High Speed Digital Design: A Handbook of Black Magic (Prentice Hall Modern Semiconductor Design Series) [ハードカバー]

Second, I should caution young engineers that the authors of this book enumerate several stratagems in high-speed design; some good, some bad. That is, not all of the tricks in later sections are sound engineering practices. Experienced engineers will be able to differentiate between sound engineering practices and hacks, and when compromises should be made. Young engineers may be lead astray too easily.

Lastly, this book is a good book if you already know something of the subject. If you had only to buy one book, I'd recommend "High-Speed Digital System Design: A Handbook of Interconnect Theory and Design Practices" ISBN: 0471360902.

After reading that book, I'd purchase this book, as this book has some practical information, for example, on choosing capacitor dielectrics, oscillators, etc., not contained in the first.

To take one example (page 134,) Johnson purports to describe problems associated with a wire-wrapped prototype processor board containing TTL devices operating at high edge rates ( 2 ns.) According to Johnson, the design engineers failed to realize that the circuits would ring excessively, making the board unusable. To "prove" this he posits a model consisting of a 30 ohm TTL driver, with a 2 ns rise time, a 4" length of wire with 89 nH of self inductance, and a 15pf load - a series LRC circuit. Yes, this circuit will ring wildly, but the model is totally incorrect. The TTL input is not considered, which has a relatively low input impedance in the low state since it is current operated. This circuit -effectively a parallel LRC - does not ring nearly as much, as any experienced engineer knows. As a reality check, remember that wire wrap was successfully used for years by thousand of engineers. To listen to Johnson, though, this technology is almost unusable. Wire wrap circuits do ring, but under his example, the real amount of overshoot/undershoot is well within the limits of TTL. Further, no real circuit produces textbook looking traces, so the role of experience is to learn what worst-case design means, and what is acceptable for good manufacturing yield. Lesson: real experience teaches you how to produce correct, functional models. An incorrect model will cause you grief.

Much could have been done here, to be useful, by way of analysis and of recommendation. The wire should have been modeled as part of a transmission line, not as a lumped element, which any high speed digital design engineer would know, and the idea of terminating a transmission line should have been introduced. This is standard fare. Even with the series LRC, instead of deriving the formula for critical damping, he instead says: "This approximation (reduce Q to .5) is derived from the solution to a second order linear differential equation describing an RLC low pass filter. First find the point at which the derivative of the solution passes through zero (a maximum point) and then evaluate the solution at that point."

Got that? Take the derivative of a solution you want to find? Any book on circuits will reduce this to the solution of a quadratic equation. Obfuscating something that's really elementary does not help produce genuine insight. But this is what Johnson does throughout the book.

Isn't it simpler to say that if you have fast rise time signals, treat most connections as transmission lines, and add termination resistors? As for a series RLC, use the formula for critical damping: R = 1/2 (sqrt(L/C))