标题: Bogatin:25年的10个经验教训 [打印本页] 作者: stupid 时间: 2010-10-7 09:49 标题: Bogatin:25年的10个经验教训 Ten lessons from 25 years of teaching electrical design.# o& t& {! ~& l3 j2 ?
' K+ f+ o; ]" o8 w0 [ I
I’ve been writing about, lecturing and teaching signal integrity topics for more than 25 years. I’ve taught more than 7,000 engineers U& s m) P3 a' q3 x
and personally mentored more than 30. Along the way, I’ve developed some sayings that crystallize important rules to consider when$ X6 n, \$ O$ p% H1 e3 F
working on signal integrity projects. Of course, these rules apply to more than just signal integrity.1 {3 b8 [6 G; j, B+ ?4 @6 ?( o
6 _, I$ [$ s- DI received a note from a recent student who jotted down my “rules.” He sent me a copy. This month, I thought I would share some of+ ?* J. t2 D4 e5 r* I& L8 Z9 d7 G
the list: + F$ z. Z" s' e( K2 }$ u+ `# e& X) {' s3 Z' Z! S
1.The most common answer to all signal integrity questions is “it depends.” And, the way to answer all “it depends” questions is! B! d. b% X! |' c5 M* k; ~
by “putting in the numbers” using rules of thumb, approximations, numerical simulations and measurements.( L- |8 b1 Z! C8 d
3 w' e& \1 D; G7 o n2.The way to separate myth from reality is by “putting in the numbers,” using rules of thumb, approximations, numerical ) T7 [) V; I$ f7 usimulation tools and measurements. All these processes are equally important and should be in the tool box of every engineer. ) G4 y* A9 T2 \! _- PThey each have a different balance between accuracy and cost to get an answer (cost as measured by time, money and % Y: n5 u) y# [/ D7 vexpertise). Use the process for each problem appropriate to your budget. : ^. Q0 d, R ~- T' L & J: u8 R6 L/ L8 w6 q. D5 u' f3.Watch out for the “Whac-a-Mole” effect. Often, changing one design feature to improve a performance metric has a negative : X) ~) J5 t0 jimpact on another performance metric. It is like the Whac-a-Mole arcade game. For example, bringing the signal path closer to h- _7 @; o5 |& h% wthe return path decreases ground bounce, but at some point, this will reduce the impedance of the interconnect and cause 3 M* ]5 h. K' H1 t0 A8 kexcessive reflection noise. 1 g0 E) q& }. k @9 B ; y# {# G# u4 J5 }. P4.The most efficient way to solve a signal integrity problem is to find its root cause. If you don’t know the root cause of a : x+ I0 K! E4 Pproblem, and it goes away when you try something, you have no idea if this was a coincidence or if this problem will creep , j' b5 k$ n5 g( `# @back in. ' F: H) L8 s e" ? 3 v# g: I1 M9 k5.Use the Youngman Principle to turn a root cause into a design guideline. This is named after Henny Youngman, a famous * v4 M; Q! E1 Z& [! zcomedian of the 20th Century. One of his jokes was, “A man goes into a doctor’s office and says, ‘Doctor, my arm hurts when I# X* n" R! o3 d' t
raise it. What should I do?’ The doctor replies, ‘Don’t raise your arm.’ ” If design feature A causes problem B, to eliminate 2 I9 g: y8 V4 A! Y* _6 pproblem B, eliminate design feature A. For example, if reflection noise is caused when the instantaneous impedance the signal $ U; A& K3 \7 O! R3 Y/ v- psees changes, engineer the instantaneous impedance to be constant down the entire interconnect.) i7 y" W7 ]$ ~9 }7 E5 b7 v9 y
! h5 k, F6 d3 Y( F6.Sometimes an OK answer NOW is better than a good answer later. You often have to make decisions without all the6 D% j; {4 Q5 D( e! A; C0 H
information you would like. This is where rough estimates are important. What is the bandwidth of an 800 Mbps DDR3 signal?2 ?) }0 G, L/ g8 {& q
It depends on the rise time, of course, but if you don’t know the rise time, do you sit and wait until someone can measure it? If 4 W/ \/ ^5 `1 N" {3 A8 [" uyou need an answer NOW!, you can use the rule of thumb that the bandwidth is about the 5th harmonic of the clock. The2 }% C' J: I- Q" g
clock is 400 MHz and the 5th harmonic is 2 GHz." B; J; J, q% y1 h1 Q% _) J' Q/ A
- X3 j/ x" U$ F8 g+ H! x$ O
7.Always evaluate the bang for the buck from a design change using a “virtual prototype.” This is a parameterized model for 2 w ?3 K5 R q: s3 z2 Dyour system, and a way of simulating its performance using this model. It will help you answer “what if” questions, and lets you ; J2 @3 S' \2 [* s1 f" Z) _measure the expected performance gain for the extra cost of a new material, design or component, before you commit to% g. v' _( R% z
hardware. ( {$ Z0 m$ B0 {3 z, {; R- T% z% Y7 v" N a) Z0 ?! W* J r# b
8.Watch out for “mink holes.” A rat hole is a convoluted path you detour down that takes away from the real goal. A mink hole is9 Z8 s: Y! p+ Z5 R K$ W2 v
a rat hole lined in mink: It feels really good while you are in it. Engineers love technical puzzles. Resist the temptation to track# { H) J6 \) _
down every little detail, or get that model bandwidth to just another few GHz. More important problems are always awaiting" G9 }1 u3 o4 @' H+ W
attention.! R! o w/ n5 K3 ^6 w
. q- V5 W$ Z3 Z9 E) o* M9.Never perform a simulation or a measurement without anticipating what you expect to see. If you are wrong, something is off in / ` m2 ^, R4 W6 h1 o4 a E( ^the problem setup, the tool accuracy, or your intuition. Either way, you will learn something by tracking down the discrepancy. : n% D5 S2 J! x7 eIf you are right, and you see what you expect, you get a nice, warm feeling that maybe you really do understand what is going n) [) A, d" r& w) i. e% k6 {: S
on.8 M | e9 b. t1 V3 J9 I4 e
4 N2 g; { R! Y" p, k, v
10.There are two kinds of engineers: those who have signal integrity problems and those who will. The corollary is, there are two 4 @/ ?9 I8 Q* |$ S0 P$ vkinds of designers: those who are designing antennae on purpose and those who aren’t doing it on purpose.作者: wowo1215 时间: 2010-10-12 15:34
再一次感叹：Bogatin太牛X了!作者: willyeing 时间: 2011-1-26 12:57
lz太牛了，纯英文，要向lz学习啊！！！作者: huangtrong 时间: 2011-2-9 15:28
牛！！！+ ]# \& \/ D: \; F& P# T
不过要让我们大多中国人能看懂吧作者: 徽之兴 时间: 2011-4-14 18:16
受教了！从方法论的高度总结了SI问题的处理原则，很好很强大！作者: xiechangqing 时间: 2011-4-18 09:21
可惜是E文的
惜是E文的