layer was significantly and gradually enriched in Si supplied by the solution24. Here except at the extreme surface, almost none of the silicon atoms of the glassy network were completely hydrolysed. These observations lead to two conclusions: first, a thermodynamic equilibrium is achieved between the surface of the altered glass and the bulk solution despite the absence of isotopic equilibrium inside the porous material; second, the altered glass cannot result from the congruent dissolution of the glass. The properties of the alteration layer next needed to be determined, to better understand the behaviour of the mobile species.

The alteration layer acts as a molecular sieve

It has long been thought that Si-rich amorphous alteration layers formed during glass or mineral corrosion could be passivating25,61. The B profile was calculated with the ImageJ software.

NMR spectroscopy

Liquid-state 29Si NMR spectra were recorded on a 400-MHz Agilent DD2 spectrometer equipped with the OneNMR 5-mm probe and without sample-spinning, following the procedure described elsewhere13. 29Si cross-polarization magic-angle spinning (CP-MAS) NMR spectra were collected on a Bruker Avance 300 MHz spectrometer (magnetic field 7.05 T) using a 4-mm (rotor outer diameter) Bruker CP-MAS Probe and sample-spinning frequency of 10,000 Hz. CPMG 29Si MAS spectra collected are described in ref. 40, co-adding typically 20–40 echoes, with a recyle delay of 20 s (no changes in lineshape were observed between 2 and 200 s). CP-MAS 29Si{1H} spectra were collected with a recycle delay of 2 s and 1H-29Si magnetization transfer ranging from 0.5 to 16 ms.

Consistency of the measurements

Consistency of the glass alteration thickness measurements was verified using four independent techniques. It was found that all the measurements are consistent within a range of ±15% (Supplementary Fig. 4).

Additional information

How to cite this article: Gin, S. et al. Origin and consequences of silicate glass passivation by surface layers. Nat. Commun. 6:6360 doi: 10.1038/ncomms7360 (2015).