Upload
others
View
18
Download
2
Embed Size (px)
Citation preview
Cosmin Iorga - NoiseCoupling.com - [email protected] 1
DesignCon 2012
Experimental Optimization ofDecoupling Capacitors in FPGADesigns by On-Die Measurementof Power Distribution ImpedanceFrequency ProfileCosmin [email protected] (805) 231-9786
Cosmin Iorga - NoiseCoupling.com - [email protected] 2
Outline
Overview of PDN Design and Optimization
Typical PDN Design Methodology
Differences Between PDN Impedance On-Die and On-PCB
Proposed PDN Design Methodology
FPGA Configuration for On-Die PDN ImpedanceMeasurement
Case Study Implemented in a Spartan 3A FPGA
Conclusions
Cosmin Iorga - NoiseCoupling.com - [email protected] 3
Overview of PDN Design andOptimization
Define power and ground planes in PCB stackup
Design mounting layout structures for decoupling capacitors
Assign physical location of decoupling capacitors on the PCB
Determine values and types of decoupling capacitors
Cosmin Iorga - NoiseCoupling.com - [email protected] 4
Typical PDN Design Methodology
Modeling and simulation toolsincrease cost
Iterative process may consumesignificant schedule time
No visibility of PDNimpedance seen by on-diecircuits
Expensive measurementinstruments (VNA …)
Cosmin Iorga - NoiseCoupling.com - [email protected] 5
Differences Between PDN ImpedanceOn-Die and On-PCB
FPGA
As seen by on-PCB measurements (example VNAone-port or two-port)
As seen by the FPGA on-die circuits - which generate transientcurrents and are affected by supply noise
Cosmin Iorga - NoiseCoupling.com - [email protected] 6
Differences Between PDN ImpedanceOn-Die and On-PCB
As seen by on-PCB measurements (example VNAone-port or two-port)
As seen by the FPGA on-die circuits - which generate transientcurrents and are affected by supply noise
Cosmin Iorga - NoiseCoupling.com - [email protected] 7
Proposed PDN Design MethodologyTypical Methodology Proposed Methodology
Cosmin Iorga - NoiseCoupling.com - [email protected] 8
Proposed PDN Design Methodology
PCB Stackup:
PCB Layout:
Place “open” capacitor pads asmany as could be fit
Use multiple pad sizes
Place pads close to FPGA asfloorplan allows
Larger pads for electrolytic ortantalum may be further apart
Reduce spacing betweenpower/ground planes
Cosmin Iorga - NoiseCoupling.com - [email protected] 10
AC Steady-State AnalysisImplemented in FPGAs
A group of FPGA configurable logic blocks is configured tofunction as a sinusoidal current source
Another group of FPGA configurable logic blocks is configuredto form a ring oscillator
(patent pending technology)
Cosmin Iorga - NoiseCoupling.com - [email protected] 11
Sinusoidal Current SourceImplemented with FPGA Logic Blocks
The sinusoidal current is obtained by summing the dynamic loadcharging and discharging currents of multiple CMOS gates, built as partof CLB blocks
∑ = SINCLB Ii
(patent pending technology)
Cosmin Iorga - NoiseCoupling.com - [email protected] 12
AC Steady-State AnalysisImplemented in FPGAs
The sinusoidal current flows through the power distribution networkand generates sinusoidal variation of the on-die power supply voltageThe sinusoidal variation of the voltage supply modulates thefrequency of the ring oscillatorA counter circuit measures the frequency of the ring oscillator
PDN impedance is calculated from the average frequency deviation(patent pending technology)
Cosmin Iorga - NoiseCoupling.com - [email protected] 13
Proposed PDN Design MethodologyTypical Methodology Proposed Methodology
Cosmin Iorga - NoiseCoupling.com - [email protected] 14
Case Study
The proposed PDN impedance measurement techniqueimplemented in a Spartan 3A FPGA on a test board
PDN impedance frequency profile measured atfrequencies up to 500MHz
Decoupling capacitors chosen by experimentaloptimization
Additional design tradeoffs studied through this technique
FPGA clock frequency chosen based on on-die PDNimpedance profile
Cosmin Iorga - NoiseCoupling.com - [email protected] 15
Case Study
Implementation of PDN Measurement Method
FPGA Clock Distribution Block
PDN Measurement Block
The PDN measurementblock needs a clock signal,which is provided by theFPGA clock distribution
The PDN measurementblock uses a serial interfaceconnected to FPGA I/Oblocks
Cosmin Iorga - NoiseCoupling.com - [email protected] 16
Case Study
Device Utilization Summary in a Spartan 3A FPGA
Cosmin Iorga - NoiseCoupling.com - [email protected] 17
Case Study
Experimental Optimization of DecouplingCapacitors
Start with “open pads” or“best guess” values fordecoupling capacitors:
C39, C43, C44, C45, C50
Cosmin Iorga - NoiseCoupling.com - [email protected] 18
Case Study
Measured on-die PDN impedance frequency profile
Measurement up to 500MHz
Not accurate below 20kHz
Accuracy can be extendedbelow 20kHz with theexpense of increasedmeasurement time
Cosmin Iorga - NoiseCoupling.com - [email protected] 19
Case Study
ExperimentalOptimization ofC45
1nF 22nF
47nF
100nF
220nF
Starts with 1nF andgradually increases up to220nF
Values larger than 220nFdo not further reduce theimpedance
Cosmin Iorga - NoiseCoupling.com - [email protected] 20
Case Study
Remaining resonance peak at220MHz
Capacitive deep at 320MHz
This resonant peak is causedby inductance and decouplingcapacitors in the package
After optimization a resonantpeak at 220MHz still remains
The intended FPGA clockfrequency of 200MHz ischanged to 320MHz wherethe PDN impedance has aminimum
Package and Die Contribution
Cosmin Iorga - NoiseCoupling.com - [email protected] 21
Case Study
Capacitors removed
Capacitors with addedmounting inductance
Example of Further PDN Optimization Study
Feasibility study ofmoving C39, C43, C44,C50 outside the FPGAregion and using full viasinstead of blind vias
Additional inductanceadded by liftingcapacitors and addingwire jumpers
Cosmin Iorga - NoiseCoupling.com - [email protected] 22
Test Case Summary
This test case has evaluated how the proposed FPGA on-die PDNimpedance measurement technique can help with:
Selection of decoupling capacitors’ values and types
Trade-off between placement of decoupling capacitors and PCBmanufacturing cost reduction (blind vias versus full vias)
Identifying that the intended FPGA core clock frequency of200MHz was too close to a PDN resonance peak at 220MHz
Locating a minimum PDN impedance at 320MHz where theFPGA clock has been set to operate
Some of these PDN improvements have not been possible with only on-board PDN impedance measurements since the significant contributors werethe interconnects and decoupling capacitors on the FPGA package and die
Cosmin Iorga - NoiseCoupling.com - [email protected] 23
Conclusions
Low cost alternative to existing PDN design methods
Uses only logic blocks commonly available in any FPGA
Can be implemented in any existing FPGA without theneed of built-in dedicated measurement circuits
Eliminates expensive test bench instruments andsimulation software tools
A case study has exemplified experimental optimizationof decoupling capacitors and has studied various PDNdesign tradeoffs