Using the Altera PDN Tool to Optimize Your Power Delivery Network Design
This application note provides guidance on the estimation of FPGA current
requirements for the design of a robust FPGA Power Delivery Network (PDN). It provides
guidance on how to improve the efficiency of the PDN with the Altera PDN Tool, with an
objective of reducing the number of PCB mounted decoupling capacitors.
Altera provides a PDN tool that is based on an Microsoft Excel spreadsheet
used to calculate an impedance profile based on user inputs. This PDN Tool helps design a
robust PDN, estimate its performance and optimize decoupling solutions and reduce cost. The
PDN Tool can be used to analyze all FPGA power supply rails.
This application note should be read in conjunction with the PDN Tool User guide, and PowerPlay Early
Power Estimators (EPE) and PowerPlay Power Analyzer.
For more information on PDN design refer to Printed Circuit Board
(PCB) Power Delivery Network (PDN) Design Methodology and on the Altera
Board Design Resource Center.
A stable power supply is the foundation of your FPGA design. It helps ensure
that the device is within electrical specifications. A robust Power Delivery Network (PDN)
that manages voltage ripple, is an important part of your power supply design.
With an increase in current loading, an insufficient PDN can result in
excessive voltage ripple, voltage drops, and VRM instability. Voltage ripple on VCC
supplies can cause brown-out conditions or timing margin reduction through power supply
noise induced jitter. This can lead to data integrity problems.
A robust PDN is important for transceiver designs. A transceivers
performance can be adversely affected by voltage ripple on its power supplies. Increased
voltage ripple on transceiver power supplies can increase the Bit Error Rate (BER) through
increased transmit jitter or reduced jitter tolerence.
Jitter induced by supply voltage ripple on General Purpose IO (GPIO) and PLL
supplies reduces timing margin on External Memory Interfaces (EMIF) such as DDR3 and DDR4.
Bit errors can be seen if timing margins are violated.
PDN Circuit Topology
The PDN tool is based on a lumped equivalent model representation of the
power delivery network topology. A schematic representation of the circuit topology,
modelled as part of the tool is shown in the figure below.
For first order analysis, the voltage regulator module (VRM) can be
modelled as a series-connected resistor and inductor. At low frequencies, up to 50 KHz,
the VRM has a very low impedance and is capable of responding to the instantaneous
current requirements of the FPGA. The ESR and ESL values can be obtained from the VRM
manufacturer. At frequencies higher than 50 KHz, the VRM impedance is mostly inductive,
that makes it incapable of meeting the transient current requirement. The on-board
discrete decoupling capacitors must provide the required low impedance from low to high
frequencies, depending on the capacitor intrinsic parasitics (RcN, CcN, LcN) and the
capacitor mounting inductance (LmntN). The interplanar capacitance between the
power-ground planes typically has lower inductance than the discrete decoupling
capacitor network, making it more effective at higher frequencies (10 MHz and higher).
The effectiveness of the decoupling capacitors is limited by the PCB spreading
inductance and the ball grid array (BGA) via inductance that a given capacitor
encounters with respect to the FPGA. To simplify the circuit topology, the PDN tool
models the distributed nature of PCB spreading, BGA inductance, and resistance with a
single lumped inductor and resistor.
Figure 1. PDN Circuit Topology
From this diagram it is clear that the key to improving the efficiency of the PCB PDN is
to reduce component mounting inductance (Lmnt), parasitic inductance (Lc) and spreading
inductance (Ls). This application note will provide guidance on methods of reducing
these inductances, and how to evaluate this in the PDN Tool.
Estimating the Current Requirements of your FPGA Design
Before designing your PDN, it is important to use accurate current
estimates for each FPGA power supply. You should consider worst case scenarios of your
design, but do not over-estimate, because it can make the PDN design more
You can estimate the current requirements of your FPGA design with the
PowerPlay Early Power Estimators (EPE) and Quartus PowerPlay Power Analyzer (PPPA). The EPE is a
spreadsheet based tool that shows the FPGA design power, based on your design
requirement input. The PPPA is a Quartus® software tool that estimates your current
requirements, based on the Fit of your Quartus project.
The PowerPlay Early Power Estimators (EPE)
and Power Analyzer userguide provides additional
information on these tools.
The current and decoupling requirements increase if the current in your
To estimate the current requirements of your design, Altera recommends
including all functional blocks that you plan to use. Include accurate use of Logic
Elements, DSP blocks, PLLs, transceivers, Hard IP, HPS, Internal Memory and IO.
Dynamic power requirements increase with clock frequency. To account for worst case current
requirement conditions, Altera recommends using the highest clock frequencies that your
design will run at. When using dynamic reconfiguration of PLLs in your design, ensure that
your estimates are based on the highest frequency you intend to use for logic clocked by
For multiple clock frequency domains in your design, it is tempting to
simplify your estimates by assuming that all clock domains in your design run at the same
highest frequency. Estimate each clock domain separately, because over estimation makes the
PDN design difficult. The EPE and Power Analyzer allow you to do this estimation.
General Purpose IO (GPIO) current requirements are dependent on the IO standard, drive
strength, termination use, frequency and toggle rate. Altera recommends using accurate IO
standard settings in your estimations.
Power requirements increase with frequency so when using transceivers
that can be reconfigured between multiple data rates, you should ensure that you use
your highest datarate for your power estimations. The estimates should use the correct,
worst case Physical Coding Sub-layers (PCSs) for your design.
Depending on protocol and datarate, transceivers use the following PCSs:
PCI Express Hard-IP
When implementing dynamic reconfiguration to switch between the Standard
and Enhanced PCS, Altera recommends using the PPPA to determine whether the Standard or
Enhanced PCS consumes more power for your design.
As an example, dynamic reconfiguration is used to switch between Altera
IP 1G/10G Ethernet IP using the Standard PCS for 1Gbs Ethernet and the Enhanced PCS for
10Gbps Ethernet. To determine worst case power estimations, use the 10Gbs Ethernet using
the Enhanced PCS.
If you intend to use Decision Feedback Equalization (DFE) or Adaptive
Equalization (AEQ) with your transceivers you should include these in your power
estimations. Adding DFE and AEQ adaption to your design increases the transceiver power
requirements and the PDN design challenges. The estimations should be based on channels
that use DFE and AEQ. If you do not use DFE and AEQ for some channels, then do not not
include them in your power estimations.
Power requirements depend on toggle % of the logic in your design. A
clock-like data pattern has higher current requirements than a PRBS pattern. A PRBS
pattern has higher current requirements than those for a grey-code pattern.
Use the highest toggle % that you expect to see in your design. The PDN
design becomes difficult if you over estimate. Although it is design dependent, Altera
assumes a 12.5% toggle percentage. Use the PPPA to analyze the toggle % of your
When using the Quartus PPPA, base your power estimations on
simulation-based input data. It provides accurate power estimations.
A PDN should have a low enough target impedance (Ztarget) to meet the
voltage ripple requirements under maximum dynamic current requirements.
Figure 2. Ztarget Formula
This formula shows:
A decrease in voltage rail decreases Ztarget
A decrease in allowed voltage ripple decreases Ztarget
An increase in dynamic current decreases Ztarget
Very low target impedances require significant effort to design a
It can be challenging to provide PCB decoupling for the FPGA core (VCC),
and transceiver VCCR_GXB, VCCT_GXB supplies. The low voltage, high current, VCC supply
results in low target impedance. The transceiver VCCR_GXB and VCCT_GXB supplies have
lower current than the VCC supply, but have low ripple requirements and hence low target
Assuming the following requirements, Ztarget can be calculated for the
following VCC, VCCR_GXB, and VCCT_GXB cases. The current transient %, and allowable
ripple % recommendations can be found on the Introduction tab of the PDN tool.
Table 1. Ztarget Calculations for VCC, VCCR_GXB, and VCCT_GXB Supplies
Nominal Voltage (V)
Allowable Ripple (%)
Current Transient (%)
Target Impedance (mR)
Additional challenges may occur when sharing supply rails with differing
current requirements such as VCCR_GXB and VCCT_GXB. The worst case combination of higher
IMax of the VCCR_GXB supply and lower ripple and high current transient of the VCCT_GXB
supply, can create very low target impedances. Using the x or x/related options in the PDN Tool helps reduce the decoupling
burden by estimating synchronous versus non-synchronous current switching.
In this application note, the VCCR_GXB and VCCT_GXB supplies are
considered separately because it results in decoupling solutions with fewer PCB
This application note focuses on the FPGA VCC, VCCT_GXB, and VCCR_GXB
supplies that have low target impedance and require additional effort to decouple.
Altera recommends that you analyze the performance of all design supplies.
A Quartus project targeting an Arria 10 10AX115N4F45I3SGE2 device,
implementing a transceiver design and core noise generators was used to produce current
requirements for the core VCC, and transceiver VCCT_GXB and VCCR_GXB supplies.
The following figure shows the transceiver channel placement of the
design used in this application note.
Figure 3. Transceiver Channel Allocation of the Example Design
The 1GbE and 10GbE capable channels in transceiver block 1E implement
transceiver PHYs capable of switching between the Standard PCS (configured for 1.25
Gbps), and the Enhanced PCS (configured for 10.3125 Gbps). To assess worst case
transceiver VCCT_GXB, and VCCR_GXB current requirements, the Enhanced PCS (configured
for 10.3125 Gbps) is the default PHY configuration.
The 9.8 Gbps, 4.9 Gbps, 2.5 Gbps capable channels in transceiver blocks
4E and 4F are implemented using the Standard PCS. To assess worst case power
requirements of these transceiver blocks, the transceiver PHY default configuration is
DFE and AEQ were not used for any channels in this design.
Multiple clock pattern and PRBS core noise generators filling 83% of the
device logic were implemented for this design. Vectorless estimation was used to
generate the PowerPlay Power Analyzer results.
You should base your power requirement estimates on your actual
Table 2. Total Current Estimates from the PPPA for the VCC, VCCP,
VCCERAM, VCCR_GXB, and VCCT_GXB Supplies
Power Supply Pin
The current estimates from the above table were entered in the PDN Tool
as shown below:
Figure 4. Power Group Configuration and Current Entry for the Example
Because of the high number of channels clocked from single ATX PLLs x/related is chosen for the VCCT_GXB and VCCR_GXB supplies in
this design. With the x/related setting, the PDN Tool assumes
the supplies have synchronous switching and hence higher dynamic current. This makes the
PDN design more challenging, but is a worst case scenario as not all channels are
synchronous. Assess if you should select x or x/related for your design.
PDN Design Optimization Study
This PDN design optimization study uses the previously determined current
estimates with a sub-optimal PCB design in terms of PDN performance. It demonstrates step
improvements in the PDN performance at each optimization stage. The goal is to improve the
PDN efficiency as observed through an increase in the effective frequency (Feffective) for
the VCC supply. It is easier to meet Ztarget with an inefficient PDN, but you should target
the highest possible Feffective for your design. The VCCT_GXB and VCCR_GXB supplies have a
fixed 70MHz frequency target. Optimizing the PDN results in the total number of PCB mounted
decoupling capacitors being reduced. The design study discusses PCB optimization and the
importance of minimizing spreading and vertical inductance in the PDN. The importance of
increasing the power and ground plane capacitance is also demonstrated. Alternative
capacitor technologies and their significant effect on PDN efficiency are then demonstrated
resulting in fewer PCB mounted capacitors. Finally, a novel high-frequency de-rating of the
VCC supply in the PDN Tool demonstrates a further dramatic reduction in the number of
required decoupling capacitors.
Initial Stackup Entry
The initial layer-stack of this design study is shown below.
Figure 5. Initial Stackup
When entered into the PDN Tool, and Auto decoupling
mode is selected for all power groups, it reports 301 capacitors are required to decouple
each of the VCC, VCCT_GXB and VCCR_GXB supplies. When the PDN Tool decoupling mode is set
to Auto, it will not add more than 301 capacitors. In reality, more
than 301 capacitors will be required to decouple the design. Fitting in excess of 903
capacitors for three supplies is unrealistic so optimization of the PCB PDN is
Auto decoupling mode is used throughout this application note.
Figure 6. Initial Number of Required Decoupling Capacitors
The following figure shows the impedance plots for VCC, VCCT_GXB and
VCCR_GXB with this initial stackup and PDN Tool configuration.
Figure 7. Initial VCC, VCCT_GXB, and VCCR_GXB Supply Impedance Plots
With this initial stackup and PDN Tool configuration, Feffective for the
VCC supply is reported to be just 10.62MHz and the target Feffective of 70MHz for VCCT_GXB
and VCCR_GXB supplies are not met. This is because the initial stackup, layer allocation,
and PDN is not optimized, resulting in an inefficient PDN.
Figure 8. Initial VCC Supply Feffective
Using the Correct Number of Power/Ground Via Pairs
Unless you change manually, the PDN Tool sets the Number of Power/Ground Via Pairs for each power group to the default value
defined on the Stackup tab. For this case it is 50 as shown by the
Figure 9. Incorrect Number of Power/Ground Via Pairs and Layer Number
PCB PDN design is highly dependent on the Number of
Power/Ground Via Pairs. This contributes to the spreading inductance (Ls) in the
PDN design. If the actual Number of Power/Ground Via Pairs in the
PCB is less than the number defined in the PDN Tool, the PDN Tool reports optimistic
results. It is important to set the correct Number of Power/Ground
Via Pairs to match your device and PCB.
Always provide a dedicated via connecting each power and ground pin on
the FPGA to the plane. For more information on the importance of the number of power and
ground pin pairs refer to Knowledge Database note.
Figure 10. Number of power and ground pin pairs
An increase in the Number of Power/Ground Via Pairs
between the plane and FPGA reduces spreading inductance (Ls) and increases the
high-frequency current delivery efficiency from the plane to the FPGA. This results in
increasing the maximum effective frequency of the PCB PDN.
Using the Correct Number of Power/Ground Via Pairs and Layer Number
Unless you change manually, the PDN Tool initially sets all power groups to
the same Layer Number . For this example it is 18. You must change
this to match your PCB design.
Long vias between the plane and FPGA have higher vertical loop inductance
than short vias. The planes placed farther from the FPGA have higher loop inductance.
Because inductance impedes high-frequency current delivery, it reduces the effective
maximum frequency of the PDN. To reduce loop inductance, you can place planes closer to the
Corrected Number of Power/Ground Via Pairs and Layer Numbers
You should use the correct number of power and ground via pairs for your
device. This can be found in the device pinout data. This example uses an AX115N_F45 device
selection from the PDN Tool and the initial power plane allocation, the Number of Power/Ground Via Pairs and Layer Number data is
corrected and entered into the PDN Tool as shown below:
Figure 11. Correct Number of Power/Ground Via Pairs and Layer Number
With the corrected Number of Power/Ground Via Pairs
and Layer Number data, the total number of required decoupling
capacitors remains at 301 for the VCC, VCCT_GXB and VCCR_GXB supplies. The VCC, VCCT_GXB,
and VCCR_GXB Impedance plots are shown below.
Figure 12. VCC, VCCT_GXB, and VCCR_GXB PDN Performance with corrected Number of
Power/Ground Via Pairs and Layer Numbers
The following figure shows that the VCC supply Feffective parameter has
increased from 10.62MHz to 30.68MHz. The increase in performance is due to an increase in
the number of power/ground vias from 50 to 197, and a reduction in spreading inductance
(Ls). The vertical loop inductance is reduced by moving the VCC supply plane closer to the
FPGA from Layer 18 to Layer 15.
Figure 13. VCC Supply Feffective with corrected Number of Power/Ground Via Pairs and Layer
In contrast, the VCCT_GXB and VCCR_GXB supply impedance plots shows a lower
frequency Zeffective and Ztarget intersect and hence worse PDN performance. This is because
the Number of Power/Ground Via Pairs was corrected from 50 to 16 and
the spreading inductance increased. The reduction in performance is despite the planes
moving closer to the FPGA from Layer 18 to Layer 9, resulting in a reduction in vertical
Moving Supplies to Optimal Layers
Reducing the spreading inductance (Ls) between the planes and the FPGA
increases the effective frequency of the VCC supply. Re-allocating the FPGA supply rails in
the layer-stack so that they have lower vertical inductance, will also show an improvement.
Moving the VCCT_GXB and VCCR GXB supplies from Layer 9 to Layer 4, and the
VCC supply from Layer 15 to Layer 9, brings them closer to the FPGA with a lower vertical
The following figure shows a comparison between the initial supply layer allocation on the
left with the new allocation on the right.
Figure 14. Original Layer Allocation (Left) Versus New Layer Allocation
As a result of these changes, the PDN Tool reports that the VCC supply
Feffective has improved from 30.68MHz to 35.61MHz.
Figure 15. VCC Supply Feffective with Better Supply Plane Allocation
The following figure shows that VCCT_GXB and VCCR_GXB supply impedances are
close to meeting the 70MHz target. However, further optimization is required because more
than 301 capacitors are still required for each of the three supplies.
Figure 16. VCC, VCCT_GXB, and VCCR_GXB Supply PDN Performance with Better
Supply Plane Allocation
Moving Power and Ground Planes Closer Together
The impedance plots imply the challenge is in meeting the target impedance
at high frequencies. Higher frequency decoupling can be achieved by increasing the
capacitance of the power and ground supply planes. This can be achieved by moving the power
and ground plane pair closer to each other.
Increasing the surface area of the power and ground plane pair also increases the
inter-plane capacitance. The default PDN Tool plane size is used throughout this
application note. You can change the plane capacitance and resistance for each power group
in the PDN Tool.
Reducing the original stackup power and ground separation to 1 mil from 4
mils should show an improvement. In the following example, in order to reduce plane
resistance, the plane thickness of the power/ground pairs has been increased to 1.2 mil.
The figure below compares the original layer-stack (left) with the new stack (right).
Figure 17. Original Stackup (Left) Versus Revised Stackup with Thinner Power and
Ground Plane Separation (Right)
The PDN Tool shows that moving the power and ground planes closer together
increases the VCC supply Feffective to 36.71MHz.
Figure 18. VCC Supply Feffective after reducing the Power and Ground Plane
Moving the power and ground planes closer together reduces the number of
decoupling capacitors for the VCCR_GXB supply to 239. More than 301 capacitors are still
required for VCC and VCCT_GXB.
Figure 19. Capacitors Required for the VCC, VCCT_GXB, and VCCR_GXB Supplies
after Reducing the Power and Ground Plane Separation
Figure 20. VCC, VCCT_GXB, and VCCR_GXB Supply PDN Performance after Reducing
the Power and Ground Plane Separation
It is clear that improvements in the PCB result in better VCC performance
and requires fewer PCB decoupling capacitors for the VCCR_GXB supply but optimization is
Move Decoupling Capacitors to the Top Surface of the PCB
In all steps so far, the decoupling capacitors are placed on the bottom
surface of the PCB and the FPGA is placed on the top.
Traditionally the decoupling capacitors are placed at the bottom side of the PCB, beneath
the FPGA and connect directly to the BGA vias. It is assumed that the best electrical
position is the closest physical location to the FPGA pins.
In the layer-stack if the power and ground plane pair are close to the
FPGA, the total vertical inductance of the vias between the capacitors and plane, and
between the plane and FPGA is less if the decoupling capacitors are placed on the top
surface of the PCB.
Assuming wide contiguous power and ground planes that have thin separation, the inductance
is much lower even though the physical horizontal distance of the capacitors placed around
the outside of the FPGA is far greater.
In this 18 Layer example, the VCCT_GXB, and VCCR_GXB supplies are allocated
to layer 4 which is closer to the FPGA than the mid point in the layer-stack. Therefore if
the capacitors are placed on the top surface of the PCB, the combined vertical loop
inductance between decoupling capacitor to plane, and plane to FPGA is lower.
Providing the PDN Tool with the freedom to place capacitors on either the
top or bottom surface of the PCB can be achieved by duplicating entries for select
capacitor values. You can allocate one to the top surface, and the other to the bottom
surface. This is shown below with the effect that the number of capacitors required for the
VCCT_GXB, and VCCR_GXB supplies is reduced to 255 and 180 respectively.
Figure 21. Capacitors Required for the VCC, VCCT_GXB, and VCCR_GXB Supplies
After Moving Capacitors to the Top Surface
The target impedance is now met up to 70MHz for the VCCT_GXB, and VCCR_GXB
supplies but the large number of high-frequency 22nF capacitors shows there are still
challenges in achieving Ztarget at high-frequencies.
Figure 22. VCC, VCCT_GXB, and VCCR_GXB Supply PDN Performance After Moving
Capacitors to the Top Surface
Using X2Y Decoupling Capacitors
As discussed, inductance is a big contributor in failing to meet target
impedance at high-frequencies. The decoupling capacitors used already specify Via On Side
(VOS) mounting which has lower mounting inductance (Lmnt), than Via On End (VOE). 0402 and
0201 capacitor case sizes that have lower Lmnt than 0603 or 0805 are also used. Previously,
significant improvements in performance were found by increasing the number of FPGA power
and ground via pairs. Using this technique with ultra-low mounting inductance X2Y
capacitors can show significant improvements. The figure below shows an example of an X2Y
capacitor mounting with two GND vias and four power vias. Effectively the spreading
inductance of the capacitor has been reduced.
Figure 23. Example of a X2Y Capacitor Mounting with two GND Vias and Four Power
Using X2Y capacitors in the PDN Tool reduces the number of capacitors
required to meet the target impedance of the VCCT_GXB, and VCCR_GXB supplies from 255 and
180 to 28 and 22 respectively.
Figure 24. Capacitors Required for the VCC, VCCT_GXB, and VCCR_GXB Supplies
When Using X2Y Capacitors
Using X2Y capacitors made a big improvement in the PDN efficiency. The
intention of this application note was to first optimize the PCB, and then consider
different capacitor technologies. The X2Y capacitors could have been used earlier in the
flow, but without the underlying PCB optimization less improvement may be seen.
The impedance plots for the VCCT_GXB, and VCCR_GXB supplies shown below
indicate a much cleaner impedance profile than before. The VCC supply still requires more
than 301 capacitors so effort is still required to meet the target impedance.
Figure 25. VCC, VCCT_GXB, and VCCR_GXB Supply PDN Performance When Using X2Y
Using Ultra–Low ESR Bulk Capacitors
Significant improvements were made using low ESL X2Y capacitors instead of
standard packaged capacitors. Some improvements can be made at low frequencies by using
ultra-low Effective Series Resistance (ESR) bulk capacitors. Ultra-low ESR bulk capacitors
can have very low resistances of 5mR when compared with approximately 50mR for standard
bulk capacitors. You can use Ultra-low ESR bulk capacitors to reduce the number of bulk
capacitors required for meeting Ztarget at low frequencies.
Enter parameters of your own user capacitors in the Library tab of the PDN Tool. In this example, 680uF and 1000uF Panasonic POSCAP
Tantalum-Polymer Capacitors with an ESR of 0.005 were used. 0.005 is entered for User5 and User6 of the Bulk Cap ESR(R) column. On the System Decap tab 680 and 1000
is entered for User5 and User6. In this case
the higher 680uF and 1000uF may also help to reduce the total capacitor count. The PDN Tool
reports that using ultra-low ESR capacitors reduces bulk capacitor count of the VCC,
VCCT_GXB, and VCCR_GXB supplies as shown below.
Figure 26. Capacitors Required for the VCC, VCCT_GXB, and VCCR_GXB Supplies
when Ultra-Low ESR Bulk Capacitors
The VCCT_GXB, and VCCR_GXB supplies now require 25 and 22 capacitors
respectively. VCC bulk capacitors are reduced from 23 to 17 when using ultra-low ESR bulk
capacitors. However the target impedance is still not met at high frequencies.
The PCB, stackup, layer allocation and capacitor use and placement are now
highly optimized. Further evaluation of the supply rail layers can still be done. The VCC
supply is placed on Layer 9, but VCCT_GXB, and VCCR_GXB are placed on Layer 4. Layer 4 is
more optimal because it has less vertical inductance.
Swapping VCC on Layer 9 with VCC, VCCT_GXB, and VCCR_GXB on Layer 4
Swapping VCC on Layer 9 with VCCT_GXB and VCCR_GXB on Layer 4 to see the
effect of moving the VCC supply closer to the FPGA at the expense of the VCCT_GXB and
VCCR_GXB supplies has the effect of improving the VCC Feffective to 38.78MHz. All three
supplies now require greater than 301 capacitors so this is not a viable solution because
the VCCT_GXB and VCCR_GXB supplies may be insufficiently decoupled. For this design
example, we keep VCC on Layer 9, and VCCT_GXB and VCCR_GXB on Layer 4.
Assessing How Much Total Capacitance Might be Required
Changing the PDN Tool Decoupling Mode from Auto to Manual allows you to enter any number
of capacitors instead of limiting you to a maximum of 301. This is useful not only for
implementing your own custom scheme, but also to estimate how much total capacitance is
required to meet the target impedance with your PCB. For example, by entering 99999
high-frequency 22nF capacitors, you can see that Feffective does not increase much. It is
very difficult to optimize the PDN any further even with this unrealistic number of
capacitors. In this example we switch back to Auto decoupling
In this case you can re-estimate and reduce where possible your power
requirements, or you can estimate your Core Clock Frequency and Current Ramp Up Period parameters for the VCC supply.
Using the Core Clock Frequency and Current Ramp Up Period Parameters
On chip noise can be split into two categories. High frequency noise and
low frequency noise. High frequency noise is generated by the transistor switching
activities and is mainly regulated by die capacitance. Low frequency noise is generated by
the average current fluctuation and has to be regulated by on package and PCB decoupling
capacitors. It takes time for the average current to ramp up or down. Some applications,
such as DSP and matrix operation, can cause current surge in a short period while current
change in normal applications may take 25 clock cycles or more to settle down. The current
transition time affects the PCB decoupling requirement and can be used to de-rate the
target impedance. For more details refer to Improving the Target
Impedance Method for PCB Decoupling of Core Power.
Some PDN tool variants allow you to add data for the Core Clock Frequency and
Current Ramp Up Period parameters using the pull-down menus. These values tell
the tool how to calculate the current ramp up period for transient events, sometimes
reducing transient current changes. The values relate to how fast the clock for the section
is running, and the length of the data pipeline. Given a transient change in the input
data, there are clock cycles in the pipeline for the algorithm to deliver the results. If
the input data change activates a broad yet short pipeline, the transient is abrupt. This
results in a large current change for the number of logic elements you are using. If the
pipeline is narrow and long, the overall change in current usage is proportionately
You can set the Core Clock Frequency parameter to a
High, Medium, Low,
or Custom set of input frequencies. The Custom
option allows you to enter a specific input frequency.
The Current Ramp Up Period parameter allows you to
specify the number of clock cycles consumed by the pipeline. You can select a High, Medium, Low, or
Custom setting. Altera recommends using a smaller value unless
you have already entered a complete design in Quartus and determined the precise value.
Using the Core Clock Frequency and Current Ramp Up Period parameters has the effect of de-rating the
target impedance at higher frequencies so that it ramps-up and is easier to meet.
In this example, entering a Medium 300MHz Core Clock Frequency and a Low 25 clock
cycle Current Ramp Up Period , the number of decoupling capacitors
required to meet the VCC supply is reduced from 301 to 37.
The Figures below show the capacitors required and impedance plot for the
Figure 27. VCC Supply Capacitors and PDN Performance When Using the Core Clock
Frequency and Current Ramp Up Period Parameters
Reducing the number of VCC supply capacitors from 301 to 37 is a big
improvement. The Core Clock Frequency and Current Ramp Up Period parameters can be applied earlier in the flow of this
application note but it is generally beneficial to optimize the efficiency of the PDN
before applying such de-rating effects.
The VCC, VCCT_GXB, and VCCR_GXB supplies are now decoupled with an
effective and acceptable decoupling solution for this PCB with the estimated current
This design study demonstrated the following improvements in PDN performance and the
reduction in the number of PCB mounted decoupling capacitors.
Table 3. Study Results showing PDN performance Improvements
and reduction in PCB mounted decoupling capacitors
PDN Design Stage
VCC Feffective (MHz)
VCC Caps Required
Original stackup, all supplies on L18, # PWR/GND vias = 50
Corrected # PWR/GND vias and layer number
Reduced spreading inductance for VCC
Increased spreading inductance for VCCR_GXB & VCCT_GXB
Reduced vertical inductance for VCC, VCCR_GXB, VCCT_GX
Reduce vertical inductance by moving supplies to more optimal
layers closer to the FPGA
Increase high-frequency capacitance by moving planes closer
Reduce vertical inductance by placing decoupling capacitors on
Reduce capacitor mounting inductance by using low ESL X2Y caps
Improve low-frequency performance by using low ESR bulk caps
De-rate VCC high-frequency requirement by using the Core Clock
Frequency and Current Ramp Up Period parameters
The diagrams below compare the original (left) and final (right) PDN Tool
number of required capacitors for the VCC, VCCT_GXB, and VCCR_GXB supplies.
Figure 28. Comparison of VCC, VCCT_GXB, and VCCR_GXB Supply Required Capacitors
Between the Original (Left) and Final (Right) PDN Designs
The diagrams below compare the performance of the original (top row) and
final (bottom row) PDN performance of the VCC, VCCT_GXB, and VCCR_GXB supplies.
Figure 29. VCC, VCCT_GXB, VCCR_GXB Supply Performance Comparison Between Original
(Top Row), and Final (Bottom Row)
It is important to analyze the performance of your PDN design given your specific PCB
configuration and FPGA current requirements.
Reductions in the decoupling complexity can be made by using accurate power
estimates for your design. Decoupling complexity is dependent on the magnitude of dynamic
current, and the dynamic current is a percentage of the IMax. Therefore over estimating the
IMax for your design can result in an excessive number of decoupling capacitors.
Decoupling capacitor savings can be made by not over-estimating the FPGA
current requirements, or with accurate use of the x or x/related settings in
the PDN Tool.
The key to optimizing your PDN design at high-frequencies is to reduce
parasitic inductance wherever possible. Increasing power and ground plane pair capacitance
also improves high-frequency performance. Reducing effective series resistance can help the
PDN performance at low-frequencies. This can be done in the following ways:
Reduce the spreading inductance from the power and ground plane pair to the FPGA by increasing the number of power and ground
vias connecting the planes to the FPGA.
Reduce the vertical loop inductance from the power and ground plane pair to the FPGA by moving them closer to the surface
of the PCB that the FPGA is mounted to.
Reduce the vertical loop inductance from the decoupling capacitors to
the power and ground plane pair by placing them on the surface of the PCB that is
closest to the planes.
Use VOS with lower mounting inductance (Lmnt) instead of VOE capacitor
Increase inter-plane capacitance of your power and ground plane pair by reducing their
dielectric thickness and increasing their surface area.
Use ultra-low (Effective Series Resistance) ESR bulk capacitors to help at low
Consider using larger vias with lower ESL to reduce via loop
Use larger diameter through hole vias for all power connections to
reduce via inductance. Micro-vias should not be used for PDN design.
Use ultra-low ESL mounting capacitors such as X2Y package styles
instead of standard 0603, 0402 or 0201 packages.
By using the Core Clock Frequency and Current Ramp Up Period parameters, it is possible to de-rate the
VCC supply at high frequencies and make the VCC supply PDN design easier.
Using the Altera PDN Tool it is possible to assess the effect of each
improvement on the PDN design. Each improvement in the PDN performance improves reliability
and saves cost through reduced numbers of PCB mounted decoupling capacitors.
Guang Chen and Dan Oh, “Improving
the Target Impedance Method for PCB Decoupling of Core Power,” in Electronic
Components and Technology Conference (ECTC), 2014 IEEE 64th.