PCIe Clock Arch
PCIe Clock Arch
APPLICATION NOTE
PCI−SIG defines three clock distribution methods for In systems using spread spectrum clocking (SSC) for EMI
PCIe standards – Common Clock, Data clock and Separate suppression, the Tx and the Rx PLLs are edge aligned and
clock architectures. Irrespective of clocking methodology, much of the jitter propagates equally to both PLLs. The
the accuracy requirement of PCIe standard is the same i.e. drawback of this clocking architecture is the clock needs to
± 300 ppm. be distributed to each PCIe endpoint in the system, which
increases the PCB real estate and sets tighter limits on clock
Common Clock Architecture jitter, clock skew requirement, and the number of signals
In this architecture, same clock reference is sourced to the routed on hardware.
transmitter (Tx) and the receiver (Rx). This clocking method This is the preferred and simple clock distribution scheme
is most common in commercially used devices for its when the design consists of single card with multiple PCIe
simplicity in implementation. It also supports end points and the same clock source needs to be transmitted
spread−spectrum clocking used for EMI (emission) with other signals through the distribution channel.
reduction.
Transmitter Receiver
TX PLL CDR
RX PLL
Refclk
Data Clock Architecture stream at the receiver. Additional clock recovery circuitry is
In this architecture, the clock signal is embedded into the used to extract the clock from signal stream at the receiver
transmitted data stream and is recovered from the data end. SSC can be implemented in this architecture.
www.onsemi.com
2
AND9202/D
Transmitter Receiver
Clock Out
Data In TX Latch RX CDR and Data
Out
TX PLL
Refclk
Separate Clock Architecture / SRIS (Separate Refclk effective jitter at the Rx is the RSS (Root Sum Square) sum
Independent Spread) of the Tx and Rx PLLs, thus the jitter requirement for
This architecture avoids transmitting the clock to all separate reference clocks is substantially tighter than for
channels by using separate clock sources at each PCIe common clock architecture. The PCIe SRIS clocking specs
endpoint, however it substantially increases the design are still in their infancy. However, as per preliminary results,
complexity since the frequency accuracy limit of ±300 ppm the PCIe Generator clocks from ON Semiconductor
without SSC still holds. This clocking method supports two conforms to the current SRIS refclk requirements (as per
types of clocking – Tx and Rx with same data rates, and Tx PCI−SIG: Separate Refclk Independent SSC Architecture
and Rx at different data rates. The later tolerates 5600 ppm draft).
difference for separate refclks utilizing the SSC (SRIS). The
Transmitter Receiver
TX PLL CDR
RX PLL
Refclk
Refclk
www.onsemi.com
3
AND9202/D
The PCIe Serializer De−serializer (SerDes) system uses a The Refclk contains jitter over wide range of frequencies.
reference clock (Refclk) to generate higher frequency clock A specific frequency band of interest will be tracked by the
from internal PLL which delivers higher bit rates. Typically receiver which is meaningful for a PCIe system – mainly the
the reference clock is multiplied by 4 to 25 times to generate jitter component tracked (not filtered) by the transfer
the bit rate frequency. Jitter on reference clock degrades the function of the CDR. Thus, PCIe systems are concerned
communication both at the Tx and Rx thereby prejudicing about this band of period jitter that is extracted and
the system performance. Thus, a precise low jitter reference calculated by mathematical functions, which differ for each
clock is essential for a good operational system. of the PCIe Gen I, II and III models. The maximum limits of
Furthermore, the Refclk jitter performance specifications the jitter values are listed in Table 1.
are stringent for higher bitrates.
The jitter on the signal is comprised of Deterministic (Dj) below shows the results on the clock output from
and a Random (Rj) component. The Dj is the non−Gaussian NB3N51034, a 25 MHz Crystal to 100 MHz/ 200 MHz Quad
probability density function bounded in amplitude and HCSL/LVDS Clock Generator after PCIe Gen I, II and III
appears due to specific causes like interference, data transfer functions or filters are applied to the cycle trend
dependence, Duty Cycle Distortion and is predictable. For data.
clocks, the Dj appears as spurs of definite amplitude in the
phase noise plot. The Rj is the Gaussian distribution
component of jitter which is unbounded and present due to
random phenomena such as thermal noise, process
variation, etc. In clocks, the Rj component appears as the
integrated phase noise over a specified frequency offset
range. Total jitter (TJ) is calculated from Dj and Rj by −
Tj = Djpeak−to−peak + 2 × n × Rjrms, ; n is based on BER of
the link (10−12 for Ethernet)
Table 1 represents the Tj value acceptable for PCIe Gen
I, II and III systems in specified frequency bands (based on
CDR transfer function).
Jitter Evaluation Method Figure 4. PCIe Jitter Level Post PCIe Jitter
In order to evaluate the PCIe jitter values from clock, a Extraction on NB3H51034 Clock
cycle trend period jitter data of 1.6 ms (approx. 200,000
cycles) is fed to the PCIe Jitter analyzer tool (A tool The red vertical line is the maximum jitter limit for each
developed by ON Semiconductor which is similar to Intel® of the PCIe generation standards. The green color indicates
Clock Jitter Tool). This extraction can also be done on the the conformance to the spec and the green bar length
clock cycles data by applying the respective transfer indicates how far the measured jitter is from the maximum
functions for each of the PCIe generations. The Figure 4 limit.
www.onsemi.com
4
AND9202/D
The jitter results of the ON PCIe clocks meet the protocol other PCIe parts in market. These margins are useful to
specification with ample margins and performs superior to overcome system challenges explained in next section.
SYSTEM CHALLENGES
To overcome the performance limitations and voltage Serializer De−serializer (SerDes) is used to transmit and
scaling problem posed by parallel bus communication receive data over the serial link. The reference clock is
systems, the industry is moving towards point−to−point sourced to the SerDes block of a system.
serial communication. In a serial bus communication, a
Transmitter Block
TXDP
TX
Serializer TX Data
TXDN
TX PLL
Refclk
Act Activity
Detect Detect
As system bandwidth requirements continue to increase, design and sound circuit design techniques, SerDes can be
SerDes are moving large amount of data between systems. virtually immune to supply noise.
The success of SerDes serial link poses a challenge for Jitter Signal integrity issues like ISI, crosstalk, impedance
performance. Power supply noise, Reference clock quality mismatch can also be dealt with impedance matching, better
and signal integrity are the main system issues that are PCB material and routing techniques.
potential pitfalls affecting the performance of a SerDes Jitter on Reference clock is the most difficult to suppress.
system. It not only multiplies and propagates directly with the
Power supply noise introduces bounded and uncorrelated transmitted signal, but also corrupts the reference clock for
jitter DJ at Refclk input, PLL, clock distribution and serial the recovered signal at the receiver. The high performance
data distribution. However, with regulated power supply PLL architecture of ON Semiconductor PCIe clock
www.onsemi.com
5
AND9202/D
generators ensures low jitter peaking and substantial margin The extremely low random jitter from ON Semiconductor
to the PCIe Gen I, II and III jitter specification limits as PCIe reference clock generator at both the transmitter and
shown in Table 2. The challenge is to create a system where the receiver is beneficial in avoiding SerDes system failure
jitter generation plus the system jitter is less than receiver due to jitter performance. Furthermore, SSC capability
tolerance. enables EMI suppression.
In designing any future scalable system with PCIe industry. In the PCIe based systems, the CPU and memory
architecture, clock tree planning is a critical task. Based on exchange data through a hub consisting of multiple PCI
number of internal and external PCIe I/Os (including the Express lanes called a PCIe Root Complex. The clock tree
future scalability) in the system and speed/ bandwidth mainly consists of several components – clock generators,
requirements, appropriate PCIe clock sources, buffers, and clock buffers (Zero delay and non−zero delay buffers) and
switches are selected. Based on the PCIe data exchange, switches.
below are few of the commonly used clock trees in the
PCIe Endpoint
PCIe Endpoint
PCIe Endpoint
100 MHz HCSL
100 MHz HCSL
100 MHz HCSL
100 MHz HCSL
100 MHz HCSL NB3N106K 100 MHz HCSL
www.onsemi.com
6
AND9202/D
PCIe Clock
External Links
ON Semiconductor offers a range of high performance performance. Some of the clock generators offer SSC
solutions in Clock Generation, clock and data Distribution, Generation for EMI suppression. In this technique, the peak
and Multiplexing/ switching to construct a complete system energy of the carrier is re−distributed to nearby frequencies
clock tree and continuously expanding the portfolio. in the spectrum, in order to reduce the radiated power. Refer
to AND9015/D A Solution for Peak EMI Reduction with
Clock Generators Spread Spectrum Clock Generators for further details.
The range of clock generators from ON Semiconductor Table 3 lists several off the shelf PCIe clock generators
provide the best in class sub picosecond phase jitter available from ON Semiconductor.
Table 3. PCIe CLOCKS FROM ON SEMICONDUCTOR WITH PARAMETRICS AND PHASE NOISE
Phase Jitter Spread
No. of PCIe Noise Cycle− Spread Frequency
Part Clock RMS cycle Spectrum Deviation
Number Description Outputs (ps) (ps) Modulation (%) Package
NB3N3002 3.3 V, Crystal to single 1 0.35 3 No TSSOP−16
output HCSL clock
synthesizer
NB3N5573 3.3 V, Crystal to dual 2 0.4 No TSSOP−16
output HCSL clock
synthesizer
NB3N51032 3.3 V, Dual output HCSL 2 0.4 2 Yes 0% TSSOP−16
clock synthesizer (SS −0.5%
compatible)
−0.75%
NB3N51034 3.3 V, Crystal to quad 4 0.4 Yes 0% TSSOP−20
output HCSL clock −0.5%
synthesizer
−1.0%
(SS compatible)
−1.5%
www.onsemi.com
7
AND9202/D
Table 3. PCIe CLOCKS FROM ON SEMICONDUCTOR WITH PARAMETRICS AND PHASE NOISE
Phase Jitter Spread
No. of PCIe Noise Cycle− Spread Frequency
Part Clock RMS cycle Spectrum Deviation
Number Description Outputs (ps) (ps) Modulation (%) Package
NB3N51044 3.3 V, Crystal to quad 4 0.4 No TSSOP−28
HCSL clock synthesizer
with individual OE
NB3N51054 3.3 V, Crystal to quad 4 0.4 Yes 0% TSSOP−24
HCSL clock −0.35%
synthesizer with I2C
−0.5%
Spread Aware Fanout and Zero Delay Buffers these high performance fanout buffers in PCIe generation I,
The differential fanout clock and data buffers provide a II and III compliant applications. Many of the devices can
choice of either 6, 8, 10 or 21 outputs and the zero delay accept single−ended or differential LVPECL, LVDS and
buffers are offered in 8, 12 and 19 outputs. These buffers are HCSL input clocks, providing flexibility by allowing
ideal for systems requiring multiple copies of high system designers to easily interface to multiple input
precision, low phase noise clocks. Typical additive phase signaling schemes. Moreover these fanout buffers are
jitter as low as 0.1 picoseconds (ps) rms (integrated over spread aware, meaning they replicate the reference clocks
12 kilohertz [kHz] to 20 megahertz [MHz]), and low with spread spectrum.
output−to−output skew of 100 ps (max) enables the use of
Table 4. PCIe FANOUT BUFFERS AND ZDBs FROM ON SEMICONDUCTOR WITH PARAMETRICS AND ADDITIVE
JITTER
Input/ Additive Output− Max
Part Output Jitter output Propagation Frequency
Number Description Ratio (ps) Skew (ps) Delay (ns) (MHz) Package
NB3N106K Clock Fanout Buffer, 1:6 1:6 0.1 100 0.8 400 QFN−24
Differential, 3.3 V, with HCSL
Outputs
NB3N108K Clock / Data Fanout Buffer, 1:8 0.1 100 0.8 400 QFN−32
1:8 Differential, 3.3 V, with
HCSL Outputs
NB3N111K Clock / Data Fanout Buffer, 1:10 0.1 100 0.8 400 QFN−32
1:10 Differential, 3.3 V, with
HCSL Outputs
NB3N121K Clock / Data Fanout Buffer, 1:21 0.1 100 0.8 400 QFN−52
1:21 Differential, 3.3 V, with
HCSL Outputs
NB3W1200L 3.3 V 100/133 Mhz 1:12 0.001 50 100, 133 QFN−64
Differential 1:12 Push−Pull
Clock ZDB/Fanout Buffer for
PCle
NB3N1200K 3.3 V 100/133 Mhz 1:12 0.001 50 100, 133 QFN−64
Differential 1:12 HCSL Clock
ZDB/Fanout Buffer for PCle
NB3W800L 3.3 V 100/133 Mhz 1:8 0.001 50 100, 133 QFN−48
Differential 1:8 Push−Pull
Clock ZDB/Fanout Buffer for
PCle
NB3W1900L 3.3 V 100/133 Mhz 1:19 0.001 50 100, 133 QFN−72
Differential 1:19 Push−Pull
Clock ZDB/Fanout Buffer for
PCle
NB3N1900K 3.3 V 100/133 Mhz 1:19 0.001 50 100, 133 QFN−72
Differential 1:19 HCSL Clock
ZDB/Fanout Buffer for PCle
www.onsemi.com
8
AND9202/D
PCIe Switch The single−pole, double throw (SPDT) PCIe Switches offer
The differential switches are designed for use in PCIe 2.0 up to six differential channels allowing it to handle up to
and PCIe 3.0 applications and support data rates of up to 8.0 three PCIe lanes, thereby enabling a single controller to
gigabits per second (Gbps). The new devices have low manage three PCIe slots. The NCNx612B family is
supply current requirements of 200 microamperes (mA) and compatible with both Display Port 1.2 and PCIe.
250 mA respectively that also help reduce heat dissipation.
Table 5. PCIe DATA SWITCHES FROM ON SEMICONDUCTOR WITH PARAMETRIC AND GENERATION
APPROPRIATENESS
VCC Max Data PCIe
Part Number Description Channels Configuration min (V) Rate Compatible Package
NCN2411 4−Channel Differential 8 8PDT 1.5 5 Gbps Gen 2 WQFN−42
1:2 Mux/Demux Switch
CONCLUSION
ON Semiconductor, a premier supplier of high clock tree can be realized with the PCIe clock synthesizers
performance silicon solutions for energy efficient to provide a clock source to the fanout buffers, and
electronics has a range of components that significantly selectively expand the clock tree using PCIe switches. The
strengthens its portfolio of products to address Peripheral superior performance of each of the ON Semiconductor
Component Interconnect Express (PCIe) applications in components in the Clock Tree alleviate the system timing
communications systems such as routers, servers, problems and strengthen the SerDes system to meet
networking equipment and ATE. Furthermore, a complete stringent clock performance requirements.
www.onsemi.com
9
AND9202/D
ON Semiconductor and the are registered trademarks of Semiconductor Components Industries, LLC (SCILLC) or its subsidiaries in the United States and/or other countries.
SCILLC owns the rights to a number of patents, trademarks, copyrights, trade secrets, and other intellectual property. A listing of SCILLC’s product/patent coverage may be accessed
at www.onsemi.com/site/pdf/Patent−Marking.pdf. SCILLC reserves the right to make changes without further notice to any products herein. SCILLC makes no warranty, representation
or guarantee regarding the suitability of its products for any particular purpose, nor does SCILLC assume any liability arising out of the application or use of any product or circuit, and
specifically disclaims any and all liability, including without limitation special, consequential or incidental damages. “Typical” parameters which may be provided in SCILLC data sheets
and/or specifications can and do vary in different applications and actual performance may vary over time. All operating parameters, including “Typicals” must be validated for each
customer application by customer’s technical experts. SCILLC does not convey any license under its patent rights nor the rights of others. SCILLC products are not designed, intended,
or authorized for use as components in systems intended for surgical implant into the body, or other applications intended to support or sustain life, or for any other application in which
the failure of the SCILLC product could create a situation where personal injury or death may occur. Should Buyer purchase or use SCILLC products for any such unintended or
unauthorized application, Buyer shall indemnify and hold SCILLC and its officers, employees, subsidiaries, affiliates, and distributors harmless against all claims, costs, damages, and
expenses, and reasonable attorney fees arising out of, directly or indirectly, any claim of personal injury or death associated with such unintended or unauthorized use, even if such claim
alleges that SCILLC was negligent regarding the design or manufacture of the part. SCILLC is an Equal Opportunity/Affirmative Action Employer. This literature is subject to all applicable
copyright laws and is not for resale in any manner.
www.onsemi.com AND9202/D
10