Abstract
A Monte Carlo technique for the solution of the Wigner transport equation has been developed, based on the generation and annihilation of signed particles (Nedjalkov et al. in Phys Rev B 70:115319, 2004). A stochastic algorithm without time discretization error has been recently introduced (Muscato and Wagner in Kinet Relat Models 12(1):59–77, 2019). Its derivation is based on the theory of piecewise deterministic Markov processes. Numerical experiments are performed in the case of a GaAs resonant tunneling diode. Convergence of the time-splitting scheme to the no-splitting algorithm is demonstrated. The no-splitting algorithm is shown to be more efficient in terms of computational effort.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Modelling of electronic transport in nanometer systems requires a theory that describes open, quantum-statistical systems which cannot be provided simply by the Schrödinger equation. Several formulations of quantum transport have been employed, such as those based on the density matrix, non-equilibrium Green’s functions, and the Wigner function.
The Wigner function is a real-valued but not necessarily positive definite quasi-distribution, representing a quantum generalization of the Boltzmann distribution function. The Wigner function formalism is attractive as it provides the expression of quantum dynamics in a phase-space formulation, directly comparable with the classical analogue, more intuitive compared with the more abstract density matrix and Green’s function approaches.
At the same time the Wigner equation can be augmented by a Boltzmann-like collision operator accounting for the process of decoherence.
The first finite-difference-based solver appears in the mid 1980s (see [6] for a review), and more efficient solvers have been developed nowadays [2, 3, 7, 11, 24, 28]. We had to wait until the beginning of 2000, to have particle Monte Carlo (MC) solvers for the Wigner equation [20, 25]. From that period until now, several papers have been published on this subject (see [27] for a review) and, recently, very interesting device simulations have been provided [4, 19, 22].
In the realm of the particle Monte Carlo methods, we have focused on the so-called signed Monte Carlo method [21]. Here, the Wigner potential is treated as a scattering source which determines the electron-potential interaction, and consequently new particles with different signs are stochastically added to the system. Recently this method has also been be understood in terms of the Markov jump process theory [12, 17], producing a class of new stochastic algorithms. Algorithms that belong to this class are a standard time-splitting algorithm and a new no-splitting algorithm that avoids errors due to time-discretization [13, 16, 18]. The goal of this paper is twofold: i) to write, for the sake of clarity and dissemination purposes, a simplified set of rules for the splitting and no-splitting algorithms to help a potential user; ii) to perform, from a mathematical point of view, numerical experiments to compare the standard time-splitting algorithm with the the new no-splitting algorithm, for a simplified GaAS Resonant Tunneling Diode (RTD) structure.
2 The signed particle Monte Carlo method
The Wigner quasi-distribution function \(f_w\) can be seen as a generalization of the semiclassical Boltzmann distribution function. If we separate the potential energy \(V_{\mathrm{tot}}\) felt by the electrons into a slowly-varying and rapidly-varying parts \(V_{\mathrm{tot}} = -e \varphi + V\), the Boltzmann-Wigner transport equation (BWTE) has the form [23]
where \(x \in \mathbb {R}^d\) is the particle position, \(k \in \mathbb {R}^d\) the wave vector (and \(\hbar k\) the momentum), \(m^{*}\) the particle effective mass, \(\varphi\) satisfying the Poisson equation
where e is the absolute value of the electric charge, \(\varepsilon _0\) the absolute dielectric constant, \(\varepsilon _r\) the relative dielectric constant, \(N_D\), \(N_A\) are the assigned doping profiles (due to donors and acceptors), and n the density
The equation includes the Boltzmann scattering operator \(\mathcal{C}(f_w)\) having the scattering rate \(\lambda (k)\) [9] and the quantum evolution term
where \(V_w\) is the Wigner potential
The Wigner potential is a non-local pseudo-differential operator which is responsible of the quantum transport, is real-valued, and anti-symmetric with respect to k. Solving the Wigner equation, from the numerical point of view, is a quite difficult task. The main complication arising in the direct solution based on finite-difference scheme is the discretization of the diffusion term \(k \cdot \nabla _x f_w\) due to the typically fast variation in the phase space. Particle-based MC techniques do not require the discretization of this term, but they need costly computational times.
According to the so-called signed particle Monte Carlo approach developed initially in [21], the quantum evolution term (4) looks like the gain term of a collisional operator in which the loss term is missing. But the Wigner potential (5) is not always positive and cannot be considered a scattering term. For this reason, it can be separated into a positive and negative parts \(V^+_w, V^-_w\) such that
In this way, we can define an integrated scattering probability per unit time as
and rewrite the quantum evolution term as the difference between gain and loss terms, i.e.
Now the term \(w(k^{\prime },k)\) is interpreted as a new scattering rate which produces, from the old particle, a new pair of particles having weight u and \(-u\). In conclusion, an initial parent particle (with sign) evolves on a free-flight trajectory and, according to a generation rate given by the function \(\gamma (x)\), two new signed particles are generated in the same position having weight u and \(-u\) respectively. The momentum of the new particles is generated with probability \(V^+_w(x,k)/\gamma (x)\). However this procedure suffers of an efficiency issue in particle generation, because \(\gamma\) usually is a rapidly oscillating function, and exponential growth of particle numbers. In order to contain the particle number, a cancellation procedure must be introduced.
3 Piecewise deterministic Markov process
A probabilistic model for the quantum evolution term of the BWTE is based on a particle system having time evolution of a piecewise deterministic Markov process [1]. Each particle is characterized by the state space
where the first component is the weight \(u_j \in \{-1 , +1\}\), the second component is the position vector, and the third component is the wave-vector. The time evolution of the particle system (10) is determined by a flow \(F(z_j(t))\) and a jump kernel \(Q(z_j(t))\). Starting at a certain state \(z_j(0)\), the process performs a deterministic motion according to the flow
and the random waiting time \(\tau\) until the next jump satisfies
There is considerable freedom in choosing the jump kernel Q. The following choice is particularly well suited for numerical purposes. We introduce a majorant \(\hat{V}_w\) such that
In the case of a “jump” in which the jth particle generates two new particles, added to the system such that
the jump kernel takes the form [18]
and the random waiting time, for the jth particle, satisfies
where
represents the generation probability. In order to assure finiteness in the integrals with respect to the wave vector, a cutoff parameter c > 0 must be introduced such that Eq. (7) is evaluated in
Then the main result is that appropriate functionals of this process satisfy a weak form of the Wigner equation [17].
4 The splitting algorithm
The time evolution of the system (10) contains both a continuous component (movement in the position space) and a jump component (generation of new particles or phonon scattering). A splitting time step \(\Delta t\) is used in order to separate the transport and the jump processes. The total simulation time is divided into tiny time intervals \(\Delta t\). Generation of new particles is dictated by Eq. (16). The probability that a particle will survive without generation during a ballistic flight of duration \(\Delta t\) is:
If
so, let be \(r \in U[0,1]\)
Since the condition (20)\(_1\) must be fulfilled during all the simulation time, then we have the following limitation for \(\Delta t\)
which must be estimated at the beginning of the simulation from the bias condition, lattice temperature, device dimensions and other input parameters. Finally, this is the generation procedure:
-
s0.
For each jth particle
-
s1.
Let be \(r \in U[0,1]\)
$$\begin{aligned} \text{ if } \quad r < 1 - \hat{\gamma }(x_j , c)\Delta t \end{aligned}$$(23)do not generate anything, next particle GOTO s0.
-
s2.
Otherwise create a random vector \(\tilde{k}\) uniformly in \(\mathbb {B}(c)\) .
-
s3.
Let be \(r \in U[0,1]\)
$$\begin{aligned} \text{ if } \quad r < 1 - \frac{ | V_w(x_j ,\tilde{k}) | }{\hat{V}_w(x_j ,\tilde{k})} \end{aligned}$$(24)do not generate anything, next particle GOTO s0
-
s4.
Otherwise generate a couple of particle
$$\begin{aligned}&(x_j, k_j + \tilde{k}, \tilde{u}), \quad (x_j, k_j - \tilde{k}, -\tilde{u}) \end{aligned}$$(25)$$\begin{aligned}&\tilde{u}= u_j \, \text{ sign } \left[ V_w(x_j ,\tilde{k}) \right] \end{aligned}$$(26)next particle GOTO s0.
For the sake of clarity, a simplified set of rules is given in Table 1 for this splitting algorithm. Phonon scattering is tackled using standard MC techniques [9]. The cancellation is performed each time the total particle number N(t) is greater than the cancellation frequency \(N_{\mathrm{canc}}\). Couples of particles which belong to the same phase-space cell with opposite affinity are canceled (see [17] for the details). The pros of this algorithm are that the flights progress synchronously in small increments, programming becomes much easier and a number of costly checking procedures, necessary for bookkeeping when dynamics calculations are done for the whole flight, may be avoided. The technique is also very effective for studying transient regimes and for vectorization, since the particles are naturally kept synchronous. The cons is that a discretization error depending on the time step is introduced, which could be minimized by decreasing \(\Delta t\) at the expense of a voracious CPU use.
5 The no-splitting algorithm
In this case we shall consider the majorant for the generation process (7) as well as one for the total phonon scattering rate
and we have the total majorant
The random waiting time (16) for all particles is:
If \(\hat{\gamma }\) does not depend on x, we have
where \(r \in U[0,1]\) and the random waiting time \(\tau\) is completely determined. In the no-splitting algorithm the transport and the generation process are not separate. Let us indicate T the next observation time. The particle system evolves on a time interval [0, T]. The system time is denoted by t. The individual time \(t_j\) of a particle indicates the moment of the last update of its position. The Drift/Generation/Scattering procedure (DGS) is the following:
-
1.
Put \(t=0\) and \(t_j=0\) , \(j=1,\ldots ,N\)
-
2.
Generate the random waiting time \(\tau\) according to Eq. (30) Put \(t := t + \tau\).
-
3.
If \(t \ge T\), then move (drift) all particle up to time T solving Newton’s equation of motion. The component \(u_j\) does not change. Erase particles which are out of the boundary and STOP the algorithm.
-
4.
Choose an index \(j \in \{1, \ldots , N\}\) uniformly. Move the jth particle up to t solving Newton’s equation of motion. If the particle is out of the boundary erase it and GOTO 2, otherwise update the individual time of the jth particle into \(t_j := t\). The component \(u_j\) does not change. The position and wave-vector are \(x_j\), \(k_0 =(k_{x0}, k_{y0}, k_{z0})\) respectively.
-
5.
Let be \(r \in U[0,1]\)
$$\begin{aligned} \text{ if } \quad r \le \frac{\hat{\gamma }}{\Gamma } \quad \text{ GOTO } \, \, 6 \quad \text{ ELSE } \quad \text{ GOTO } \, \, 10 \end{aligned}$$(31) -
6.
Create a vector \(\tilde{k}\) according to the probability density
$$\begin{aligned} \frac{1}{2 \hat{\gamma }} \hat{V}_w(x,k) , \quad k \in \mathbb {B}(c) \end{aligned}$$(32) -
7.
Check if the creation event is fictitious. Let be \(r \in U[0,1]\)
$$\begin{aligned} \text{ if } \quad r \le 1 - \frac{|V_w(x_j,\tilde{k})|}{\hat{V}_w(\tilde{k})} \end{aligned}$$(33)GOTO 2
-
8.
Generate the pair of particles
$$\begin{aligned} \left( \tilde{u}(u_j, x_j, \tilde{k}), x_j , k_j + \tilde{k} \right) , \quad \left( -\tilde{u}(u_j, x_j, \tilde{k}), x_j , k_j - \tilde{k} \right) \end{aligned}$$(34)where
$$\begin{aligned} \tilde{u}(u, x, k) = u \, \text{ sign } V_w(x,k) \end{aligned}$$(35)The individual times of these particles equal t. Put \(N := N +2\).
-
9.
If the total number of particles exceeds a certain bound,
$$\begin{aligned} N \ge N_{\mathrm{canc}} \end{aligned}$$(36)then move all particles up to time t according to Newton’s equations. Erase particles which are out of the boundary. Pairs of particles with similar positions and wave-vectors, but with opposite signs, are removed from the system. GOTO 2
-
10.
check/perform phonon scattering, GOTO 2
A set of rules for this algorithm is given in Table 2. The main advantage is that no discretization error due to the time step is introduced.
6 The resonant tunneling diode
Our goal is to perform numerical experiments with the splitting and no-splitting algorithms introduced in the previous section. We have considered the resonant tunneling diode structure introduced in [26], but with a simplified physics setup. The device simulated is a 1-D (in space) RTD of total length \(L = 150 \, \text{ nm } ,\) which consists of two barriers (with \(b=\) 3 nm, \(a=\) 0.3 eV) surrounding a quantum well (with \(b_w = 5\) nm). The barrier structure is centered in a 30 nm lightly doped (\(N_D = 10^{16} \text{ cm}^{-3}\)) spacer region that is connected to 60 nm highly doped (\(N_D^+ = 10^{18} \text{ cm}^{-3}\) drift regions on either side). The quantum well region is symmetric with respect to the mid-point L/2, as shown in Fig. 1, where the distance between each barrier center and the mid-point is
The device is considered to be entirely GaAs (with \(m^{*}= 0.067\)), with scattering mechanisms due to polar optical phonons within a single \(\Gamma\) band [9]. The temperature is 300 K. In our simulation we do not use degenerate statistics in scattering rates and boundary conditions, as done in [26].
Let us suppose that the confining potential V depends only on the longitudinal dimension x. Then we can separate the vector \(\textsf {x} = (x, y, z)\) in its x longitudinal component and the transversal one, as well as for \(\textsf {k} = (k_x, k_y, k_z)\) i.e.
In this case, it is possible to consider a Wigner function of the type
where the longitudinal variables come from the single-dimensional definition, imposed by the fact that the potential depends only on x, while the transversal variables are introduced by the other parts of the Hamiltonian accounting for phonons. Since
the Wigner potential (5), with \(d=3\), can be written
The vectors \(\textsf {x}_\perp ^{\prime } =(y^{\prime },z^{\prime })\) and \(\textsf {k}_\perp =(k_y, k_z)\) belong to the same plane, hence
and finally we have
The total barrier potential is one-dimensional and it is the sum of the single barrier potentials
where
After some manipulations we get
with \(k \rightarrow k_x\). The deltas in the formula (46) mean that the \(k_y, k_z\) of the particle do not change during the creation process. The majorant (13) in this case is
and the creation probability (17) writes
7 Numerical results
We have used an uniform grid in (x, k)-space
with \(N_x = 251\), \(k_{\mathrm{max}} = - k_{\mathrm{min}} = k_{\perp \max }\) = 4 \(\text{ nm}^{-1}\), \(N_k = 33\), \(N_{k_\perp } = 9\). Other parameters used are:
where c is the cutoff in Eq. (18), \(N_{\mathrm{ini}}\) the initial particle number, \(N_{\mathrm{canc}}\) the cancellation frequency, \(N_r\) the repetition number. In the following we shall consider ohmic boundary conditions.
The current at the contact is usually obtained by counting charges which are in the contact mesh. But the instantaneous current at the contact is also due to the induced current due to the electron flow, which can be evaluated using to the Ramo-Shockley theorem [5] and, in the 1D case, yields
where \(N_{\mathrm{elec}} = \rho _{\mathrm{elec}} \times \, \hbox {Area}\) where \(\rho _{\mathrm{elec}}\) is the total particle density and Area the cross section area. Then the current density is easily defined as \(J = I/\hbox {Area}\). Figure 2 shows the current versus the bias voltage, for some splitting time steps \(\Delta t\) and in the no-splitting case. The convergence of the splitting algorithm to the no-splitting one is clearly shown for \(\Delta t \rightarrow 0\). Figure 3 shows the CPU time for both methods: a gain factor of 3 for the no-splitting case can be estimated for the smallest \(\Delta t = 0.03125\) fs. Figures 4 and 5 show the potential energy V and density, respectively, obtained with the splitting algorithm, corresponding to the peak (on-resonance) and valley (off-resonance) of Fig. 2. The results presented have been obtained using a high performance cluster based on Intel Xeon E5-2620 V3 2.4 Ghz and 32 Gb RAM.
Potential versus position, obtained with the no-splitting algorithm, corresponding to the peak (on-resonance) and valley (off-resonance) of Fig. 2
Charge density versus position, obtained with the no-splitting algorithm, corresponding to the peak (on-resonace) and valley (off-resonance) of Fig. 2
8 Conclusions
The Boltzmann-Wigner transport equation has been solved by using for the quantum evolution term the signed-particle Monte Carlo method, where a new pair of particles characterized by a sign is generated randomly and added to the system. This generation mechanism has been recently interpreted in terms of the Markov jump process, producing a class of new stochastic algorithms [17]. One of these algorithms has been implemented without time discretization error, and it has been applied to a resonant tunneling diode. The results are compared with the corresponding splitting algorithm, showing an excellent agreement as well as a low-cost computational effort. Future research will develop this MC methodology for the simulation of realistic devices, such as silicon nanowires according to the guidelines in [14, 15] and new emerging materials such as graphene [8, 10].
References
Davis, M.H.A.: Markov Models and Optimization. Chapman & Hall, London (1993)
de Put, M.L.V., Soree, B., Magnus, W.: Efficient solution of the Wigner-Liouville equation using a spectral decomposition of the force field. J. Comput. Phys. 350, 314–325 (2017)
Dorda, A., Schürrer, F.: A WENO-solver combined with adaptive momentum discretization for the Wigner transport equation and its application to resonant tunneling diodes. J. Comput. Electron. 284, 95–116 (2015)
Ellinghaus, P., Weinbub, J., Nedjalkov, M., Selberherr, S.: Analysis of lense-governed Wigner signed particle quantum dynamics. Phys. Status Solidi RRL 11(7), 1700102 (2017)
Houlet, P., Awano, Y., Yokoyama, N., Hamaguchi, V.: Accurate Monte Carlo modeling of terminal currents in short semiconductor devices by using a generalized Ramo-Shockley theorem. In: Proceedings of IWCE, pp. 218–221 (1998)
Kosina, H.: Wigner function approach to nano device simulation. Int. J. Comput. Sci. Eng. 2(3–4), 100–118 (2006)
Lee, J.H., Shin, M.: Quantum transport simulation of nanowire resonant tunneling diodes based on a Wigner function model with spatially dependent effective masses. IEEE Trans. Nanotechnol. 16(6), 1028–1036 (2017)
Luca, L., Mascali, G., Nastasi, G., Romano, V.: Comparing kinetic and MEP model of charge transport in graphene. J. Comput. Theor. Transp. 49(7), 368–388 (2020)
Lundstrom, M.: Fundamentals of Carrier Transport. Cambridge Univ. press, Cambridge (2000)
Mascali, G., Romano, V.: A hierarchy of macroscopic models for phonon transport in graphene. Physica A 548, 124489 (2020)
Morandi, O., Schürrer, F.: Wigner model for quantum transport in graphene. J. Phys. A Math. Theor. 26, 265301 (2011)
Muscato, O.: A benchmark study of the signed-particle Monte Carlo algorithm for the Wigner equation. Commun. Appl. Ind. Math. 8(1), 237–250 (2017)
Muscato, O.: Wigner Monte Carlo simulation of a double potential barrier. In: Nicosia, G., Romano, V. (eds.) Scientific Computing in Electrical Engineering. SCEE 2018. Mathematics in Industry, pp. 161–170. Springer, Berlin (2020)
Muscato, O., Castiglione, T.: A hydrodynamic model for silicon nanowires based on the maximum entropy principle. Entropy 18(10), 368 (2016)
Muscato, O., Castiglione, T., Di Stefano, V., Coco, A.: Low-field electron mobility evaluation in silicon nanowire transistors using an extended hydrodynamic model. J. Math. Ind. 8, 14 (2018)
Muscato, O., Di Stefano, V.: Wigner Monte Carlo simulation without discretization error of the tunneling rectangular barrier. Commun. Appl. Ind. Math. 10(1), 20–30 (2019)
Muscato, O., Wagner, W.: A class of stochastic algorithms for the Wigner equation. SIAM J. Sci. Comput. 38(3), A1438–A1507 (2016)
Muscato, O., Wagner, W.: A stochastic algorithm without time discretization error for the Wigner equation. Kinet. Relat. Models 12(1), 59–77 (2019)
Nedjalkov, M., Ellinghaus, P., Weinbub, J., Sadi, T., Asenov, A., Dimov, I., Selberherr, S.: Stochastic analysis of surface roughness models in quantum wires. Comput. Phys. Commun. 228, 30–37 (2018)
Nedjalkov, M., Kosik, R., Kosina, H., Selberherr, S.: A Wigner equation for nanometer and femtosecond transport regime. In: Proceedings of the IEEE Conference on Nanotechnology, pp. 277–281. IEEE (2001)
Nedjalkov, M., Kosina, H., Selberherr, S., Ringhofer, C., Ferry, D.K.: Unified particle approach to Wigner-Boltzmann transport in small semiconductor devices. Phys. Rev. B 70, 115319 (2004)
Nedjalkov, M., Weinbub, J., Ballicchia, M., Selberherr, S., Dimov, I., Ferry, D.: Wigner equation for general electromagnetic fields: The Weyl-Stratonovich transform. Phys. Rev. B 99, 014423 (2019)
Querlioz, D., Dollfus, P.: The Wigner Monte Carlo Method for Nanoelectronic Devices. Wiley, Hoboken (2010)
Shao, S., Lu, T., Cai, W.: Adaptive conservative cell average spectral element methods for transient Wigner equation in quantum transport. Commun. Comput. Phys. 9(3), 711–739 (2011)
Shifren, L., Ferry, D.: Particle Monte Carlo simulation of Wigner function tunneling. Phys. Lett. A 285, 217–221 (2001)
Shifren, L., Ringhofer, C., Ferry, D.: A Wigner function-based quantum ensemble Monte Carlo study of a resonant tunneling diode. IEEE Trans. Electron Devices 50(3), 769–773 (2003)
Weinbub, J., Ferry, D.: Recent advances in Wigner function approaches. Appl. Phys. Rev. 5, 041104 (2018)
Xiong, Y., Chen, Z., Shao, S.: An advective-spectral-mixed method for time-dependent many-body Wigner simulations. SIAM J. Sci. Comput. 38(4), B491–B520 (2016)
Acknowledgements
The author acknowledges the support of the project “Developing a Computational Framework for Quantum Information and Communication Technologies”, Piano di incentivi per la ricerca di Ateneo 2020/2022 - linea 2, Università degli Studi di Catania.
Funding
Open access funding provided by Università degli Studi di Catania within the CRUI-CARE Agreement.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares that he has no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Muscato, O. Wigner ensemble Monte Carlo simulation without splitting error of a GaAs resonant tunneling diode. J Comput Electron 20, 2062–2069 (2021). https://doi.org/10.1007/s10825-021-01734-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10825-021-01734-3