on-chip bypass caps

Apparently it is done. There are patents for paving over the top of a chip with power distribution metalization, dielectric, and ground. The numbers make sense.

I'd just never heard of this before. It's mentioned in an Altera appnote. This is great, and largely negates their own silly guidelines for on-pcb bypassing, which we routinely ignored anyhow.

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin
Loading thread data ...

responding to

formatting link
, rolfo wrote: I just did a similar analysis for a Xilinx Zynq FPGA and found more than 3uF in the die/package for just the 1.0V core supply. Details of the method are here:
formatting link

Reply to
rolfo

(snip)

If you are interested in which tools, you might be interested in:

formatting link

Which teaches how some of the tools work, though it is a little late to start now if you want a certificate.

-- glen

Reply to
glen herrmannsfeldt

I've been progressively cranking down the number of bypass caps on my boards, and the on-chip caps reinforce that trend.

I did manage to recently design an analog circuit, a photodiode amplifier, that went unstable from too little bypassing. It includes a 1 MHz high-gain TIA and a beefy cable driver on a tiny board, and I used one ceramic cap on each supply. Oops.

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

I would be wary of doing that in general, unless supported by measurements of the actual supply ripple (and stability tests in the case of LDO regulators). The only reason I used to put supply bypass caps on the chip was (a) because the bondwire inductance makes off-chip capacitors increasingly ineffective at frequenies much above 1GHz, and (b) to make use of an empty space in the layout, (which one first tries to avoid by making the chip smaller if the padframe permits), or (c) if the power rail comes from an on-chip LDO, and bringing the capacitance on-chip allows deletion of the pad, bondwire and pin.

Capacitance is much cheaper per microfarad when it is not on the chip, so if the capacitance would be equally effective off-chip (e.g. frequency content of supply current being mostly

Reply to
Chris Jones

On a sunny day (Thu, 10 Apr 2014 16:46:11 +1000) it happened Chris Jones wrote in :

Yes, and liability, if plane with that stuff crashes, and no decoupling caps where manufacturer says there should be, the design is defective. And it is silly to save a few cents on a very expensive board that are only made a few of, and engineering is the main cost.

I have learned that a long time ago: Do Not Save On Parts. And however you slice it, the customer pays for parts, more parts higher price, more profit. He (J.L.) is not selling mass market consumer stuff, and even then reliability is important, guarantee repairs are very expensive, look at the recent car recalls for a simple ignition switch.

Reply to
Jan Panteltje

I've seen boards that have 3000 bypass caps. Caps+solder is a lot of weight!

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

On a sunny day (Thu, 10 Apr 2014 07:29:18 -0700) it happened John Larkin wrote in :

I never seen that much, and I am sure X or A do not recommend that many around their chips. This was about what the manufactireer says there should be.

Reply to
Jan Panteltje

The 3000-cap board that I saw went into an Anritsu memory tester system. It was being assembled in Hamamatsu, Japan in a tin-roof shed full of turret-feed Panasonic auto pick-and-place machines, with one visible employee.

Some of the FPGA vendors recommend a cap, or sometimes three caps, for every power ball on a BGA. I generally use two or three caps per power rail, per FPGA.

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

Nonsense. FPGAs, in particular, recommend a stupid number of decoupling caps. They claim it's needed for the absolute worst-case use of their part (which never happens). There is something to be said for "engineering".

Perhaps *you* only make a few.

That's certainly what Slowmanomics advocates. Reality is *quite* different.

There is a big difference. I doubt that JL is intentionally burying rapports of customer problems for political reasons. POTUS isn't covering for him, either.

Reply to
krw

On a sunny day (Thu, 10 Apr 2014 13:54:16 -0400) it happened snipped-for-privacy@attt.bizz wrote in :

Well, you obviously only design for non critical systems.

;-)

Reply to
Jan Panteltje

Am 08.04.2014 22:09, schrieb Phil Hobbs:

I played around with a spectrum analyzer & tracking gen. and noted the results, waiting for Xilinx ISE some years ago.

Those funny sockets are at he end: <

formatting link
>

Cheers,

Gerhard

Reply to
Gerhard Hoffmann

It could just be that their parasitic extraction tool told them capacitance was 501.596683nF and Marketing didn't understand about significant figures.

Theo

Reply to
Theo Markettos

how large was this board?

Reply to
Cydrome Leader

Very nice, thanks.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Pretty big, maybe 20" square.

--
John Larkin         Highland Technology, Inc 

jlarkin att highlandtechnology dott com 
 Click to see the full signature
Reply to
John Larkin

You obviously don't *design* anything.

Reply to
krw

0-.htm ,

nly made a few of,

Most people do. There aren't that many manufacturers of high volume product s, but many more people cater for the much more numerous niche markets.

price, more profit.

I hadn't realised that I'd spawned a school of economics.

And krw telling us about "reality" is a trifle surreal. Krw's idea of of "r eality" is what he knows, which he knows to be absolutely correct, even whe n it isn't.

bility is important, guarantee repairs are very expensive, look at the rece nt car recalls for a simple ignition switch.

I wonder where that all came from? Most likely some Tea Party political myt h that krw is silly enough to take as gospel.

JL won't have a dealer in every state to fix things that were wrong when th ey were shipped. Getting somebody out to each customer that bought a defect ive board is indeed very expensive. Service engineers aren't cheap and not everybody is willing to work as a service engineer - the Cambridge Instrume nts joke was that the service engineer's tool kit included a do-it-yourself divorce kit.

--
Bill Sloman, Sydney
Reply to
Bill Sloman

Only krw could be far enough out of touch with reality to think that Jan Panteltje doesn't design anything. The quality of his design skills may be debatable, but he does seem to be very busy turning his concepts into reality.

--
Bill Sloman, Sydney
Reply to
Bill Sloman

I do that sort of thing with TDR.

formatting link

formatting link

And here's the TDR of a PCB power plane, against the adjacent ground plane, in a multilayer PCB.

formatting link

The flat pulse is the 50 ohm hardline. It slams down to about 1 ohm when the TDR step hits the PCB plane.

For all practical purposes, the power plane is a perfect capacitor, and adding bypass caps, pretty much anywhere, increases its capacitance. So there's no big incentive to put the bypass caps very close to ICs.

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.