The role of di-boron diffusion in evolution of B diffusion profiles has been investigated. We find that boron pair (B s -B i ) diffusion can become as important as boron-interstitial pair (B s -Si i ) diffusion when both boron concentration and annealing temperature are very high, leading to concentration-dependent B diffusion. Our simulated B diffusion profiles with dramatic shouldering are in excellent agreement with experimental ones reported by Schroer et al. ͓Appl Boron doping is an essential ingredient in the fabrication of silicon-based semiconductor devices. As gate dimensions shrink to nanometer scales ͑р100 nm͒, it becomes critical to gain precise control of doping profiles. Consequently, a great deal of effort is being devoted to understanding and controlling transient enhanced diffusion ͑TED͒ of boron during implantation and postimplantation annealing.While it is understood that a mobile boron-silicon interstitial pair (B-Si i ) plays an important role in B TED, 1,2 still little is known about underlying reasons for the enhancement ͑or the retardation͒ of B diffusion at high concentrations of boron (Ͼ10 18 cm Ϫ3 ) ͑or impurities, such as carbon or oxygen͒.We were particularly intrigued by the diffusion profiles ͑Fig. 1͒ determined by Schroer et al. 3 using secondary ion mass spectroscopy ͑SIMS͒. Their results show clearly a concentration-dependent behavior; that is, B diffusion is enhanced as the B concentration increases. They implanted boron at energies ϳ500 eV with a dose of 10 15 cm Ϫ2 into a p-type, epitaxially grown ͑epi͒ silicon layer on Si͑001͒. Then, the substrate was annealed at 1200°C. The concentrations of oxygen and carbon in the epi-Si layer are typically less than 10 15 cm Ϫ3 . Hence, impurities are likely to play an insignificant role in determining the doping profiles in these experiments. In addition, high temperature annealing at 1200°C results in fast dissolution of B clusters formed at the very early stages of annealing. Therefore, the density of immobile large boron clusters, if any, is too low to influence diffusion profile evolution. This suggests that only Si interstitials and mobile B species should be considered in explaining these experiments. In the absence of concentrationdependent and/or transient effects, single component diffusion should lead to a Gaussian distribution ͑once the diffusion profile is fully developed͒, but the experimental results 3 in Fig. 1 differ substantially from a simple Gaussian. B TED appears to be enhanced with increasing B concentration, leading to shouldering in the diffusion profiles. The shouldering behavior is consistent with a previous experimental observation 4 that shows B diffusion enhancement at high B concentration; that is, 10 B diffusion increased at the presence of high concentrations (Ϸ10 19 cm Ϫ3 ) of background boron 11 B. In fact, the concentration-dependent B diffusion has been explained by the variation of charged defect concentrations under extrinsic conditions ͑i.e., Fermi level shift effect, vide infra͒. 10 In this letter...