* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Computational Electronics Relaxation Time Approximation
Survey
Document related concepts
Cross section (physics) wikipedia , lookup
Path integral formulation wikipedia , lookup
Euler equations (fluid dynamics) wikipedia , lookup
Noether's theorem wikipedia , lookup
Perturbation theory wikipedia , lookup
Navier–Stokes equations wikipedia , lookup
Equation of state wikipedia , lookup
Electron mobility wikipedia , lookup
Equations of motion wikipedia , lookup
Relativistic quantum mechanics wikipedia , lookup
Derivation of the Navier–Stokes equations wikipedia , lookup
Partial differential equation wikipedia , lookup
Monte Carlo methods for electron transport wikipedia , lookup
Transcript
Computational Electronics Relaxation Time Approximation Prepared by: Dragica Vasileska Associate Professor Arizona State University Relaxation-time approximation Analytical solutions of the Boltzmann equation are possible only under very restrictive assumptions [1]. Direct numerical methods for device simulation have been limited by the complexity of the equation, which in the complete 3-D time-dependent form requires seven independent variables for time, space and momentum. In recent times, more powerful computational platforms have spurred a renewed interest in numerical solutions based on the spheroidal harmonics expansion of the distribution function [2]. To-date, most semiconductor applications have been based on stochastic solution methods (Monte Carlo), which involve the simulation of particle trajectories rather than the direct solution of partial differential equations. Most conventional device simulations are based on approximate models for transport which are derived from the Boltzmann equation, coupled to Poisson's equation for selfconsistency. In the simplest approach, the relaxation time approximation is invoked, where the total distribution function is split into a symmetric term in terms of the momentum (which is generally large) and an asymmetric term in the momentum (which is small). In other words, f (r , k , t ) = f S (r , k , t ) + f A (r, k , t ) . (1) Then, for non-degenerate semiconductors (1 - f) ≈ 1, the collision integral may be written ∂f ∂t = ∑ [ f (k ') S (k ', k ) − f (k ) S (k , k ') ] coll k ' =∑ k' [ f S (k ')S (k ',k )− fS (k )S (k,k ')] ( ∂fS ∂t )coll + ∑ [ f A (k ')S (k ',k ) − f A (k )S (k,k ') ] k' ( ∂f A ∂t )coll (2) We now consider two cases: ∂f ∂f f S = f0 , f A = 0 → = S = 0 ∂t coll ∂t coll (a) Equilibrium conditions: (b) Non-equilibrium conditions when f A ≠ 0 . In this case, we must consider two different situations: Low-field conditions, where f S retains its equilibrium form with TC = TL . In this case - ( ∂f S ∂t )coll = 0 . High-field conditions when TC ≠ TL and f S does not retain its equilibrium form. In - this case ( ∂f S ∂t )coll ≠ 0 . In all of these cases, a plausible form for the term ( ∂f A ∂t ) coll is fA ∂f A ∂ = −τ t coll f , where τf (3) is a characteristic time that describes how the distribution function relaxes to its equilibrium form. With the above discussion, we may conclude that - fA ∂f A ∂f ∂ = ∂ = −τ t t coll f At low fields: coll . - fA ∂f S ∂f A ∂f S ∂f ∂ = ∂ + ∂ = ∂ −τ t t coll t coll t coll f At high fields: coll . To understand the meaning of the relaxation time, we consider a semiconductor in which there are no spatial and momentum gradients. With the gradient terms zero, the BTE becomes f − f0 f ∂f ∂f A = =− A =− ∂t ∂t scatt τf τf (4) f ∂f f + = 0 ∂t τ f τ f (5) i.e. The solution of this first-order differential equation is f (t ) = f 0 + [ f (0) − f 0 ] e − t /τ f . (6) This result suggests that any perturbation in the system will decay exponentially with a characteristic time constant τf . It also suggests that the RTA is only good when [ f (0) − f 0 ] is not very large. Note that an important restriction for the relaxation-time approximation to be valid is that τf is independent of the distribution function and the applied electric field. Solving the BTE in the Relaxation Time Approximation Let us consider the simple case of a uniformly doped semiconductor with a constant electric field throughout. Since there are no spatial gradients, ∇r f = 0 . Under steady-state conditions we also have ∂f ∂t = 0 . With the above simplifications, the BTE reduces to F ⋅∇p f = 1 ∂f F ⋅∇k f = h ∂t coll (7) For parabolic bands and choosing the coordinate system such that the electric field is along the zaxis, one can expand the distribution function into Legendre polynomials ∞ f ( z, p ) = f 0 ( z, p) + ∑ g n ( E ) Pn (cos θ ) n =1 (8) 2 3 1 where P0 = 1 , P1 = cos θ , P2 = 2 cos θ − 2 , … . In the above expressions, θ is the angle between the applied field (along the symmetry axis), and the momentum of the carriers. For sufficiently low fields, we expect that only the lowest order term is important, so that f ( p ) ≅ f 0 ( p ) + g1 ( p) cos θ = f 0 ( p ) + f A ( p ) . (9) Substituting the results on the LHS of the BTE and using parabolic dispersion relation gives e LHS = − E ⋅∇k [ f 0 ( p ) + g1 ( p ) cos θ ] h ∂f ∂f e ≈ − E ⋅∇k f 0 ( p ) = −eE ⋅ v 0 = −eEv cos θ 0 ∂ε ∂ε h (10) where, as previously noted θ is the angle between the electric field and v. We now consider the collision integral on the RHS of the BTE. Substituting the first order approximation for f(p) gives ∂f ∂t = ∑ [ S (k ', k ) f0 (k ') − S (k , k ') f 0 (k ) ] coll k' + ∑ [ S (k ', k ) g1 (k ') cos θ '− S (k , k ') g1 (k ) cos θ ] k' S (k ', k ) g1 (k ') cos θ ' = − g1 (k ) cos θ ∑ S (k , k ') 1 − S (k , k ') g1 (k ) cos θ k' f (k ) g1 (k ') cos θ ' = − g1 (k ) cos θ ∑ S (k , k ') 1 − 0 f 0 (k ') g1 (k ) cos θ k' (11) where in the last line of the derivation we have used the principle of detailed balance. We can further simplify the above result, by considering the following coordinate system: z k E θ α θ’ k’ y x Figure 1. Coordinate system used in the rest of the derivation. Within this coordinate system, we have: k = (0, 0, k ) k ' = (k 'sin α cos ϕ , k 'sin α sin ϕ , k 'cos α ) E = (0, E sin θ , E cos θ ) (12) which leads to: E ⋅ k ' = Ek 'cos θ ' = Ek ' ( sin α sin ϕ sin θ + cos α cos θ ) . (13) The integration over φ will make this term vanish, and under these circumstances we can write: cos θ ' = tgθ sin α sin ϕ + cos α → cos α cos θ (14) To summarize: ∂f ∂t coll f (k ) g1 (k ') = − g1 (k ) cos θ ∑ S (k , k ') 1 − 0 cos α f 0 (k ') g1 (k ) k' (15) Note that for the relaxation approximation to be valid, the term in the brackets inside the summation sign should not depend upon the distribution function. • Consider now the case of elastic scattering process. Then k = k' and f 0 (k ) g1 (k ') =1 f 0 (k ') g1 ( k ) . Under these circumstances: ∂f ∂t = − g1 (k ) cos θ ∑ S (k , k ')(1 − cos α ) = − coll k' g1 (k ) cos θ f (k ) =− A τ m (k ) τ m (k ) Hence, when the scattering process is ELASTIC, the characteristic time τf equals the momentum relaxation time. • If the scattering process is ISOTROPIC, then S(k,k’) does not depend upon α . In this case, the second term in the square brackets averages to zero and ∂f ∂t = − g1 (k ) cos θ ∑ S (k , k ') = − k' coll g1 (k ) cos θ f (k ) =− A τ (k ) τ (k ) Thus, in this case, the characteristic time is the scattering time or the average time between collision events. To summarize, under low-field conditions, when the scattering process is either isotropic or elastic, the collision term can be represented as − f A / τ f (k ) , where in general τ f (k ) is the momentum relaxation time that depends only upon the nature of the scattering process. Following these simplifications, the BTE can thus be written as g (k ) cos θ ∂f −eEv cos θ 0 = − 1 τ m (k ) ∂E (16) ∂f g1 (k ) = eEvτ m (k ) 0 ∂E (17) or The distribution function is, thus, equal to ∂f f (k ) = f 0 (k ) + eEv cos θτ m (k ) 0 ∂E ∂f = f 0 (k ) + eEvzτ m (k ) 0 ∂E (18) ∂f 0 1 ∂f 0 = To further investigate the form of this distribution function, we use ∂E hvz ∂k z . This leads to f (k ) = f 0 (k ) + ∂f e Eτ m (k ) 0 h ∂k z (19) The second term of the last expression resembles the linear term in the Taylor series expansion of f(k). Hence, we can write e f (k x , k y , k z ) = f 0 k x , k y , k z + Eτ m (k ) h (20) To summarize, the assumption made at arriving at this last result for displaced Maxwellian form for the distribution function is that the electric field E is small. Hence, the displaced Maxwellian is a good representation of the distribution function under low-field conditions References 1 P. A. Markowich and C. Ringhofer, Semiconductor Equations (Springer-Verlag, Wien New York). 2 C. Ringhofer, “Numerical Methods for the Semiconductor Boltzmann Equation based on Spherical Harmonics Expansions and Entropy Discretizations”, Transp. Theory and Stat. Phys., Vol. 31, pp. 431-452 (2002).