Five Myths about the PDN

Transcription

1 Slide -1 A copy of the slides is available on search VL-180 or PPT-180 Five Myths about the PDN Eric Bogatin, eric@bethesignal.com Signal Integrity Evangelist Bogatin Enterprises March 2009 Slide -2 For More Information Feature articles and columns Signal integrity public classes Online lectures Podcasts Coming soon: Hands on labs (using the Virtual Lab Bench) IEEE Professional Development Certification Program Published by Prentice Hall, 2004 Contact : info@bethesignal.com

2 Slide -3 Copyright 2009 by Bogatin Enterprises, LLC All rights reserved. No material contained in this presentation may be distributed or reproduced in whole or in part without the written permission of Bogatin Enterprises. Please respect the great deal of effort that has gone into the preparation of these lectures and use these materials for your personal use only. Bogatin Enterprises, LLC W. 110th Terr. Olathe, KS v: f: e: Slide -4 Outline The PDN What s a myth The general process to separate myth from reality Some examples

3 The Power Delivery Network (PDN) Slide -5 All the interconnect from Voltage Regulator Module (VRM) to pads on the chip Purpose: Provide stable voltage to chip pads from DC to > BW of the signals Provide low impedance return path for signals To help mitigate EMI emissions Courtesy of Altera Corp Slide -6 The Consequence of Not Getting it Right PDN 50 mv/div 2 µsec Vdd Z PDN Z chip VRM chip Clock period is 3 nsec Voltage drop on the pads of the chip may exceed the noise threshold Design Goal: do everything possible to reduce the impedance of the PDN

4 Some Common PDN Myths Slide -7 A myth: a statement that may apply to one situation, but is incorrectly generalized to all situations 1. The power and ground planes provide a lot of decoupling capacitance, especially at high frequency 2. Bigger capacitance is always better 3. Use as many decoupling capacitors as you can afford 4. The higher capacitance of thin laminates allows you to eliminate some of the capacitors from a board 5. If a signal crosses a split plane, just add a capacitor to provide a low impedance return path 6. It s important to bring the vias on each capacitor as close together as possible 7. It s better to route the Vcc connection from a capacitor directly to the power pin with a surface trace 8. Just add 3 capacitors, all with the same value to each power pin 9. Just add 3 capacitors with decade values to each power pin: 100 nf, 10 nf, 1 nf 10. Place the power and ground planes in the middle of the board 11. Swiss cheese holes in planes dramatically increase the series inductance 12. Capacitors under the BGA are always better than on the same layer Slide -8 What is the most common answer to all signal integrity questions? Are three capacitors per power pin enough? Can I use a 10 mil thick dielectric? it depends How do we answer it depends questions?

5 Slide -9 Put in the Numbers! Quantifying is the only way to distinguish Myth from Reality Calculation Measurement Analysis Characterization Slide -10 Three Analysis Tools for Every Tool Box Accuracy Rules of Thumb: Feeds your intuition, useful for order of magnitude estimating Self inductance ~ 25 nh/inch 1 st order approximations: Analytic approximations, useful for quick estimates and early design tradeoffs 2d 1 L self = 5d + in nh w + t 2 Cost required to get the answer Numerical simulation: field solver, parasitic extraction, SPICE, IBIS simulations, can base a design on this (such as Ansoft, Agilent, Mentor ) expertise money time

6 Slide -11 PDN Design Principles in 4 Minutes PDN Vdd Z PDN Z chip VRM chip Design Goal: Keep the impedance the chip sees looking into the PDN below some target impedance value Select capacitor values, and design board interconnects for lowest impedance, below ~ 100 MHz Important Features of the PDN Slide -12 Impedance, Ohms 1E1 1 1E-1 1E-2 Lowest frequency: VRM Z target VRM Bulk capacitors Ceramic decoupling capacitors Bulk decoupling capacitors: Electrolytic, tantalum Board level PDN design PCB Planes Package IC package limit Chip Highest frequency: On-chip capacitance 1E-3 1E2 1E3 1E4 1E5 1E6 1E7 1E8 1E9 freq, Hz 1E10

7 Slide -13 Meeting Cost-Performance Tradeoffs is Hard In the ideal world: Three habits for PDN design: Use power and ground planes on adjacent layers with thin dielectric, close to the surface Short surface traces for decoupling capacitors measured Select the number and size of capacitors to sculpt the impedance profile simulated HyperLynx In the real world: Too many voltage of rails Not enough planes Non-optimal shapes, Swiss cheese effect Limited location and space for capacitors FUD: fear, uncertainty, doubt 4 C s: confusing, contradictory, conflicting, complex of Bogatin Enterprises Courtesy Mentor Graphics LLC 2009 Slide -14 Myth #1: Throw as Many Capacitors on the Board as You Can Afford 1E1 in d uc ta nc e Target impedance = 0.4 Ohms VRM 1E-2 1E-3 Bulk C On-chip C = 50 nf em 1E-1 Sy st Impedance Target impedance = 2 Ohms 1 Typical PDN impedance profile, with no board caps 1E2 1E3 1E4 1E5 1E6 1E7 1E8 1E9 1E10 Hzboard caps are used Sometimes it doesn t matter howfreq, many Maybe enough bulk capacitors with VRM Maybe enough on-package capacitance Maybe enough on-chip capacitance Bogatin Enterprises LLC 2009

8 Slide -15 Myth #2: Bigger Capacitors are Always Better Impedance, Ohms 1E5 2E5 1E3 1E1 1E-1 1E-3 1E-5 1E3 1E4 1E5 1E6 1E7 1E8 1E9 freq, Hz i Z = ω C 1E10 C = 1 nf C = 10 nf C = 100 nf C = 1000 nf Why not just use a single big capacitor and be done? Slide -16 Measured Behavior of Real, 220 nf, 0603 MLCC Capacitor Sample courtesy of X2Y 4E0 Impedance, Ohms 1 1E-1 Measured impedance Capacitor s impedance > ~ 10 MHz is all about its ESL 1E-2 1E6 1E7 1E8 2E8 freq, Hz

9 Slide -17 Changing a Capacitor s C has no impact on its High Frequency Impedance ZRLC_sim R R6 R=1 TOhm I_AC SRC4 Iac=polar(1,0) A Freq=freq SRLC SRLC1 R=R1 Ohm L=L1 nh C=C1 nf Varying C, L and R Dominated by C Dominated by L Dk D R Slide -18 Myth #3: Power and Ground Planes Provide Significant Low Inductance Capacitance h A C = ε Dk 0 h A C = ε Dk 0 h A ~ h ε 0 = pf/in Dk = 4 A = area in in 2 h = dielectric thickness, inches C = capacitance in pf Dk = 4 A = area in in 2 h = dielectric thickness, mils C = capacitance in nf Suppose A = 10 in 2, h= 3 mils, what is C planes? Is this a lot or a little? Compared to what? A 10 C = = = 3.3 nf h 3

10 V dd Slide -19 Intrinsic On-Chip Decoupling Capacitance When output is low: n channel is on, p channel is off, gate capacitance of p channel acts as decoupling capacitance p channel For 0.15 micron process, gate capacitance ~ 10 ff/micron 2 n channel Changing the on-chip capacitance V ss On 10 mm x 10 mm chip, ~ 10 7 micron 2 gate area or 100 nf, low inductance capacitance Trends: thinner gate oxide, higher gate capacitance, larger die- can be > 400 nf! Slide -20 Some Capacitor Position Myths Capacitors should go underneath the BGA Capacitors should go on same surface as the BGA Route direct surface trace from capacitor to Vcc pin Bring capacitor as close to the package as possible Bring vias to capacitor as close together as possible Add multiple vias in each capacitor pad Goodness is measured by the resulting impact on ESL

11 Slide -21 Low Frequency Loop Inductance Loop L = capacitor trace mounting inductance + via loop inductance down to cavity + spreading inductance in the planes to the BGA Slide -22 Summary of First Order Approximations for Loop Inductance Len trace Len cap (all dimensions in mils) 2 1 h top h planes s D Capacitor trace inductance Via pair loop inductance Spreading inductance L L L trace vias spread B = 32 x h top = 10 x h 2 x Len w trace top = 21x h trace Len + w 2s x ln ph D planes B x ln ph D ph cap cap

12 Slide -23 Myth #4: Use a Direct Surface Trace from Capacitor to BGA s h Case 1: inches away Filled surface plane h cavity = 5 mils Case 1 Case 2: inches away w = 20 mils h cavity = 5 mils Case 2 Case 3 s h Case 3: 0603, 0.5 inches away Short surface traces Cavity 10 mils below surface h cavity = 5 mils Only use direct surface connection when surface is a copper filled power plane, with adjacent ground NEVER use narrow surface trace Slide -24 Myth #5: Capacitors Should Always Go Underneath the BGA Which location is better? s h it depends Cap on top Cap on bottom Cap on top If cavity is thick, bottom surface mounting has advantage If cavity is thin, top surface may be lower inductance

13 Slide -25 The Origin of Myths: A solution to one custom design doesn t always apply to ALL designs 0.5% > > 15% Noise tolerance 1 mohm > > 10 Ohms Target impedance 1 > 10 Total number of voltage rails and number per layer 1 > > 30 Size, shape of each power/gnd layer 2 > 30 Number of board layers no no Optimized stack up? Got the important information? yes yes Follow These Rules Slide -26 Eric s Philosophy Question authority and experts. The only way to separate myth from reality is by putting in the numbers: rules of thumb, approximations, numerical simulations, measurements on well-characterized test vehicles. Every designer should be empowered with the skill, tools, and technologies to find their own optimized cost/performance solution for each specific design.

14 Slide -27 Slide -28 For More Information Feature articles and columns Signal integrity public classes Online lectures Hands on labs (using the Virtual Lab Bench) IEEE Professional Development Certification Program Published by Prentice Hall, 2004 Contact : info@bethesignal.com