2

I am slightly confused in deciding decoupling/bypass capacitor for DDR2 power supply pins. Some recommendations mention using 100nF and some mention using 10nF. I know that lower capacitance is more effective at higher frequency, but when I think of functions of bypass capacitor, I also think about the sudden requirement of charge during switching, where I believe 100nF can be more stable compared to 10nF (as it can store more charge for more time) in maintaining a steady voltage. Am I correct in my understanding?

I am using a single/discrete DDR2 RAM IC from Samung.

Any advice would be much appreciable.

LoveEnigma
  • 425
  • 2
  • 7
  • 15
  • 1
    Probably 100nF will have lower impedance at all frequencies of interest and thus will be more effective as bypass. The SRF may be lower for 100nF than 10nF, but the impedance is still lower up into very high frequencies. If you haven't already, you should check whether the processor has suggestions for what capacitors to use (or maybe there is a reference design you can copy). – user57037 May 28 '15 at 07:21
  • @mkeith: Thank you for the explanation. The reference schematic of microcontroller with DDR2 interface uses 100nF, but I thought to check one step further to ensure that I get it right, as relying on reference schematic alone is not good. When is SRF of 10nF significant to consider for bypassing purpose? – LoveEnigma May 28 '15 at 07:40
  • 1
    possible duplicate of [Why does ST recommend 100nF decoupling caps for a 72MHz MCU? (And not 10nF)](http://electronics.stackexchange.com/questions/134969/why-does-st-recommend-100nf-decoupling-caps-for-a-72mhz-mcu-and-not-10nf) – Oleksandr R. May 28 '15 at 11:23
  • Thank you, but I did refer that post prior to posting. I wanted to know more specifically for DDR2 decoupling, as DDR2 frequency is much higher than 72MHz. I will refer the post again though, to ensure I did not miss anything. – LoveEnigma May 28 '15 at 11:42
  • 1
    In bypassing, the SRF is kind of incidental. The point is to maintain low impedance. If you ever use ceramic capacitors to implement a VHF or UHF filter, then you definitely need to worry about SRF, because the filter behavior will deviate from the ideal as the frequency approaches or exceeds SRF. This isn't necessarily a bad thing. It could possibly improve your filter. But you need to be aware of it or model the parasitic inductance when designing or simulating your filter. – user57037 May 29 '15 at 04:00
  • @mkeith: Okay, thanks much for the information. I really appreciate it. One more thing: When we say low value capacitors are more effective at higher frequencies what do you really mean? Low values = eliminate high frequency noise, isn't it? But as you mentioned, low impedance is what is required for bypassing. So, is it only this "low impedance" characteristic that is important and helps filter out the high frequency noise from the supply voltage AND keeps the voltage stable by providing instantaneous charge when needed? – LoveEnigma May 29 '15 at 05:55

1 Answers1

6

The "new rules" for decoupling with modern small MLCC X7R caps is to use the physically smallest cap with the larger capacitance, i.e. 0402 100nF. I'll try to find a reference for this assertion (something I stumbled across a year or two ago, but the gist of it was that, because of the shrinking scales of smaller MLCCs and the lesser impact of parasitic inductance (so long as they're located/placed in ideal locations) a single smallest-possible-size large-capacitance cap performed better.

This is in contrast to through-hole & even larger-scale SMT, where conventional wisdom (and moreso for particularly sensitive designs that warrant it) was (still is) to have 2 or more decoupling caps (i.e. 100n, 10n & even 1n), which is appropriate due to the parasitic inductances of their larger physical size.

Again, I'll try to dig up the research I read about this, as I'm sure some here will dump upon me from a great height for suggesting this :)

Techydude
  • 4,089
  • 15
  • 23
  • Thank you. I am getting a little confused about bypassing/decoupling, even though it is not new to me. I mean even in high frequency applications like DDR2, which in my case is 300MT/s (data rate), they recommend using 100nF capacitors. So I keep wondering when to use 10nF then? I understand that 10nF is more effective in filtering noise at higher freq., but I am still confused when to use small value caps. For my above question, my gut feeling says to use 100nF. One more question - what caps we place close to ICs is bypassing and caps at regulator output is decoupling? or is it vice versa? – LoveEnigma May 28 '15 at 08:22
  • 2
    @LoveEnigma the only reason to use smaller capacitors is when doing so allows one to use a physically smaller package, so that the parasitic inductance will be reduced. If you can find a large-value capacitor with the same physical construction as a small-value and with the same parasitics, the large-value one is preferable because it will have the same high-frequency response while also providing lower impedance at low frequencies. There are quite a few questions about decoupling on this site, so I would recommend you review those. – Oleksandr R. May 28 '15 at 11:19
  • 1
    This topic can get complicated quickly, & as @Oleksandr said, check out several of the other answers to similar questions on this site. As I said, small (i.e. 0603, 0402 or 0201) & 100nF should be fine. 'bypass caps' & 'decoupling caps' are the same thing, used in several scenarios. Also do some research on PCB routing techniques for bypass caps, it makes a lot of difference in your calibre of application. – Techydude May 28 '15 at 11:32
  • @OleksandrR. Okay, thank you again! I got it better now. For further reading, I will refer other posts. :) – LoveEnigma May 28 '15 at 11:45
  • 1
    here's one reference to modern decoupling approaches, this one specifically from TI with regard to BGA packages. you will surely find other references from most of the top-tier silicon vendors, either general in nature, or targeted to specific package constraints. http://processors.wiki.ti.com/index.php/General_hardware_design/BGA_PCB_design/BGA_decoupling – Techydude May 28 '15 at 23:36
  • This matches what I learned in an EMC seminar some 10 years ago or so. But the analytical proof is to just look at the typical impedance curve. When comparing 100nF and 10nF of the same series, the 100nF will have lower impedance up to very high frequencies. This low impedance is what you want for good bypass. In the past, people got hung up on SRF, without realizing impedance was the best metric. – user57037 May 29 '15 at 03:55
  • 1
    @Techydude , thanks for pointing this document out! TI has broken the link from his comment. Here is the link, that is not presently broken: https://www.ti.com/lit/SPRABV2 – PF4Public Aug 02 '19 at 19:36