|
|
Eric 大師在 DesignCon 2013 community 的這篇發文正好有說明Tighter coupling and Looser coupling 問題。. m. h/ I( `5 z2 l
http://www.designconcommunity.com/author.asp?section_id=2607: n. b$ X4 d d
9 u# h" N& w7 l* i9 y( E
Coupled Differential Pairs ：
2 u8 d9 \$ O$ E! uBefore you read on, think about your answer to the question, which is better for differential pairs: tight or loose coupling? When I teach classes to a large group, I always ask this question and the answers never cease to astound me. There are just as many strong proponents for each answer, and some of them are very adamant. When I ask why they choose one or the other, the answers fall into three categories:4 v5 M! Q% q0 X' E3 K: S
9 \7 @9 u( E: k7 w0 N: d
1. It’s what we’ve always done, it worked in the last design, it’s what our experts tell us, we changed it once nd the design didn’t work so we changed it back.
" w8 | H4 g/ ]' F: u; k2. Tighter coupling reduces cross talk, tighter coupling reduces EMC problems, tighter coupling makes it easier to match lengths.4 H6 v1 S6 f: k" Y4 o6 u
3. Looser coupling has higher bandwidth, looser coupling has less manufacturing variation, looser coupling is easier to route, looser coupling is easier to add compensation lengths.
" Q1 i# Z3 a4 xThe first category of reasons is never a good reason for a design rule. Every design is custom. It may not be true that because one design rule worked in the previous design that it will also work in the next design. In fact, the same design may not even work with just a die shrink change in one of the chips.
9 e( e" ?. ?3 p& l, E1 {# l0 @% W8 C6 X/ E) S- m
The real answer to the question of which is better, tight or loose coupling, has only one answer. The most common answer to all signal integrity problems is "it depends." The only way to answer "it depends" questions is to "put in the numbers" with rules of thumb, approximations, or numerical simulations.2 \5 h0 k/ |4 g0 K, a
9 D; |9 O5 D6 P+ E3 |) y
All engineers should be empowered with the knowledge, skill, and tools to answer "it depends" questions. At the very least, it’s important to know what the answer depends on. With experience and confidence, all engineers can answer "it depends" questions.; D E% v6 n. u
. q7 T" n7 i2 W, f( S5 mWhile all the reasons listed above for or against tight or loose coupling have some element of truth, when you put in the numbers, there are only two first-order factors driving the answer.* K: g8 ]* U% }; k& Z+ U7 e
' |. P3 p+ a+ C MFor microstrip or stripline transmission-line structures, tighter coupling will always result in higher interconnect density. That allows for fewer board layers or a smaller board size. {: g1 T6 `+ h& C
& c7 O5 a' w( t( q0 V9 C* M
If you care about cost, tighter coupling should always be your first choice. It’s not about cross talk, but about interconnect density.2 a }( H; L9 N
0 c2 U. o/ i& G; NThe downside to tighter coupling is that to achieve the target differential impedance, tighter coupled pairs requires a narrower line width than if all features were kept the same and the lines were pulled apart as loosely coupled. The downside of tighter coupling is higher resistive loss from the narrower line width, which means higher attenuation.
, ?. A/ I3 I, G I, i, T& d7 @0 P
$ q7 Z3 n( g6 p$ z" h6 D3 ^If you care about attenuation, you’ll want to consider loosely coupled lines.: _, Z* L5 o, I8 m( |0 [# p
0 {2 F t' M2 w! e6 b m- ]4 m2 I. ]
How much is the line width reduced by tight coupling? Only a 2D field solver can really answer this question." G0 k0 V4 W1 j9 Y" P+ }0 W
; ~+ V3 Y" `2 tAs the two lines in a differential pair are brought closer together, the differential impedance of the pair will decrease. To raise the impedance back to the target value, the line width needs to be decreased. Using a 2D field solver, we can literally map out design space for a differential pair: how the line width needs to change as the line to line spacing changes, to maintain a constant differential impedance.* P' P. i9 B$ g3 d' y1 k6 x
! _! x" M- {! k& c+ q0 Y2 SIn Figure 1, we show the design space for a stripline differential pair that was designed for 100Ohm target impedance when the two lines were far apart. The conditions were: total thickness between the planes of 13 mils, half ounce copper and Dk of 4. As long as you always walk this line (that’s why I refer to this as the “Johnny Cash” principle), your differential pair will always have a 100Ω differential impedance.7 b) S- f% s7 S7 q1 j
Between loosely coupled and tightly coupled pairs, the line width decreases from about 5 mils to 3.5 mils. This is a 30 percent decrease in line width, which would contribute to a 30 percent increase in conductor loss. When conductor loss dominates, this can contribute to a 30 percent decrease in a channel's interconnect signal bandwidth.5 E2 v+ X$ z* f/ [7 b' M
5 H6 C% v5 m. J3 M6 z. k
If loss is important, you’ll want to consider tweaking every design knob to decrease loss. That’s when loosely coupled lines should be considered.
, U$ `% |& ?3 g& i* D" |9 W
6 `" N* g+ ~+ x5 m# ?Before you make an important design decision, put in the numbers to see just how large or small the “bang for the buck” might be.
8 T, i$ ]6 P/ i( j1 q2 P, n |
|
/1 
发表于 2012-10-26 17:07
楼主