Hi,

Does anyone know of any guidance as to how wide very long straight
tracks should be to avoid manufacturing difficulties or breakage due to
board flexing (or any other potential problem I haven't thought of)? I
have to run a couple of signal tracks a distance of 500 mm and I'm
guessing that my usual 0.2 mm width would be a bad idea, but I don't
want to make them excessively wide as that would add unwanted capacitance.

Regards,

Robert

* Plain text email - safe, readable, inclusive. *

Depending on what you are doing with the signals, the resistance of that trace, or it's impedance may be more of an issue than breaking through flexing.

If flexing really will be an issue, it's more likely to fail at the ends of the long trace if that trace terminates squarely into a pad or similar. Usually in these locations, solder mask clearance cuts directly across the trace. This abrupt change can lead to increased stresses on the copper, from stress risers. This may be more prevalent with dry film masks vs. LPI for example.

In the above case, failure would be possible from a high number of flexing cycles (high cycle fatigue), but only if the board is subject to lots of flexing or vibration.

Failure at the termination of the trace, or solder mask actually, might be more likely than anywhere else. Tear dropping the trace would probably help.

Dan

On 1/9/24 09:41, Robert via groups.io wrote:

Hi,
Does anyone know of any guidance as to how wide very long straight
tracks should be to avoid manufacturing difficulties or breakage due to
board flexing (or any other potential problem I haven't thought of)??? I
have to run a couple of signal tracks a distance of 500 mm and I'm
guessing that my usual 0.2 mm width would be a bad idea, but I don't
want to make them excessively wide as that would add unwanted capacitance.
Regards,
Robert

You are talking transmission lines at >25mm length in modern circuitry.

That subject cannot be adequately covered here. See a good disertation on "microstrip" layout and design them as the transmission lines they actually are. The end result will be much more stable in operation.

* Plain text email - safe, readable, inclusive. *
.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

The lines concerned are I2C clock and data, and yes, I will be
considering them as transmission lines. However, there's no point
laying something down that is electrically just fine if the board cannot
be manufactured reliably or the tracks break during assembly (it will
get bolted down, but before that happens someone will have to pick it up
and move it into position).

Chinese PCB houses can routinely achieve 0.15 mm on normal sized boards.
So am I worrying about nothing, because they will in fact have no
trouble with a track (say) 0.2 mm wide and 500 mm long, if that's what I
would choose from an electrical viewpoint?

Regards,

Robert.

* Plain text email - safe, readable, inclusive. *

开云体育

0.2 mm sounds OK, but it's only eight thousandths of an inch. Any defect, even a change of hardness/annealing, in the copper creates a stress-concentration point, leading to a crack. I think you should make the tracks as wide as possible, for good reliability.

I wouldn't have thought you would need to treat I2C lines as
transmission lines, the data rate is quite slow (even at high speed
I2C you are still only at 4MHz). It is not as though you are laying
out a memory array where a heap of parallel address and data lines
need to be synchronised to a double data rate clock, and so edge
timings are highly critical. The I2C transfer clocking is designed to
deal with a small amount of ringing on the clock and data lines.
Considering it was originally designed to be run in a wire harness in
a TV set, it should be quite robust.

But as to track width on a PCB, as you are not constrained by
requiring impedance controlled tracks, make them wide enough that they
will be robust in manufacturing of the PCB (i.e. don't create
under-etching problems, corner peel problems, etc) so that the PCB
house minimises rejection rates, and then during assembly &
integration the tracks are robust enough to withstand knocks and rough
handling without causing micro-cracks in the tracks. Personally I
would go for something like 0.5mm, only thinning when needed to
squeeze through gaps between pins and pads.

On 1/11/24 13:52, Alan Pearce via groups.io wrote:

I wouldn't have thought you would need to treat I2C lines as
transmission lines, the data rate is quite slow (even at high speed
I2C you are still only at 4MHz). It is not as though you are laying
out a memory array where a heap of parallel address and data lines
need to be synchronised to a double data rate clock, and so edge
timings are highly critical. The I2C transfer clocking is designed to
deal with a small amount of ringing on the clock and data lines.
Considering it was originally designed to be run in a wire harness in
a TV set, it should be quite robust.
But as to track width on a PCB, as you are not constrained by
requiring impedance controlled tracks, make them wide enough that they
will be robust in manufacturing of the PCB (i.e. don't create
under-etching problems, corner peel problems, etc) so that the PCB
house minimises rejection rates, and then during assembly &
integration the tracks are robust enough to withstand knocks and rough
handling without causing micro-cracks in the tracks. Personally I
would go for something like 0.5mm, only thinning when needed to
squeeze through gaps between pins and pads.

And every time you do that, there's a huge bump in the impedance at that point. If you don't have a T.D.R., make it or get one. I've made one a few times, and made it work well enough to tell a tower crew looking for a burnout in a transmission line which joint to take apart to find the top or bottom of a line fire. Saves at least a day getting the fire damage repaired and the tv station back on the air.

I2c is the most ticklish transmission they ever threw at us, designed to work at ttl voltage levels, the drivers have so little surplus power that it can't be properly terminated to make it a real transmission line. That limits it to under 40 feet. src terminated, maybe 50 feet. With more modern fasrer circuitry.

On Thu, 11 Jan 2024 at 16:39, John Woodgate <jmw@...> wrote:

0.2 mm sounds OK, but it's only eight thousandths of an inch. Any defect, even a change of hardness/annealing, in the copper creates a stress-concentration joint, leading to a crack. I think you should make the tracks as wide as possible, for good reliability.

On 2024-01-11 16:01, Robert via groups.io wrote:

The lines concerned are I2C clock and data, and yes, I will be
considering them as transmission lines. However, there's no point
laying something down that is electrically just fine if the board cannot
be manufactured reliably or the tracks break during assembly (it will
get bolted down, but before that happens someone will have to pick it up
and move it into position).

Chinese PCB houses can routinely achieve 0.15 mm on normal sized boards.
So am I worrying about nothing, because they will in fact have no
trouble with a track (say) 0.2 mm wide and 500 mm long, if that's what I
would choose from an electrical viewpoint?

Regards,

Robert.

* Plain text email - safe, readable, inclusive. *

.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

Yeah, but if you have a 'transmission line' you would avoid needing to
vary the track width or doing other things to create impedance bumps.

And using I2C to go 40-50 feet is taking it way beyond what it was
designed for. Going those sort of distances one uses other techniques,
or runs the I2C much slower.

This is all going very off-topic, but I understand the main problem is
ending up with edges on the I2C bus that have an RC time constant that
is too slow, so one needs to drop the pull-up resistance, and eventually
it wont be possible to drive the bus without some sort of line driver.

NXP have published an application note on how to run I2C along a (long)
transmission line, such as CAT5 cable, using a line driver:



The narrower I can make the track, the higher the characteristic
impedance, and therefore the easier it will be to drive (perhaps
allowing a line driver to be avoided). But that's no good if it can't
be reliably manufactured. Failing authoritative information, I would
probably go with 0.4 mm, similar to what Alan suggested, and a prayer to
the PCB goddess. I would prefer not to have to rely on divine
intervention, so I can blame something more substantive if it doesn't
work, but all I've found so far relates to current capacity, not
manufacturability.

Regards,

Robert

* Plain text email - safe, readable, inclusive. *

On 1/12/24 07:00, Alan Pearce via groups.io wrote:

Yeah, but if you have a 'transmission line' you would avoid needing to
vary the track width or doing other things to create impedance bumps.
And using I2C to go 40-50 feet is taking it way beyond what it was
designed for. Going those sort of distances one uses other techniques,
or runs the I2C much slower.

Tell that to a now defunct maker of a quite expensive device used to synchronize a satellite signal which may have doppler effects from the satellite drifting around in its orbital "box" that takes the incoming signal well outside the then current NTSC standards. Well over 15G$ in it, had a remote control so we could rack mount the main unit. The remote came with an admonition that it could only use 40 feet of cable, and would not connect at 41 measured feet. I had to string a 36 foot cable up out of the operator console, hang it in the air near the ceiling to make it work. Squawks to the maker did no good. It wasn't running that fast but I made line interfaces that converted it to low voltage balanced lines. Running that cable nearly 60 feet in the cable trays it never made a mistake. Now I can buy that interface for $3 on ebay, and I use it in my cnc machinery in the garage. Now I'm 22 years retired, that whole control room has been torn out for an all digital system that handles 8 hidef channels in and 8 channels out 24/7. And I'm glad I retired when I did.

On Fri, 12 Jan 2024 at 00:12, Gene Heskett <gheskett@...> wrote:

On 1/11/24 13:52, Alan Pearce via groups.io wrote:

I wouldn't have thought you would need to treat I2C lines as
transmission lines, the data rate is quite slow (even at high speed
I2C you are still only at 4MHz). It is not as though you are laying
out a memory array where a heap of parallel address and data lines
need to be synchronised to a double data rate clock, and so edge
timings are highly critical. The I2C transfer clocking is designed to
deal with a small amount of ringing on the clock and data lines.
Considering it was originally designed to be run in a wire harness in
a TV set, it should be quite robust.

But as to track width on a PCB, as you are not constrained by
requiring impedance controlled tracks, make them wide enough that they
will be robust in manufacturing of the PCB (i.e. don't create
under-etching problems, corner peel problems, etc) so that the PCB
house minimises rejection rates, and then during assembly &
integration the tracks are robust enough to withstand knocks and rough
handling without causing micro-cracks in the tracks. Personally I
would go for something like 0.5mm, only thinning when needed to
squeeze through gaps between pins and pads.

But you are ten ignoring the laws of physics, the time sensitivity, and E=MC2 actually start from zero speed. I am constantly amazed at the intelligent people who think that time dialation vs speed has a floor below which it does not apply. If you had a sensitive enough clock, which a cesium beam is, you can measure the time difference of one clock is stationary and another is traveling down the interstate at 70 mph. One of the experiments to check Einsteins theory, we put a cesium beam clock in a 747 and flew it around the world. On taking it back and setting it down beside the stationary version and the moving clock had lost, to quite a few decimal places of what Einsteins theory said it would. That's good enough for me. Use a balanced line, properly terminated.

And every time you do that, there's a huge bump in the impedance at that
point. If you don't have a T.D.R., make it or get one. I've made one a
few times, and made it work well enough to tell a tower crew looking for
a burnout in a transmission line which joint to take apart to find the
top or bottom of a line fire. Saves at least a day getting the fire
damage repaired and the tv station back on the air.

I2c is the most ticklish transmission they ever threw at us, designed to
work at ttl voltage levels, the drivers have so little surplus power
that it can't be properly terminated to make it a real transmission
line. That limits it to under 40 feet. src terminated, maybe 50 feet.
With more modern fasrer circuitry.

On Thu, 11 Jan 2024 at 16:39, John Woodgate <jmw@...> wrote:

0.2 mm sounds OK, but it's only eight thousandths of an inch. Any defect, even a change of hardness/annealing, in the copper creates a stress-concentration joint, leading to a crack. I think you should make the tracks as wide as possible, for good reliability.

On 2024-01-11 16:01, Robert via groups.io wrote:

The lines concerned are I2C clock and data, and yes, I will be
considering them as transmission lines. However, there's no point
laying something down that is electrically just fine if the board cannot
be manufactured reliably or the tracks break during assembly (it will
get bolted down, but before that happens someone will have to pick it up
and move it into position).

Chinese PCB houses can routinely achieve 0.15 mm on normal sized boards.
So am I worrying about nothing, because they will in fact have no
trouble with a track (say) 0.2 mm wide and 500 mm long, if that's what I
would choose from an electrical viewpoint?

Regards,

Robert.

* Plain text email - safe, readable, inclusive. *

.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

Unless you're getting your board made in Joe's workshop under the railway arches, you shouldn't need to worry about the manufacturability of boards with 0.2mm tracks. They're routinely used in PC motherboards, for example.

Consider for a moment tracks the on a flexi-PCB. They never break with casual flexing. Copper is very malleable, not brittle. Neglecting undercutting due to over-etching, there is no reason width would have much influence on whether the trace would be likely to break when the board was bent. The force per unit width will be the same. Of course, absolute amount of undercutting is a function of the trace thickness. So a 2oz Cu weight trace might arguably be more liable to breakage than for 0.5oz. The presumption, above, doesn't take that into account.

Track breakages are most likely where they meet the pad, because you have stress points. For thin tracks meeting big pads, consider adding teardrops. I can't remember when KiCad got them, but the feature is in version 7. For the same reason, you might avoid 90° corners.

--
Regards,
Tony

If you are trying to drive I2C over long distances, consider using a PCA9600 on each end of the run.? It is a lot harder to blow up than some other buffer chips (e.g. PCA9617).? You can also use higher signal voltage levels (e.g. 12V), but it is tolerant to those higher voltages even when using low-voltage signaling.

On 1/12/24 11:31, steves via groups.io wrote:

If you are trying to drive I2C over long distances, consider using a PCA9600 on each end of the run.? It is a lot harder to blow up than some other buffer chips (e.g. PCA9617).? You can also use higher signal voltage levels (e.g. 12V), but it is tolerant to those higher voltages even when using low-voltage signaling.

Interesting Steve, although it is now way past being even historical, but when did this higher voltage version become available? My unpleasant run-in with it was in the latter half of the 90's and was my first and last, as I forbid the purchase of anything using it. The first USB had much the same problems, with a 5' limit to the length of cable. That quickly led to the use of hub chips in both ends of the cable. Cheap, had enough power to drive terminations if the cabling was good enough. Attention to the terminations soon made 10 meter cables possible, just bring the shekels for quality cabling.

It seems every new class of interface designers have to learn about the real world limits their prof didn't teach them about, often at the expense of their parent companies commercial failure. Some learned, and some are flipping burgers at in&out today.

Its amazing what can be done with a properly terminated transmission line. I was a bench tech at Oceanographic Engineering in 1959 and suddenly found myself in the middle of a project to put TV cameras on the Trieste in prep for its one and only dive into the Marianas trench northwest of the philipines. We were building what was then the smallest TV camera, to be towed thru sewers to inspect for storm water ingress. that was before the first video tape recorder so due to the size of to gondola on the Trieste we were limited to bringing in the video, and displaying it on 5" b&w monitors which were recorded ether with a Leica camera, but 35mm film spools were bulky so most of the pix you saw later were shot with a minox camera. But the Trieste gondola only had 6, 16 gauge std Packard automotive wires thru its wall, so we had to get video for 2 cameras, and lights, and 2 pan & tilt things. Thru those 6 wires. So the first thing we did was get a 1000 gallon stock watering tank, threw in 300 lbs of salt and pumped it full of mission bays brackish water since we had a fishing dock in the back yard to simulate the pacific ocean. Then went to NAPA and bought a 100' roll of that wire and unrolled it in the tank. Setup sweepers to scan that wire to 10 MHz. As long as the test gear was grounded to the tank it looked usable. we redesigned the video for a 40 ohm cable and it worked. The rest is old old history now. As a bench tech, I was a fly on the wall but I was there. The external pressure against the wall of that gondola with Lt. Walsh and Jacques Cousteau in it at the bottom of the trench? 18,000 psi. Those guys in the carbon fiber tube diving on the Titanic, signed their own death warrant, they never felt a thing because the collapse only took a millisecond.

Solve problems is what real engineers do and I was blessed by having the opportunity to learn from them.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

On 1/12/24 07:00, Alan Pearce via groups.io wrote:

Yeah, but if you have a 'transmission line' you would avoid needing to
vary the track width or doing other things to create impedance bumps.
And using I2C to go 40-50 feet is taking it way beyond what it was
designed for. Going those sort of distances one uses other techniques,
or runs the I2C much slower.

I should also mention an extremely fast setup called SPI which is totally dependent of the slew of rise and falls of the signals, to make it work, typicaly src terminated as in the case of the connection between an rpi4b's gpio pins, and a mesa 7i90HD interface card that runs my biggest lathe, an 85 yo Sheldon 11x54. Exchanging 32 bit packets of data over a 3 wire buss, typically a piece of std ribbon cable. That stuff has a typical impedance of 122 ohms. And it runs at the clock speeds of the system so the pi transmits at 42.66 megabaud, and the mesa card answers at 25 megabaud. The way I did it was to turn the pi upside down which made the ribbon cable just over an inch long. I first had a pi3b in there, but when the faster rpi4b came out swapped the card, keeping the sd card. I don't believe it has made a mistake in 8 years.

It may be a transmission line, but its very short, I've heard rumors it still works with a 1 foot cable. Using std 6" jumper wires didn't work unless I put a 10 pf cap on one of the gpio pins to slow the rise and fall times, discovered when I put a scope probe on the line and that made it work. That was far enough back up the log it was an analog scope, the usual dual trace 100 mhz Hitachi. I bought the digital storage version when it came out, then bought Siglents best a couple years ago, a 4 colored trace 350 mhz model. You can look at the ringing of a signal edge and see the clipping of the negative overshoot being squared off by the conduction of the isolation substrait diode in the ic's input at a -.6 volt level. I had read about it, but that is the first time I'd actually seen it. Actual conduction time was about 1.3 nanoseconds. You'd be amazed at what you can see with a fast enough scope.

Take care guy.

On Fri, 12 Jan 2024 at 00:12, Gene Heskett <gheskett@...> wrote:

On 1/11/24 13:52, Alan Pearce via groups.io wrote:

I wouldn't have thought you would need to treat I2C lines as
transmission lines, the data rate is quite slow (even at high speed
I2C you are still only at 4MHz). It is not as though you are laying
out a memory array where a heap of parallel address and data lines
need to be synchronised to a double data rate clock, and so edge
timings are highly critical. The I2C transfer clocking is designed to
deal with a small amount of ringing on the clock and data lines.
Considering it was originally designed to be run in a wire harness in
a TV set, it should be quite robust.

But as to track width on a PCB, as you are not constrained by
requiring impedance controlled tracks, make them wide enough that they
will be robust in manufacturing of the PCB (i.e. don't create
under-etching problems, corner peel problems, etc) so that the PCB
house minimises rejection rates, and then during assembly &
integration the tracks are robust enough to withstand knocks and rough
handling without causing micro-cracks in the tracks. Personally I
would go for something like 0.5mm, only thinning when needed to
squeeze through gaps between pins and pads.

And every time you do that, there's a huge bump in the impedance at that
point. If you don't have a T.D.R., make it or get one. I've made one a
few times, and made it work well enough to tell a tower crew looking for
a burnout in a transmission line which joint to take apart to find the
top or bottom of a line fire. Saves at least a day getting the fire
damage repaired and the tv station back on the air.

I2c is the most ticklish transmission they ever threw at us, designed to
work at ttl voltage levels, the drivers have so little surplus power
that it can't be properly terminated to make it a real transmission
line. That limits it to under 40 feet. src terminated, maybe 50 feet.
With more modern fasrer circuitry.

On Thu, 11 Jan 2024 at 16:39, John Woodgate <jmw@...> wrote:

0.2 mm sounds OK, but it's only eight thousandths of an inch. Any defect, even a change of hardness/annealing, in the copper creates a stress-concentration joint, leading to a crack. I think you should make the tracks as wide as possible, for good reliability.

On 2024-01-11 16:01, Robert via groups.io wrote:

The lines concerned are I2C clock and data, and yes, I will be
considering them as transmission lines. However, there's no point
laying something down that is electrically just fine if the board cannot
be manufactured reliably or the tracks break during assembly (it will
get bolted down, but before that happens someone will have to pick it up
and move it into position).

Chinese PCB houses can routinely achieve 0.15 mm on normal sized boards.
So am I worrying about nothing, because they will in fact have no
trouble with a track (say) 0.2 mm wide and 500 mm long, if that's what I
would choose from an electrical viewpoint?

Regards,

Robert.

* Plain text email - safe, readable, inclusive. *

.

Cheers, Gene Heskett.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

.

Cheers, Gene Heskett, CET.
--
"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.
- Louis D. Brandeis

The project changed as various engineers reported concerns, both potential
and actual. Instead of one huge board there are now four relatively small
boards, two at the top and two at the bottom. Top-to-bottom comms is a
logical extension of the CAN bus, whilst the I2C bus will run the width of
each board, a distance that doesn't give me any cause for concern, either
in terms of signal length or physical length. So thank you for your
suggestions, but to my relief the potential problem has gone away.

The project is a larger version of a product that has already been built,
and that did use one large board that took the I2C over a significant
length. I used 0.4 mm track width, and all was OK, though being a
demonstrator it hasn't been through environmental testing. So maybe the
roasting is yet to come, both for the board and for myself :)

I wouldn't want to run SPI over anything other than a short distance, as
that would be begging for trouble IMHO. I have in the past run I2C over
many metres of cable with no special drivers (just carefully designed
filtering), but it was only low speed.

Regards,

Robert.