Welcome to the MICS 6001C at HKUST(GZ), 2024 Fall

Course Information

Course Description

Being able to customize hardware architecture to application’s exact needs, FPGA-based accelerators deliver better efficiency than general architectures such as CPUs. This course explores the latest advances in FPGA-based accelerators for computation-intensive applications. It covers FPGA architecture fundamentals, underlying structures, and key technologies for high-performance programming. Using C/C++ via HLS tools, the course aims to agile FPGA development. It showcases recent FPGA acceleration achievements across various domains, including DNNs. Learners will engage with multiple design examples, starting with basic designs and offering opportunities for further exploration. Optionally, this course will include accelerator/algorithm co-design, being an extremely important and promising research topic.

Course Schedule

Week Date (Friday) Topics Assignments Paper Reading Recommended Readings
Week 1 Sept 6 1. Course Introduction
  1. Domain Specific Architectures | | | Domain-specific hardware accelerators. Communications of the ACM 2020. | | Week 2 | Sept 13 | 1. FPGA Architecture
  2. Hardware Programming Languages | | | Three Ages of FPGAs: A Retrospective on the First Thirty Years of FPGA Technology, Proceedings of the IEEE, 2015 | | Week 3 | Sept 20 | 1. HLS Overview
  3. Vitis HLS Tutorial (Bring your laptop) | Lab 1 Released | | FPGA HLS today: successes, challenges, and opportunities. TRETS, 2022. | | Week 4 | Sept 27 | 1. Loops and Optimizations in HLS
  4. Memory and Streaming in HLS | | 2 pres. | TAPA: a scalable task-parallel dataflow programming framework for modern FPGAs with co-optimization of HLS and physical design. TRETS, 2023. | | Week 5 | No Class (Oct 4) | National Day Holiday | | | | | Week 6 | Oct 11 | 1. HLS Workflow
  5. Paper Presentation | | 2 pres. | Allo: A Programming Model for Composable Accelerator Design. PLDI, 2024 | | Week 7 | Oct 18 | Chip Forum 2024 | | | | | Week 8 | Oct 25 | Project Proposal Presentation | Lab 1 Due Lab 2 Released | 1 pre. | A survey of quantization methods for efficient neural network inference. Low-Power Computer Vision. | | Week 9 | Nov 1 | 1. Neural Networks Fundamentals
  6. Data Quantization | | 2 pres. | Tpu v4: An optically reconfigurable supercomputer for machine learning with hardware support for embeddings. ISCA, 2023 | | Week 10 | Nov 8 | 1. Convolution Operations in Hardware
  7. Systolic Arrays and their Applications | | 2 pres. | Graphit: A high-performance graph dsl. OOPSLA 2018. | | Week 11 | Nov 15 | 1. Advanced DSP Techniques
  8. Domain Specific Languages | | 2 pres. | ReGraph: Scaling graph processing on HBM-enabled FPGAs with heterogeneous pipelines. MICRO, 2022. | | Week 12 | Nov 22 | 1. Hardware-Accelerated Graph Analytics
  9. Paper Presentation | | 2 pres. | Polardb serverless: A cloud native database for disaggregated data centers." SIGMOD, 2021. | | Week 13 | Nov 29 | 1. Database Acceleration
  10. FPGAs In Cloud Computing | Lab 2 Due | 3 pres. | | | Week 14 | Dec 6 | Final Project Presentation | | | |

Paper Presenting

  1. Please propose the paper you intend to present as early as possible
  2. You can exchange the paper presentation time with someone else.
  3. Link:https://docs.google.com/spreadsheets/d/1omcmors_WhQhUB7WKqTEY6a47YsLl5COF5RpkvfY9Cc/edit?usp=sharing

[https://docs.google.com/spreadsheets/d/1omcmors_WhQhUB7WKqTEY6a47YsLl5COF5RpkvfY9Cc/edit?usp=sharing](https://docs.google.com/spreadsheets/d/1omcmors_WhQhUB7WKqTEY6a47YsLl5COF5RpkvfY9Cc/preview?usp=sharing)

Collecting Project Information

  1. Please fill your project information in shared Google Sheets below.
  2. As a gentle reminder, it's advisable to start your project as early as possible.
  3. Link:https://docs.google.com/spreadsheets/d/1Efc9X8AnJznC2G7o7LoDW5EZJ2cASg8VBFOldrzdMSY/edit?usp=sharing

[https://docs.google.com/spreadsheets/d/1Efc9X8AnJznC2G7o7LoDW5EZJ2cASg8VBFOldrzdMSY/edit?usp=sharing](https://docs.google.com/spreadsheets/d/1Efc9X8AnJznC2G7o7LoDW5EZJ2cASg8VBFOldrzdMSY/preview?usp=sharing)