Does device order matter? When?

[brett]

Hi
I'm using a new PCB design tool and it routes the PCBs acording to connection networks.  The results often look very different to the schematic (I usually make PCB look a lot like the schematic, so this is freaking me out a little).

Usually, I would pay attention to device order.  e.g. put a large electrolytic cap (470uF) closer to the power supply than the small film one (0.1uF).
But does order matter?  The trace has minimal resistance, and in the case of pedals the current draw is very small. 

The only exception I can think of is for amps that might oscillate at very high frequencies (ike JFET input op-amps), where decoupling must be kept close.  But I suppose it can be "downstream" of the device (?).

Or am I missing something?
Thanks for any advice.

[R.G.]

I'm using a new PCB design tool and it routes the PCBs acording to connection networks.  The results often look very different to the schematic (I usually make PCB look a lot like the schematic, so this is freaking me out a little).

Making a PCB laid out like the schematic is usually a mistake, for reasons I covered, albeit indirectly in my PCB layout book.

Usually, I would pay attention to device order.  e.g. put a large electrolytic cap (470uF) closer to the power supply than the small film one (0.1uF).
But does order matter?  The trace has minimal resistance, and in the case of pedals the current draw is very small. 

Sometimes order matters, sometimes it doesn't. But that has already made you think "OK, so when does it matter, R.G.?".

No trace is a perfect conductor. It's a resistor with some series inductance and parallel capacitance. While these are usually negligible, they are not always. For instance, JFET and MOSFET source followers can oscillate at VHF to UHF frequencies from the parasitic inductances and capacitances of the traces leading to their gate lead. You can't even see that on a typical 20MHZ scope. And we won't talk about what can happen from parasitics on unbuffered CMOS input leads. The best trace is a zero length one. And for that, you rearrange the parts to minimize the amount of traces on the board.

Then too, you have to know your parts. A 100uF cap looks like an inductor at over a few MHZ. You *can't* bypass a high frequency capable chip or transistor with one of these. It takes a good, low ESR ceramic. Then there are the parasitics on resistors.

Bottom line - you have to know all you can about your parts, and still worry about what currents flow through which ground traces and where the decoupling caps force transients in to tiny local loops instead of broadcasting from large loops.

As you might guess, it's unlikely that a program placing parts and traces will know this stuff.

[Taylor]

I'm using a new PCB design tool and it routes the PCBs acording to connection networks.  The results often look very different to the schematic (I usually make PCB look a lot like the schematic, so this is freaking me out a little).

Making a PCB laid out like the schematic is usually a mistake, for reasons I covered, albeit indirectly in my PCB layout book.

Usually, I would pay attention to device order.  e.g. put a large electrolytic cap (470uF) closer to the power supply than the small film one (0.1uF).
But does order matter?  The trace has minimal resistance, and in the case of pedals the current draw is very small. 

Sometimes order matters, sometimes it doesn't. But that has already made you think "OK, so when does it matter, R.G.?".

No trace is a perfect conductor. It's a resistor with some series inductance and parallel capacitance. While these are usually negligible, they are not always. For instance, JFET and MOSFET source followers can oscillate at VHF to UHF frequencies from the parasitic inductances and capacitances of the traces leading to their gate lead. You can't even see that on a typical 20MHZ scope. And we won't talk about what can happen from parasitics on unbuffered CMOS input leads. The best trace is a zero length one. And for that, you rearrange the parts to minimize the amount of traces on the board.

Then too, you have to know your parts. A 100uF cap looks like an inductor at over a few MHZ. You *can't* bypass a high frequency capable chip or transistor with one of these. It takes a good, low ESR ceramic. Then there are the parasitics on resistors.

Bottom line - you have to know all you can about your parts, and still worry about what currents flow through which ground traces and where the decoupling caps force transients in to tiny local loops instead of broadcasting from large loops.

As you might guess, it's unlikely that a program placing parts and traces will know this stuff.

This is one of those posts that doesn't so much teach me, as illuminate to me how much I don't know. I wish your book was downloadable. Every time I think to buy it, I get bummed because I don't want to wait 2 weeks to get it from Small Bear.

[brett]

2 weeks.  Jumpin' Jehosophat that's quick.
Surface mail to Oz is typically 4 to 6 weeks.  Maybe longer if it's container cargo, like my mountain bike (8 weeks). 

Maybe the book could be made "payment, then download an e-copy"?

[Taylor]

Maybe the book could be made "payment, then download an e-copy"?

Right, of course I meant I'd pay for it. It's the waiting that frustrates me. I'm a member of a generation that doesn't know how to deal with non-instant information (perhaps something I should work on?)

[R.G.]

This is one of those posts that doesn't so much teach me, as illuminate to me how much I don't know. I wish your book was downloadable. Every time I think to buy it, I get bummed because I don't want to wait 2 weeks to get it from Small Bear.

I know that feeling. However, the devil makes me want to ask - was the first time you felt bummed that it would take two weeks to get there more than two weeks ago? 

[Taylor]

I know that feeling. However, the devil makes me want to ask - was the first time you felt bummed that it would take two weeks to get there more than two weeks ago? 

At least I admit the folly of my generation.

Still, you should look into doing a PDF version.

[iaresee]

Still, you should look into doing a PDF version.

Why? Then it starts showing up on torrent sites. Besides, paper is beautiful. Why screw with a medium that works so well?