Toward the end of a long day walking and talking all over the floor at DesignCon 2003, I was participating in a four-way panel discussion organized by Henri Merkelo of atSpeed Technologies. The subject was high-speed serial links. We got the usual questions from the audience about how fast can we go, what technologies will we deploy next, and are we ready for optical backplanes. Then, out of the blue, someone asked a terrific question. For me, it made the whole panel worthwhile.
Terry Morris of Hewlett-Packard wanted to know: "In the present time, not the future, imagine you are the technical leader of a group designing a complete 10-Gbps system." He explained that you are given one chance, and one chance only, to address your team. Then you must go away and let them complete their work. What top directives, he asked, would you give to your team members?
I gave this question a great deal of thought. As a manager, you get to say lots of things to your team, but if you boil the scenario down to just the most important items, what would they be?
Morris' question reminded me of a question once posed to the famous physicist, Richard Feynman. Someone brought up the notion of a worldwide holocaust in which our descendants are reduced to the status of cavemen. Then, he asked Feynman, "If you could pass on only one technical concept to the survivors, to help them rebuild their society, what would it be?"
Feynman replied, "Everything is made of atoms." He felt that this one idea was so essential to the understanding of all physical phenomena—from chemistry to material science to biology—that it was the most helpful thing he could say to future scientists.
While I was thinking, the guy next to me on the panel stood up. He was a salesman from a major connector company. He began speaking, "Well, we just came out with this new backplane connector..." As he blathered on, I tried hard to think of something as basic, and as universal, as Feynman's answer. Here are the three rules that emerged:
- Measure everything. Take no specification for granted. This idea implies that you must have test equipment available to see and measure the performance of your hardware at the relevant frequency of operation.
- Sit with your layout people. (Actually, I said sit on your layout people, which is a little harsh.) Get to know them. Check in every day. Watch while they place your components. Layout specialists are resourceful and usually don't call if they run into trouble with the design; they just do something to fix what they perceive is a problem. That's why you need to be there.
- Make your hardware testable. Build in margining and testing features from the start that make it possible to verify the correctness of your design. If you need to probe a location, put some probe points on the layout. If you need a loop-back path for initial testing, design a special backplane that makes that scenario possible.
Merkelo offered these good points of his own, based on his company's trademark philosophy of measurement-based design:
- Education is crucial, particularly on the issue of total chip-to-chip performance. Make sure the team considers the overall performance of the system from chip to chip, and not just as isolated parts of the system, such as connectors and pc boards.
- Characterization and measurement procedures are always susceptible to certain inaccuracies. You must understand these errors.
- You must provide tools and training in how to use those tools for integrating the individual characterizations into a complete chip-to-chip framework.