Ieee Lvds 设计标准
Ieee Lvds 设计标准
3-1996
Sponsor
Microprocessor and Microcomputer Standards Committee
of the
IEEE Computer Society
Abstract: Scalable Coherent Interface (SCI), specified in IEEE Std 1596-1992, provides computer-bus-like
services but uses a collection of fast point-to-point links instead of a physical bus in order to reach far higher
speeds. The base specification defines differential ECL signals, which provide a high transfer rate (16 bits
are transferred every 2 ns), but are inconvenient for some applications. IEEE Std 1596.3-1996, an
extension to IEEE Std 1596-1992, defines a lower-voltage differential signal (as low as 250 mV swing) that
is compatible with low-voltage CMOS, BiCMOS, and GaAs circuitry. The power dissipation of the
transceivers is low, since only 2.5 mA is needed to generate this differential voltage across a 100 Ω
termination resistance. Signal encoding is defined that allows transfer of SCl packets over data paths that
are 4-, 8-, 32-, 64-, and 128-bits wide. Narrow data paths (4 to 8 bits) transferring data every 2 ns can
provide sufficient bandwidth for many applications while reducing the physical size and cost of the interface.
The wider paths may be needed for very-high-performance systems.
Keywords: backplane, bus, cable, differential, low-power, point-to-point, scalable, signal
Use of an IEEE Standard is wholly voluntary. The existence of an IEEE Standard does not imply that there are no other
ways to produce, test, measure, purchase, market, or provide other goods and services related to the scope of the IEEE
Standard. Furthermore, the viewpoint expressed at the time a standard is approved and issued is subject to change
brought about through developments in the state of the art and comments received from users of the standard. Every
IEEE Standard is subjected to review at least every five years for revision or reaffirmation. When a document is more
than five years old and has not been reaffirmed, it is reasonable to conclude that its contents, although still of some
value, do not wholly reflect the present state of the art. Users are cautioned to check to determine that they have the
latest edition of any IEEE Standard.
Comments for revision of IEEE Standards are welcome from any interested party, regardless of membership affiliation
with IEEE. Suggestions for changes in documents should be in the form of a proposed change of text, together with
appropriate supporting comments.
Interpretations: Occasionally questions may arise regarding the meaning of portions of standards as they relate to
specific applications. When the need for interpretations is brought to the attention of IEEE, the Institute will initiate
action to prepare appropriate responses. Since IEEE Standards represent a consensus of all concerned interests, it is
important to ensure that any interpretation has also received the concurrence of a balance of interests. For this reason
IEEE and the members of its technical committees are not able to provide an instant response to interpretation requests
except in those cases where the matter has previously received formal consideration.
NOTE — Attention is called to the possibility that implementation of this standard may require use of subject matter covered
by patent rights. By publication of this standard, no position is taken with respect to the existence or validity of any
patent rights in connection therewith. The IEEE shall not be responsible for identifying all patents for which a
license may be required by an IEEE standard or for conducting inquiries into the legal validity or scope of those
patents that are brought to its attention.
Authorization to photocopy portions of any individual standard for internal or personal use is granted by the Institute
of Electrical and Electronics Engineers, Inc., provided that the appropriate fee is paid to Copyright Clearance Center.
To arrange for payment of licensing fee, please contact Copyright Clearance Center, Customer Service, 222 Rosewood
Drive, Danvers, MA 01923 USA; (508) 750-8400. Permission to photocopy portions of any individual standard for
educational classroom use can also be obtained through the Copyright Clearance Center.
ii
Introduction
[This introduction is not a part of IEEE Std 1596.3-1996, IEEE Standard for Low-Voltage Differential Signals (LVDS) for Scalable
Coherent Interface (SCI).]
The demand for more processing power continues to increase, and apparently has no limit. One can usefully saturate
the resources of any computer by merely specifying a finer mesh or higher resolution for the solution to a physical
problem such as hydrodynamics or 3-D graphics. This demand leads engineers and scientists in a desperate search for
more powerful and faster computers.
To economically obtain this kind of computing power, it seems necessary to use a large number of processors
cooperatively. This cooperation is provided by the Scalable Coherent Interface (SCI), a high-speed packet
transmission protocol that efficiently provides the functionality of bus-like transactions (read, write, lock, etc.).
However, the initial physical implementations are based on emitter coupled logic (ECL) signal levels, which consume
more power than is practical in the low-cost workstation environment. The initial specification’s 1 Gbyte/s bandwidth
(16-bit data path) may be overly expensive in the workstation environment. It may be more cost effective to use a
narrower data path of sufficient bandwidth. The combination of a high-speed transmission environment and efficient
protocols can provide the link for multiple processors to cooperate in a low-cost workstation environment.
The initial developers of this standard came from the Working Group that developed the SCI protocol (IEEE Std
1596-1992). The ECL signal levels defined for the SCI were effective in getting the standard implemented quickly
and are practical for high-performance applications. They are less well suited, however, to using SCI in low-cost
workstations. The obvious low-cost solution is to integrate the transceivers into the controller and implement both
in CMOS. This integration will satisfy the space and power requirements of the workstation and personal
computing market.
Eventually, a lower voltage swing will be needed in order to get higher speeds than ECL signal levels can provide. This
standard can provide the basis for increasing parallel signal switching frequency into the gigahertz range.
Committee Membership
The following is a list of participants in the IEEE Project 1596.3 Working Group. Voting members at the time of
publication are marked with an asterisk (*).
iii
The following persons were on the balloting committee that approved this document for submission to the IEEE
Standards Board:
When the IEEE Standards Board approved this standard on 21 March 1996, it had the following membership:
*Member Emeritus
Also included are the following nonvoting IEEE Standards Board liaisons:
Paula M. Kelty
IEEE Standards Project Editor
iv
CLAUSE PAGE
1. Overview .............................................................................................................................................................1
1.1 Scope.......................................................................................................................................................... 1
1.2 Objectives................................................................................................................................................... 1
1.3 Strategies.................................................................................................................................................... 2
1.4 Design models............................................................................................................................................ 2
v
IEEE Standard for Low-Voltage
Differential Signals (LVDS) for Scalable
Coherent Interface (SCI)
1. Overview
1.1 Scope
This standard specifies a process-technology-independent low-voltage (less than 1 V swing) point-to-point signal
interface as optimized for IEEE Std 1596-1992 [B1],1 which uses a differential driver connected to a terminated
receiver through a constant-impedance transmission line. The interface will be optimized for CMOS processes, while
being compatible with other IC processes, including GaAs and BiCMOS. The specification should support a transfer
rate of at least 200 mega-transfers/second.
In addition, the specification will define encodings for transporting SCI packets over narrow and wide data paths
(4-, 8-, 32-, 64-, and 128-bits, rather than the 16 bits defined by IEEE Std 1596-1992) using these signals.
1.2 Objectives
The primary goal of this standard is to create a physical layer specification for drivers and receivers and signal
encoding suitable for use with the SCI as specified by IEEE Std 1596-1992 in low-cost workstation and personal
computer applications. Other objectives include the following:
Technology independence. Specifications should allow designs to be implemented in a variety of integrated-
circuit technologies.
CMOS compatible. Signal voltage levels and other specifications should be compatible with digital CMOS
processes operating from 2 V through 5 V power supply levels.
Backplane and cable applications. Specifications should be optimized for connections between boards
contained within one chassis and short (less than 5 m) chassis-to-chassis interconnects. Longer connections
are not prohibited, provided they meet specified signal loss and ground shift criteria for proper receiver
operation. Connector and cable specifications are beyond the scope of this standard.
Scalable. The original 16-bit-wide SCI data path should be supplemented by 4- and 8-bit-wide data paths to
support a variety of cost/performance ratios. Support for 32-, 64-, and 128-bit-wide data paths will also be
addressed.
1
The numbers in brackets preceded by the letter B correspond to those of the bibliography in annex A.
1.3 Strategies
The basic design strategies selected by this standard include the following:
Low-voltage swing. To minimize power dissipation and enable operation at very high-speed, low-swing
(400 mV maximum) signals are specified.
Differential signals. Small signal swings require differential signaling for adequate noise margin in practical
systems.
Self-terminated. To minimize board real estate and costs, and to maximize clock rates, each receiver is
assumed to provide its own termination resistors.
Uniform ground. The standard assumes that the ground potential difference between driver and receiver is
kept small by the system design. The mechanism for constraining the ground potential difference is beyond
the scope of this standard.
The most controversial decision was to use differential signals, which at first appears to double the number of signal
lines. The pin-count overhead is actually much less than this, since reliable single-ended schemes require many more
ground signals (many high-speed chips and/or backplanes provide one ground for every two signal pins) and run
significantly slower. Other design benefits associated with differential signals include the following:
Constant driver current. The transmitter consumes a (near) constant current when driving the links; the
current remains the same, but is routed in the opposite direction when the signal value changes. This
simplifies the design of power-distribution wiring.
Constant link current. The net signaling current in a differential link is (nearly) constant, which greatly
simplifies system design. The links are unidirectional and transmitters always drive a differential signal per
table 2–1 or table 2–2. Reversing or stopping links would cause the net common-mode signaling current to
change, creating system noise.
Low power. A low signal current can be used, since much of the induced noise and ground-bounce appears as
a common-mode signal.
Simple board design. Although differential signals must be carefully routed on adjacent matched tracks, they
are usually less sensitive to imperfections in the transmission line environment.
Low EMI. Differential signals minimize the area between the signal and the return path. In addition, the equal
and opposite currents create canceling electromagnetic fields. This dramatically reduces the electromagnetic
emissions.
Low susceptibility to externally generated noise. Though these links generate little noise, other parts of the
system may. Differential signals are relatively immune to this noise.
The SCI-LVDS link model assumes unidirectional operation (the driver always at one end of the link, the receiver at
the other), and that a clock signal is sent along with the data as though it is just another data bit.
Both edges of the clock are used to delimit data, so the maximum transition rate of the clock is the same as the
maximum transition rate of the data signals. This clock flows through the link at the same velocity as the data, and is
to be used as the time reference for sampling the data.
In most applications, the received sampled data will need to be synchronized to the receiver’s local clock. If the
transmitter’s clock and the receiver's clock are independent, and thus perhaps at slightly different frequencies,
occasional symbols will need to be inserted or removed from time to time in an elasticity buffer in order to maintain
synchronization.
The transmission system shall ensure that the setup and hold requirements of the receiving latches are met, in order to
avoid incorrect data sampling and triggering metastable states. The receiver can observe the timing of the received
clock relative to its own clock in order to choose an appropriate sampling time.
By carefully adhering to these assumptions, the SCI signaling protocol becomes independent of distance or delay. The
maximum distance is limited by signal skew, caused by slight differences in propagation velocity from one signal to
another, and by attenuation and distortion of the signals.
Because these signals are unidirectional, it is relatively easy to reshape and time-align them in order to transmit them
greater distances. However, this may introduce timing jitter, which can make it impossible for the receiver to anticipate
clock transitions with sufficient accuracy for reliable operation.
In addition to extending the signal encoding to parallel widths not included in IEEE Std 1596-1992, this standard
specifies driver and receiver parameters only. However, a system must interconnect these components to be useful. The
interconnect termination is specified in the receiver portion of this standard. The interconnect is beyond the scope of
this standard because of the many options possible. The interconnect could include bond wires, packages and pins,
printed circuit board, cables, connectors, multi-chip modules, wafer scale integration, or any combination of the
previous options in one driver-to-receiver signal path. This signal path is important to the correct operation of a system
implementing LVDS signals and is therefore discussed in general terms in this standard.
At the high data rates this standard supports, it is important to consider the transmission line aspects of the signal path.
The high-frequency components of the 300 ps transition times make the parasitic reactive signal path components
important. Familiar concepts, such as the receiver input capacitance, are overshadowed by the parasitic inductance of
signal path elements that shape the waveform. If the signal delay through a signal path section is greater than the
allowed minimum transition time, 300 ps, that section must be analyzed as a transmission line with associated
characteristic impedance and delay. Impedance discontinuities through connectors, pins, solder pads, and bond wires
to the IC itself cause reflections that degrade the signal integrity.
The receiver and its package input impedance need to match the signal transmission line impedance. This serves to
minimize noise-causing reflections that create data errors. Given typical CMOS process tolerances, this generally
implies the use of active devices to adjust the terminating resistance until it matches an external reference. Integrating
the terminating impedance onto the receiver chip complicates the design and manufacturing, but the trade-off is
simplified board layout and better signal integrity.
2. Document notation
Several key words are used to differentiate between different levels of requirements and options, as follows:
2.1.1 expected: A key word used to describe the behavior of the hardware or software in the design models assumed
by this specification. Other hardware and software design models may also be implemented.
2.1.2 may: A key word that indicates flexibility of choice with no implied preference.
2.1.3 shall: A key word indicating a mandatory requirement. Designers are required to implement all such mandatory
requirements to ensure interoperability.
2.1.4 should: A key word indicating flexibility of choice with a strongly preferred alternative. Equivalent to the phrase
is recommended.
Many bus and interconnect-related technical terms are used in this document. These terms are defined below:
2.2.1 backplane: A board that holds the connectors into which SCI modules can be plugged. In ring-based SCI
systems, the backplane may contain wiring that connects the output link of one module to the input link of the next.
Usually the backplane provides power connections, power status information, and physical position information to the
module.
2.2.2 board: The physical component that is inserted into one of the backplane slots. Note that a board may contain
multiple nodes.
2.2.3 byte: Eight bits of data. Syn: octet.
2.2.4 differential voltage signal: The voltage difference between the true and complementary signals from a driver
with two single-ended outputs whose signals always complement each other. Differential signals are also referred to as
“balanced signals.”
2.2.5 driver: An electrical circuit whose purpose is to signal a binary state for transmitting information. Also referred
to as a “generator” in international standards.
2.2.6 flag: A signal used to delimit packets in parallel-signal-transmission implementations.
2.2.7 ground potential difference voltage: The voltage that results from current flow through the finite resistance and
inductance between the receiver and driver circuit ground voltages.
2.2.8 idle symbol: A symbol that is not inside a packet and is therefore not protected by a CRC. Idle symbols serve
to keep links running and synchronized when no other data are being transmitted. The idle symbol also contains
flow-control information.
2.2.9 jitter: Refers to the time-uncertainty of a transitioning edge recurring in a repetitive signal. This uncertainty is
only with respect to other edges in that signal. Jitter is commonly measured using random bit patterns and
accumulating an eye pattern to show the worst-case difference in transitions.
2.2.10 LVDS: An abbreviation for low-voltage differential signal.
2.2.11 offset voltage: The driver offset voltage is the average dc voltage generated by the differential driver;
Vos = (Voa + Vob) / 2.
2.2.12 packet: A collection of symbols that contains addressing information and is protected by a CRC. A subaction
consists of two packets: a send packet and an echo packet.
2.2.13 physical interface: The circuitry that interfaces a module’s nodes to the input link, output link, and
miscellaneous signals.
2.2.14 receiver common-mode voltage: The combination of three components: 1) the driver-receiver ground
potential difference (Vgpd); 2) the longitudinally coupled peak noise voltage measured between the receiver circuit
ground and the signal transmission media with the driver end shorted to ground (Vnoise); 3) the driver offset voltage.
2.2.15 receiver differential noise margin high: The tolerable signal voltage variation from any source that still
results in the receiver producing a logic high output state when the driver is stimulated by a logic high input.
Differential noise margin high is calculated by subtracting the receiver’s minimum differential high input voltage from
the driver's minimum high differential output voltage; Vodh(min) – Vidh(min).
2.2.16 receiver differential noise margin low: Tolerable voltage variation to guarantee that the receiver produces a
logic low output when the driver is stimulated by a logic low input; Vidl(max) – Vodl(max).
2.2.17 SCI: See:Scalable Coherent Interface
2.2.18 Scalable Coherent Interface (SCI): An abbreviation for the Scalable Coherent Interface standard, IEEE Std
1596-1992 .
2.2.19 signal line: An electrical or optical information-carrying facility, such as a differential pair of wires or an
optical fiber, with associated driver and receiver, carrying binary true/false logic values.
2.2.20 skew: The difference in time that is unintentionally introduced between changing signal levels (incident edges)
that occur on parallel signal lines. This difference results in an uncertain position with respect to time among parallel
signals.
2.2.21 symbol: Refers to data within an SCI packet. A 16-bit unit of data accompanied by flag information. The flag
information may be explicitly present as a 17th bit, or implied by the context. Symbols are transmitted one after
another to form SCI packets or idles. The particular physical layer used to transmit these symbols is not visible to the
logical layer.
2.2.22 sync packet: A special packet that is used heavily during initialization and occasionally during normal
operation for the purpose of checking and adjusting receiver circuit timing.
3. Electrical specifications
A LVDS interface, (figure 3-1), has a low-voltage swing (400 mV single-ended maximum), is connected point-to-point,
and achieves a very high data rate (500 Mbits per second per signal pair) and reduced power dissipation. Power is low
because signal swings are small: a minimum of 2.5 mA are sent through a 100 Ω termination resistor. This sharply reduced
power dissipation enables an important advance: integrating the line termination resistors, interface drivers and receivers,
and the processing logic in the same integrated circuit.
Switching speed is high because the driver load is an uncomplicated point-to-point 100 Ω transmission line
environment. Switching speed is also high because interface devices are all on the same piece of semiconductor
material, reducing the skew due to process, temperature, and supply variations between signal pairs. Connected in
serial or parallel pairs, the LVDS interface forms a link used to transfer packets between integrated circuits, such as
SCI nodes. For example, figure 3-2 shows circuit boards with LVDS links connected in a ring. The ring is implemented
on a printed circuit board (PCB) similar in mechanical function to a multidrop bus backplane. The difference is that
fewer PCB layers are needed to make the point- to-point connections. The PCB is simplified by eliminating the
multidrop bus lines, as there is no need to route around interlayer vias used to make mechanical and electrical
connections.
The data path can be serial or parallel with 1-, 4-, 8-, 16-, 32-, 64-, or 128-bits, depending on the needs of the user
(see annex A, Signal encoding, for all widths except 1 and 16, which are defined in IEEE Std 1596-1992, clause 6,
Physical layer).
Electrical specifications and skew specifications are optimized for 2–5 V supply voltages. The full range of
semiconductor process technologies can be used to implement LVDS. It is intended that the specification be
interoperable for all these technologies. The rapid trend toward reduced power supply voltage was considered in
providing for signals that can be compatible with future system requirements.
The physical environment of point-to-point connections between circuit boards is further divided into two categories.
The first (a general purpose link) is for circuit boards that need to operate with tolerance for Vgpd (table 3-1). This tolerance
(approximately ± 1 V for a 2.5V powered system) is for a general purpose system. The second (a reduced-range link)
is for boards mounted on a PCB or similar environment that will guarantee less than 50 mV Vgpd (table 3-2). In this
environment, the differential signal is reduced by reducing the driver current. This reduces the power at both driver and
receiver. This is a special consideration for subsystem implementations such as IEEE Std 1596.4-1994 RamLink.
The backplane environment implies short interconnects with controlled Vgpd. The use of cables implies that all the
skew and signal quality requirements will be met by the cables, and the system designer will account for the worst-case
Vgpd and provide appropriate safeguards. The scope of the electrical specification is the differential interface of drivers
and receivers. The transmission media specification, whether cables or printed circuits, is beyond the scope of this
standard.
The driver output, when properly terminated, results in a small-swing differential voltage. The relation between the
single-ended outputs and the differential signal is shown in figure 3-3. The differential driver is made up from two
single-ended outputs. These outputs alternate between sourcing and sinking a constant current. The differential voltage
level is determined by the load resistance. The dc load seen by the driver is the receiver input impedance in parallel
with the differential termination, 100 Ω, which dominates. The case where the current source is providing 4 mA is
shown in figure 3-3, where the outputs are switching the current at a 50% duty cycle.
The receiver threshold limits are shown in figure 3-3, in relation to the single-ended signals that arrive at the receiver
inputs. When the magnitude of the voltage difference exceeds the receiver threshold, then the receiver is in a
determined logic state. For the purpose of this standard, a differential voltage greater than or equal to Vidth(max) is a
logic high, and less than or equal to Vidth(min) is a logic low.
Ground shift margin is built in by confining the output to a range of Vol to Voh (e.g., this allows approximately 1 V of
ground shift between a driver and receiver that are powered from 2.5 V supplies). Therange of allowable dc output
levels for driver output voltages Voa and Vob is illustrated in figure 3-4. Measurement of the voltages Voa, Vob and the
differential output voltage Vod is illustrated in figure 3-5.
The driver output shall always be terminated in compliance with this specification. The unterminated driver output
voltage shall not exceed 2.4 V. Note that the receiver may be exposed to the unterminated driver output voltage briefly
when a cable from the driver is being connected to the receiver—the cable will be charged to the unterminated driver
output voltage.
Figure 3-3 —Maximum driver signal levels shown for 1.2V Vos
The following driver dc output voltage limits refer to figure 3-4 and figure 3-5, and shall apply for a load resistance
R = 100 Ω connected as shown in figure 3-5.
Ideally, the amplitude and common-mode voltage of the steady-state differential output would not change, but in
practical designs, both change. The output of a driver whose differential voltage (Vod) and driver offset voltage (Vos)
change when the output changes state is shown in figure 3-6. The definition for Vod and Vos is shown in figure 3-7.
The definition of ∆Vos and ∆Vod are explicitly stated by taking into account the varying voltage levels of the single
ended outputs when in the different logic states. This can be expressed by equation 1 and equation 2.
where
∆V od = V od ( high ) – V od ( low )
(2)
where
The driver dc output voltage limits refer to figure 3-6 and figure 3-7, and shall apply for a load resistance R = 100 Ω
connected as shown in figure 3-7.
To ensure that the driver circuit does not damage itself or other parts of the electronics, limits on the output currents
when shorted mutually and to ground are imposed.
When the driver output terminals are short-circuited to the driver circuit ground, neither current magnitude(Isa or Isb)
shall exceed the specified value in table 3-1 or table 3-2 as appropriate. The test circuit is shown in figure 3-8.
When the driver terminals are short-circuited to each other, the current magnitude shall not exceed the specified value
in table 3-1 or table 3-2 as appropriate. The test circuit is shown in figure 3-9.
A difference between driver output impedance and signal path impedance causes reflections of incident edges arriving
at the driver output from the transmission media. These waves that oppose the signal direction come from two sources:
reflected signals and common-mode noise coupled onto the interconnect. To prevent common-mode noise reflected
from the driver output from becoming a differential signal, the output impedance of the inverting and non-inverting
outputs should be closely matched.
The upper limit of the output impedance should be as close to the transmission line impedance as possible. The lower
limit on output impedance should not be much less than the impedance of the interconnect. An output impedance
significantly less than the interconnect impedance will generate negative differential reflections. The upper limit can
be above the interconnect impedance; however, large reflections will cause ringing and noise problems on the lines.
The amplitude of the reflected signal is the product of the incident wave amplitude and the reflection coefficient (ρ),
as specified by equation 3.
Vreflected = ρ × Vincident
(3)
The reflection coefficient is determined by the transmission line impedance and the driver output impedance, as
specified in equation 4.
Zd – Z t 2 ( Zd )
ρ = ----------------- = ----------------- – 1
Z d + Zt Zd + Z t
(4)
where
Zt is transmission line impedance and Zd is driver impedance. It follows that reflections less than 10% of the incident
wave will result when equation 5 is satisfied.
2Z t ( Z outa – Zoutb )
0.1 > ---------------------------------------------------------
( Z outa + Z t ) ( Z outb + Zt )
(5)
where
Zouta and Zoutb are the respective output impedance of the complementary differential drivers.
Figure 3-10 illustrates a means of measuring the reflected voltage difference due to mismatched driver output
impedance. The driver would be tested when driving differential high and then driving differential low. This is a means
to test the dynamic output impedance, which may differ from the static output impedance. The dynamic impedance is
important in the very high frequency operation for which the driver is designed to work. Since it is difficult to test the
dynamic output impedance in a production environment, this parameter can be tested, verified, and guaranteed for a
design.
Figure 3-11 illustrates a means of measuring the static driver output impedances, which have min/max as well as
mismatch constraints. For each of the two possible output signal values (0 and 1), the output voltages shall be
measured with driver-load common-mode (Vcm) voltages of 1.0 V and 1.4 V, yielding measured voltages of Voalo,
Voblo, Voahi, and Vobhi.
The values of these output voltages are based on the absolute and relative impedances of the drivers. The difference in
a single output voltage (for 1.0 V and 1.4 V output-load voltages) is affected by the value of the driver’s output
impedance on this signal. The difference in the differential output voltages (for 1.0 V and 1.4 V output-load voltages)
is affected by the matching of the driver's a and b output impedances. These values shall be within the constraints
specified in table 3-3.
10% matching of
|(Voahi–Vobhi)−(Voalo–Voblo)| mV 0 20
reflections coefficients
The driver output leakage currents (Ixa and Ixb) are measured under power-off conditions, Vcc=0 V, as shown in
figure 3-12. With the voltage on the driver output terminals between 0 V and 2.4 V, with respect to driver common,
these currents shall not exceed the value specified in table 3-1 or table 3-2.
The receiver input signal is measured differentially, figure 3-13. The receiver output state is determined by a
differential input signal greater than +Vidth or less than −Vidth, within the permitted Vi range. The table 3-1 and table
3-2 receiver specifications are the same except for the common mode operating range and the on chip termination
tolerance. The receiver common mode voltage input range, over which it must meet all other specifications, is reduced
in table 3-2 because that is intended for operation in a well-controlled environment. The termination is allowed to have
greater variance because it is intended to operate in a more controlled environment with less common-mode noise. For
simplicity, the remaining receiver specification discussion here will apply directly to the general purpose specification.
The analogous explanation will apply to the table 3-2 specification by substituting the appropriate numerical limits.
The ability to accept voltages outside a Vi range is desirable because it increases noise immunity to ground potential
difference and interconnect-coupled noise. The upper limit to the differential swing is given to ensure that receiver
skew specifications are maintained for this range of input signals throughout the receiver common mode range. The
range of allowable dc input levels for receiver input voltages, Via and Vib, is illustrated in figure 3-14. Measurement of
the voltages Via, Vib, and the differential input voltage Vid is illustrated in figure 3-15.
The signal common-mode level for table 3-1 is shown in figure 3-14. The receiver common-mode input voltage, Vicm,
will be an alternating voltage depending on three superimposed conditions: the driver output condition, voltages
induced on the interconnect, and reflections caused by common-mode termination imperfections.
This voltage can be expressed by accounting for the varying levels during both logic states. Equation 6 expresses the
relationship of the four input voltages resulting from the three conditions previously stated.
V ia + Vib
Vicm = ---------------------
2 (6)
A differential termination resistor (connected across the receiver inputs) should be integrated onto the receiver die.
Integration is feasible because the power dissipated in the termination is less than 2 mW per receiver, and a single
value can match the point-to-point interconnect characteristic impedance. Since an alternate termination scheme
would be more useful in certain applications, there are modifications allowed to this requirement. For example, a
common-mode termination to 1.2 V would be desirable when the ground potential difference between receiver and
driver cannot be guaranteed to meet the limits specified in table 3-1. This allows the ground potential difference to be
divided between the driver and the receiver. This common-mode termination can still use integrated termination
resistance values of 50 Ω from each input to a common externally accessible pin. This does require an additional pin
as shown in figure 3-16; however, all parallel channels can bus this common-mode termination together. Figure 3-16
shows both alternatives and defines the differential input impedance as the impedance measured across the receiver
inputs.
Note that the receiver input capacitance should be designed to be as low as possible. Details of integrating the
termination impedance are left to the circuit designer's discretion, but the termination should not limit the high-
frequency, 250 MHz operation of the receiver.
The unpowered receiver impedance is not specified, since active circuits (which require power) are expected to be used
to implement the on-chip termination resistance. However, when unpowered, the magnitude of the receiver’s leakage
current (the sum of the two input currents) shall not exceed 1.0 mA.
The threshold hysteresis is important in receiver design to eliminate the possibility of oscillating receiver output when
the differential input is undefined (see figure 3-17). The undefined input can occur when the receiver inputs are
unconnected, when the connected driver is powered down, or when transitioning between defined values. The 25 mV
minimum hysteresis means that an input signal must change by more than this value to change the receiver output
state. A known output condition for an open or shorted receiver input (failsafe) is implementation-dependent and
beyond the scope of this standard.
3.3 AC specifications
A basic goal for this specification is to preserve the high transfer rate of the ECL physical layer (1 bit/2 ns for each
differential signal pair) while drastically reducing power dissipation. The skew and transition time limits specified in
this subclause correspond to those of the base SCI standard.
The 500 ps maximum transition time specified in this subclause is equivalent to a 0.3 V/ns slew rate for a 250 mV
differential signal. This is the minimum guaranteed slew rate for the signals specified in this standard. The transition
time is most critical through the receiver threshold region because the high-speed comparator design needs
unambiguous inputs to function efficiently. The 0.3 V/ns slew rate in the threshold region shall guarantee reliable
receiver switching.
The fast transitions contain high-frequency components that directly affect the electromagnetic radiation (EMR)
created by the signal. Normally these high frequencies would create electromagnetic compatibility (EMC) problems.
The differential signal advantage is produced by single-ended signals simultaneously rising and falling. These edges
will generate equal and opposite electromagnetic fields that cancel each other and reduce generated fields and
radiation. Therefore, it is important to have both single-ended channels switching at the same time and at the same
slew rate; i.e., no skewed single-ended transitions.
Undershoot, overshoot, and fast rise times can generate noise, crosstalk, and electromagnetic interference. Although
the driver specifications in figure 3-19 and figure 3-20 reduce these problems, careful transmission-line design is
important to maintain signal integrity.
The receiver specified for table 3-1, the general purpose specification, is intended to operate over a common-mode
voltage range that will allow for about ± 1 V ground-potential difference between the driver and receiver power
supplies. The common-mode input voltage, Vicm, is the average of Via and Vib measured with respect to the receiver
ground potential. The equation expressing this value is given in 3.2.6. The receiver specified in table 3-1 and table
3-2 must maintain the sensitivity and skew specifications throughout this common-mode voltage range [see figure
3-14 for Vicm(max) and Vicm(min) definitions].
System interconnects comprised of parallel signals must account for skew. This standard includes skew specifications
because limiting skew is important for correct data transfers in a system. For example, the SCI physical system can be
thought of as a data pipeline. Bytes of data can be lined up in the physical layer, a virtual FIFO. Since more than one
transmission can be contained in the data path at any time, it is not the total delay time but the timing skew that is
important to reliable SCI data transfer. The skew is extremely important because it directly affects the sample window
(setup and hold time) available to the receiver logic. These skew specifications are based on a total allowable skew
tolerance that will still provide an adequate sample window for receiver capture logic. Since the bit width is nominally
2 ns, reliable data transfer assumes that a 600 ps skew will be the maximum allowed.
The physical layer skew is defined for the purpose of this standard as the difference in time that is unintentionally
introduced between changing signal levels (incident edges) that occur on parallel signal lines. This difference results
in an uncertain position with respect to time between parallel signals. Jitter is not specified in this standard because the
skew specification accounts for jitter as well. Jitter is the error of the time of transitions occurring in a serial
transmission line. The range of parallel skew includes uncertainties that would be called jitter in a serial specification.
A parallel clock signal always accompanies SCI data signals. This clock is specified to be 250 MHz and to have a duty
cycle that is greater than 45% and less than 55%. Since it would be impractical to force the circuit designer to vary the
clock duty cycle to the minimum and maximum limits, testing would be done with the clock as generated. It would be
tested under the same conditions as the data signals, since it is implied that they are on the same die. The requirement
is for the clock to comply with the 45–55% duty cycle first, and then test all skew parameters to that clock. Since “any
two signals” is specified for skew measurements, the relation of all signals to clock is implied because “any two”
includes all pairs for skew measurement.
From the perspective of the receiver, the maximum clock-to-data signal skew is more important than the maximum
data-to-data signal skew. However, from a manufacturing perspective, minimal skew design techniques are unlikely
to treat the clock signal as special; therefore, skew measurements are not based upon it. Note that sophisticated
receivers can measure min/max data-signal skews and dynamically adjust the clock signal delay, to reduce the
effective clock-to-data skew to a little more than half of the “any-two” skew specification.
However, testing for the range of skew between the first and last signal may be inconvenient, so a more constraining
specification could be used to simplify testing: require all data signals to be within ± 300 ps of the clock.
The physical transmission mechanisms do not differentiate between the clock and other signals, so any given
transmission link might happen to result in the clock arriving as the latest or the earliest signal. Many of the receiver
technologies appropriate for very-high-frequency operation do not use the incoming clock to sample the incoming
data, which must be resampled and shifted to the local clock domain. Such receivers merely monitor changes in clock
phase as the source clock drifts relative to the receiver's own clock, and thus do not care whether the clock transition
is centered relative to the data transitions or at one extreme or another. Additionally, at these baud rates, receivers will
often need to incorporate automatic deskewing circuits to function in real-world environments. For example, SCI
protocols provide sync packets for dynamic skew compensation (see [B.1]). Dynamic skew compensation will need to
be used in the circuit designer's systems if the cables and connectors used cannot assure meeting the 600 ps maximum
skew specification.
The single-ended pairs that make up the differential signal are shown in figure 3-22. This picture shows the single ac
receiver specification called tskew. Here, tskew is pictured as only one receiver differential input to another receiver
differential input. It is the total allowable skew among parallel channels. The receiver must be able to correctly sample
the logic state on all parallel channels when there is this much difference between the transitioning signals as they
arrive at the inputs to the high-speed comparator. The tskew includes that resulting from driver, interconnect,
packaging, induced noise, etc. Differential signals Vidm and Vidn would be measured for the difference in delay at the
receiver input, using the test circuit shown in figure 3-15.
Since LVDS receivers are expected to be integrated on VLSI parts, it is difficult to measure skew for difference in
delay through a comparator. For this reason, the receiver skew is specified as a maximum tolerable timing difference
(as observed on the receiver's inputs) for which the data will be sampled correctly. This specification implies that the
internally sampled data values are accessible during testing.
When generating differential signals, two skews are important. Skew 1 is the skew between the high-to-low and low-
to-high transitions of complementary single-ended channels (see figure 3-23). This skew can be the result of different
propagation delays between complementary drivers, or different slew rates of the driver outputs. It is always measured
at the Vos as defined in figure 3-7 for the single-ended signals. This skew creates EMR as discussed in 3.3.1.
where tpHLA/B and tpLHA/B are the propagation delays on outputs A and B for high to low and low to high.
Skew2 is the skew between any differential signals as they can be measured at the driver output. It results from circuit
mismatches for complementary drivers and layout or packaging differences. As shown in figure 3-24, it is measured
between parallel channels. Skew2 is always measured between any two parallel signals, at the 0 V differential point.
If tpdiff[i] is the differential delay of Vod through the LVDS driver i, and assuming that the (probably inaccessible)
inputs to the driver are simultaneous, then
where m is any one of the parallel channels and n is any other channel.
The diagram in figure 3-25 shows a representative breakdown of a typical SCI signal path. The backplane could also
represent a cable segment of the signal path. An estimate of how the skew budget could be allocated for each of the
signal path elements up to the package input is given in table 3-4. The chip inside the package also has to tolerate the
additional package-to-chip wiring skews.
Annex A Bibliography
(Informative)
This document has been developed with point-to-point interconnects, such as the following standard, in mind:
[B1] IEEE Std 1596-1992, IEEE Standard for Scalable Coherent Interface (SCI) [ANSI].2
2
This standard is currently in international balloting as an international Draft, where it bears the designation ISO/IEC DIS 13961. IEEE publications
are available from the Institute of Electrical and Electronics Engineers, 445 Hoes Lane, P.O. Box 1331, Piscataway, NJ 08855–1331, USA.
(Normative)
The SCI encoding specifies how packet types, packet lengths, and idle symbols are uniquely identified. Although the
logical encoding is specified in terms of 16–bit symbols, the physical encoding layers support several data-path
widths. Distinct physical encoding layers can be supported without changing the logical protocols, if they define how
conversions between the physical and logical encodings are performed.
For 16-bit-wide links, one 16-bit SCI symbol is transmitted in each data-transfer period. In addition, a clock signal is
needed to define symbol boundaries (the data should be stable when sampled), and a flag signal is used to locate the
starting and ending symbols of packets.
The zero-to-one transition of the flag signal is used to mark the beginning of each packet, and the one-to-zero
transition of the flag signal specifies the approaching end of each packet. The flag signal returns to zero for the final 4
symbols of send packets and for the final symbol of an echo packet as illustrated in figure B.1. A zero always
accompanies the CRC of any packet, so the zero-to-one transition can be used to identify the start of the next packet
(even when there is no idle symbol between them).
The sync packets are used for initial synchronization of the physical receivers and for dynamic compensation of skew.
Sync packets are always a pattern of one-bits the full width of the link, followed by zero-bits the full width of the link.
Note that for P18 and wider encodings, the clock may have either a zero-to-one or a one-to-zero transition at the point
in the sync packet where all other signals have a one-to-zero transition. (Packets may start at either clock transition in
these wider encodings.) To simplify the design of the SCI interface hardware, the idle symbols are always 16 bits wide
or the full width of the data path depending on which is larger.
It should be noted that the 1- and 16-bit encodings, which are specified in IEEE Std 1596–1992 and therefore not
included in this annex, can also use this LVDS differential signaling standard.
The 16-bit SCI symbols need not be sent in a single physical-clock-signal transition; multiple data-transfer cycles can
be used. Encoding for use on physical links that are narrower than the 16-bit SCI logical symbol width use the polarity
of the clock signal or extra transitions on the flag signal to mark the beginning of the logical symbol.
On a byte-wide (8-data-bit) interface, half of an SCI symbol is sent in each of two data-transfer intervals. The clock
signal changes before each subsymbol; the low and high clock-signal values identify the first and last subsymbol
respectively. The flag line transitions occur at most once per symbol, as illustrated in figure B.2.
In this illustration, e0.a and e0.b bytes are the most- and least-significant bytes of the e0 (echo-packet) symbol; el.a
and el.b bytes are halves of the el symbol; e2.a and e2.b are halves of the e2 symbol; e3.a and e3.b are halves of
the e3 symbol; i.a and i.b bytes are the most- and least-significant bytes of the i (idle) symbol.
The sync packet is 16 clock-periods long (8 symbols), to simplify conversions between P18 and P10 encodings, as
illustrated in figure B.3.
The 16-bit symbol may also be sent four bits at a time, using four subsymbols to form a logical SCI symbol. The clock
and flag signals are combined, since the overhead of the extra signal is significant and 10-bit cables may then be used
for two 5-bit links, one in each direction. The physical clock signal has a high-to-low transition at the start of each
symbol; the flag value is derived by sampling the clock in the middle of each symbol period, as illustrated in
figure B.4.
A high logical flag has a low-to-high clock-signal transition after the first subsymbol of each symbol. A low logical
flag has a low-to-high clock-signal transition after the third subsymbol of each symbol.
The e2.a, e2.b, e2.c, and e2.d subsymbols are the most- through least-significant quarters of the e2 (echo-packet)
symbol; the e3.a, e3.b, e3.c, and e3.d subsymbols are the most- through least-significant quarters of the e3 (echo-packet)
symbol; the i.a and i.b subsymbols are the two most-significant quarters of the following idle symbol.
The sync packet is 16 clock-periods long (8 symbols), to simplify conversions between P18 and P5 encodings, as
illustrated in figure B.5.
Encoding for links that are multiples of the 16-bit SCI logical symbol include the clock signal, provide an additional
sync signal (to mark the beginning of a sync packet), and have one flag signal for each 16-bit data symbol. Transitions
of the physical clock signal mark the boundaries of data-transfer intervals. The idle symbols are always the width of
the link.
A sync packet consists of one data-transfer interval with ones on the sync signal, all flag signals and all data signals,
followed by seven data-transfer intervals of zeroes on all these signals.
On a 4-byte (32-bit) interface, two 16-bit SCI symbols are sent for each physical-clock-signal transition. There is a
total of 36 signal pairs: clock, sync, and two sets of (flag + 16 data), as illustrated in figure B.6.
A sync packet, which remains 8 data periods in length, is illustrated in figure B.7.
On an 8-byte (64-bit) interface, four 16-bit SCI symbols are sent for each physical-clock-signal transition. There are a
total of 70 signal pairs: clock, sync, and four sets of (flag + 16 data) signals, as illustrated in figure B.8.
On a 16-byte (128-bit) interface, eight 16-bit SCI symbols are sent for each physical-clock-signal transition. There are
a total of 138 signal pairs: clock, sync, and eight sets of (flag + 16 data) signals, as illustrated in figure B.9.
(Informative)
A driver design model is illustrated in figure C.1. This simplified model is only intended to give an idea of how to
implement a near-constant-current differential driver. It is not necessarily a manufacturable or cost-effective design.
Crb is intended to operate as a current sink; however, at any particular operating condition it is equivalent to a
resistance Rb. The voltage on Cra is adjusted so that the attached transistor behaves like a voltage follower, providing
the appearance of a resistance Ra connected to 2.4 V. The voltage on Crc is adjusted so that these transistors behave like
termination resistors of value Rc. Rt is the receiver termination.
Values are selected and/or adjusted so that Ra+Roa and Rb+Rob are closely matched, where the switching transistors
have impedance of Roa and Rob. When so matched, the driver's output characteristics are illustrated in figure C.2.
Although constrained by the Ro (driver’s output impedance) specification, the value of Rc may be significantly greater
than the transmission line impedance of 50 Ω, to minimize driver power dissipation.
The receiver design model includes a terminating resistance (90–110 Ω), as illustrated in figure C.3.
The design of the receiver can be simplified by recognizing that the amplifier impedance need only be large compared
to the receiver’s termination impedance (100 Ω). Also, it may be easier to take advantage of the fact that SCI signals
are always clocked; a combined amplifier/latch (i.e., sense amp) may be easier to build than an amplifier followed by
an independent latch.
The signal-transmission model, including the driver, transmission line, and termination resistance within the receiver,
is illustrated in figure C.4.