刚才忘传文章,对不起

jinht

Differential Trace Impedance


*-----------------------------------------------------*
               High-Speed Digital Design
                           
                 *On-Line Newsletter*
                           
          Dr. Howard Johnson, Vol. 5  Issue 2
*-----------------------------------------------------*

A RARE EVENT -- YOU MAY BENEFIT

   A rare, last-minute cancellation has opened a window
   in  my  schedule  for  private  seminars.  A private
   seminar  gives  you  and  your  co-workers a perfect
   opportunity  to  learn  the   latest  in  high-speed
   digital design within the context of YOUR designs.

   If  you  are  located in  the  Southwest US and have
   30  or  more  digital engineers at your site, please
   write   to  hsdd@sigcon.com   by   Feb 15  for  more
   information about this opportunity.
   

*----------------------(QUESTION)---------------------*

Differential Trace Impedance
Mitch Morey, San Diego, CA

   I'm  working  on  a board with 100-ohm  differential
   signaling   that  I  would  like   to   design   for
   microstrip  routing. I've used the Polar  Instrument
   calc,  the ADS LineCalc software, and have  got  two
   additional   stackup  constructions  for   our   fab
   houses, and have talked to numerous people on this.
   
   Here  are  the  recommendations I have  gathered  so
   far. All configurations represent
   
   - 100-ohm differential microstrips,
   
   - operating at 2.4Ghz speeds,
   
   - using a 5mil-thick FR4 dielectric,
   
   - with an underlying solid copper reference plane.
   
   
   
.005" lines with .005" edge to edge (fab shop 1)

.004" lines with .008" edge to edge (fab shop 2)

.005" lines with .008" edge to edge (ADS LineCalc)

.006" lines with .0065" edge to edge (ADS LineCalc #2)

.016" lines with .016" edge to edge (engineer #1)

etc. etc.

   
   
   ???**!!*!???
   

*--------------(REPLY FROM DR. JOHNSON)---------------*

   Thanks  for  your  interest  in  High-Speed  Digital
   Design.
   
   What  you need is a piece of software called a  two-
   dimensional   E&M   field  solver.    This   program
   calculates   the   magnetic  and   electric   fields
   surrounding   your  traces,  and  from   that   data
   extracts  the impedance and delay. This is the  best
   way  to  do  impedance calculations. The good  field
   solvers  allow  you  to  specify  the  trace  width,
   height,  spacing,  thickness,  dielectric  constant,
   AND  they  allow  you to overlay the  trace  with  a
   solder-mask layer.
   
   I'm  not sure what ADSlinecalc uses, but if it's not
   a  two-dimensional field solver you shouldn't  trust
   its  results. I have reason to distrust the accuracy
   of the examples you have provided.
   
   First  let  me  give you some general principles  to
   help  you understand what's happening and then  I'll
   rule out a couple of the solutions below.
   
   The  first  thing  you  need to  know  is  that  the
   patterns  of  electric field lines in  a  dielectric
   medium  follow  the  same  shapes  as  patterns   of
   current  flow  in  solid  conductors.  This   sounds
   pretty  obtuse,  but it's going to  help  you  in  a
   major  way,  because it will help you  SEE  what  is
   happening when you change the trace geometry.
   
   Follow  me  for a minute on this mental  experiment.
   Start  with a microstrip trace of length X.  I  want
   you   to  mentally  replace  the  dielectric  medium
   surrounding  this  trace with a slightly  conductive
   solid  material.  Now imagine that  you  connect  an
   ordinary  ohmmeter between the trace and the  ground
   plane.  The  value of DC resistance you  measure  in
   this experiment will be exactly proportional to  the
   IMPEDANCE  of the trace. I hope you can now  imagine
   what  would happen if you press the trace closer  to
   the  ground  plane. Can you see that  the  impedance
   must  go  DOWN,  because there is now less  material
   between  the trace and the reference plane?  If  the
   trace  is pressed down to the point where it  nearly
   touches  the  reference  plane,  its  resistance  to
   ground (i.e., impedance) approaches zero.
   
   What  about  doubling  the  width?  This  adjustment
   practically  doubles  the conductive  surface  area,
   substantially   lowering  the  resistance   to   the
   reference  plane  (i.e., impedance).  I  like  using
   this  DC  analogy because most engineers find  it  a
   lot  easier to imagine simple patterns of DC current
   flow  than  they  do high-frequency  electromagnetic
   fields.   The  constant  of  proportionality   isn't
   important -- I just want you to see what's going  to
   happen as you make various adjustments.
   
   So  far  I've  shown  two things that  decrease  the
   impedance in microstrips:
   
      (1) Moving a trace closer to the reference
            plane decreases its impedance.
                           
      (2) Fattening a trace (i.e., increasing its
            width) decreases its impedance.
                           
   And the converse statements are also true,
   
     (3) Moving a microstrip further away from the
       reference plane increases its impedance.
                           
      (4) Shaving down the trace width increases
                    its impedance.
                           
   Stripline  traces are a little more  complicated  in
   that  you  must account for the distance  from  your
   trace  to both top and bottom reference planes.  The
   general   result  for  offset  striplines  is   that
   whichever  plane lies closest to the trace  has  the
   most  influence  on  the  impedance.  Smack  in  the
   middle the planes are both equally important.
   
   Let's now imagine a differential configuration  with
   TWO  traces.  Connect the ohmmeter BETWEEN  the  two
   traces  (from one to the other). The resistance  you
   read  now  will be proportional to the  DIFFERENTIAL
   impedance  of two-trace configuration.  [NOTE:  one-
   half  the differential impedance is DEFINED  as  the
   odd-mode impedance.]
   
   If  your two traces are set far apart, and they have
   the  same  dimensions  as in the  first  experiment,
   your  new  differential measurement will be  exactly
   TWICE  as  great  as before. If  you  draw  out  the
   patterns  of  DC current flow you can see  why.  For
   *widely*  separated traces the current flows  mostly
   from   one  trace  straight  down  to  the   nearest
   reference plane, then it shoots across the plane  to
   a  position  underneath the second trace,  and  from
   there it leaks back up to the second trace. As  this
   current  flows, it encounters a resistance  R1  when
   leaving   the   first   trace,   practically    zero
   resistance  flowing  across  the  plane,  and   then
   another amount R1 as it flows back up to the  second
   trace. The total resistance encountered is 2*R1.
   
     (5) The differential impedance of two widely
      separated traces equals twice the impedance
           to ground of either trace alone.
                           
   Now  let's  see  what  happens to  the  differential
   impedance  as you slide the two traces towards  each
   other.  When  they  get  close  enough,  significant
   amounts  of  current begin to flow directly  between
   the  traces.  You  still get the same  old  currents
   going  to  and  from  the reference  plane,  but  in
   addition to those currents you have now developed  a
   new  pathway  for  current,  direct  from  trace-to-
   trace.  This additional current pathway acts like  a
   new  resistance  in  *parallel* with  the  original,
   widely-spaced  current pathways.  The  new  parallel
   pathway  lowers  the differential impedance  of  the
   configuration. You may conclude:
   
     (6) The differential impedance of a tightly-
     spaced pair is less than twice the impedance
           to ground of either trace alone.
                           
   If  the  traces are moved so close that they  nearly
   touch,   the   differential  resistance  (impedance)
   approaches   zero.  In  general,  the   differential
   impedance  is  a  monotonic function  of  the  trace
   separation.
   
        (7) All other factors being equal, the
     tighter the inter-trace spacing, the less the
                differential impedance.
                           
   I  view  any  decrease in impedance as  an  annoying
   side-effect  of close spacing. If I could  re-design
   the  universe,  I'd  try  to  make  it  not  happen.
   Fortunately,  you can counteract the  annoying  drop
   in  impedance  by  shaving down the  width  of  your
   traces.  If you shave off just enough width you  can
   push  the impedance back up to where it belongs.  In
   this  way, the trace separation and trace width  can
   be made somewhat interchangeable.
   
         (8) To maintain constant impedance, a
     reduction in spacing must be accompanied by a
      reduction in trace width (or an increase in
                    trace height).
                           
   With  these eight rules in mind, let's now  look  at
   the specific recommendations you have been given.
   
   With   your   5-mil   dielectric,   the   individual
   impedance  of  a 16-mil trace on FR-4 already  falls
   below  50  ohms, so the differential impedance  will
   be  less  than  100 ohms regardless of what  spacing
   you   use.  You  can  there  rule  out  the   16-mil
   configuration. I suspect your engineer #1  may  have
   been  thinking about using a thicker dielectric than
   what you propose.
   
   The  two  ADSlinecalc  results  conflict  with  each
   other. Staring from a pair 5-mil wide with an  8-mil
   space,  INCREASING the trace width to  6  mils  will
   lower  the impedance, and DECREASING the spacing  to
   6.5  mils will lower it even further. Therefore, one
   of  these results must be wrong. They cannot both be
   100-ohm  solutions. Therefore, I  suspect  something
   is  either  wrong with your copy of ADSlinecalc,  or
   (dare I say it) your use of the tool.
   
   Here's  some data from a commercial 2-D field solver
   (HyperLynx). All these combinations should give  you
   a  100-ohm  differential microstrip impedance  under
   the following conditions:
   
      - Dielectric thickness = 5 mil
      
      - Relative permittivity at 1 GHz = 4.3
      
      -  Trace  thickness = 1/2-oz cu  +  1-oz  plating
      (1.5-oz total)
      
      -  No  solder  mask   (***when your  vendor  adds
      solder  mask he or she will somewhat  reduce  the
      trace   width   to  compensate  for   the   extra
      dielectric material above the traces.
      
   Each  combination  is listed as  a  pair  [xW,  yS],
   where  x  is  the  finished, plated trace  width  in
   mils,  and  y  is the finished, plated  edge-to-edge
   separation in mils.
   
   100-OHM DIFFERENTIAL MICROSTRIP DIMENSIONS
   (    All at h=0.005", Er=4.3. 1.5-oz Cu.
   
   w     s
   8     14
   7     11
   
                        6     7
                        5     5
                           
   Whatever  you choose to do, insist that  your  board
   fabrication  shop place differential impedance  test
   coupons  on your panels and test each one to  verify
   that you are getting the correct impedance.
   
   Below   I   have  attached  a  listing  of   100-ohm
   stripline configurations. Obviously, this is just  a
   small sampling of all possible combinations. In  the
   chart,  dimension "b" is the interplane  separation,
   "h"  the  height  of  the trace  above  the  nearest
   plane,  "w"  the trace width, and "s"  the  edge-to-
   edge  separation,  all in mils. Parameter  rSKIN  is
   the  effective AC skin-effect resistance at  1  GHz,
   Z0  is  the (high-frequency value of) characteristic
   impedance, and the column marked "dB/in.  @1GHz"  is
   the  trace  attenuation due to the skin  effect,  in
   dB/in.,  at  a frequency of 1 GHz. Don't  forget  to
   also  consider dielectric losses. All the  stripline
   configurations  assume  1/2-oz  Copper  on  an  FR-4
   substrate with Er=4.3 at 1GHz.
   
   I hope this information is useful to you.
   

Best Regards,
Dr. Howard Johnson

   100-OHM DIFFERENTIAL STRIPLINE DIMENSIONS
   (    All at Er=4.3. 1/2-oz Cu.)
   
b     h     w     s     rSKIN     Z0     (dB/in.@1GHz)
10     3     3     40     3.501      99.03    0.1522
10     4     3     7      3.235     100.4     0.1389
10     5     3     7      3.218     101.2     0.1371
10     5     4     40     2.76       94.59    0.1258
14     4     3     5.5    3.191     101       0.1361
14     4     4     12     2.738     100.3     0.1177
14     5     3     4.5    3.136     100.1     0.135
14     5     4     7.5    2.599     100.5     0.1116
14     5     5     40     2.329     99.51     0.1011
14     7     3     4.5    3.112     101.8     0.1317
14     7     4     6.5    2.556     100.6     0.1096
14     7     5     13     2.236     100.9     0.09574
14     7     6     40     2.006      94.98    0.09126
20     5     3     4.4    3.141     101       0.134
20     5     4     6.5    2.592     100.7     0.1111
20     5     5     11     2.271     100.6     0.09746
20     5     6     40     2.087      98.41    0.0916
20     7     3     3.9    3.132     101.1     0.1336
20     7     4     5.2    2.547     100.9     0.1089
20     7     5     7      2.165     100.7     0.0929
20     7     6     10     1.908     100.3     0.08218
20     7     7     19     1.75      100.7     0.0751
20     7     8     40     1.613      96.34    0.07242
20     10    3     3.7    3.143     100.6     0.1347
20     10    4     5      2.54      101.6     0.1079
20     10    5     6.5    2.148     101.4     0.0915
20     10    6     8.5    1.876     100.7     0.08054
20     10    7     12     1.682     100.5     0.07236
20     10    8     25     1.562     100.3     0.06735
30     5     3     4.3    3.147     100.7     0.1347
30     5     4     6.3    2.595     100.7     0.1112
30     5     5     10     2.268     100.7     0.09729
30     5     6     22     2.087     100.2     0.08996
30     6     3     4      3.139     101       0.1339
30     6     4     5.4    2.564     100.7     0.1099
30     6     5     7.5    2.195     100.7     0.09415
30     6     6     11.2   1.955     100.6     0.08399
30     6     7     20     1.81      100.4     0.07795

*-----------------------------------------------------*

   Join our next public seminar:
   
           San Jose, Feb. 4-5, 2002.

   A  full  schedule of our public seminar  cities  and
dates  appears at http://signalintegrity.com.  Register
today!

   
   
   If  you  have an idea that would make a  good  topic
   for   a   future  newsletter,  please  send  it   to
   hsdd@signalintegrity.com.
   
   To  unsubscribe  from this list: send  an  email  to
   hsdd@sigcon.com with "unsubscribe"  in  the  subject
   line.  To  subscribe to this list: send an email  to
   hsdd@sigcon.com  with  "subscribe"  in  the  subject
   line.  Include  your name and email address  in  the
   body  of  the message. (NOTE:  If you received  this
   message, you do not need to re-subscribe.)
   
   Newsletter Archives:
   http://www.sigcon.com/newsletter.htm
   
   Copyright 2002, Signal Consulting, Inc.
   All Rights Reserved.