Reading Guide: An AI Approach to Integrated Circuit Floorplanning

Long before machine learning models were placing macros on modern chips, researchers were asking whether a computer could reason about a chip floorplan the way an experienced designer does. An Artificial Intelligence Approach to Integrated Circuit Floorplanning, by Marwan A. Jabri, is a compact record of that effort. Published by Springer in 1991 as volume 66 of the Lecture Notes in Engineering series, it runs to about 150 pages and grew out of doctoral research the author carried out between 1984 and 1987. This guide will help you decide whether it belongs on your desk and how to get the most from it.

Get the book: View An AI Approach to IC Floorplanning on Amazon

What this book is (and who it's for)

This is a research monograph, not a textbook. It reports one researcher's investigation into knowledge-based (expert-system style) approaches to integrated circuit floorplanning: the early physical-design step where the major blocks of a chip are given positions, shapes and rough sizes before detailed placement and routing begin. In the preface, the author explains that the volume looks at the problem from four angles: an overall floorplanning strategy, knowledge engineering in the IC design domain, the development of knowledge-based floorplanning processes, and new floorplanning algorithms that fit inside that strategy.

It suits three kinds of readers:

• Graduate students in VLSI or electronic design automation (EDA) who want to see how floorplanning was framed as a reasoning problem rather than a pure optimization problem.
• Engineers and researchers working on AI for chip design who want historical context for today's learning-based placement work.
• Historians of AI and expert systems interested in how knowledge engineering was applied to a real engineering domain in the 1980s.

You will get more from it if you already know the basics of digital IC design flow and have met terms such as cells, blocks, aspect ratio, routing channels and wire length. Some familiarity with rule-based systems will also help.

Before you start

No software setup is needed, but a little preparation pays off:

• Refresh the physical design flow: partitioning, floorplanning, placement, routing. Any introductory VLSI physical design text or lecture notes will do.
• Review what a rule-based or knowledge-based system is: a knowledge base of facts and heuristics, an inference mechanism, and a way of capturing an expert's know-how.
• Remember the era. The work predates today's EDA tools and modern deep learning, so judge its methods against the computing resources and design sizes of the mid-1980s.

How to read it

Because the book follows the four perspectives the author sets out, a sensible route is:

1. Start with the preface and introduction. They explain why floorplanning was treated as a design problem that benefits from expert knowledge, and they set out the overall plan of the work. Read these carefully; everything else hangs on them.
2. Read the strategy material next. Understand the overall floorplanning strategy before diving into individual techniques. Sketch it as a flow diagram as you go.
3. Work through the knowledge engineering material. Pay attention to what kinds of design knowledge are captured and how they are represented. This is the part that best rewards a modern reader interested in AI.
4. Treat the algorithms as a second pass. On a first reading, aim to understand what each algorithm is for and where it sits in the strategy. Come back for the details once the overall picture is clear.
5. Finish with the conclusions. Note which claims are about results and which are about future directions, then compare them with how the field actually developed.

If you are short on time, the introduction, the strategy and the knowledge engineering parts give you the core argument. EDA practitioners may prefer to read the algorithm material more closely and skim the knowledge engineering background.

Key ideas to take away

• Floorplanning is a design decision, not only a calculation. The book treats it as a task where experienced designers rely on judgment about data flow, control flow and block relationships.
• Expert knowledge can be captured and reused. A central theme is how a designer's heuristics can be turned into explicit knowledge a program can apply.
• Strategy and algorithms work together. Knowledge-based reasoning sets the overall approach, while specialized algorithms handle well-defined sub-problems.
• Good floorplans reflect good architecture. The preface recalls the idea that a well-designed circuit tends to have a clean, sensible floorplan, a useful lens for reading the rest of the book.
• History matters for AI in EDA. Many questions raised here, such as how to combine domain knowledge with search, reappear in current machine learning work on chip layout.

Practice and discussion questions

1. What makes floorplanning a good candidate for a knowledge-based approach compared with, say, detailed routing?
2. What kinds of designer knowledge does the book try to capture, and which would be hardest to write down as rules?
3. How does the overall strategy divide work between knowledge-driven reasoning and conventional algorithms?
4. Which assumptions in the book depend on the circuit sizes and technology of the 1980s, and which still hold?
5. If you rebuilt this system today, where would you use machine learning instead of hand-written knowledge, and where would you keep explicit rules?
6. How would you evaluate whether an automatically generated floorplan is "good"? List the measures you would use.

Pair it with

• VLSI physical design fundamentals. A current physical design textbook or course fills in the detailed placement and routing steps that follow floorplanning.
• Expert systems and knowledge engineering. Background on rule-based systems will make the knowledge engineering sections easier to judge. In our store, Industrial and Engineering Applications of Artificial Intelligence and Expert Systems (IEA/AIE-92) collects papers from roughly the same era.
• AI in design. Also on our shelf, Artificial Intelligence in Design '92 shows how researchers applied AI to design problems more broadly in the early 1990s.

Closing thoughts

This short monograph rewards a patient reader. Read it as a window into how AI researchers first tried to teach machines the craft of chip layout, and you will come away with a sharper sense of what today's tools inherited and what problems remain open. If it fits your research or your curiosity, get An AI Approach to IC Floorplanning on Amazon, or browse more titles on the BooksPlusAi shelf.

As an Amazon Associate, BooksPlusAi may earn from qualifying purchases.

Sources

• KIT Library catalog record: An Artificial Intelligence Approach to Integrated Circuit Floorplanning (Springer, 1991)
• Amazon product page (publisher description and preface excerpt)