Bypass Caps for digital IC decoupling - diyAudio
Go Back   Home > Forums > Source & Line > Digital Source

Digital Source Digital Players and Recorders: CD , SACD , Tape, Memory Card, etc.

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 12th July 2002, 10:28 PM   #1
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Post Bypass Caps for digital IC decoupling

Hi there,

Perhaps some of you remember the Spice simulations I posted concerning the various supply decoupling configurations of digital ICs. These simulations concerned the impedance of the bypass network, mainly based on Pete Goudreau's triplet, and its possible improvements.

But simulations don't tell the whole story, and I wanted to go a little bit further by making measurements. The problem is I don't have access to a network analyzer to measure impedances of bypass configurations. But I had the opportunity to use a spectrum analyzer. Thus I designed a little circuit which allowed me to measure the noise spectrum on the supply pin of a digital IC.
Basically, the circuit is composed of a canned 50MHz crystal oscillator, driving an 8-bit divider, namely a SMT 74HC393 in SO package. Both the XO and the 393 have their own regulators (78L05), which are fed by two 9V dry cells, the only supply common point being the ground. The PCB I designed provides room to test different packages for decoupling caps around the 74HC393. Noise on the 393's supply is monitored by the spectrum analyzer though two SMA coaxial connectors : one on the supply pin of the IC (output ot the regulator) and the other on the reg's input. The inputs of the regulators are decoupled with a 220nF/X7R/0805 on the solder side. The PCB drawing is attached here, and the story goes on on next post
Attached Images
File Type: jpg pcb.jpg (10.7 KB, 3963 views)
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 12th July 2002, 10:31 PM   #2
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Arrow Still here ?

I began to make measurements with the simpliest and widely used decoupling scheme : a 100nF/25V/0805/X7R cap, and the results were quite awful (top curve of the attached image). All the peaks you can see are generated by the clock (50 MHz), and by all the frequencies related to the counter, down to 195 kHz (50MHz/256), and all their harmonics.... Quite a hairy spectrum I must confess I was a little bit discouraged by the view of all these frequencies on a supply line

Then I slowly moved (adding and removing small SMT caps without breaking the pads, stacking caps, etc..., is not a bed of roses ), making measurements at every step. First I added a 'lytic, three types : a Panasonic 120uF/25V/FC, a Rubycon 100uF/25V/ZA, and a Sanyo 47uF/6V3/OsCon. Globally, it didn't lead to significant improvements, but it improved a bit. To my surprise, if the Panasonic FC was sligthly worse than the others, I had hard times deciding between ZA and OsCon. Curves were pretty close to each other, and the highly praised OsCon was not so good (in this application) I finally settled on the ZA, mainly because if I have to use a lot of them, they cost far less .

Now keeping the Rubycon ZA on the PCB, I stacked a 10nF/X7R/0805 on the top of the 100nF/X7R/0805. Nothing significant. Removed both of them, and replaced 'em by 100nF/X7R/0612 (first) and 10nF/X7R/0612, which was stacked on the top of the previous after measuring the 100nF alone. For people who haven't read the previous posts, these are 1206-sized SMT caps, but with the long side metallized. They are characterized by a much lower inductance. Yeah ! The spectrum region above 450MHz showed nice improvements.

Finally, I added the 22uF/10V/Y5V/1210 high-value ceramic cap from Murata. Wow !! Couldn't believe my eyes the first time I plugged the circuit in the analyzer... I checked the circuit with a scope, everything was OK. I even removed the cap, remeasured, soldered the cap again. Same amazement ! On the attached image, the bottom curves compare the original 100nF/0805 decoupling (red curve) with the - hold your breath - 100uF/ZA // 100nF/0612 // 10nF/612 // 22uF/1210/Murata (green curve) - breathe Amazingly efficient, isnt'it ? Everything in the high-frequency region has disappeared, and the other peaks are lowered by about 30dBm...(Yes I know, I'd have to convert it to Volts, but I'm too lazy. I'll do it later, but if you can't wait, 0dBm correspond to a power of 1mW into 50 Ohms, i.e. 223.6 mV)

I now have to see if I can remove one or two of the other caps of the quadruplet without degrading noise rejection. It was planned, but I miss the time...

Well, I'm sure you'd like to see the intermediate curves, don't you ? I haven't had the time to do it, but if you're interested, I'll make a little document with all the configs I tested and the corresponding curves. If it fits the size limit for the forums, I'll post it, otherwise drop me an e-mail and I'll send it to you.

I have another PCB version in the pipeline, without the through-hole 'lytic, in order to test the efficiency of the SMT 'lytics, but, given the results I wonder if it's worth the testing.

Last thing : I have results concerning the noise at the input of the regulator, with and w/o coils, chokes and three-lead caps, but I'm falling asleep now...

Feel free to comment.
Attached Images
File Type: jpg bypass.jpg (64.7 KB, 4142 views)
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 13th July 2002, 05:16 AM   #3
tiroth is offline tiroth  United States
diyAudio Member
 
Join Date: Dec 2001
Location: Pittsburgh, PA, USA
Great work! Time to start investing in those high-value ceramics.
  Reply With Quote
Old 13th July 2002, 11:35 AM   #4
hifiZen is offline hifiZen  Canada
diyAudio Member
 
hifiZen's Avatar
 
Join Date: May 2001
Location: Mountain View, CA
Wow, very interesting results. Many thanks for your hard work, and for posting your findings! I look forward to seeing your results for the coil / choke / 3-lead device. Have you tried ferrite beads as well?

__________________
- Chad.
  Reply With Quote
Old 13th July 2002, 05:43 PM   #5
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Thanks Gentlemen.

Well, concerning the coils/etc... I made tests without any filtering (for comparison), with a SMT ferrite choke (given for an impedance of 600R@100MHz), a 47uH SMT coil, and a 3 leaded SMT filter given for an attenuation better than 40dB above 5 MHz... Sorry, no ferrite bead, but the ferrite choke can be assumed as a bead, its transfer function being roughly the same. I will post detailed results soon, but I won't disclose any secrets by saying that the better the decoupling at the Vreg's output, the lower the noise at its input .
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 16th July 2002, 02:32 PM   #6
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Default A little improvement

Hello, happy taxpayers.

I made some additional measurements just to see the influence of each cap on the noise rejection. I've got interesting results, but the best one so far is obtained with the following configuration :

100uF/ZA // 22uF Murata // 100nF/X7R/0612 // 220nF/X7R/0805.

The latter cap is soldered on the solder side of the pcb, right on the output and GND pins of the 78L05. Results are shown on the attached picture, where the black curve is the "optimum" configuration obtained before, and the red one is the improved one.

I begin to get tired of testing caps , so I'll stop here... for the moment. I wish I could test Pete Goudreau's triplet, but I can not source any 220/270nF in 0612 package here in France (nobody's perfect ). Anyone knowing where I could buy some ?

If someone's interested in having all the curves for the configurations I tested, feel free to ask (drop me an e-mail). The document with all the stuff is too large to be posted here (pdf, 400k). Datafiles are also available if curves are too fuzzy.

Last, these are "only" measurements . I suppose it has a positive effect on the digital audio sound, but real tests are still to come. Anyone (apart from me) ? Should be nice to compare...

Enjoy.
Attached Images
File Type: jpg 07_13.jpg (31.2 KB, 3680 views)
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 16th July 2002, 03:50 PM   #7
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Default Pre-Regulator filtering...

I almost forgot

While testing various bypass configurations, I also tested some pre-regulator filtering. Reg's input was bypassed to ground by a 220nF/X7R/0805, and I removed the 22uF Murata chip to allow a higher noise level to pass through the reg towards its input.

Noise was measured on the battery, thus I monitored the noise reinjected by the reg in the supply lines. But must I add that the filtering devices work in the other way ?

The three devices I tested are from Murata, but I do not make money on them . They were :

A SMT Ferrite bead, as they call it (have you ever seen a bead with no hole ?), namely the BLM21BD601S, which is supposed to have an 600R impedance at 100MHz.
A SMT 1210 47uH coil, 1.3R DC, rated for 170mA, from the LQH32C series,
And a three-lead cap, the NFM41P, 200nF, withstanding 2 Amps (!)

The attached pic compares these devices. Yes I know, it's a bit noisy , but try to follow the different markers. The black curve (9V_18_dBV.dat) is the noise without any filtering, the red one (9V_17_dBV.dat) is the 47uH coil, the blue one (9V_16_dBV.dat) is the chip ferrite bead, and the purple curve is the three-lead cap. It's not obvious, but all the devices seem more or less to work . The ferrite "bead" (blue/square) is the worst, it even induces resonances above the initial level(200MHz). The coil (red/cross) has a nice behavior, especially above 500MHz. My favourite is the three-leaded cap (purple/diagonal cross), which performs quite nicely (almost nothing above 500MHz). Pick yours !

One last thing. This mainly reflects the noise going through the reg from output to input, and therefore it is better to efficiently bypass the output of the reg to avoid reinjecting noise on the supply line.

We can also imagine a combination of the different devices, either in series or in //, feel free to test

Just my $0.04
Attached Images
File Type: jpg 9v_18_15_16_17.jpg (43.8 KB, 3625 views)
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 16th July 2002, 05:58 PM   #8
diyAudio Member
 
Join Date: Feb 2002
Location: Germany
Hi Francois (still haven't found out how to do the c thing on a German keyboard),

great work!!! Having just gotten back from vacation in France, I have enough extra energy to embark on a few nitpicking comments:

1) Why doesn't the oscillator can get a ceramic supply cap? Is it inside the can? Most cans I know have very insufficient internal bypassing.

2) The HC is already quite noisy. If you need an even better signal generator, go for an AC part.

3) I think the layout is suboptimal. So while I am not surprised that the caps close to pin 14 had no great influence, I am a little surprised about the effectiveness of the 22 uF cap.
Reason: the IC draws current between pins 7 and 14. The current supplied by the 100 nF caps close to pin 14 has to travel around the case to reach pin 7. Worse yet, the current supplied by the 22 uF has to travel through the ground pad of the regulator (where it mucks up its reference - a goo d thing the regulator is so slow!) and then take a detour aroung the clk pin of the can before it can reach pin 7.
You could route the signal between pins 6 and 13 around the case or on the other side of the board. This would give you a continous ground plane underneath the IC minimizing the path and hence both inductance and radiation of EMI. Then both ceramic caps should be placed where the ground and output pins of the regulator are right now so that the ground connection of the caps is right at the ground plane. Alternatively, you could use vias at pins 7 and 14 and place the caps on the other side of the board, directly underneath the IC.
The bead or 3 pin filter (which I use very frequently) belongs between the decoupling caps and the regulator, as close to the decoupling caps as possible and with the center lead of the filter touching the ground plane that extends underneath the IC. The electrolytic could be placed in the same position as the ceramics, but on the other side of the board. In your layout, the regulator, the electrolytic and the lead going to the electrolytic serve as antennae that radiate the supply noise of the IC.

4)Where did you get the 22 uF capacitor and how much were they? I have seen the ads by Murata but wasn't able to buy them anywhere. RS components used to offer 4u7 murata caps but when I tried to get them, they stayed on the backorder list for months until they were finally cancelled. The subsequent catalogs did not even contain them any more. I use Murata caps Y5V - GRM 40 in 0805 case. The largest I can get are 10 V/2u2 and 16 V/1u0. The price is reasonable, 9 and 5 Euros for a pack of 50.
Do you have any idea why they are so much more effective than your 100 and 10 nF 0612 caps? It can't be the capacitance at these high frequencies. Is the ESL, ESR so much better for the large parts?

5) SMD ferrite coil: I am a little wary of using wound coils, even if they are on a ferrite core because the interwinding capacitance capacitance will shunt the inductive and dissipative part at very high frequency. I am surprised this is not the case here. Maybe Murata found a good way to make the windings. I don't think it can be a straight pass at 47 uH.

6) Some people advocate the use of 1 nF NPO or COG caps in parallel to the 100 nF cap. Worth a try?


Curious to hear your answers!

Eric
  Reply With Quote
Old 16th July 2002, 08:56 PM   #9
diyAudio Member
 
ftorres's Avatar
 
Join Date: Apr 2001
Location: Limoges, France
Hi Eric,
nice to have you back. How was your trip to France ? Rainy ?

Well, 'll try to answer...

2) I've taken what I had in my junkbox. I'm not a great fan of AC chips for audio, and an HC seemed a good choice. I still wanted to have noise on the supply.

1 & 3) I've only shown the visible face of the iceberg. The pcb is double-sided, with an almost complete ground plane underneath. The XO has a 100nF/X7R/0805 right on its supply pin, and the 120uF, which is also connected to the XO's reg output, is bypassed by a 220nF/1206. Otherwise, I agree, the layout is suboptimal, mainly because I wanted to have some room to be able to solder and desolder, and my iron's tip must find its way, and my fat fingers too . I know, it's not the best for minimizing inductances and paths, but I had to make some compromissions :-). Somehow, I don't know if you have counted the following direct ground connections from topside to bottom groundplane : ground of regs, ground of HC's electrolytic, ground of XO (near pin 7), those last two being metallized through-holes. I was reluctant to add a via to pin 14, since it adds inductance, which I had already too much.

I've not tested the ferrite bead between caps and reg. My aim was to provide the IC's supply pin with the most resistive path to ground, and adding the bead in series seems to add some inductance to reg's output impedance... But it's worth testing.

4) Got the 22uF from Radiospares in France (www.radiospares.fr) Reference 312-3128, for 15 Euros (excluding VAT) the bunch of ten, or 1.5 Euros each. Yes, it's a little bit pricey, but far less than an OS-Con. I think that their performances is due to their low inductance, below 1 nH. Yes, I know, 0612 do have a low ESL too, but the high cap value of the Murata helps to remove the resonances of the small caps, and smoothes the impedance of the network. But I think the 100nF or 220nF are mandatory IMHO. I have curves where the 0612 caps have been removed, leaving the 100uF/ZA and 22uF/Murata alone, and results are not that best. If you're interested, I can send you the file.

5) Don't know Just wanted to test'em since a lot a people commonly use a 47uH or 10uH coil to filter their digital supply lines.

6) Shall we try ? I'm afraid to add a high frequency resonance that can not be easily smoothed, but why not ?

Hope I've answered you...
__________________
François
"Learning French is trivial: the word for horse is cheval, and everything else follows in the same way."
  Reply With Quote
Old 16th July 2002, 09:23 PM   #10
diyAudio Member
 
Join Date: Feb 2002
Location: Germany
In that case vias close to pin 7 and the ground connections of all caps would seem adequate. At the same time you could probably cut the connections to the topside ground.

I don't quite get the point about a via close to pin 14 resulting in higher inductance. A typo?

Cheers,

Eric
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
bypass caps for bypassing mylar caps? crippledchicken Multi-Way 4 27th February 2008 04:24 AM
Are Polystyrene caps suitable for digital circuit decoupling/bypassing? homemade Parts 8 29th September 2006 03:37 PM
Caps for digital decoupling?? imperfectcircle Digital Source 5 28th July 2006 05:39 PM
Decoupling caps-Alternatives to tantalum capacitors in digital circuits? Dr.H Digital Source 20 21st February 2005 05:01 PM


New To Site? Need Help?

All times are GMT. The time now is 02:01 AM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright ©1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2