CyberTraining: Pilot: Employing Proper Orthogonal Decomposition (POD) and High-Performance Computing (HPC) in Advanced CI

Part of NSF Initiative on Workforce Development for Cyberinfrastructure (CyberTraining)

Contact

mailDaqing Hou
[email protected]

CyberTraining 2022

Introduction

The cohort participated in our 2022 CyberTraining workshop will be divided into 2 or 3 teams based on the mutual interests and complementary skills that each team possesses. Each team will be led by a PI and assisted by the TA to implement a POD project of their choice in C++. It is expected that each team is able to successfully complete the POD coding and perform the POD simulation of their selected project.

Learning Outcomes

Trainees are expected to demonstrate mastery of the POD method, ability to collect data using FEniCS, and implement POD using algorithms from PETSc and SLEPc in C++. More specifically, the projects can be divided into four milestones as follows:

1. Collect data by modifying the provided sample FEniCS code to fully describe the selected problem in the simulation domain with appropriate BCs and excitations

2 Prepare a code in C++ and use PETSc and SLEPc to perform the method of snapshots and solve a discrete eigenvalue problem to generate POD modes

3 Perform integrations of the POD modes and their gradients in C++ using tools PETSc and SLEPc

4 Solve ODEs to perform POD simulation in C++ using solvers in PETSc.

Project 1: 3D thermal simulation of a small IC

This project will use the FinFET NAND gate as the selected block to generate the POD modes. A different IC consisting of several NOT and NAND gates will be used to train the block. Also, the heat conduction equation will be implemented in FEniCS for the IC structure to collect the data. The team will run through the step-by-step procedure and eventually perform POD simulation of the selected block embedded in the IC with different input excitations, compared against the FEniCS simulation.

Project 2: 3D thermal simulation of a CPU

This project will implement the heat conduction equation in a CPU to study the thermal profile over an entire CPU heated by different dynamic power maps. The team will perform FEniCS simulation to train a CPU. The POD model for the CPU will be validated compared to FEniCS simulations. A good agreement with FEniCS with only 3 or more modes in the POD model is supposed to be observed. Also, the hot-spot distribution in the chip will be analyzed, compared to FEniCS simulation.

Project 3: Quantum eigenvalue problem of a 2D nanostructuree

This project will develop a POD model to solve wave functions in a 2D quantum structure, similar to a superlattice quantum dot structure, influenced by electric field. For simplicity, boundaries are assumed far away from the quantum dots and the wave function nearly vanishes on the boundaries.