This section is not intended as a promotion of LVDS logic. It is a primer on how to interpret the specifications of any differential logic family.
LowVoltage Differential Signaling (LVDS) is a good example of a highspeed differential logic family. The LVDS standard was generated in 1995 by the IEEE. [62]
The LVDS standard contemplates both generalpurpose and shortrange applications. Here I will confine my remarks to the generalpurpose version of the standard. It defines a differential data path operating at data transfer rates in the range of 200 to 500 MHz with a sourcesynchronous clock and various bus widths up to 128 bits.
Table 6.5 lists some of the key performance specifications for LVDS generalpurpose transceivers.
6.13.1 Output Levels
LVDS being a differential logic family, there are two (complementary) outputs per logic signal. The nominal steadystate operating conditions for these outputs are 1.0 and 1.4 volts, for the low and high states respectively. When one wire goes to 1.0 volts , the other goes to 1.4, and vice versa.
Table 6.5. LVDS GeneralPurpose Link Specs ( adapted from ANSI/IEEE P159631995)
Transmitter specifications 


Signal 
Parameter 
Conditions 
Min 
Max 
units 
V oh 
Output voltage high, either wire 
R load = 100 W ± 1% 
1475 
mV 

V ol 
Output voltage low, either wire 
R load = 100 W ± 1% 
925 
mV 

V od 
Output differential voltage 
R load = 100 W ± 1% 
250 
400 
mV 
R 
Output impedance, singleended 
V cm =1.0V and 1.4V 
40 
140 
W 
V os 
Output offset voltage 
1125 
1275 
mV 

D V os 
Change in VOS between 0 and 1 states (this specification defines the AC commonmode output voltage) 
R load = 100 W ± 1% 
25 
mV 

t rise , t fall 
V od rise/fall time 20% to 80% 
R load = 100 W ± 1% 
300 
500 
ps 
Receiver specifications 

V i 
Input voltage range, either input 
V gpd < 925 mV 
2400 
mV 

V idth 
Input differential threshold 
V gpd < 925 mV 
“100 
+100 
mV 
V hyst 
Input differential hysteresis 
V idthh “ V idthl 
25 
mV 

R in 
Receiver differential input impedance 
” 
90 
110 
W 
C in 
Not specified 

Implementation specifications 

Pcb skew allocation 
Worst case 
50 
ps 
You may decompose this situation into a steadystate commonmode component of 1.2 volts, plus a changing differential voltage of ±0.4 volts. [57] The differential voltage varies from “0.4 V to +0.4 V, so we say that the differential peaktopeak voltage is 800 mV. The peaktopeak voltage on either wire alone would be 400 mV.
[57] Or an evenmode voltage of 1.2 volts plus an oddmode voltage of ±0.2 volts.
The standard requires that the steadystate differential voltage representing either a zero or one state be at least 250 mV, but no larger than 400 mV when the outputs are loaded differentially by 100 ohms . It also requires that no output ever fall below 925 mV or exceed 1475 mV under the same conditions.
Concerning output levels under other loading conditions, the standard provides little guidance. The only hints come in the form of stated constraints on output impedance. From the standard it is impossible to determine, for example, the output levels that would result from loading the outputs with a 75ohm differential load. Nor is it possible to determine something else that the IBIS community has long sought in these sorts of documents: a precise specification of the shape of the rising and falling edge.
POINT TO REMEMBER
6.13.2 CommonMode Output
The output offset voltage (DC bias) can change by 25 mV when switching from the one state to the zero state. This is a peaktopeak change. The AC amplitude range of the commonmode voltage is therefore ±12.5 mV.
The AC amplitude of the differential output voltage lies somewhere in the range of ±250 to ±400 mV.
When considering certain radiation and crosstalk problems, it is handy to know the approximate ratio between commonmode and differentialmode emissions coming out of a driver. For LVDS, the ratio of common to differential amplitudes can be as poor as 12.5/250 = 5%.
POINT TO REMEMBER
6.13.3 CommonMode Noise Tolerance
Looking at the commonmode operating range of the receiver (called V i in the specification), it is apparent that the receiver can tolerate inputs anywhere in the range of 0 to 2400 mV. From these numbers you can derive how much commonmode noise the system will tolerate. First assume a driver is at its lowest permissible level (925 mV). Now figure out how much commonmode noise you can add in the negative direction without the receiver input falling outside its guaranteed operating range. The answer is “925 mV. Next assume the driver is at its highest permissible level (1475 mV). Now figure out how much commonmode noise you can add in the positive direction without the input falling outside its guaranteed operating range. The answer is +925 mV. LVDS therefore tolerates a commonmode difference ( V gpd ) between the ground potential at the driver and the ground potential at the receiver as great as ±925 mV.
If you exceed the commonmode operating range of the receiver, all bets are off. It could do anything (see Section 6.7, "CommonMode Range").
POINT TO REMEMBER
6.13.4 DifferentialMode Noise Tolerance
Next let's look at the differential noise margin. In the worst case the transmitter differential output may be as small as 250 mV. At the same time, the receiver threshold may be offset by as much as 100 mV. The difference between these two figures is the differential noise margin, which is 150 mV. As a percentage of the signal swing on either wire (400 mV pp), the 150 mV figure represents 37%. A transmitter with a larger output swing would enjoy an even bigger percentage noise margin. These are excellent noise margins for digital logic. Most singleended logic families have noise margin percentages on the order of only 10% to 15%.
POINT TO REMEMBER
6.13.5 Hysteresis
A receiver with hysteresis has two input switching thresholds, one used for positivegoing signals and one used for negativegoing signals. The positivegoing threshold is always set a little higher than the negativegoing threshold. Once the input crosses into positive territory, the receiver automatically switches to the negativegoing threshold so that the signal must turn around and descend below the negative threshold before it can cause another switching event. Once the signal crosses below the negative threshold, the receiver flips back to using the positive threshold. A receiver equipped with hysteresis requires a certain amount of signal change before flipping to the other state.
The hysteresis feature is intended to avoid annoying selfoscillation that might happen with slowly changing, but very clean, inputs. The feature is usually implemented as a form of limited positive feedback from the receiver output back to own its input. With hysteresis, the inputs tend to switch quickly and firmly, once they reach an acceptable level, and then stay there. It's a good feature.
LVDS inputs have a guaranteed amount of hysteresis. You still, however, shouldn't supply any of these parts with a slowly moving input, because in that case as the input slowly sweeps through the transition region, any crosstalk that happens to exceed the hysteresis switching range will cause glitches (and therefore more noise) in the receiver.
POINT TO REMEMBER
6.13.6 Impedance Control
The LVDS specification goes to a lot of trouble to control ringing and reflections. This is one of the strongest provisions of the specification. Excellent control of ringing and reflections makes it possible to obtain firstincidentwave switching in most LVDS applications.
LVDS uses a bothends termination strategy to control reflections. Each LVDS transmission line is terminated first at the source and again at the end of the line. [58]
[58] Variants of LVDS are available in which the drivers can enter a tristate (highimpedance) mode, and the receivers do not incorporate terminations. These versions are suitable for building multidrop bus structures.
The source impedance of the driver is constrained to the range of 40 to 140 W . That is a ratio of only 3.5:1 from highest value of allowed output impedance to the lowest. To a boardlevel analog designer, this sounds easy, because you can go out and buy very accurate lumpedelement resistors. To a chip designer, it's a nightmare. You just can't control the absolute value of R DS (ON) or the absolute value of transconductance very accurately.
The achievement of a 3.5:1 output impedance specification represents a major accomplishment for the chip industry, one which I am sure will pay off in terms of higher volume, given the user friendly advantages of bothends termination (see box "BothEnds Termination").
The worstcase reflection coefficient at the transmitter, assuming it is coupled to a perfect 100 W transmission line, will therefore be the worse of these two numbers:
Equation 6.26
Equation 6.27
The input impedance of the receiver is constrained to the range 90 to 110 W . The LVDS specification recommends (but does not require) that this be implemented as a builtin terminator, placed inside the integrated chip package right at the die. From a signal integrity perspective, that would definitely be the best place to put it. Initial LVDS implementations, however, did not do this. Due in part to the difficulty of fabricating accurate onchip resistances, early implementations of LVDS left the 100ohm termination as an external component.
If you have to design with external terminations, use a 100 W ±10% external terminating resistor in a lowinductance package (0805 or smaller package) directly attached to the transmission line at the input terminals of the package, with very small pads (for low parasitic capacitance ). [59]
[59] Smaller components , if properly installed, have less parasitic series inductance and also less parasitic shunt capacitance, than larger ones. There is an electromagnetic theorem that says that if you shrink a configuration (component and layout together) in all three physical dimensions by a factor of k , then the inductance of that configuration also shrinks by a factor of k . If, as is common, you shrink the component length and breadth by factor k but do not change the height (i.e., the spacing from surface layer to the underlying reference plane), then the inductance shrinks, but not by a factor of quite as much as k (more like k 2/3 ).
The worstcase reflection at the terminator, assuming a line impedance of precisely 100 W , will be the worse of these two numbers:
Equation 6.28
Equation 6.29
Multiplying together the worstcase transmitter and receiver reflection coefficients, G MIN RO [6.27] and G MIN RIN [6.29], shows that the amplitude of any residual reflections in the transmission structure (meaning anything that arrives after the initial step edge) can in no case exceed 2.25% of the initial signal amplitude. Therefore, you can expect solid firstincident wave switching performance from this system, assuming a perfect implementation with perfect 100 W differential transmission lines.
You may be wondering how far the line impedance may stray from the ideal value of 100 W while still guaranteeing firstincidentwave switching. Figure 6.34 reveals the answer. This figure shows the magnitude of the residual reflections remaining after the arrival of the firstincident waveform. The figure is a compilation of four different constraint lines corresponding to different combinations of worstcase high and low R o interacting with worstcase high and low R in . At various specific values of trace impedance, one constraint or another takes precedence, which accounts for the segmented appearance of the curve.
Figure 6.34. Residual reflection after arrival of initial step edge for terminated LVDS logic with worstcase transmitter and receiver impedances, as a function of trace impedance.
The chart indicates that a line impedance of 100 W ±10 would produce an initial residual reflection no greater than 5% of the incoming step amplitude. A plusorminus 20ohm tolerance would increase the initial residual to no greater than 7%. LVDS logic, because it uses a bothends style termination, tolerates a fairly wide range of line impedances.
The existence of a significant residual reflection may not by itself endanger the performance of a particular link, depending on the polarity and timing of the arrival of the reflected signal power. Only timedomain simulation can tell.
This figure assumes the transmitter and receiver impedances are purely resistive. Any significant reactances at either the transmitter or receiver will further degrade the performance.
Both Ends TerminationThe bothendstermination strategy uses a source termination at the transmitter and an end termination at the far end of the line. This is what I call the axmurderer approach to reflection control. An ax murderer never takes just one whack at a problem. He keeps working and working on the problem until it is completely, finally solved . That's how bothends termination works. The main signal, once launched from the transmitter, proceeds toward the far end at full speed. Upon arrival, the end termination metes out one terrible, swift chop that reduces the incoming signal to little more than a small reflected bit of rubble . This surviving reflection retreats posthaste towards the safety of the transmitter, where the source terminator delivers another fatal blow. Only a very small remaining fraction of the original signal amplitude survives this doublechop to crawl back toward the far end a second time. The secondincident wave amplitude is the product of the original transmitted signal size, the reflection coefficient at the far end, and the reflection coefficient at the near end. In a reasonably welldesigned system this amplitude is small enough to simply ignore. That's the big advantage of bothendsterminated systems, there are hardly any reflections to worry about. Another advantage of bothends termination is its high tolerance for obstacles in the middle of the transmission line (like vias). Singleend terminated transmission lines have a lower tolerance for obstacles. In a singleend terminated line, there is always some reflection pattern that can take one bounce off the obstacle , and one additional bounce off some unterminated end, and wind up at the receiver. In a bothendsterminated design the reflection coefficient at each end of the line is small, so all reflection modes are damped, even ones that bounce off of obstacles in the middle of the line. 
POINT TO REMEMBER
6.13.7 Trace Radiation
In the LVDS specification the number D V OS defines the degree of balance between the two complementary outputs. It calls for a peaktopeak commonmode (or evenmode) content in the transmitted signal of no more than 25 mV. Compared to the peaktopeak signal level on either of the two signal wires (400 mV), that's a relative commonmode content of 6.25%.
The commonmode content limits the degree of attainable radiated field cancellation to a value of “24 dB (=20log(0.0625)). You can easily achieve this amount of cancellation at all frequencies up to 1 GHz by placing the differential traces at any separation of 0.5 mm or less (see Section 6.11.3). Unless you need to save the circuit board space, it is not, in this author's opinion, worth the effort trying to cram LVDS traces closer together than 0.5 mm.
In individual circumstances with particularly wellbalanced transmitters it is possible to get better cancellation, but you can't depend on always having parts that beat the specification.
POINT TO REMEMBER
6.13.8 Risetime
I'm glad to see a specification for the minimum risetime. That' s a big help when dealing with all manner of highspeed phenomena, especially the calculation of crosstalk. I offer my sincere thanks to all the standards weanies who voted for this provision.
6.13.9 Input Capacitance
The last of the receiver specifications is the input capacitance. Sadly, this specification is lacking. The closest we get in the standard to addressing the input capacitance is a vague statement that the input capacitance "should not limit the highfrequency, 250MHz operation of the receiver." That's nice, but it's not a specification. Standards like this leave open the possibility of receivers that meet the spec as written, but don't interoperate .
6.13.10 Skew
The LVDS committee did a lot of work on clocktodata skew. They carved out an overall skew budget, defining a permissible amount of skew for each signal in an LVDS link. Their budget assumes a link architecture that includes two pcbs, each with a connector and each plugged into some sort of backplane media.
In this architecture the specification pcb designers need to worry about is the pcb skew number of 50 ps. If every data and clock signal in an LVDS link is matched to within this amount of delay, the timing for the link as a whole should work.
Keep in mind that skew accumulates as your signal progresses. If your signal must traverse more than two connectors, the skew budget for each is less than in a simpler system.
If you are using an FR4 dielectric, the 50ps delay number gives you an allowance for about 1/4 inch of line length imbalance between any two signals in an LVDS link. This figure is definitely achievable, but don't depend entirely on your autorouter ”you need to take a close look at the final artwork to make sure you've stayed under the limit.
The LVDS specification does not make any specific reference to the degree of skew imbalance permitted between the two wires of an individual differential signal. My rule of thumb is that the skew imbalance in any differential pair should be kept to less than 1/10 of the risetime.
POINT TO REMEMBER
6.13.11 FailSafe
LVDS components from National Semiconductor include a failsafe circuit in the receivers. This feature shuts off the output in the event the input is disconnected (zero differential input). This feature is permitted by the standard, but not required, so check carefully if you will be mixing different vendors to make sure they all do it in a compatible fashion.
Figure 6.35 illustrates how failsafe is implemented in the National LVDS logic family.
Figure 6.35. Switching levels for National LVDS logic family.
The figure depicts the mandated differential thresholds, V IH and V IL , for an LVDS receiver. The manufacturer of the receiver in Figure 6.35 has created a part with better control over the input threshold than the mandated minimum. The actual thresholds, V TH (+) and V TH ( “), are specified at ±30 mV. The close tolerance of the actual thresholds is exploited to create the failsafe feature.
The failsafe feature is created by forward biasing the inputs. When the input is disconnected from any source (the transmitter is turned off or unplugged), biasing resistors R2 and R3 trickle enough current through the external endtermination resistor R1 to forwardbias the input by 50 mV. This level is above the actual component threshold, so the receiver output stays locked at 1.
When the input is connected to a source with a differential output impedance of 100 ohms, the current from resistors R2 and R3 forwardbiases the input by only half as much, or only about 25 mV, shifting V TH (+) and V TH ( “) to new worstcase values of +55 mV and “5 mV respectively. These values remain well within the mandated limits of ±100 mV.
Some applications require a greater margin of safety for the failsafe feature. For example, let's say you are making a twistedpair communication link. When the transmitter is powered off, you may expect more than 25 mV of differential noise. This can be implemented by adding two new resistors, R4 and R5, in parallel with the existing bias resistors R2 and R3, but outboard of the IC package. The new resistors can be sized to enforce an arbitrary amount of offset in the case the transmitter is disconnected. One disadvantage of this technique, if taken to an extreme, is that the failsafe bias current may be large enough to disturb normal operations.
The circuit in Figure 6.36 fixes this problem. In the event the transmitter is powered off or unplugged, the failsafe resistors R4 and R5 provide a large amount of bias current. In the event the transmitter is connected and powered on, you can pick values for R6 and R7 in the transmitter that will source an equal and opposite amount of current, canceling the offset. Resistors R2 through R5 appear in parallel with the differential impedance of the termination network and must be taken into account when selecting values for R1 and trace impedance. The same applies for resistors R6 and R7 at the source.
Figure 6.36. Combining external bias resistors at receiver and transmitter creates zero bias during operational mode.
POINT TO REMEMBER
For further study see: www.sigcon.com
Fundamentals
Transmission Line Parameters
Performance Regions
FrequencyDomain Modeling
Pcb (printedcircuit board) Traces
Differential Signaling
Generic BuildingCabling Standards
100Ohm Balanced TwistedPair Cabling
150Ohm STPA Cabling
Coaxial Cabling
FiberOptic Cabling
Clock Distribution
TimeDomain Simulation Tools and Methods
Points to Remember
Appendix A. Building a Signal Integrity Department
Appendix B. Calculation of Loss Slope
Appendix C. TwoPort Analysis
Appendix D. Accuracy of Pi Model
Appendix E. erf( )
Notes