EECS 473 Advanced Embedded Systems. Lecture 9: Groups introduce their projects Power integrity issues

Transcription

1 EECS 473 Advanced Embedded Systems Lecture 9: Groups introduce their projects Power integrity issues

2 Final proposal due today Final proposal I should have signed group agreement now. I should have feedback by Tuesday night to all groups. Expect sooner actually.

3 Where we are; where we are going Labs 1-3 done, lab 4 due this week PCB lab Much less conceptual it s about learning a tool Entering full-time project mode. Have midterm and 2 homework assignments before project due. Everything else is project (and lecture). HW1 posted by the end of the day. Due 10/18 (13 days from now) I expect it will take ~4 hours It s a nice prep for the midterm.» Also practice midterms are posted. You should be putting in ~15 hours/week into the project. The more you put in now, the less you have later. And some things (reorders, PCB redos) just take time can t cram. You really (really) want time to debug! Large part of project grade based upon fully working. More complex projects get a bit more slack, but Project due a week before classes end (design expo)

4 Restart on electrical issues Talked about PCBs last time. Finished on sizing traces for power/ground Talk today about power/ground issues. Not melting the traces (pretty easy) Keeping the voltage constant (or constant enough) Pretty hard. That idea is called power integrity We ll need to review some very basic electronics along the way. 215

5 Understanding Power and Energy Trace width for a given current Example #1 Say you have 1 amp current in an ambient max temp of 75 degrees Your wire has a thickness of (fairly standard) So you need a widith of around 40 mils. What is Example #2?

6 EECS 215/Physics 240 review Background issue #1: Inductance An inductor resists the change in the flow of electrons The light bulb is a resistor. The wire in the coil has much lower resistance (it's just wire) so what you would expect when you turn on the switch is for the bulb to glow very dimly. What happens instead is that when you close the switch, the bulb burns brightly and then gets dimmer. And when you open the switch, the bulb burns very brightly and then quickly goes out.

7 EECS 215/Physics 240 review Background issue #2: Capacitance A capacitor resists the change of voltage When you first connect the battery, bulb lights up and then dims If you then remove the battery and replace with a wire the bulb will light again and then go out.

8 EECS 215/Physics 240 review Background issue #3: Impendence Impedance (symbol Z) is a measure of the overall opposition of a circuit to current, in other words: how much the circuit impedes the flow of current. It is like resistance, but it also takes into account the effects of capacitance and inductance. Impedance is measured in ohms. Impedance is more complex than resistance because the effects of capacitance and inductance vary with the frequency of the current passing through the circuit and this means impedance varies with frequency! The effect of resistance is constant regardless of frequency.

9 Impedance EECS 215/Physics 240 review A look at impedance (with capacitors, inductors and resistors vs. frequency) Pure inductor Cap/resistor Pure cap E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 Frequency Notice the log scales!

10 Power Integrity Power Integrity In order to get digital electronics to work correctly, they need a minimum* voltage differential. If we get below that, the devices might Be slow (and thus not meet setup times) Lose state Reset or halt Just plain not work. Even a very (very) short power droop can cause the chip to die. In my experience, this is a really common problem. Keeping power/ground constant and noise/droop free is Power Integrity *and maximum. Don t forget this is really a range though we usually talk about the minimum

11 Power Integrity So? We need the Vcc/Ground differential to be fairly constant. But rapid changes in the amount of current needed will cause the voltage to spike or droop due to inductance. We basically want a no-pass filter. That is we don t want to see any signal on the Vcc/Ground lines. The obvious thing? Add a capacitor That should keep the voltage constant, right? The problem is we need to worry about a lot of frequencies AND capacitors aren t ideal.

12 Power Integrity Lots of frequencies Even fairly slow devices these days are capable of switching at very high frequencies. Basically we get drivers that have rise and fall times capable of going 1GHz or so. This means we generally have to worry about frequencies from DC all the way to 1GHz. Because our chip may be varying its draw at rates up to that fast.

13 Power Integrity Non-ideal devices. ESR is Effective Series Resistance ESL is Effective Series Inductance Ceff is the effective capacitance. How does quantity effect these values? Obviously impendence will be varying by frequency.

14 Power Integrity Other things can add to ESR/ESL Generally a bad solder job can make ESR/ESL worse. Packaging has an impact wires have inductance so surface-mount packages preferred Pads can have an impact

15 Power Integrity Power Distribution Network Talked a lot about keeping the power supply voltage constant. Should think of situation as follows: Input PDN Processor Output PDN If the processor drops 3.3V and uses 100mA, what is it s effective resistance? If the power supply is 3.3V, the processor uses 100mA and the total resistance of the PDN (Power distribution network) is.01ω, what voltage does the processor really see?

16 Consider an FPGA with the following characteristics Acceptable voltage range is from 2.65 to 2.75V Max current is 5A. What is the largest impedance we can see on the PDN and still have this work?

17 Power Integrity Impedance Given the previous table.. Decoupling Impedance vs Frequency Z(pup) Z(tant) Z(1uF) Z(0.1uF) Z(0.01uF) Z(pcb) ZT Z(LICA) E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 Frequency

18 Impedance Power Integrity Removing the PCB Decoupling Impedance vs Frequency Z(pup) Z(tant) Z(1uF) Z(0.1uF) Z(0.01uF) Z(pcb) ZT Z(LICA) E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 1.E+09 Frequency

19 artcatid=0&clmid=65&artid=85396&pg=3&_pf_=1 Power Integrity But wait VRM Voltage regulator module bulk bypass (tantalum) and decoupling capacitors (ceramic). These capacitors supply instantaneous current (at different frequencies) to the drivers until the VRM can respond. However sets of different capacitors cause problems!

20 Power Integrity Other power integrity issues Of course, one source of power integrity problems is coming from the processor Power supply just can t keep up with processor varying (what we just did) But there are other problems. And these are issues introduced by the PCB designer. Don t be that guy/gal.

21 Power Integrity Connecting ground poorly One big issue is that people think of ground as, well, ground. It isn t. Only one point is 0V. Everything else has a higher voltage. Wires aren t perfect. It s really easy to make this mistake. Classes like EECS 215 basically encourage it. Better to think of things as return path not ground. And yes, you can make the same mistake with power, but people do that a lot less often. Partly because we often have different Vcc levels on the board. But mostly because we just think of power and ground differently.

22 Power Integrity Consider the following Consider the figure on the right. Why is the top picture wrong? Let s consider the case of A being DC motor that runs at 120 Watts (12V 10A). B is processor drawing 100mA Wire from A to PSU return is 15cm long, 400mils wide. What is the voltage at the ground? 0.1A 3.3V 12V 10A 0.02Ω Top figure from The Circuit Designer s Companion. If you are going to do PCB design much, buy and read this book.

23 Power Integrity Review: Power integrity (1/2) Processors and other ICs have varying current demands Sometimes at frequencies much greater than the device itself runs at Why? So the power/ground inputs need to be able to deal with that. Basically we want those wires to be ideal and just supply how ever much or little current we need. If the current can t be supplied correctly, we ll get voltage droops. How much power noise can we accept? Depends on the part (read the spec). If it can run from 3.5V to 5.5V we just need to insure it stays in that range. So we need to make sure that given the current, we don t end up out of the voltage range. Basically need to insure that we don t drop too much voltage over the wires that are supplying the power!

24 Power Integrity Review: Power integrity (2/2) So we need the impedance of the wires to be low. Because the ICs operate at a wide variety of frequencies, we need to consider all of them. The wires themselves have a lot of inductance, so a lot of impedance at high frequencies. Need to counter this by adding capacitors. Problem is that the caps have parasitic inductance and resistance. So they don t help as well as you d like But more in parallel is good. Each cap will help with different frequency ranges. We also can get a small but lowparasitic cap out of the power/ground plane. Finally we should consider antiresonance*. * provides a very nice overview of the topic and how to address it.

25 Power Integrity Power Integrity (PI) summary Power integrity is about keeping the Vcc/ground difference constant and at the value you want. Covered two issues: Many devices that sink power do so in pulses Due to internal clocks and time-varying behavior Need caps to keep value constant But parasitic ESR/ESL cause problems So lots of them==good» Reduce ESR/ESL» Increase capacitance. Anti-resonance can cause problems! Need Spice or other tools to model.» Will do a bit of this next time Also, need to watch return paths Can easily bump up your ground level Cuts into your margin for the work above

26 Additional reading Very nice coverage of ESR and impedance in a non-idea capacitor. Touches on the fact that ESR varies by frequency! Very readable and short! Nice spice models of real capacitors. A much more academic treatment of ESR. Mildly amusing and useful (who doesn t like magic smoke?)

27 Up next Next week: Batteries Linear Regulators Week after: Fall break (HW due Wednesday after) Exam review Week after that Milestone meetings Introduction to special purpose processors (mostly DSPs)