Atomistic implementation of defect diffusion and interdiffusion in SiGe heterostructures P. Castrillo, M. Jaraiz, R. Pinacho, J. E. Rubio University of Valladolid Spain Atomistic implementation of defect diffusion and interdiffusion in SiGe heterostructures Outline • Motivation • Physical models • Simulation scheme • Simulation results • Conclusions Castrillo et al (University of Valladolid), EMRS 2009 Strain for CMOS performance “Since introduction at the 90nm node, strain has become a central performance enhancement element for the standard CMOS flow” • mobility ↑ • drive currents ↑ 45 nm technology: Si1-xGex with x ∼0.3 in embedded S/D Ge for post-Si technologies? • Higher carrier mobilities make Ge a potential candidate SiGe: alloys, strain and more • A new degree of freedom : from Si to Ge, from 0 to ±4.2% • The vast world of semiconductor heterostructures… Authin et process al., Intel Tech. Journal (2008) • And new issues technologies Castrillo et al (University of Valladolid), EMRS 2009 Simulation of SiGe processing • Some of the new issues: – Composition dependence of diffusivities – Si-Ge interdiffusion – Strain-induced diffusion anisotropy No just a – Modification of band structure recalibration – Dopant segregation – Alloy inhomogeneities • Drawbacks for continuum simulators: – Many equations (many mechanisms, 3D…) – Parameters (difficult correlation with microscopic properties) Castrillo et al (University of Valladolid), EMRS 2009 Atomistic Process Simulators • Based on non-lattice Kinetic Monte Carlo approach • Atomistic description of defects and dopants. • Can handle sizes and times for device processing. • Direct plug of ab-initio or basic experimental data. • Inherently 3D. • Computation time increases moderately with number of mechanisms. • More advantageous for deep submicron, far from equilibrium,… • Our simulator: DADOS. Jaraiz et al, IEDM (2007) Castrillo et al (University of Valladolid), EMRS 2009 Outline • Motivation • Physical models • Simulation scheme Self-diffusion Strain effects Interdiffusion Dopant diffusion • Simulation results • Conclusions Castrillo et al (University of Valladolid), EMRS 2009 Diffusion events in SiGe Point-defect diffusion Si and Ge self-diffusion Interdiffusion Dopant diffusion Si I B Ge V Si Ge • Diffusion activated by Vs and Is. • Transport capacities: DCV and DCI • Charge states (V0, V-,V=,I0,..) : DCV = ∑ DCVq Castrillo et al (University of Valladolid), EMRS 2009 Self-diffusion in Si and Ge 1.E-11 1.E-12 1.E-13 D (cm2/s) 1.E-14 Lines: model 1.E-15 Symbols: experimental 1.E-16 Ge in Si Bracht et al, PRL (1998) Ge self-diff Si self-diff 1.E-17 Ural et al, PRL (1999) Si in Ge 1.E-18 Shimizu et al, PRL (2007) Werner et al, PRB (1985) 1.E-19 Vogel et al, JPC (1983) 1.E-20 Strohm, PhD thesis (2002) 1.E-21 Huger et al, APL (2008) 6 7 8 9 10 11 12 1e4/T 13 14 Zangenberet al, PRL (2001) Silvestri et al, SST (2006) • In general, DSi ≠ DGe Castrillo et al (University of Valladolid), EMRS 2009 Self-diffusion in SiGe • Defects could move differently Si and Ge atoms • Different self-diffusivities in Si1-xGex: Castrillo et al (University of Valladolid), EMRS 2009 Dependence on Ge content: Ge Si (-/=) (0/-) (0/-) (+/0) 1.05 0.6 V 1.0 0.35 I Approximation: energies and entropies linear with Ge content (0/-) (-/=) (0/-) (+/0) 0.4 0.2 V 0.6 0.0 I Castrillo et al (University of Valladolid), EMRS 2009 Ge and Si self-diffusion in SiGe Experimental: Bracht PRL (1998) Ural PRL (1999) Werner PRB (1985) Vogel JPC (1983) Strohm, Thesis (2002) Zangenber, PRL (2001) Silvestri SST (2006) Castrillo et al (University of Valladolid), EMRS 2009 DGe in SiGe: effective Eact and prefactor Is at high T Vs at low T Only Vs • Multiple components: ⇒ Effective values of Eact and prefactor ⇒ Prefactor dependence on x Experimental: Strohm , PhD thesis (2002) Zangenber et al, PRL (2001) Castrillo et al (University of Valladolid), EMRS 2009 DGe/DSi ratio Comparison with experimental results for different Si1-xGex compositions Symbols: Kube et al MSSP (2008). Dotted: fit to data for x=1 Castrillo et al (University of Valladolid), EMRS 2009 Extrinsic self-diffusion • For Ge, extrinsic effect attributed to V= Dotted line: estimated from Naganawa et al, APL (2008) Symbols: Werner et al, PRB (1985) Castrillo et al (University of Valladolid), EMRS 2009 Strain • Biaxial strain Example: SiGe on Si Relaxed Si1-xGex a0 > 0.543nm Biaxially compressed Si1-xGex Strain: ε (compressive ε<0) Stress: σ = Y·ε a0 Relaxed Silicon a0 = 0.543nm a0 Castrillo et al (University of Valladolid), EMRS 2009 Strain effects on diffusion • Biaxial compression Eform(V) ↓ ⇒ DCV ↑ Eform (I) ↑ ⇒ DCI ↓ • Diffusion jump anisotropy growth direction For neutral defects in plane direction • Changes in charge levels ⇒ ∆DC q − Band changes: hydrostatic + band splitting − Charge level assumed to scale as the hydrostatic ∆Egap Castrillo et al (University of Valladolid), EMRS 2009 Strain effects on diffusion Activation volumes Si0.9Ge0.1 (0.21%) Cfr.: Aziz et al, PRB (2006); Daw et al, PRB (2001); Diebel, IEDM (2003) (0.21%) Experiments: Zangenberg et al, PRL (2001) Castrillo et al (University of Valladolid), EMRS 2009 Interdiffusion • Si and Ge diffusion ⇒ interdiffusion • Relation between DSi, DGe and Dinter: Ge fraction Nernst-Planck equation 0.16 as-grown 870ºC, 1.5h DADOS sim 0.12 0.08 0.04 x→0 (Si) ⇒ Dinter ≈ Dge x→1 (Ge)⇒ Dinter ≈ Dsi Si0.5Ge0.5 ⇒ limited by the slowest 0 10 20 30 40 50 Depth (nm) Cfr.: Aubertine et al, JAP (2003) Castrillo et al (University of Valladolid), EMRS 2009 Effect of strain on interdiffusion Dependence of Dinter with composition (T=800 ºC) substrate: Silicon DCV substrate: Si0.44Ge0.56 DCV Dinter Dinter DCI substrate DCI substrate • Slope change due to strain-induced DCI-DCV crossing Castrillo et al (University of Valladolid), EMRS 2009 Interdiffusion Comparison with experiments in SiGe superlattices Average SL composition Pseudosubstrate composition Experiments: Meduna et al, SST(2007); Ozguven et al, APL (2007,2008) Castrillo et al (University of Valladolid), EMRS 2009 Dopants Example: Boron • Diffusion: D B ≈ D 0Bi [BI0 ] [ B- ] Ediff(B) ≈ Eform (I0) + Em(BI0) - Eb(BI-) + (e(BI-)-Ei) + (Ei - EF) composition and strain “extrinsic” • Segregation: Eform(B-) = Eform(B0) + (e(B-) - Ei) + (Ei - EF) composition and strain “extrinsic” • No need for B-Ge pair model assumption, ready for other dopants. Castrillo et al (University of Valladolid), EMRS 2009 Dopant in unstrained diffusion Si1-xGex Antimony Boron Experimental data: Kringhøj et al, PRL (1996); Laitinen (2004); Fair (1981); Kuo et al, APL (1995); Uppal et al, JAP (2004) • Quadratic dependence of Ediff with Ge content Castrillo et al (University of Valladolid), EMRS 2009 Outline • Motivation • Physical models • Simulation scheme • Simulation results • Conclusions Castrillo et al (University of Valladolid), EMRS 2009 Atomistic diffusion scheme • Atomistic handling of defects and dopants: particles • Continuum handling of Fermi level and strain: boxes nGe1 nGe2 • Ge atom counters per box • Particle properties depend on box properties: x (within a radius), strain, Fermi level (if charged)… • Jump rate (may be anisotropic) drift • Charge state updated every jump • Jump probability rejection as a function of ∆(Eform+Em) • Boxes (x, strain and Fermi level) updated from time to time Cfr: Martin-Bragado et al, JAP (2005) Castrillo et al (University of Valladolid), EMRS 2009 Charged defects in heterostructures • Correct formulation of the forces on charged defects in heterostructures drift Castrillo et al (University of Valladolid), EMRS 2009 1019 cm-3 uniformly B-doped Si/ Si0.7Ge0.3 abrupt structure at T = 900ºC. • Effect of band-bending on extrinsic diffusivity E nergy (eV) • Si/SiGe heterostructure bands at process temperatures 0.6 0.4 EC eBi(0,-) 0.2 Ei 0.0 EF -0.2 EV -0.4 0 D iffus ivity (cm2/s ) Band-offsets and extrinsic diffusivity 10 1E -14 x (nm) 20 30 1E -15 1E -16 1E -17 0 10 20 x (nm) 30 Castrillo et al (University of Valladolid), EMRS 2009 Interdiffusion scheme • Driven by Is & Vs • Example. I: box1 box2 1 2 ⇒ ⇒ (and analogous for Is) Nernst-Planck equation by components (Vs or Is) • Small modification of α coming from strain relaxation Castrillo et al (University of Valladolid), EMRS 2009 Outline • Motivation • Physical models • Simulation scheme • Simulation results • Conclusions Castrillo et al (University of Valladolid), EMRS 2009 KMC simulations 0.5 0.25 0.4 0.2 Ge fraction Ge fraction Interdiffusion 0.3 0.2 as-grown 920ºC, 1h DADOS sim 0.1 0.15 0.1 0.05 0 as-grown annealed 845ºC, 9h DADOS sim 0 0 10 20 30 Depth (nm) 40 50 00 10 10 20 20 30 30 40 40 50 50 Depth (nm) Depth (nm) Experiments: Xia et al, JAP 2007; Aubertine et al, JAP 2005 Castrillo et al (University of Valladolid), EMRS 2009 KMC simulations: Interdiffusion 0.2 0.1 0 0.2 0.1 0 0 10 20 30 40 relaxed 0.3 0 10 20 30 40 Depth (nm) Si0.7Ge0.3 on Si0.7Ge0.3 Ge fraction tensed 0.3 Ge fraction 0.3 Ge fraction Si0.7Ge0.3 on Si0.44Ge0.56 Si0.7Ge0.3 on Si compressed 880ºC, 90min as-grown annealed DADOS sim 0.2 0.1 Experiments: Xia et al, APL (2006); JAP (2007) 0 0 10 20 30 40 Depth (nm) Castrillo et al (University of Valladolid), EMRS 2009 KMC simulations Interdiffusion 800ºC, 120min Experiments: Xia et al, JAP (2007) 0.6 0.6 0.5 0.5 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 Depth (nm) 40 Ge fraction 0.6 Ge fraction Ge fraction as-grown annealed DADOS sim 0.4 0.3 0.2 0.4 0.3 0.2 0.1 0.1 0 50 0 0 50 0 10 20 30 Depth (nm) 40 10 20 30 40 50 Depth (nm) Castrillo et al (University of Valladolid), EMRS 2009 Simulation results Boron diffusion: Si vs. SiGe Initial Experiment Simulation T = 860ºC, t = 30 min Experiment: Kuo et al, APL (1993) Unstrained Silicon Strained Si0.83Ge0.17 • Retarded by both Ge content and compressive strain Castrillo et al (University of Valladolid), EMRS 2009 KMC simulations Boron segregation to SiGe 96 hours 850ºC Strained Si0.9Ge0.1 Dots: Experiment: Lever et al, JAP (1998); Lines: simulation Castrillo et al (University of Valladolid), EMRS 2009 Conclusions • Comprehensive description of diffusion in SiGe • From point defects to self-, inter- and dopant diffusion • Connection of the basic microscopic phenomenology with measurable magnitudes • Parameter calibration with a wide range of experimental conditions and ab-initio calculations • KMC implementation in the atomistic DADOS simulator • Good agreement between simulated profiles and experiments Castrillo et al (University of Valladolid), EMRS 2009