Physical mechanisms of transient enhanced dopant diffusion in ion-implanted silicon

Abstract

Implanted B and P dopants in Si exhibit transient enhanced diffusion (TED) during annealing which arises from the excess interstitials generated by the implant. In order to study the mechanisms of TED, transmission electron microscopy measurements of implantation damage were combined with B diffusion experiments using doping marker structures grown by molecular-beam epitaxy (MBE). Damage from nonamorphizing Si implants at doses ranging from 5×1012 to 1×1014/cm2 evolves into a distribution of {311} interstitial agglomerates during the initial annealing stages at 670–815 °C. The excess interstitial concentration contained in these defects roughly equals the implanted ion dose, an observation that is corroborated by atomistic Monte Carlo simulations of implantation and annealing processes. The injection of interstitials from the damage region involves the dissolution of {311} defects during Ostwald ripening with an activation energy of 3.8±0.2 eV. The excess interstitials drive substitutional B into electrically inactive, metastable clusters of presumably two or three B atoms at concentrations below the solid solubility, thus explaining the generally observed immobile B peak during TED of ion-implanted B. Injected interstitials undergo retarded diffusion in the MBE-grown Si with an effective migration energy of ∼3.5 eV, which arises from trapping at substitutional C. The concept of trap-limited diffusion provides a stepping stone for understanding the enormous disparity among published values for the interstitial diffusivity in Si. The population of excess interstitials is strongly reduced by incorporating substitutional C in Si to levels of ∼1019/cm3 prior to ion implantation. This provides a promising method for suppressing TED, thus enabling shallow junction formation in future Si devices through dopant implantation. The present insights have been implemented into a process simulator, allowing for a significant improvement of the predictive modeling of TED.