Objective

You will be implementing four compiler/accelerator falvors for DNN inference. There will be four mini projects through the quarter and one final project. I value innovation above all and the second part is your execution of the ideas. As researchers, you should be to develop an idea to a submission and meet the conference deadlines.

Projects

The projects will be as follows:

HW0: Machine Learning Programming
HW1: ONNX -> Verilog
HW2: ONNX -> Weight-embedded Verilog
HW3: Quantized weight-embedded Verilog
Final Project: ONNX -> Your Time-Multiplexed Programmable Accelerator

Collaboration

You are required to work in groups of two.

Project proposal

For the Final Project you are required to email a half page project proposal to the instructor and the TA by 12:00am on TBD. The proposal should convince the instructor:

What research problem the project is solving.

Why this specific research problem is important.

How the team will solve the research problem.

Final Project Due date

TBD

Submission

Each group should submit a comressed file containing the following items through gradescope on the due date of the Final Project before the class starts. Note that the mini projects just require report and source code:

The project report in pdf format.

Source code with instructions describing how it should be run

The original proposal in pdf format.

The final presentation.

Final report

The project report should be in pdf format with no more than six pages of primary content using the ACM SIGPLN conference template with 11 point font size. You are allowed unlimited space for the citations and appendices, starting on page seven, but your report should be complete and understandable without reading this extra material. You are free to use either the word or the LaTeX template. However, I personally prefer the LaTeX template. There are many helpful material on how to write a good conference paper on the ACM SIGPLAN author center. I strongly encourage you to read the provided material.

Grading

Your project will be graded on a scale of one to 30 with ten points for the report, ten points for the implementation, and ten points for the presentation. If your work is ready to submit to a conference, you get 10 bonus points. In case of survey projects, the grade for report and presentation will be scaled by a factor of 1.5x.

Report

5 points: Organized communication of the ideas in a conference paper format.
1 point: Grammar and spelling (1 point).
2 points: Clarity and grace, i.e., clear, well organized arguments in well organized paragraphs.
2 points: Empirical support for the proposed technique (4 bonus point).

Presentation

4 points: Clearly presenting key ideas from the work.
4 points: Clear, well-organized slides.
1 point: Drawing high-level insights from the empirical results.
1 point: Motivating future work.

Implementation

You should demonstrate that your implementation works. You will get a score between zero to ten based on the completeness of your implementation.

CSE 240C: Advanced Microarchitecture (Fall 2018)

Instructor: Hadi Esmaeilzadeh

Email: hadi [AT] eng [DOT] ucsd [DOT] edu

Office: CSE 3228

TA:

Soroush Ghodrati: soghodra [AT] eng [DOT] ucsd [DOT] edu

Office hours: TBD