Title:
NEURAL NETWORK-BASED METHOD FOR SOLVING MULTI-SUB-BAND BOLTZMANN TRANSPORT EQUATION
Document Type and Number:
WIPO Patent Application WO/2022/267749
Kind Code:
A1
Abstract:
The present invention belongs to the field of electronic design automation, and disclosed is a neural network-based method for solving a multi-sub-band Boltzmann transport equation. In the present invention, a Boltzmann-Schrödinger-Poisson system is firstly established with regard to a nanoscale transistor electron transport problem; meshing is performed in time and position space and wave vector space on a simulation device; a training dataset is generated by self-consistently solving the multi-sub-band Boltzmann transport equation by means of a deterministic method; the Boltzmann transport equation is rewritten into a neural partial differential equation to embed a neural network therein, and form a neural Boltzmann equation; the difference between an equilibrium electron distribution function and a current electron distribution function is used as an input of the neural network, and same is inputted as a part to be approximated of a scattering term on a right end of the neural Boltzmann equation; and the neural network is trained on the basis of a time series modeling policy. The trained neural network can enable a solution of the neural Boltzmann transport equation to be used as an accurate approximate solution of electron distribution, greatly improving computational efficiency while guaranteeing accuracy.
More Like This:
Inventors:
LIU JIE (CN)
DING XI (CN)
DING XI (CN)
Application Number:
PCT/CN2022/093255
Publication Date:
December 29, 2022
Filing Date:
May 17, 2022
Export Citation:
Assignee:
UNIV HUNAN (CN)
International Classes:
G06F17/11
Foreign References:
| CN110414024A | 2019-11-05 | |||
| CN102945295A | 2013-02-27 | |||
| JPH08335225A | 1996-12-17 |
Other References:
TONG JIANNONG, ZOU XUECHENG, SHEN XUBANG : "An Analytic Model of Boltzmann Equation in Extremely Scaled MOSFETs", COMPUTER & DIGITAL ENGINEERING, vol. 32, no. 5, 20 October 2004 (2004-10-20), pages 4 - 7, XP093016723
Download PDF: