Loading [MathJax]/jax/output/SVG/jax.js
Research article Special Issues

Acceleration of solving drift-diffusion equations enabled by estimation of initial value at nonequilibrium

  • Received: 27 January 2024 Revised: 28 March 2024 Accepted: 02 April 2024 Published: 15 April 2024
  • In this study, a novel method enabled by estimation of initial value guess at nonequilibrium was proposed to accelerate drift-diffusion equations in semiconductor device simulation. The initial value guess was obtained by solving analytical model about electrical potential with the decoupling algorithm. By obtaining the initial value directly at the target bias voltage, the proposed method eliminated time-consuming bias ramping process in the classical method starting from the equilibrium state, thereby accelerating the whole process. The method has been applied to a junction barrier Schottky (JBS) diode for validation. Numerical results showed that the proposed method achieves convergence within 10 iterations at several reverse bias voltages, achieving significant reduction of iteration number compared to the classical method using the bias ramping process. It demonstrated that the proposed method holds high feasibility to facilitate the semiconductor device property prediction in relatively regular device structure in the case of low current. With further improvements, this method can also be applied to more complex devices.

    Citation: Chunlin Du, Yu Zhang, Haolan Qu, Haowen Guo, Xinbo Zou. Acceleration of solving drift-diffusion equations enabled by estimation of initial value at nonequilibrium[J]. Networks and Heterogeneous Media, 2024, 19(1): 456-474. doi: 10.3934/nhm.2024020

    Related Papers:

    [1] Shouyun Cheng, Lin Wei, Xianhui Zhao, Yinbin Huang, Douglas Raynie, Changling Qiu, John Kiratu, Yong Yu . Directly catalytic upgrading bio-oil vapor produced by prairie cordgrass pyrolysis over Ni/HZSM-5 using a two stage reactor. AIMS Energy, 2015, 3(2): 227-240. doi: 10.3934/energy.2015.2.227
    [2] Wasinton Simanjuntak, Kamisah Delilawati Pandiangan, Tika Dwi Febriyanti, Aryani Putri Islami, Sutopo Hadi, Ilim Ilim . Catalytic upgrading of palm oil derived bio-crude oil for bio-hydrocarbon enrichment using protonated zeolite-Y as catalyst. AIMS Energy, 2024, 12(3): 600-616. doi: 10.3934/energy.2024028
    [3] Vincent E. Efeovbokhan, Augustine O. Ayeni, Osuvwe P. Eduvie, James A. Omoleye, Oladotun P. Bolade, Ajibola T. Ogunbiyi, Victoria N. Anyakora . Classification and characterization of bio-oil obtained from catalytic and non-catalytic pyrolysis of desludging sewage sample. AIMS Energy, 2020, 8(6): 1088-1107. doi: 10.3934/energy.2020.6.1088
    [4] Xianhui Zhao, Lin Wei, James Julson . First stage of bio-jet fuel production: non-food sunflower oil extraction using cold press method. AIMS Energy, 2014, 2(2): 193-209. doi: 10.3934/energy.2014.2.193
    [5] Ahmad Indra Siswantara, Illa Rizianiza, Tanwir Ahmad Farhan, M. Hilman Gumelar Syafei, Dyas Prawara Mahdi, Candra Damis Widiawaty, Adi Syuriadi . Analyzing temperature distribution in pyrolysis systems using an atomic model. AIMS Energy, 2023, 11(6): 1012-1030. doi: 10.3934/energy.2023048
    [6] Lihao Chen, Hu Wu, Kunio Yoshikawa . Research on upgrading of pyrolysis oil from Japanese cedar by blending with biodiesel. AIMS Energy, 2015, 3(4): 869-883. doi: 10.3934/energy.2015.4.869
    [7] Lifita N. Tande, Valerie Dupont . Autothermal reforming of palm empty fruit bunch bio-oil: thermodynamic modelling. AIMS Energy, 2016, 4(1): 68-92. doi: 10.3934/energy.2016.1.68
    [8] Sunbong Lee, Shaku Tei, Kunio Yoshikawa . Properties of chicken manure pyrolysis bio-oil blended with diesel and its combustion characteristics in RCEM, Rapid Compression and Expansion Machine. AIMS Energy, 2014, 2(3): 210-218. doi: 10.3934/energy.2014.3.210
    [9] Hasanudin Hasanudin, Wan Ryan Asri, Utari Permatahati, Widia Purwaningrum, Fitri Hadiah, Roni Maryana, Muhammad Al Muttaqii, Muhammad Hendri . Conversion of crude palm oil to biofuels via catalytic hydrocracking over NiN-supported natural bentonite. AIMS Energy, 2023, 11(2): 197-212. doi: 10.3934/energy.2023011
    [10] Boua Sidoine KADJO, Mohamed Koïta SAKO, Kouadio Alphonse DIANGO, Amélie DANLOS, Christelle PERILHON . Characterization and optimization of the heat treatment of cashew nutshells to produce a biofuel with a high-energy value. AIMS Energy, 2024, 12(2): 387-407. doi: 10.3934/energy.2024018
  • In this study, a novel method enabled by estimation of initial value guess at nonequilibrium was proposed to accelerate drift-diffusion equations in semiconductor device simulation. The initial value guess was obtained by solving analytical model about electrical potential with the decoupling algorithm. By obtaining the initial value directly at the target bias voltage, the proposed method eliminated time-consuming bias ramping process in the classical method starting from the equilibrium state, thereby accelerating the whole process. The method has been applied to a junction barrier Schottky (JBS) diode for validation. Numerical results showed that the proposed method achieves convergence within 10 iterations at several reverse bias voltages, achieving significant reduction of iteration number compared to the classical method using the bias ramping process. It demonstrated that the proposed method holds high feasibility to facilitate the semiconductor device property prediction in relatively regular device structure in the case of low current. With further improvements, this method can also be applied to more complex devices.



    With the rapid technological advancement today, the access to realistic 3D human shapes is of great importance in both computer vision and graphics, and has various applications in different industries including virtual game design, film making, bioinformatics[41], healthcare[38], and especially, those related to garment design. Some applications involve fitting predictions, virtual try-on simulations[13,16,23,34] or size recommendations[5,6,7], that help to recommend relevant clothing which would fit specific occasions or fashion trends for online customers. Such applications require a critical ingredient on digital transformation from humans bodies to digital 3D shapes, such that the shapes maintain some of the main features from human bodies.

    The traditional approaches to access reliable digital information of a human body are through laser range scanners[1], stereo reconstruction[18,22,33] or structured light methods for 3D sensing[11,20,24]. However, considering the cost of data storage, network transmission and expensive scanning equipment, it is rather unpractical to scan individuals for each application. Hence many studies have been done to generate 3D human shapes based on partial input information. These prior systems can be mainly classified into three types: marker-based systems, silhouette-based systems and measurement-based systems. Marker-based system estimates dynamic 3D human body shapes by capturing a sparse set of marker positions. These techneqiues proceed by using a single static scan or multiple scans and a marker motion capture sequence of the person[4]. For the static case, silhouette-based system estimates human body shapes based on a set of input images by fitting the silhouette in each view [3,8,9,15]. Some apporaches also combine with machine learning that build a correlation between a training dataset of 3D body shapes and a set of 2D images, and then predict a shape based on the correlation[14].

    Although marker-based systems and silhouette-based systems could yield satisfactory reconstructions on 3D human body shapes under tight dresses or naked human shapes, most of the schemes are so computationally expensive and the results are easily affected if heavy or loose clothes are worn. To overcome these difficulties, a great deal of efforts have been devoted to the investigation of simple and fast measurement-based systems [17,32,35]. Typically, one considers the landmarks or circumferences from the human structures at specific locations as characteristic values. Since such characteristic values are linear or curvilinear, they are relatively invariant to articulation changes than those silhouettes measurements. If the set of characteristic values is well selected, one can achieve meaningful estimation for both global and local body shapes. Kart et al. (2011) built a system which only requires to input some personal information, such as weight, height and age as well as a 2D photograph. The decision algorithm then determines the human shape according to the measurements and the body mass index (BMI)[12]. Seo et al. [35] presented a human body generation apporach by taking the anthropometric measurements, e.g., stature, crotch length, arm length, neck girth, chest/bust girth, underbust girth, waist girth and hip girth as input. They derived the relationship between the input characteristic values and the preprocessing database of 3D scanned data of human body models by using radial basis interpolation. At run-time, the system generates new human body shapes from the user input characteristic values by fitting the template model onto each scanned data.

    In this paper, we develop a completely novel methodology for the geometric body generation, which fulfils the following two basic requirements: (ⅰ) the geometric body generation is automatically determined by the input characteristic sets; (ⅱ) the predicted geometric shape fits for all input characteristic values and moreover it can well approximate the exact geometric body possessing the aforesaid characteristic values. The proposed method follows a machine-learning flavour that generates the inferred geometric body with the customized characteristic parameters from a training dataset. The training dataset consists of some preprocessed body shapes associated with appropriately sampled characteristic parameters. One of the critical ingredients and novelties of our method is the borrowing of inverse scattering techniques in the theory of wave propagation to the body generation. This is done by establishing a delicate one-to-one correspondence between a geometric body and the far-field pattern of a source scattering problem governed by the Helmholtz system. It in turn enables us to establish a one-to-one correspondence between the geometric body space and the function space defined by the far-field patterns. Hence, the far-field patterns can act as the shape generators. The shape generation with prescribed characteristic parameters is achieved by first manipulating the shape generators in the function space and then reconstructing the corresponding geometric body from the obtained shape generator by a stable multiple-frequency Fourier method. The proposed method is in sharp difference from the existing methodologies in the literature, which usually treat the human body as a suitable Riemannian manifold and the generation is based on non-Euclidean approximation and interpolation. In fact, in all of the literature mentioned earlier on manifold learning of body generation, one typically uses Principal Component Analysis (PCA) or Principal Geodesic Analysis (PGA). PCA and PGA are used for optimal reduction of the data and thus efficient deformation by computing statistics on Euclidean manifolds or non-Euclidean manifolds can be achieved; see [35,10] and the references therein for more relevant discussion. In our new approach, the shape generator enables us to train the learning dataset via the algebraic operations in the shape space directly without dealing with the deformation of the manifold meshes between geometric shapes.

    The rest of the paper is organized as follows. In section 2, we provide rigorous mathematical formulations of characteristic values and shape space. Section 3 introduces the notion of shape generator via the inverse source scattering associated with the Helmholtz system. In section 4, we present the mathematical setup of the geometric body generation from a machine-learning perspective. Section 5 is devoted to the development of the new method for the shape generation. In section 6, we present several two- and three-dimensional numerical examples to show the effectiveness and efficiency of our method. The paper is concluded in section 7 with some relevant discussion.

    In this section, we present some preliminary knowledge on the shape manifold theory that shall be needed in our subsequent study of body generation. Generally speaking, a geometric shape or a geometric body is a topological n -manifold, nN , equipped with certain shape descriptors, which give the full information to describe the geometric shape. We call such shape descriptors as characteristic values. We have the following formal definition.

    Definition 2.1. Let D be a topological n -manifold with nN . Let ΛD:={λ(j)}jC be a set of parameters associated with D that are invariant with respect to isometric deformations and are independent to the parametrizations of D . Here, the cardinality C might be finite or infinite. ΛD is said to be a characteristic set of D if it uniquely determines D . D and its characteristic set ΛD , written as the (D,ΛD) is referred to as a geometric shape or a geometric body.

    Clearly, Definition 2.1 includes much general geometric objects. However, for the present study, we are mainly concerned with the case that D can be embedded into Rd , d=2,3 , as a bounded domain. That means, we exclude some interesting cases such as D is a Riemannian surface with boundary in R3 . Nevertheless, our study is general enough to include the human body as a specific case.

    In Definition 2.1, the set of characteristic values is typically a set of measurements which gives a systematic characterization of the size, shape and composition of a geometric object for us to determine the shape of the object. For example, when considering a rectangular object, once can introduce a set of characteristic values containing its height, width and length, which provide all details to determine a unique rectangular shape. Expanding the same idea to human body shapes, one could also use characteristic sets to represent them. There are many different ways to represent a human shape. We would try to group those characteristic values into four main catagories, including Eucidean distance, geodesic distance, circumference and ratio. The Eucidean or geodisic distance is linear or curvilinear distance between two points on the human model, such as stature, crotch length, arm length, shoulder breadth etc. The circumference can be computed by the horizontal girth of the body, such as neck girth, chest/bust girth, under-bust girth, waist girth, hip girth, etc. The ratio can be information of weight, Body Mass Index, muscle and fat rate. In spite of the above characteristic values, one can also consider some pure measurements such as age or gender as characteristic values.

    The full set of characteristic values gives the complete information of a geometric shape without lossing any information. It is easy to imagine that the cardinality of a set of characteristic values depends on the complexity of a shape. Hence, the number of characteristic values required can be considered as the dimensionality of the geometric shape. For those complicated objects, like human shapes, it may require infinite set of characteristic values for accurate formulations. Due to practical reasons, one can consider a reasonable truncation of an infinite characteristic set into a finite one for a complicated geometric shape. In doing so, we can consider our study in the following product space

    S:=D×V, (2.1)

    where D is composed of all the bounded domains in Rd and V is an M -dimensional vector space containing the characteristic values. In fact, in the present study, the characteristic values are usually real numbers and one can take V=RM with MN . S is referred to as the geometric shape space. According to the (approximate) one-to-one correspondence between a geometric shape and its characteristic values in Definition 2.1, we readily see that all the shape information can be obtained by a single point of this M -dimensional vector space V . By adjusting the characteristic values, we can obtain new geometric shapes and this is a key ingredient in our human body interpolation.

    In the previous section, we introduce the important notion of shape space for our study. We proceed to introduce another critical ingredient, shape generator, for our subsequent study of the geometric body generation. In fact, the generation of a new geometric shape shall be based on algebraic interpolation of exemplar models from the shape space. If the algebraic operations are to be conducted directly in the shape space, dealing with geometric deformations of manifolds, one would certainly encounter very complicated and tedious calculations and manipulations because of the lack of global parametrizations for the non-Eucidean shapes involved. The shape generator can overcome this challenge by bridging the geometric shape space and the function space. To that end, we next introduce the inverse scattering problem in finding an active source from its generated far-field pattern.

    Let f:RdC be a function having a compact support, f=χDφ , where DRd is a bounded domain and φL(Rd) . The set D is the external shape of f while φ describes the intensity of the source at various points in D . We assume that φ and D do not depend on the wavenumber kR+ . In other words we are considering monochomatic scattering. The source f produces a scattered wave uH2loc(Rd) given by the unique solution to

    (Δ+k2)u=f,limrrd12(rik)u=0, (3.1)

    where r=|x| for xRd . The limit in (3.1) is known as the Sommerfeld radiation condition which characterizes the outgoing nature of the radiating wave. By the limiting absorption principle (cf. [21]), the solution to (3.1) can be computed as follows,

    u=(Δ+k2)1f=limε+0(Δ+(kiε)2)1f=limε+0Rdeixξˆf(ξ)|ξ|2(kiε)2 dξ, (3.2)

    where

    ˆf(ξ):=Ff(ξ)=(2π)dRnf(x)eiξx dx (3.3)

    signifies the Fourier transform of f . Inverting the Fourier transform in (3.2), one has the following integral representation,

    u=(Δ+k2)1f:=i4(k2π)d22Rd|xy|2d2H(1)d22(k|xy|)f(y) dy, (3.4)

    where H(1)(d2)/2 is the first-kind Hankel function of order (d2)/2 . Stationary phase applied to (3.4) yields that

    u(x)=eik|x||x|(d1)/2Cd,kRdeikˆxyf(y) dy+O(|x|d2),|x|, (3.5)

    where ˆx:=x/|x|Sd1 , xRd{0} , and

    Cd,k=i8π(k2π)d22e(d1)π4i.

    The far-field pattern of u is given by

    u(ˆx,k;f):=Cd,kRdeikˆxyf(y) dy=(2π)dCd,kFf(kˆx)L2(Sd1). (3.6)

    It is obvious that u is (real) analytic in both ˆx and k . Hence, if u(ˆx,k) is known on any open portion of Sd1×R+ , then it is known on the whole set by analytic continuation.

    The inverse source scattering problem is concerned with the recovery of f=φχD by knowledge of u(ˆx,k;f) for (ˆx,k)Σ , where Σ is an open subset of Sd1×R+ . According to our discussion above, without loss of generality, we always assume that Σ=Sd1×R+ in what follows. The inverse source problem arises in a variety of important applications including detection of hazardous chemicals, medical imaging, photoacoustic and thermoacoustic tomography, brain imaging, artificial intelligence in gesture computing and others. We refer to the recent articles [25,26,39,40] by two of the authors of this article for some recent developments on the inverse source problem.

    Next, let us consider a specific case by assuming a source supported in a domain D with a constant density 1 . Then clearly by (3.6), there is a one-to-one correspondence between D and u(Σ;D):={u(ˆx,k;1χD)}(ˆx,k)ΣL2(Sd1×R+) in the sense that for two domains D1 and D2 ,

    u(Σ;D1)=u(Σ;D2)if and only ifD1=D2. (3.7)

    Based on (3.7), we next introduce

    Definition 3.1. For a geometric shape (D,ΛD)S ,

    u(Σ;D):={u(ˆx,k;1χD)}(ˆx,k)ΣL2(Sd1×R+)

    defined via the Helmholtz system (3.1) is called a shape generator for D .

    Remark 3.1. By Definition 3.1, a geometric body D can be completely determined by a shape generator u(Σ) . Since u(Σ) is from a function space, this paves the way for the new body generation through function interpolations.

    Remark 3.2. By (3.6), we know the far-field pattern is actually the Fourier transform of the source density up a dimensional constant. However, introducing the shape generator via the inverse scattering approach shall provide more physical insights in our study, and moreover it enables us to borrow ideas from the inverse scattering literature of recovering the geometric shape D from the associated far-field pattern. This also paves the way of extending the idea by using other inverse scattering models that have such one-to-one correspondence between geometric shapes and far-field patterns; see more relevant discussion in section 7.

    In this section, we introduce the mathematical formulation of the geometric body generation for our study from a machine learning perspective. For a geometric shape (D,ΛD) with the associated shape generator u(Σ;D) , the pair of the high dimensional variables, written as {(ΛD,u(Σ;D))} , is referred to as an input-output pair. Let {(ΛDi,u(Σ;Di))}iN with N={0,1,,Npair} be a set of input-output pairs associated with the characteristic sets {ΛDi}iN . Here the input characteristic sets ΛDi are introduced as

    ΛDi:={λ(j)i}jC,iN, (4.1)

    with C={1,,M} . In (4.1), the notation λ(j)i represents the characteristic value of the i -th geometric shape Di in the j -th direction of its characteristic set. Here, the cardinality N is finite. The training dataset of the geometric body generation is introduced to be

    Z:={(ΛDi,u(Σ;Di))}iN (4.2)

    with

    Λ=(λ(1),,λ(M))RM and u(Σ)L2(Sd1×R+). (4.3)

    The training dataset consists of certain pre-sampled geometric shapes with statistically well selected characteristic values. The corresponding shape generator of a specific body in the training dataset can also be pre-calculated and stored. The main goal of our study is to first infer a learning model from the training dataset, TZ:RML2(Sd1×R+) that fulfils the following requirements:

    1. It fits the training data well in the sense that

    TZ(ΛDi):=ˆu(Σ;Di)u(Σ;Di),DiZ. (4.4)

    2. It can be used to infer the shape generator for a given new shape with prescribed characteristic values, namely,

    ˆu(Σ;Dnew):=TZ(ΛDnew), (4.5)

    and with a statically well selected training dataset, it is justifiable to expect that

    ˆu(Σ;Dnew)u(Σ;Dnew), (4.6)

    where u(Σ;Dnew) is the shape generator for Dnew .

    If a learning model can be achieved that fulfils the two requirements as described above, then the body generation can be proceeded as follows. For a given new set of characteristic values, one first generates the learned shape generator as in (4.5). By a certain inverse scattering approach, one can then reconstruct the (approximate) shape Dnew from the corresponding shape generator ˆu(Σ;Dnew) . In the next section, we shall develop the two critical ingredients in the body generation procedure described above, namely, the learning model and the reconstruction method. To be more definite and specific, we first introduce the following definition from a machine learning perspective.

    Definition 4.1. (Body learning model) Given a training dataset

    Z:={(ΛDi,u(Σ;Di))}iN. (4.7)

    Let H be a compact subset of L2(Sd1×R+) . TZH (with specified coefficients C ) is said to be the best fit learning model associated with the training dataset Z if it is the minimizer of the following optimization problem,

    minTZH1Npair+1iNTZ(ΛDi)u(Σ;Di)2H. (4.8)

    According to Definition 4.1, the choice of the learning subspace H plays a critical role. However, we note that the shape generator is actually (real) analytic in all of its arguments. Hence, instead of solving the computationally costly optimization problem (4.8), we can make use of the functional interpolation to produce a well-rounded shape learning model. This is one of the main advantages of introducing the shape generator through the inverse scattering model. In the next section, for a given training dataset as in (4.7), we shall derive a learning model using the cubic B-spline interpolation through the use of the high-dimensional data-points (4.4). For the reconstruction of the approximate body shape from the shape generator obtained through the learning model, we shall make use of a multiple-frequency Fourier method, and it can also produce an efficient and stable recovery. Throughout, we assume that the characteristic values in the training dataset is statistically well selected and it is not the focus of the present article.

    In this section, we develop the details of our scheme for the geometric body generation following the general discussion made in the previous section. We first derive the learning model through the functional interpolation of the high-dimensional data in the training dataset. To that end, we present some preliminary knowledge on the cubic B-spline, and we also refer to [2,19,36,37] for more relevant discussion on the cubic B-spline.

    Consider the training dataset (4.2). Let the M sets of unique grids in the directions of {λ(j)}jC

    Δj={λ(j)0,,λ(j)kj}jC,kjN, (5.1)

    define on the intervals [aj,bj] as M sets of points λ(j)gj[aj,bj]R , where gj{0,,kj} and aj=λ(j)0<λ(j)1<<λ(j)kj=bj,j=1,,M . Here, kj is the greatest number of distinct characteristic values in the j -th direction of the characteristic set. We remark that if the characteristic values of the training dataset are all collected in distinct values, then kj is actually the last index of the training dataset, Npair . However, the training dataset might be collected in such a way that some body shapes may possess the same characteristic value in the j -th direction, and hence kj is usually smaller than Npair .

    With the above notation, the training dataset stored as the array in (4.2) can be represented as elements on the grid mesh corresponding to the characteristic numbers λ(j) as described above. The interpolation data are the corresponding shape generators and are written as

    u(Σ;Di):=Ug1,g2,,gML2(Sd1×R+),iN, (5.2)

    where gj=0,,kj,j=1,,M. In the subsequent study, we shall stick to the same notation g1,g2,,gM to represent the linear indexing. The following example demonstrates a real application for the human body generation.

    Example 5.1. The training dataset consists of 20 bodies with two characteristic values as consideration, say, height and relative weight. Here, the height and relative weight are the two directions of the grids, λ(height) and λ(weight) . Suppose the heights of the sampled bodies are given by 1.5 m, 1.6 m, 1.7 m, 1.8 m, 1.9 m and the relative weights of the sampled bodies are given by 60%, 80%, 100%, 120%. Then the first grid Δheight={λ(height)0,λ(height)1,λ(height)2,λ(height)3,λ(height)4}={1.5,1.6,1.7,1.8,1.9} and the second grid Δweight={λ(weight)0,λ(weight)1,λ(weight)2,λ(weight)3}={0.6,0.8,1,1.2} . The interpolation data are actually stored as listed in Table 1; e.g., U0,0=u(Σ;D1),U0,3=u(Σ;D4),U1,3=u(Σ;D8) .

    Table 1.  Training dataset with the given characteristic grids (λ(height),λ(weight)).
    2nd grid
    λ(weight)0 λ(weight)1 λ(weight)2 λ(weight)3
    λ(height)0 (ΛD1,u(Σ,D1))
    =(1.5,0.6,U0,0)
    (ΛD2,u(Σ,D2))
    =(1.5,0.8,U0,1)
    (ΛD3,u(Σ,D3))
    =(1.5,1,U0,2)
    (ΛD4,u(Σ,D4))
    =(1.5,1.2,U0,3)
    1st gridλ(height)1 (ΛD5,u(Σ,D5))
    =(1.6,0.6,U1,0)
    (ΛD6,u(Σ,D6))
    =(1.6,0.8,U1,1)
    (ΛD7,u(Σ,D7))
    =(1.6,1,U1,2)
    (ΛD8,u(Σ,D8))
    =(1.6,1.2,U1,3)
    λ(height)2 (ΛD9,u(Σ,D9))
    =(1.7,0.6,U2,0)
    (ΛD10,u(Σ,D10))
    =(1.7,0.8,U2,1)
    (ΛD11,u(Σ,D11))
    =(1.7,1,U2,2)
    (ΛD12,u(Σ,D12))
    =(1.7,1.2,U2,3)
    λ(height)3 (ΛD13,u(Σ,D13))
    =(1.8,0.6,U3,0)
    (ΛD14,u(Σ,D14))
    =(1.8,0.8,U3,1)
    (ΛD15,u(Σ,D15))
    =(1.8,1,U3,2)
    (ΛD16,u(Σ,D16))
    =(1.8,1.2,U3,3)
    λ(height)4 (ΛD17,u(Σ,D17))
    =(1.9,0.6,U4,0)
    (ΛD18,u(Σ,D18))
    =(1.9,0.8,U4,1)
    (ΛD19,u(Σ,D19))
    =(1.9,1,U4,2)
    (ΛD20,u(Σ,D20))
    =(1.9,1.2,U4,3)

     | Show Table
    DownLoad: CSV

    Let S3(Δj),j=1,,M be a function subspace of C2([aj,bj]) consisting of one dimensional, complex-valued functions in the direction of λ(j),j=1,,M on the bounded interval [aj,bj] . The function in S3(Δj),j=1,,M is piecewise polynomial of degree 3 on every subinterval [λ(l)gj1,λ(j)gj] , where gj=1,,kj,j=1,,M . Then we introduce a function subspace of multidimensional and complex-valued C2([aj,bj]) functions as

    S3(Δ1,,ΔM), (5.3)

    on each rectangular grid

    Ig1,g2,,gM:=jC[λ(j)gj,λ(j)gj+1] (5.4)

    for all 0gjkj1,j=1,,M that are piecewise polynomials of degree 3 on every interval. For easy reference we provide the definition of B-splines.

    Definition 5.1. The sets of kj+k,j=1,,M , B-spline basis functions {Bl,k(λ(j))}kj+kl=1 of degree k of the function space Sk(Δj) are defined based on concurrent boundary knots vectors with Cox-deBoor recurrence[19],

    Bl,k(λ(j))=λ(j)λ(j)l1kλ(j)l1λ(j)l1kBl1,k1(λ(j))+λ(j)lλ(j)λ(j)lλ(j)lkBl,k1(λ(j)) (5.5)

    with

    Bl,0(λ(j))={1if λ(j)lλ(j)<λ(j)l+10else, (5.6)

    for l=1,,kj+k , where k=3 is for the cubic B-spline and λ(j)l are elements of the knot vectors, satisfying the relation λ(j)l<λ(j)l+1 .

    All methods are in the following using splines with a k+1 regular knot vector, and the interior knots are the grid points. Based on Definition 5.1, we next introduce a general learning model for the geometric shape generation through the multidimensional cubic B-spline interpolation.

    Body learning model Ⅰ. Given the training dataset Z:={(ΛDi,u(Σ;Di))}iN , the learning model TZS3(Δ1,,ΔM) at Λ=(λ(1),,λ(M)) for the geometric body generation associated with the sets of the grids {Δj}jC is defined as follows

    TZ(Λ)=k1+3g1=1kM+3gM=1cg1,g2,,gMjCBjgj,3(λ(j)),λ(j)[aj,bj], (5.7)

    which satisfies the following conditions by (4.4)

    TZ(ΛDi)=k1+3g1=1kM+3gM=1cg1,g2,,gMjCBjgj,3(λ(j)i)=u(Σ;Di), (5.8)

    where cg1,g2,,gM with gj=0,,kj,j=1,,M are the coefficients to be determined from the training dataset Z , Bjgj,3(λ(j)),jC are the B-spline basis functions of degree 3 defined in (5.5), and kj,j=1,,M is the number of different characteristic values in each direction.

    Remark 5.1. In learning model Ⅰ, (5.7) presents a general form of the learning model TZ for the geometric body generation associated with the non-uniform grids {Δj}jC . The learning model eventually generates a B-spline interpolation with the associated spacing for each segment. If the training dataset consists of equidistant grids, we can derive a faster and easier learning model and this shall be provided in the next subsection.

    In this subsection, we derive a learning model for a special case with the training dataset consisting of equidistant grids. By (5.1), for the M set of equidistant grids Δj={λ(j)0,,λ(j)kj}jC with additional conditions

    λ(j)gj=aj+gjhj,hj=bjajkj,,gj=1,,kj, (5.9)

    the B-spline basis function βk(t) of degree k is a symmetrical, bell-shaped function constructed from k+1 times self-convolution of the β0(t) basis function of degree zero which is a centered rectangle around origin [37]

    β0(t)={1,12<t<1212,|x|=120,otherwise, (5.10)
    βk(t)=β0(t)β0(t)(k+1) times. (5.11)

    The centered symmetric B-spline of degree k has an explicit expression [36]

    βk(t)=1k!k+1j=0Ck+1j(1)j(t+k+12j)k+, (5.12)

    where the function x+ is defined as follows

    x+={x,forx>0,0,otherwise. (5.13)

    In this paper, we are particular intereted in the cubic B-spline. By (5.12), the closed-form representation of the cubic B-spline basis function can be also expressed as

    β3(t)=16{(2|t|)31<|t|2,46|t|2+3|t|3,|t|1,0,elsewhere, (5.14)

    which is used for preforming the interpolation. Then we choose the interpolation kernels to be

    Lgj(λ(j))=β3(λ(j)),kj=1,,Npair+3, (5.15)

    as the basis of S3(Δj) in the λ(j) -direction such that L={L1,L2,,Lkj+3} is a basis of the (kj+3) -dimensional space S3(Δ) and hence, the basis of the jC(kj+3) -dimensional space S3(Δ1,,ΔM) in the directions of λ(1),,λ(M) is given by

    {Lg1Lg2LgM|gj{1,,Npair+3},jC}. (5.16)

    Based on (5.16), (5.1) and (5.9), we next introduce the learning model for the uniform case.

    Body learning model Ⅱ. Given the training dataset Z:={(ΛDi,u(Σ;Di))}iN , the learning model TZS3(Δ1,,ΔM) at Λ=(λ(1),,λ(M)) for the geometric body generation associated with the sets of equidistent grids {Δj}jC defined in (5.9) is defined as follows

    TZ(Λ)=k1+3g1=1kM+3gM=1cg1,g2,,gMjCLjgj(λ(j)), (5.17)

    which is required to satisfy the following conditions

    TZ(ΛDi)=k1+3g1=1kM+3gM=1cg1,g2,,gMjCLjgj(λ(j)i)=u(Σ;Di), (5.18)

    where cg1,g2,,gM with gj=0,,kj,j=1,,M. are the coefficients to be determined from the training dataset Z , Ljgj(λ(j)) are B-spline basis functions of degree 3 defined in (5.15) and kj,j=1,,M . is the number of different characteristic values in each direction.

    In learning models Ⅰ and Ⅱ, the learning functionals for the non-uniform and uniform case are respectively considered in the jC(kj+3) -dimensional space S3(Δ1,,ΔM) . jC(kj+3) interpolation conditions are required to determine the coefficients ci1,i2,,iM in the training models. However, there are only jC(kj+1) shape generators to specify jC(kj+1) conditions in (5.8) or (5.18). To obtain a unique correlation between the characteristic values and the shape generator, we need to add jC2=2M conditions, which define the second-order derivatives of the spline function at the boundary aj and bj to be equal to 0 and lead to a natural spline.

    With the learning models Ⅰ and Ⅱ established in the previous subsections, for an input new set of characteristic values ΛDnew associated with a new geometric body Dnew , the unknown shape generator can be generated as follows,

    TZ(ΛDnew)=k1+3g1=1kM+3gM=1cg1,g2,,gMjCLjgj(λ(j)new)=ˆu(Σ;Dnew)u(Σ;Dnew), (5.19)

    where the coefficients cg1,g2,,gM in (5.19) could be determined by solving the natural spline problem.

    In this subsection, we briefly outline the Fourier method for the reconstruction of geometry shape Dnew by using the shape generators u(Σ;Dnew) .

    Define the periodic Sobolev space by

    Hσ(Rd)={gL2(Rd):(1+|ξ|2)σ2ˆg(ξ)L2(Rd),},

    where σ1 , ξZd and ˆg(ξ) denote the Fourier coefficients of g . Suppose that fHσ(Rd) has a compact support in domain V0=(a/2,a/2)d,(a>0) , then the Fourier transform of f is represented by

    ˆfξ=1adV0f(x)¯ϕξ(x)dx, (5.20)

    where the overbar stands for the complex conjugate and the Fourier basis functions are given by

    ϕξ(x)=exp(i2πaξx),ξZd.

    Definition 5.2 (Admissible wavenumbers and observation directions). Let μ be a sufficiently small positive constant such that 0<μ<1 and

    ξ0:={(μ,0),d=2,(μ,0,0),d=3,

    then the admissible set of wavenumbers is defined by

    K:={2πa|ξ|:ξZ3{0}}k0,

    correspondingly, the admissiable set of observation directions is given by

    X:={ξ|ξ|:ξZ3{0}}ˆx0,

    where k0=2π|ξ0|/a and ˆx0=ξ0/|ξ0| for ξ=0Zd .

    Due to suppS⊂⊂V0 , for ξZ3{0} , the far-field pattern defined in (3.6) can be written as

    u(ˆx,k;f)=Cd,kRdeikˆxyf(y) dy=Cd,kV0eikˆxyf(y) dy, (5.21)

    where kK and ˆxX depend on ξ . Combining (5.20) and (5.21), one has

    ˆfξ=1adCd,ku(ˆx,k;f),ξZ3{0}. (5.22)

    For ξ=0 , using the Fourier expansion of f , we derive that

    u(ˆx0,k0;f)=Cd,kV0eik0ˆx0yf(y) dy,=Cd,kV0(ˆf0+ξZd{0}ˆfξϕξ)¯ϕξ0(y) dy=Cd,kV0ˆf0¯ϕξ0(y) dy+Cd,kξZd{0}ˆfξV0ϕξ(y)¯ϕξ0(y) dy=adCd,ksin(μπ)μπˆf0+Cd,kξZd{0}ˆfξV0ϕξ(y)¯ϕξ0(y) dy,

    which implies

    ˆf0=μπadsinμπ{u(ˆx0,k0;f)Cd,kξZ3{0}ˆfξV0ϕξ(y)¯ϕξ0(y) dy}. (5.23)

    Therefore, the Fourier method is to approximate f by a truncated Fourier expansion

    fN=ˆf0+1|ξ|Nˆfξϕξ,

    where NN+ denotes the truncation order and the Fourier coefficients are given by (5.22) and (5.23). Hence the domain Dnew is determined since the set Dnew is the external shape of f .

    Next, we investigate the stability of the proposed Fourier method. In practical computation, there exists some noise between the shape generators u(Σ;Dnew) and the predictions ˆu(Σ;Dnew) , which satisfies

    ˆu(Σ;Dnew)u(Σ;Dnew)L2δu(Σ;Dnew)L2,

    where δ>0 denotes the noise level. Noting that (ˆx,k)X×KΣ , then the approximation of f from predicted shape generators is given by

    fδN=ˆfδ0+1|ξ|Nˆfδξϕξ,

    where

    ˆfδξ=1adCd,kˆu(ˆx,k),ξZ3{0}, (5.24)
    ˆfδ0=μπadsinμπ{ˆu(ˆx0,k0)Cd,kξZ3{0}ˆfδξV0ϕξ(y)¯ϕξ0(y) dy}. (5.25)

    Theorem 5.1. Let f be a compactly supported function in Hσ(Rd) , σ1 , with suppf⊂⊂V0 , then we have the following estimate

    fδf2L2(V0)Cδ+C(τd+τ2)δ42+d.

    where dτR+ and C is a constant which depends on f,a,d,μ .

    Proof. Using the Plancherel theorem, we have

    fδf2L2(D)=fδf2L2(Rd)=1a2dRd|ˆfδξˆfξ|2 dξ=1a2d(|ξ|N|ˆfδξˆfξ|2 dξ+|ξ|>N|ˆfδξˆfξ|2 dξ), (5.26)

    where NN+ . Due to fHσ(Rd) , that is,

    (1+|ξ|2)α2ˆfξL2(Rd),|α|σ.

    It means that both |ξ|ˆfξ and |ξ|ˆfδξ are bounded in L2(Rd) , so we can find N>0 , such that

    |ξ|>N|ˆfδξˆfξ|2 dξ1N2|ξ|>N|ξ|2|ˆfδξˆfξ|2 dξ<C1N2, (5.27)

    where C1>0 is a constant. For 1|ξ|N , from (5.22) and (5.24), we have

    |ˆfδξˆfξ|=1adCd,k|ˆu(ˆx,k)u(ˆx,k)|δadCd,k|u(ˆx,k)|=δadCd,k|Cd,kV0f(y)eikˆxy dy|δad(V0|f(y)|2 dy)12(V0|eikˆxy|2dy)12=fL2(V0)ad/2δ,

    which implies

    1|ξ|N|ˆfδξˆfξ|2 dξC2(2N+1)dδ2, (5.28)

    for C2=f2L2(V0)/ad . Define ξ=(ξ1,ξ2)Z2 or ξ=(ξ1,ξ2,ξ3)Z3 , by a straight forward calculation, one finds that

    V0ϕξ(y)¯ϕξ0(y) dy={0,|ξ||ξ1|,adcosξ1πsinλπ(ξ1λ)π,|ξ|=|ξ1|.

    For ξ=0 , using (5.23), (5.25), (5.27), (5.28) and the last equation, it derives that

    |ˆfδ0ˆf0|μπadCd,ksinμπ|ˆuδ(ˆx0,k0)uδ(ˆx0,k0)|+μπadsinμπ1|ξ|N|ˆfδξˆfξ||V0ϕξ(y)¯ϕξ0(y) dy|+μπadsinμπ|ξ|N|ˆfδξˆfξ||V0ϕξ(x)¯ϕξ0(x) dy|C3δ+C2(2N+1)dδ+C1N, (5.29)

    where C3=μπfL2(V0)/(ad/2sinμπ) . Hence, substituting (5.27), (5.28) and (5.29) into (5.26), it deduces that

    fδf2L2(V0)Cδ2+CNdδ2+CN2,

    where C=max{2C1,2d+1C2,C23}/a2d . Furthermore, if we take N=τδ22+d with τd in Theorem 5.1, then it holds that

    fδf2L2(V0)Cδ+C(τd+τ2)δ42+d.

    Let N=[τδ22+d,τd] , here [X] denotes the largest integer that is smaller than X+1 . From definition 5, the truncated wavenumbers and observation directions can be written as

    KN:={2πa|ξ|:1|ξ|N}k0,XN:={ξ|ξ|:1|ξ|N}ˆx0.

    Thus, the truncated Fourier expansion of f from the predictions {ˆu(Σ;Dnewq)}qQ takes the form

    fδN:=ˆfδ0+1|ξ|Nˆfδξϕξ(x), (5.30)

    where

    ˆfδξ=1adCd,kˆu(ˆx,k),1|ξ|N,ˆfδ0=μπadsinμπ{ˆu(ˆx0,k0)Cd,k1|ξ|NˆfδξV0ϕξ(y)¯ϕξ0(y) dy}. (5.31)

    Motivated by above discussion, we are ready to present our novel modeling methodology for geometric shape in Rd,d=2,3 , see Algorithm 1.

    Algorithm 1: Inverse-scattering-based geometric body generation scheme
    1: Select the parameter N , the set of admissible wavenumbers KN and the set of admissible observation directions XN .
    2: Given a training dataset Z:={(ΛDi,u(ˆx,k;Di))}iN for ˆxXN,kKN , obtain the coefficients cg1,,gM of the learning model TZ by solving the problem of the natural spline interpolation.
    3: Given the characteristic sets ΛDnew , predict the new shape generators {ˆu(ˆx,k)} for ˆxXN,kKN with the use of the learning model TZ .
    4: Compute the Fourier coefficients ˆfδ0 and ˆfδξ defined in (5.31) for 1|ξ|Nt .
    5: Select a sampling mesh Th in a region V0 . For each sampling point zjT , calculate the imaging function fδN defined in (5.30). Dnew is obtained as the external shape of fδN .

     | Show Table
    DownLoad: CSV

    In this section, several numerical examples of geometric body generation are conducted to show that the proposed method is effective and efficient. Here the geometric body includes general geometric shape and human body shape.

    The proposed algorithm is implemented by using Matlab 2016. The shape generator {u(ˆx,k;Di)}iN is obtained by solving the direct problem of (3.1). To avoid the inverse crime, we use the quadratic finite elements on a truncated spherical domain enclosed by a PML layer. The mesh of the forward solver is successively refined till the relative error of the successive measured scattered data is below 0.1% . Then artificial shape generators are generated by applying the Kirchhoff integral formula to the scattered data. Thus the training dataset is given by

    {(ΛDi,u(ˆxj,kj;Di)):ˆxjXN,kjKN},

    where j=1,2,,(2N+1)d and i=1,2,,Nt denotes the i -th geometry shape. In what follows, we set τ=2 (τ=3 ) for d=2 (d=3 ) and δ=1% , then we have N=20 (N=19 ) for d=2 (d=3 ).

    Next, we present the implementation of interpolation. The characteristic value set is given by {ΛDi} , where ΛDi={λ(1)i,λ(2)i,,λ(M)i} has M variables. For a fixed wavenumber kj , we use cubic spline interpolation to obtain the coefficients ci1,iM from the characteristic value ΛDi and the shape generator u(ˆxj,kj;Di) . Therefore, given a new characteristic value ΛDnew={λ(1)Dnew,λ(2)Dnew,,λ(M)Dnew} , we obtain the predicted shape generators {ˆu(ˆxj,kj;Dnew)},j=1,2,,(2N+1)d .

    Finally, we specify details of recovering the geometry. As discussed above, reconstructing the geometry shape is equally to reconstructing the source function f . In the discrete formula, the domain V0 is divided into a uniform mesh with size 100×100 in two dimensions and size 100×100×100 in three dimensions. Further, the approximated Fourier series fδN are computed at the mesh nodes Tj,j=1,,1003 in (5.30). Thus, the geometry shape Dnew is approximated by the boundary of the imaging results fδN .

    In the first example, we aim to reconstruct a kite shaped domain with scale changing. The kite shaped domain is parameterized by

    x(t)=(β1(cost+0.65cos2t0.65), 1.5β2sint),t[0,2π],

    where β1 and β2 are scale factors (characteristic values) with β1,β2[0.5,1.8] . The training dataset consists of 14×14 different scale domains, i.e., β1 and β2 uniformly distributed on [0.5,1.8] with M=142 . Next, we consider four sets of different scale factors which are not covered by the training data. In this numerical experiments, the imaging results with different characteristic values are shown in Figure 1, where the black dotted lines denote the exact boundary. It is clear that the reconstructions are very closed to the exact domains.

    Figure 1.  Contour plots of reconstructed kite shape with different scale factors (β1,β2) : (a) (0.93,1.76) , (b) (1.58,0.87) , (c) (0.95,0.95) , (d) (1.65,1.65).

    In the second example, we aim to recover multi-domain with different scale factors. The apple shaped domain is parameterized by

    y(t)=β1((0.5+0.4cost+0.1sin2t)/(1+0.7cost))(cost, sint)t[0,2π],

    and the rounded triangle shaped domain is parameterized by

    z(t)=β2(1+0.15cos3t)(cost, sint)t[0,2π],

    where β1[1,2] and β2[0.5,1.5] are scale factors (characteristic values) for different domains. The training dataset consists of 11×11 different scale domains, that is, β1 and β2 uniformly distributed on [1,2]×[0.5,1] . Similarly, we give four sets of different scale factors which are not covered by the training data. Figure 2 shows the the reconstruction of multi-domain with different characteristic values via contour plots, where the black dotted lines denote the exact boundary. It demonstrates very good imaging performance of the approach.

    Figure 2.  Contour plots of reconstructed mixed shape with different scale factors (β1,β2) : (a) (1.13,1.13) , (b) (1.94,0.53) , (c) (1.88,0.94) , (d) (1.96,1.47).

    In the third example, we verify the proposed method by using a set of artificial experiments on rectangular solid. The training dataset consists of 125 rectangular solids with different height, width and length. Here the height, width and length are uniformly distributed on [1,2] with 5 amounts, i.e., 1,1.25,1.5,1.75,2 . Here, We consider four sets of different height, width and length of rectangular solids which are not covered by the training data. The imaging results with different characteristic values are shown in Figure 3, where the black dotted lines denote the shadows of the exact cube boundary. Due to discontinuities of the source, there is Gibbs phenomena on the boundary of the rectangular solids. On the whole, given the characteristic values, our proposed method is valid for determining the geometry shape.

    Figure 3.  Isosurface plots of rectangular solid with different height, width and length, where the isosurface value is 1 . The sets of height, width and length are as follows: (a) (1.9,1.2,1.4) , (b) (1.2,1.5,1.8) , (c) (1.6,1.8,1.3) , (d) (1.8,1.8,1.8).

    In last example, we consider a challenging case and verify the proposed method by using a set of synthetic experiments on 3D human body shape. The training dataset consists of 25 bodies which are generated by the MakeHuman soft [42]. MakeHuman soft is an open source software designed to create 3D virtual human models using 3D morphing technology. Providing the main parameters gender, age, muscle mass, weight, height, proportion and ethnicity, etc. of any specific character, the intermediate human shape can be reproduced with its large and own database. In this experiment, we consider two characteristic values, i.e., height and relative weight. Define the exact weight by EW and standard weight by SW , then the relative weight RW is calculated by

    RW=EWSW×100%.

    Here the height of the body is given by 1.5m,1.6m,1.7m,1.8m,1.9m and the relative weight of the body is given by 60%,80%,100%,120%,140% . Some human body shapes in the training dataset are presented in Figure 4. In addition, we choose two characteristic values of human body which are not covered by the training data. The first body's height is 1.55m and the relative weight is 130% . The second body's height is 1.85m and the relative weight is 110% . Figure 5(a) and Figure 6(a) present the exact body shape with the given characteristic value. Figure 5(b) and Figure 6(b) show the prediction of the human body shape with the given characteristic value. The results show that our method is efficient to predict the human body shape.

    Figure 4.  Isosurface plots of some training body data. Relative weight: the left column is 60% ; the center column is 100% , the right column is 140% ; height: the top row is 1.50m , the center row is 1.70m , the bottom row is 1.90m.
    Figure 5.  The relative weight is 130% , the height is 1.55m . (a) Exact body, (b) reconstruction body.
    Figure 6.  The relative weight is 110% , the height is 1.85m . (a) Exact body, (b) reconstruction body.

    In this paper, we develop a machine-learning method in generating a geometric body shape through prescribing a set of characteristic values of the body. The generation is mainly based on a given training dataset consisting of certain pre-selected body shapes with statistically well-sampled characteristic values. A major novelty and critical ingredient of our study is the borrowing of inverse scattering techniques in the theory of wave propagation to the geometric shape generation. We introduce the notion of shape generator which establishes a one-to-one correspondence between the geometric shape space and the function space consisting of the multiple-frequency far-field patterns associated with the time-harmonic source scattering problem. The shape generator plays an intermediate role in the geometric shape generation. First, the training dataset of geometric shapes is converted into a subset of the function space consisting of the corresponding shape generators. Then a learning model is derived through a functional interpolation of the aforementioned shape generators. For a given set of characteristic values, one then uses the learning model to obtain the shape generator of the underlying geometric body and finally reconstructs it through a multiple-frequency Fourier method.

    In the current article, the shape generator is introduced through an inverse source scattering model where we make use of the one-to-one correspondence between a geometric shape and the multiple-frequency far-field pattern associated with a compactly-supported acoustic source. One may consider to introduce the shape generator through other inverse scattering models, e.g., the inverse acoustic obstacle scattering model (cf. [28,29,31]) or the inverse electromagnetic scattering model (cf. [27,30]). In doing so, one may achieve other geometric shape generation schemes that are suitable for different applications.

    The work of H. Liu was supported by the FRG and startup grants from Hong Kong Baptist University, Hong Kong RGC General Research Funds, 12302017, 12301218 and 12302919.

    The authors declare no conflict of interest.



    [1] D. Vasileska, S. M. Goodnick, G. Klimeck, Computational Electronics: Semiclassical and Quantum Device Modeling and Simulation, Boca Raton: CRC press, 2017. https://doi.org/10.1201/b13776
    [2] SILVACO International, ATLAS User's Manual: Device Simulation Software, 2019.
    [3] P. Farrell, N. Rotundo, D. H. Doan, M. Kantner, J. Fuhrmann, T. Koprucki, Drift-diffusion Models, in J. Piprek, Handbook of Optoelectronic Device Modeling and Simulation: Lasers, Modulators, Photodetectors, Solar Cells, and Numerical Methods, Vol. 2, Boca Raton: CRC Press, 2017,733–772. https://doi.org/10.4324/9781315152318
    [4] S. Selberherr, Analysis and Simulation of Semiconductor Devices, Vienna: Springer, 2012. https://doi.org/10.1007/978-3-7091-8752-4
    [5] R. E. Bank, D. J. Rose, W. Fichtner, Numerical methods for semiconductor device simulation, IEEE Trans. Electron Devices, 30 (1983), 1031–1041. https://doi.org/10.1109/T-ED.1983.21257 doi: 10.1109/T-ED.1983.21257
    [6] S. J. Polak, C. Den Heijer, W. H. A. Schilders, P. Markowich, Semiconductor device modelling from the numerical point of view, Int. J. Numer. Methods Eng., 24 (1987), 763–838. https://doi.org/10.1002/nme.1620240408 doi: 10.1002/nme.1620240408
    [7] R. D. Lazarov, I. D. Mishev, P. S. Vassilevski, Finite volume methods for convection-diffusion problems, SIAM J. Numer. Anal., 33 (1996), 31–55. https://doi.org/10.1137/0733003 doi: 10.1137/0733003
    [8] C. Chainais-Hillairet, J. G. Liu, Y. J. Peng, Finite volume scheme for multi-dimensional drift-diffusion equations and convergence analysis, ESAIM. Math. Model. Numer. Anal., 37 (2003), 319–338. https://doi.org/10.1051/m2an:2003028 doi: 10.1051/m2an:2003028
    [9] S. C. Han, S. M. Hong, Deep neural network for generation of the initial electrostatic potential profile, 2019 International Conference on Simulation of Semiconductor Processes and Devices (SISPAD), IEEE, Udine, Italy, 2019, 1–4. https://doi.org/10.1109/SISPAD.2019.8870521
    [10] X. Jia, H. An, Y. Hu, Z. Mo, A physics-based strategy for choosing initial iterate for solving drift-diffusion equations, Comput. Math. Appl., 131 (2023), 1–13. https://doi.org/10.1016/j.camwa.2022.11.029 doi: 10.1016/j.camwa.2022.11.029
    [11] K. W. Lee, S. M. Hong, Acceleration of semiconductor device simulation using compact charge model, Solid-State Electron., 199 (2023), 108526. https://doi.org/10.1016/j.sse.2022.108526 doi: 10.1016/j.sse.2022.108526
    [12] Q. Zhang, Q. Wang, L. Zhang, B. Lu, A class of finite element methods with averaging techniques for solving the three-dimensional drift-diffusion model in semiconductor device simulations, J. Comput. Phys., 458 (2022), 111086. https://doi.org/10.1016/j.jcp.2022.111086 doi: 10.1016/j.jcp.2022.111086
    [13] J. W. Slotboom, Computer-aided two-dimensional analysis of bipolar transistors, IEEE Trans. Electron Devices, 20 (1973), 669–679. https://doi.org/10.1109/T-ED.1973.17727 doi: 10.1109/T-ED.1973.17727
    [14] M. A. der Maur, M. Povolotskyi, F. Sacconi, A. D. Carlo, TiberCAD: A new multiscale simulator for electronic and optoelectronic devices, Superlattices Microstruct., 41 (2007), 381–385. https://doi.org/10.1016/j.spmi.2007.03.011 doi: 10.1016/j.spmi.2007.03.011
    [15] P. Farrell, D. Peschka, Nonlinear diffusion, boundary layers and nonsmoothness: Analysis of challenges in drift–diffusion semiconductor simulations, Comput. Math. Appl., 78 (2019), 3731–3747. https://doi.org/10.1016/j.camwa.2019.06.007 doi: 10.1016/j.camwa.2019.06.007
    [16] S. P. Chin, C. Y. Wu, A new methodology for two-dimensional numerical simulation of semiconductor devices, IEEE Trans. Comput. Aided Des. Integr. Circuits Syst., 11 (1992), 1508–1521. https://doi.org/10.1109/43.180264 doi: 10.1109/43.180264
    [17] G. Sabui, P. J. Parbrook, M. Arredondo-Arechavala, Z. J. Shen, Modeling and simulation of bulk gallium nitride power semiconductor devices, AIP Adv., 6 (2016), 055006. https://doi.org/10.1063/1.4948794 doi: 10.1063/1.4948794
    [18] R. Eymard, T. Gallouët, R. Herbin, Finite volume methods, Handb. Numer. Anal., 7 (2000), 713–1018. https://doi.org/10.1016/S1570-8659(00)07005-8 doi: 10.1016/S1570-8659(00)07005-8
    [19] C. Chainais-Hillairet, Entropy method and asymptotic behaviours of finite volume schemes, In: J. Fuhrmann, M. Ohlberger, C. Rohde, Finite Volumes for Complex Applications Ⅶ-Methods and Theoretical Aspects, Cham: Springer, 77 (2014), 17–35. https://doi.org/10.1007/978-3-319-05684-5_2
    [20] J. J. H. Miller, W. H. A. Schilders, S. Wang, Application of finite element methods to the simulation of semiconductor devices, Rep. Prog. Phys., 62 (1999), 277. https://doi.org/10.1088/0034-4885/62/3/001 doi: 10.1088/0034-4885/62/3/001
    [21] H. K. Gummel, A self-consistent iterative scheme for one-dimensional steady state transistor calculations, IEEE Trans. Electron Devices, 11 (1964), 455–465. https://doi.org/10.1109/T-ED.1964.15364 doi: 10.1109/T-ED.1964.15364
    [22] H. C. Elman, D. J. Silvester, A. J. Wathen, Finite Elements and Fast Iterative Solvers: With Applications in Incompressible Fluid Dynamics, 2 Eds., New York: Oxford University Press, 2014. https://doi.org/10.1093/acprof:oso/9780199678792.001.0001
    [23] R. Radhakrishnan, J. H. Zhao, A 2-dimensional fully analytical model for design of high voltage junction barrier Schottky (JBS) diodes, Solid-State Electron., 63 (2011), 167–176. https://doi.org/10.1016/j.sse.2011.06.002 doi: 10.1016/j.sse.2011.06.002
    [24] L. D. Benedetto, G. D. Licciardo, T. Erlbacher, A. J. Bauer, S. Bellone, Analytical model and design of 4H-SiC planar and trenched JBS diodes, IEEE Trans. Electron Devices, 63 (2016), 2474–2481. https://doi.org/10.1109/TED.2016.2549599 doi: 10.1109/TED.2016.2549599
    [25] M. Mehrota, B. J. Baliga, Very low forward drop JBS rectifiers fabricated using submicron technology, IEEE Trans. Electron Devices, 40 (1993), 2131–2132. https://doi.org/10.1109/16.239813 doi: 10.1109/16.239813
  • This article has been cited by:

    1. Tharifkhan Shan Ahamed, Susaimanickam Anto, Thangavel Mathimani, Kathirvel Brindhadevi, Arivalagan Pugazhendhi, Upgrading of bio-oil from thermochemical conversion of various biomass – Mechanism, challenges and opportunities, 2021, 287, 00162361, 119329, 10.1016/j.fuel.2020.119329
    2. Qun Wang, Bin Qin, Xiaohua Zhang, Xiaoling Xie, Li'e Jin, Qing Cao, Synthesis of N-doped carbon nanosheets with controllable porosity derived from bio-oil for high-performance supercapacitors, 2018, 6, 2050-7488, 19653, 10.1039/C8TA07563H
    3. Jia Wen Chong, Nishanth G. Chemmangattuvalappil, Suchithra Thangalazhy-Gopakumar, 2023, Chapter 2, 978-981-19-4846-6, 33, 10.1007/978-981-19-4847-3_2
    4. Anisha GS, Shyni Raphel M, 2023, 9780124095489, 10.1016/B978-0-323-93940-9.00057-8
    5. Zengtong Deng, Yuanjing Chen, Xun Hu, Wei Deng, Mortaza Gholizadeh, Song Hu, Sheng Su, Long Jiang, Jun Xu, Kai Xu, Yi Wang, Jun Xiang, Investigation on bio-oil pyrolysis with Ni/Al2O3 blending: Influence of the blended catalyst on coke formation, 2024, 358, 00162361, 130274, 10.1016/j.fuel.2023.130274
    6. Wasinton Simanjuntak, Kamisah Delilawati Pandiangan, Tika Dwi Febriyanti, Aryani Putri Islami, Sutopo Hadi, Ilim Ilim, Catalytic upgrading of palm oil derived bio-crude oil for bio-hydrocarbon enrichment using protonated zeolite-Y as catalyst, 2024, 12, 2333-8334, 600, 10.3934/energy.2024028
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1738) PDF downloads(90) Cited by(0)

Figures and Tables

Figures(6)  /  Tables(4)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog