Primary radiation damage in silicon from the viewpoint of a machine learning interatomic potential

Abstract

Characterization of the primary damage is the starting point in describing and predicting the irradiation-induced damage in materials. So far, primary damage has been described by traditional interatomic potentials in molecular dynamics simulations. Here, we employ a Gaussian approximation machine-learning potential (GAP) to study the primary damage in silicon with close to ab initio precision level. We report detailed analysis of cascade simulations derived from our modified Si GAP, which has already shown its reliability for simulating radiation damage in silicon. Major differences in the picture of primary damage predicted by machine-learning potential compared to classical potentials are atomic mixing, defect state at the heat spike phase, defect clustering, and re-crystallization rate. Atomic mixing is higher in the GAP description by a factor of two. GAP shows considerably higher number of coordination defects at the heat spike phase and the number of displaced atoms is noticeably greater in GAP. Surviving defects are dominantly isolated defects and small clusters, rather than large clusters, in GAP's prediction. The pattern by which the cascades are evolving is also different in GAP, having more expanded form compared to the locally compact form with classical potentials. Moreover, recovery of the generated defects at the heat spike phase take places with higher efficiency in GAP. We also provide the attributes of the new defect cluster that we had introduced in our previous study. A cluster of four defects, in which a central vacancy is surrounded by three split interstitials, where the surrounding atoms are all 4-folded bonded. The cluster shows higher occurrence in simulations with the GAP potential. The formation energy of the defect is 5.57 eV and it remains stable up to 700 K, at least for 30 ps. The Arrhenius equation predicts the lifetime of the cluster to be 0.0725 µs at room temperature.

Review: Machine learning for advancing low-temperature plasma modeling and simulation

“…Hamedani et al 112 used an ML interaction potential to conduct an MD study on the radiation damage and sputtering of Si. The utilized ML potential, specifically the Gaussian approximation potential (GAP), 113 was originally trained to equilibrium properties and, hence, complemented with a DFT repulsive potential (i.e., DMol 114 ) for short range interactions to resolve the evolution of the collision cascade accurately.…”

“…For example, the Ziegler-Biersack-Littmark potential 214 and all-electron DFT repulsive potential, DMol, have been merged with ML interaction potentials (e.g., GAP), which were trained to equilibrium properties, to account for screened nuclear repulsion during high-energy collisions when radiating Si or Al. 112,117,215,216 Caro et al 217,218 conducted MD simulations with the GAP to study the ion beam deposition of C thin films. Studying a complementary sputter deposition can be readily achieved by introducing Ar þ ions to the process, whose interactions are sufficiently well described by traditional means.…”

“…Hamedani et al 112 . used an ML interaction potential to conduct an MD study on the radiation damage and sputtering of Si.…”

Radiative and Magnetically Stimulated Evolution of Nanostructured Complexes in Silicon Surface Layers

“…However, technological operations of the manufacturing of electronic devices [17][18][19], operating under extreme external fields [20][21][22][23][24], can stimulate the creation of defect complexes based on point defects (e.g., vacancies and interstitial silicon atoms) [25][26][27] and their further interaction and clustering [28,29].…”

RADIATIVE AND MAGNETICALLY STIMULATED CHANGE OF THE ELECTRICAL PROPERTIES OF THE SiO2-Si STRUCTURE WITH A HIGH CONTENT OF DISLOCATION COMPLEXES

“…Разом з тим, технологічні операції виготовлення електронних пристроїв [10,11], експлуатація в екстремальних зовнішніх полях [12] можуть стимулювати створення внутрішніх дефектних комплексів на онові власних точкових дефектів (вакансій та міжвузлових атомів кремнію) [13] та їх подальшу взаємодію з дислокаціями та кластеризацію [14].…”