Presentation - Hartley - Diff Pairs

Download as pdf or txt
Download as pdf or txt
You are on page 1of 40

The Truth about Differential

Pairs in High Speed PCBs


IPC DC - RTP Chapter
- PCB Carolina September 2nd, 2009
_____________________
Rick Hartley
L-3 Avionics Systems, Inc.
[email protected]

Read Books Not IC App Notes


Circuit Application notes
produced by IC manufacturers should be assumed
Wrong until Proven Right!
Lee W. Ritchey
2

- Transmission Lines A Transmission Line is any Pair or Wires or


Conductors used to Move Energy From point A
to point B
Usually of Controlled Size and in a Controlled
Dielectric to create a Controlled Impedance (Zo).

- Transmission Lines Capacitance is formed by 2 conductive surfaces separated by an insulator.


Control Electric Field in Transmission
Line by maintaining tight coupling
between the Trace and Return Path.
Inductance is property of a circuit
which allows Energy Storage in a
Field Induced by Current Flow.
Tight coupling between forward and
return path are secret to lowering
Inductance in Circuit.
4

- Transmission Lines 2 Layer Microwave Style PC Board L1- Routed Signal, routed Power and poured Ground copper.
L2- Ground.

Where does signals return current flow?


5

- Differential Pairs First used to transmit signals between bays of super


computers or bays / racks on the factory floor.
Single ended lines (shown here) connecting large
bays can have ground currents flowing even when
signals are not present, creating offset errors.
Bay 1

EARTH

Bay 2

V
6

- Differential Pairs In its basic form a Diff Pair consists of two


transmission lines, containing equal amplitude
but opposite polarity signals.
The Receiver of the Differential Signals is a
Crossing Detector -

- Differential Pairs -

When transmitted in a Twisted Wire Pair

Have tight containment of E and H Fields.


Lines use each other as a path for return current.
Have tightly controlled impedance.
Input ignores Unit-to-Unit Ground Offsets.
Input has Common Mode Rejection of RF Noise.

How does this apply to PC Boards?


8

- Differential Pairs Transmitted in a PC Board 2 Signals in 2 Transmission Lines.


Most of the Return Current for each line is in the
Gnd / Pwr Planes of PC Board.

Advantages vs Single Ended signals Input ignores Unit-to-Unit Ground Offsets.


Will still function even with 10 to 15 dB of loss.
Some logic families can tolerate 20 dB of loss.
Input has Common Mode Rejection of RF Noise
(if and only if noise is induced equally into both
lines).
9

- Differential Pairs Differential Impedance ZDIFF = 2 x ZODD


ZODD = Z0 - ZCOUPLING

Fields That
Determine Z0

Fields That
Determine ZCOUPLING

Advantages and Disadvantages to High ZCOUPLING


Use 2D Field Solver or Good Equation Based Tool
10

- Differential Pairs Typically Routed Together, Side-by-Side (Edge


Wise Coupled) or on Adjacent Layers (BroadSide Coupled), at Approx the Same Length.
Intent is to get same RF Noise on each Line.
Minimized Skew (Within Limits of Skew Budget).

Edge Coupled gives Greater Impedance Control,


Either Characteristic or Odd Mode.
Broad Side yields Higher Routing Density.
11

- Differential Pairs To illustrate the concerns over impedance control


in Broadside Coupled lines, Polar (in their Field
Solvers), shows the potential offset.

Typical shift of 5 mils can affect impedance by


10% or more (in addition to the normal +/-10%).
12

- Differential Pairs Should Traces be Tightly Coupled?


Reasons given for Tight Coupling Return current from one line travels in other line.
Reduces unwanted coupling from other signals.

Tight Coupling does create different line widths


in Pair than loose coupling.
Advantage: Narrower line per given impedance.
Disadvantage: Narrower line per given impedance.

Narrower line is easier to route but can be hard


to manufacture and harder to control impedance
during manufacturing . AND
13

- Differential Pairs ZCOUPLING example (24 long routes)Source: Polar Field Solver

4 mil lines - 6.5 mil separation 6 mils above plane - 100


7 mil lines - 14 mil separation 6 mils above plane - 100

Source:
Speeding
Edge

4 mil lines - 3.125 Gb/S

7 mil lines - 3.125 Gb/S

Difference? --- Skin Effect!


14

- Differential Pairs Tightly Coupled lines Require constant spacing


to maintain impedance close to 100 .
Separating lines to route through a 1 mm pitch
BGA pin field will increase impedance.
ZDIFF = 100
4 mil line - 6.5 mil separation 6 mils above plane.

ZDIFF = 131

39.4 mils
separation

ZDIFF = 108

39.4 mils
separation

ZDIFF = 100
7 mil line - 14 mil separation 6 mils above plane.

15

- Differential Pairs 4 mil Lines - 6.5 mils Separation -

16

- Differential Pairs 4 mil Lines - 39.4 mils Separation -

17

- Differential Pairs 7 mil Lines - 14 mils Separation -

18

- Differential Pairs 7 mil Lines - 39.4 mils Separation -

19

- Differential Pairs Tightly Coupled traces in the Differential Pair


work fine if Skin effect loss is not an issue (Routes are fairly
short, ie- Under 10 to 12 inches - or frequency of
circuit is under 1 GHz)
Lines do Not have to be separated during routing, to
weave through a pin field (this could be a problem
for BGAs with 1 mm or less pin pitch.
The Narrower lines are not TOO narrow and can be
fabricated, and impedance controlled without
difficulty.
20

- Differential Pairs Routing between planes Eliminates Coupling from External EM Fields.
Lowers Cross Talk, However ---Diff Pair
18 mils

A
6=H

B
6.5

Aggressor Line
H

Cross Talk from Aggressor to Line B is 12%.


Cross Talk, Aggressor to Line A is 2%.
Tight Coupling does NOT make Cross Talk even.
Common Mode Rejection will not happen!!!
21

- Differential Pairs With loosely coupled diff pair, Cross Talk from
Aggressor to Line B is still 12%. Cross Talk
from Aggressor to Line A is 1%.
Diff Pair
18 mils
6=H

A
7

B
14

Aggressor Line
H

Cross Talk in the loosely coupled pair is only


very slightly worse than the tightly coupled lines.
Secret to Cross Talk control- Keep Aggressor far
away from the Differential Pair.
22

- Differential Pairs Following table lists Cross Talk in the lines at


various increments of H (height above plane) Trace B
to Aggressor
2H
3H
4H
5H
6H

Coupling - A and
B - Tight Lines.
B= 3.5% - A= 1.1%
B= 2.0% - A= 0.5%
B= 0.8% - A= 0.3%
B= 0.3% - A= 0.15%
B= 0.1% - A= 0.08%

Coupling - A and
B - Loose Lines.
B= 3.5% - A= 0.3%
B= 2.0% - A= 0.16%
B= 0.8% - A= 0.12%
B= 0.3% - A= 0.08%
B= 0.1% - A= 0.04%

Around 5H the coupling Difference is Low


Enough that jitter from cross coupling will be
minimized. At 6H difference is virtually gone.
23

- Differential Pairs With Microstrip, the loosely coupled diff pair,


Cross Talk from Aggressor to Line B is 16%.
Cross Talk to Line A is 2.5%.
Diff Pair
A
6=H

Aggressor Line

B
14

Line A to Line B Cross Talk Difference in Microstrip is much more severe than Stripline.
Keeping Aggressor far away is even more critical
in outer layer transmission lines.
24

- Differential Pairs Following table lists Cross Talk in the Microstrip


lines at various increments of H Trace B
to Aggressor
1H
2H
3H
4H
5H
6H

Coupling - A and
B - Tight Line.
B= 16.0%- A= 4.3%
B= 7.0% - A= 2.7%
B= 3.6% - A= 1.8%
B= 2.7% - A= 1.5%
B= 1.8% - A= 1.1%
B= 1.3% - A= 1.0%

Coupling - A and
B - Loose Line.
B= 16.0%- A= 2.5%
B= 7.0% - A= 1.7%
B= 3.6% - A= 1.4%
B= 2.7% - A= 1.1%
B= 1.8% - A= 1.0%
B= 1.3% - A= 0.9%

Since Microstrip Cross Talk is greater, coupling


Difference is about the same at 6H as Stripline
Coupling at around 4H..
25

- Differential Pairs Cross Talk in Broadside Coupled Lines A


18 mils

Aggressor Line

B
7

6=H

In the Differential Pair above, coupling into


B is 12% and coupling into A is 8%.
2H
3H
4H

B = 4.5%
B = 2.0%
B = 0.8%

A = 3.0%
A = 1.4%
A = 0.6%
26

- Differential Pairs Length Matching (Timing) Requirements Lengths of the 2 lines to be the Same within
Limits set by Circuit Timing requirements.

Dont follow an App Note, which states to


route the two lines within 5 mils (or less)
of one another!
Look at circuit Timing!!!
27

- Differential Pairs Length Matching Simple solution Require that the two paths be length matched as
close as possible.
This usually results in extreme difficulty routing the
PCB. Take Designability into account to arrive at a
circuit that is both routable and fully functional.

Best solution Know how much timing skew is available within the
IC logic families used and at the circuit frequency.
Match lengths close enough to Satisfy Skew Budget!
28

- Differential Pairs Length Matching (Timing) Requirements Primary consideration, to keep Jitter to a minimum.
Jitter - Movement in time of the data crossing point
from bit to bit with respect to the data path clock.
Jitter is Lowest when 2 signals cross straight
portion of the rising and falling edges.
Limits of Linear
part of Rising Edge

Also Limits of
Timing Skew

All 3 sets of Waveforms will produce the Same


Timing Jitter .. approximately Zero(0)!!!
29

- Differential Pairs Signals below will still be detected as a proper


crossing event - But, Edges
that skew beyond limits of
Linear Signal Edge WILL
produce high Jitter!!!
Crossing event moves around in time, relative to
clock, due to uncertainty associated with Low
Slope portion of waveforms.
Jitter from this will be excessive and will likely
render the circuit unusable.
Stay within Skew Limits of rising signal edges.
30

- Differential Pairs Skew limits are set by the Fastest rise and fall
times of the signals at the Receiver.
Multiply fastest edge time by prop velocity in PC
board [6 mils(.15mm) / Ps in most dielectrics].
Result yields the Skew Budget Circuit Speed
256MHz
266MHz
1.2 Gb/S
2.4 Gb/S
3.125 Gb/S
10.0 Gb/S

Available Skew
390 pS
375 pS
160 pS
80 pS
62 pS
20 pS

Length in mils (mm)


2340 mils (59.4 mm)
2250 mils (57.2 mm)
960 mils (24.4 mm)
480 mils (12.2 mm)
372 mils (9.45 mm)
120 mils (3.05 mm)
31

- Differential Pairs What happens when designers get overly zealous


about length matching of transmission lines!!!

This was a wide, single ended bus, operating at approx 500 MHz,
with margin of +/- 0.75 inch, yet lines were held to +/- 5 mils length.
32

- Differential Pairs How does PCB Material impact Skew Mismatch?


With some styles of Fiberglass weave, the yarn pitch can
cause the 2 signal lines of the differential pair to sit on areas
of vastly different Er.

With 3, 4 & 5 mil lines, closely spaced, on a layer with


1080 or 106 style glass, one of the diff pair lines may sit on
an epoxy area and one on a fiber yarn.
33

- Differential Pairs Skew Mismatch in Materials (Continued) The two lines may see Er difference as large as 0.5
Since signal Prop Delay = Er / c (11.78 inch/ns), an Er
difference of 0.5 can cause skew of 10 ps / inch.
With Gbit switching rates, lines longer than a few inches
will see more skew than their budget allows.

Per previous slide, problem only occurs with loosely


woven fiberglass (ie - 106 or 1080).
With tight weave, Er differences average out over
very short distances.

34

- Differential Pairs Solution to Skew Mismatch in Materials 1) Do Not spec materials with loosely woven fabric, such as
106 or 1080 (This is nearly impossible when designing high
layer count or thin boards).
2) Have PCB fabricated with glass on a 45o bias, relative to the
diff pair lines. (Extremely expensive, as the panel (18 x 24,
etc,) will yield many fewer boards.)
3) Do Not route differential pairs in long, straight lines.
Change direction often (can be Very difficult).
4) Spec flat, tightly woven glass, such as the types from
Dielectric Solutions, LLC (Also low loss & low DK).

35

- Differential Pairs Concerns with minor timing mismatch There are a couple caveats to the skew issues
discussed If Differential Pair routes through a Non-Grounded
connector, signals must be aligned within a few pSec
or fields DO NOT couple tightly and noise or EMI
will likely result.
This is true whether routing board-to-board or into a
cable or twisted pair!

The second concern with misalignment of signals


has to do with possible Bit Error Rate issues.
36

- Differential Pairs When signals do Not cross at midpoint, there is


short time interval when current must flow into
or out of Vref terminal.
If Termination is a single 100 ohm resistor there
is no connection from the lines to Vref.
With No path to Vref, current is not available and
one signal edge slows down.
At low data rates (LVDS or 266MHz, etc), edge
degradation from slowing is not an issue.
If data rates are high (short bit intervals) this
issue can affect Bit Error Rate.
37

- Differential Pairs Certainly an issue at 2.4 Gb/S and higher rates.


Solutions Thevenin termination on each end of line (increases
power consumption and uses excess real estate).
Best solution, two 50 termination resistors, with
common pins connected through small capacitor to
Vref.
50
10pF
50

At gigabit rates and higher, may be necessary to


locate Terminations on die (to preserve signal
integrity). Look for these type ICs.
38

- Vias in Differential Pairs Via Discontinuity A via Stub (shown in pink) is a Capacitive Discontinuity when length is close to 1/4 (in DK)
of a signal. When distance for energy to travel
into the stub (green
arrow), Reflect and
travel back (Red
arrow) is 1/2 ,
180o reflections on
the line affect signal Jitter and Bit Error Rates.
This happens in a 0.100 Stub at 30 Gb/Sec.
(Image Source: Dr. Eric Bogatin)

39

- Vias in Differential Pairs Only discontinuities that are Electrically Long


will cause ringing in circuits Vias in line are NOT a discontinuity in MOST circuits!
If length of via section is shorter than 1/20 , via is ZERO
discontinuity (0.100 via section has no effect up to 60 Gb/s).
Vias longer than 1/20 are extremely small discontinuity, if
non-functional pads are removed.

Via stubs are only a discontinuity in very thick boards


AND at GHz frequencies.
A 0.100 stub is a problem above 15 GHz (30 Gb/s).
Back-drilling Eliminates Stub!!!
Removing Non-Functional pads lower capacitance 30 - 40%.
40

You might also like