Paper 1

CS 641 Paper, Dr. Lawlor

I'd like you to write a research paper. It doesn't have to be a completely novel topic (no paper actually is), and it doesn't have to describe only your own work (referencing other people's work is always recommended), but it has to look like a research paper, and consist of 100% your own writing,

Elements of a Research Paper

Every single research paper on the planet has:

A title. This is surprisingly tough to pick, since it needs to be short, unambiguous, and attention-grabbing.
A list of authors. This is you (for now).
An abstract, a short one-paragraph blurb describing why people should read your paper. I usually write this last (bottom up). Some people write this first (top down).
An introduction, basically a longer version of the abstract.
Comparisons and citations to prior work.
A detailed quantitative comparison of your and prior work.
Conclusion section, where you describe what you learned doing the work, and what remains for the future.
A references section, where you expand your citations.

Examples:

Bill Dally has some very widely cited computer architecture papers. (For example, his 2007 Compilation for Explicitly Managed Memory Hierarchies.)
See any of my publications. These are much less widely cited, and not many cover computer architecture, but they are a good model for my understanding of what publications are supposed to look like!
Generate your own examples using SCIgen, which is uncannily accurate at the broad scale, despite being machine-generated grammatically correct nonsense.

Research or Applied Topics

These topics are research of the form "Learn and present what people have done about X" rather than applied work of the form "Build a new X".

Write up the rationale, history, and advantages/disadvantages of any current hardware topic, such as:

Pipelining, especially the very deep piplines of the Pentium 4
Out-of-order execution
Register renaming
Branch prediction, branch history, and execution speculation
Cache prefetching and out-of-order loads and stores

Describe how the design limitations and goals of nonstandard computing platforms differ from conventional computing, such as:

High-performance computing systems, such as Blue Gene or anything on the Top500 list.

Consumer game consoles, such as the PlayStation 3, Xbox 360, or Wii.
Embedded systems, such as cell phones or microwave ovens.

Pick any hardware-related article from Ars Technica. Explain what they're talking about in detail.
Pick a CPU architecture from sandpile.org. Compare this architecture's hardware design, in terms of achievable performance, with competing architectures.
Describe performance counters, which are useful for understanding code performance and pipelining (see PCL)
Describe a strange fabrication substrate or nonstandard computing scheme, such as Biological Computing, Quantum Computing, self-organizing DNA or polymer nanofabrication, etc.
Describe a novel data storage architecture, such as perpendicular bit recording, magnetoresistive memory, nanowire memory.
Describe a semiconductor or PCB fabrication process in detail, such as the problems encountered during deep-submicron photolithography, or the interelationship between planarization and metal layers in CMOS fabrication.
Describe the historical evolution of some computer architecture, such as SPARC or Motorola's 68000.
Explore the decline and fall of some popular computer architecture, such as the VAX ("All the world's a VAX!" Or, er, it was...) Be sure to cite Percy Shelly's poem Ozymandias.

Here are some applied project possibilities. As long as it's even tangentially related to something we covered in class, it counts. Do keep in mind that the end product is not code, it's a detailed document describing prior work on similar topics, explaining how your implementation is different, and analyzing the performance of your code.

SIMD Within A Register:

Something similar to Sean Anderson's Bit Twiddling Hacks
Something similar to our own Jim Long's portable Cray Bioinformatics Library

x86 SSE/AVX or ARM VFP:

Write and tune a real implementation of any interesting algorithm: long vector adds, dot product, matrix-vector multiply, fractal rendering, etc.

GPU:

Write anything interesting on the GPU, using any of the GPU interfaces: GLSL, CUDA, OpenCL.

Deadlines

Give me a detailed topic outline by Tuesday, February 28.
Give me a rough draft on Thursday, March 8.
Give me an anonymous review draft on Tuesday, March 20. I will send these around the class for anonymous peer review (double blind).
Reviewer feedback must be returned by Tuesday, March 27.
Give us a presentation and final draft by Thursday, March 29.