Caution: JLC's .5oz inner layer may have caused issues for me.

[EEVblog]

Unless you can pin to down to an actual measurable PCB fault, I think it may be premature to blame the PCB, as it could be something like a marginal component tolerance or over-sensitivity to something which could come back to bite you later.
Random reboots don't sound like much a PCB issue to me.

Yep.
It would have to be something really exotic like a thermal issue causing a problem with a marginal hairline via or track connection or something.
You can safely bet money it's not the PCB at fault.
What are your design rules?

[mikeselectricstuff]

For a problem like random reboots, my bet would be something like marginal capacitance on a crystal oscillator or similarly sensitive node, maybe coupled with excessive track length which is slightly affected by variations between PCBs - e.g. lamination differences affecting capacitance.
Could even be something as basic as a floating input pin.

[thinkfat]

I can easily imagine a crystal oscillator with the external load caps just on the edge. On some boards, with some capacitors, it will oscillate stably, on others only sometimes. Stuff like that will drive you crazy, especially since it's hard to observe.

[Siwastaja]

Oh, again this, "I don't know what I'm doing but need a scapegoat quickly to get out of the trouble I'm in, so let's blame the most improbable factor just 'cause I feel like doing that, with absolutely no proof, not even circumstantial".

Similar to blaming compiler for weird, hard-to-find bugs by default. There are people who do that.

Get the grip and continue investigating. Need more hours staring at scope. Such is this business. That's why we are paid for it.

You need to actually reproduce a scope trace showing difference in signal levels in the supposedly same net to prove this. Solder the "probe" wires to the tracks directly to eliminate bad measurement connections. Also solder to tracks (scrape off some soldermask), not component pads, to bypass possible poor component pin solder joints.

[Rat_Patrol]

Seems to me you're blaming (and shaming) JLC without any actual evidence.

In addition, you seem to discard the possibility there might be a problem in your gerbers or design based on the assumption JLC is to blame. That's a dangerous approach. I personally would not be shipping products unless I knew *exactly* what the issue was. If JLC didn't flag a problem with your design, perhaps you should run it through some other vendors online verification to see if they catch anything. Or double check your design rules and DRC.

Just moving on is asking for trouble IMHO.

My order from PCBWay with the same gerber files sent are coming out just fine. I just put in a larger order from a vendor in South Korea, they didn't contact me yet about any issues with the gerbers.

Maybe there is something in my specs that doesn't get along with JLC that I missed, but as I said: I'm not interested in perusing this further when other fab houses can seemingly do what I need done without issue. If my gerbers were not within the capabilities of JLC, I would have hoped they would have let me know and not put out sub-standard parts.

For me, I don't have interest in "fixing" the issue since other PCB shops seemingly don't have an issue.

I still may use JLC for simpler 2 sided PCB prototypes, but no longer for my 4 layer boards.

[wraper]

Maybe there is something in my specs that doesn't get along with JLC that I missed, but as I said: I'm not interested in perusing this further when other fab houses can seemingly do what I need done without issue.

You refuse to understand there may be a problem with your design which makes it marginally stable. And PCB itself may have no issues.

[Rat_Patrol]

For a problem like random reboots, my bet would be something like marginal capacitance on a crystal oscillator or similarly sensitive node, maybe coupled with excessive track length which is slightly affected by variations between PCBs - e.g. lamination differences affecting capacitance.
Could even be something as basic as a floating input pin.

I actually thought that too. I was hell bent it had something to do with the crystal. I adjusted the crystal caps up and down pretty far (to the extreme points where the micro would not boot) and never found a value that fixed it.

Maybe it was something in my specs that didn't get along with JLC's tolerances, but I was never given any warnings of the sort when I made the orders.

[Microdoser]

you seem to discard the possibility there might be a problem in your gerbers or design

I recently had an issue where later designs of a board, one where I used discrete components instead of a daughterboard, did not work. It was a PCB to switch between other components using I2C. I checked with my scope and signals were getting through. For previous designs, I had shortened the cables and fitted my own 2.54mm Molex connector on the ends. Lately, I had decided to use the standard supplied cable even though it was a little too long because I didn't want to take my time up just crimping cables. None of the boards worked at all. I wondered whether I had made the right design for the discrete components, whether the extra length on the cable was affecting the signals or something else. I tried various values for pull-up resistors and some other changes but to no avail. I even considered manufacturing errors but there were no shorts and all signal paths were good. On the plus side, after many hours staring at a scope, I now have very good signal integrity

It took me a little while (too long) to realise that I had made two mistakes. One, I had swapped SDA and SCL on the board design, but the second mistake was I had done the same with the cables I had adjusted...

There was much self-chastisement and gnashing of teeth.

What are your design rules?

Personally, I make sure all my boards use the design rules supplied by JLCPCB for exactly this reason. I also make sure that if there are any holes for pins that they are oversized by the manufacturing tolerances plus a very little bit.

Maybe it was something in my specs that didn't get along with JLC's tolerances, but I was never given any warnings of the sort when I made the orders.

Well, unless you supply your specs to JLCPCB and ask them to check your specs against their tolerances (and pay them to do so) then, of course, they won't warn you about a problem they can't possibly predict.

They will just make the boards according to your gerbers, to within their stated tolerances, and any design problems that arise are beyond their control.

Of course, if the boards you received are out of their stated tolerances then you have a good case for getting a refund or replacement.

Are the boards you received out of their tolerance ranges?

[Rat_Patrol]

Maybe there is something in my specs that doesn't get along with JLC that I missed, but as I said: I'm not interested in perusing this further when other fab houses can seemingly do what I need done without issue.

You refuse to understand there might be a problem with your design which makes it marginally stable. And PCB itself may have no issues.

All my boards are tested at temps of -20c and +75c, put on a vibration table, and get at least 2 hours of operation before they ship. Only my orders from JLC were failing at such a rate.

[Rat_Patrol]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

[lutkeveld]

If you order with the same gerber, but different manufacturing options, its not really a fair comparison.
There might be something critical in your design that doesnt work on 0.5oz, but works on 1.5oz.

Before making this claim, its good to do a fair side by side comparison, with the same gerbers and ordering options.

[coppice]

For a problem like random reboots, my bet would be something like marginal capacitance on a crystal oscillator or similarly sensitive node, maybe coupled with excessive track length which is slightly affected by variations between PCBs - e.g. lamination differences affecting capacitance.
Could even be something as basic as a floating input pin.

Another common one is power is going through a really thin trace somewhere. Its so easy to miss a power trace that was treated like a signal trace during layout. It happens quite often, and boards get into production like that. One batch of boards may have traces a little on the fat side, and the resistance of the thin trace is not a killer. On the next batch the traces may be a little on the thin side, and the resistance causes quirks during moments of high current draw.

[mikeselectricstuff]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

Plausible if something is already very marginal - decoupling etc. on a power supply.
Copper thickness on signal traces is highly unlikely to be an issue unless frequencies (or timing dependencies) are well into the several 100s of MHz range

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

You need to understand exactly what "fail" means - you should be adding some debug instrumentation to the software to see if there is a particular place they are stopping to get a better idea of exactly what is failing.
Also try replacing the firmware with something very simple to narrow down the hardware that may be involved, though marginan failures can often be dependent on many interacting factors, so a better approach may be to cut out parts one by one to see if there is a point at which there is an obvious difference in behaviour.

[mkstevo]

I can only say I've had very good, consistent results with JLCPCB. I've ordered hundreds of boards from them. Some have been unusually shaped, some four layer, mostly two layer.

Of all the many boards we've had, one batch were manufactured "incorrectly". The problem boards had four plated holes in the corners for mounting screws. I decided on a whim to have these holes plating shaped octagonally, rather than circular. When delivered the octagonal holes had been partially rotated by about 30 degrees. This wouldn't have been a problem but for the fact that the shape for the copper planes on the top and bottom of the PCB had not been rotated to match. Thus the points of the hole plating lined up perfectly with the flat sections on top and bottom copper planes. This allowed the points to connect to both top and bottom copper layers, with the through hole plating connecting top to bottom. Top copper plane? 5V. Bottom copper plane? 0V. Thus my 5V and 0V were shorted together. Oops!

JLCPCB remanufactured the PCBs for free after I raised an issue, not before I had changed the hole shape to circular and they were of course perfect. Never used octagonal holes again. I did notice that all octagonal holes had the copper portion rotated by the same amount. In Eagle (version 7 at least) the pads for resistors and capacitors are octagonal and all these pads had been rotated but there was enough clearance between these and the ground planes for it not to be an issue. Just looked at some of the more recent boards I had delivered and they don't have any of the pads rotated.

[drussell]

Seems to me you're blaming (and shaming) JLC without any actual evidence.

In addition, you seem to discard the possibility there might be a problem in your gerbers or design based on the assumption JLC is to blame. That's a dangerous approach. I personally would not be shipping products unless I knew *exactly* what the issue was. If JLC didn't flag a problem with your design, perhaps you should run it through some other vendors online verification to see if they catch anything. Or double check your design rules and DRC.

Just moving on is asking for trouble IMHO.

My order from PCBWay with the same gerber files sent are coming out just fine. I just put in a larger order from a vendor in South Korea, they didn't contact me yet about any issues with the gerbers.

Maybe there is something in my specs that doesn't get along with JLC that I missed, but as I said: I'm not interested in perusing this further when other fab houses can seemingly do what I need done without issue. If my gerbers were not within the capabilities of JLC, I would have hoped they would have let me know and not put out sub-standard parts.

For me, I don't have interest in "fixing" the issue since other PCB shops seemingly don't have an issue.

Let me get this straight...  You have found an issue with a recent production run your product where using one particular PCB production run variant seems to show intermittent stability issues in the finished units.

Yet, instead of figuring out where the actual problem lies, which could quite possibly be something very marginal or outright defective in your original design that could well make the end products vulnerable to operational stability issues, you're seemingly content to just blame it on a "bad" PCB instead of realizing that something in your design is likely on the verge of totally failing to run properly all the time?!!   

Gee, I hope I never buy one of your gizmos...

[Ice-Tea]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

Check your power delivery.

- Fails on 0.5oz, works on 1.5oz
- Only fails when heavily loaded

Points, at least for me, to insufficiently wide power traces or planes/traces necked down by rows of vias or something. Let me also restate that it points to a marginal design. If tomorrow another supplier has a -10% tolerance on the copper thickness or on the copper width of the power trace or... you may well find yourself in trouble again. You need to understand why it works on a 1.5oz board and not on 0.5oz.

If you don't figure it out, it will remain a project that works by luck rather than by design.

[floobydust]

I had similar issues as OP with a Canadian fab- not JLC. Batch of 250 boards and about 25 had problems. Circuits didn't work or intermittent problems when flexed.
Traced it down to micro-shorts mostly between the ground-pour and adjacent traces, or between traces. PCB testing does not give 100% coverage, "flying lead" is just what you think.

When I dug in, there were tiny specs of copper here and there, I think from dust or lint at the photo-lithography stage. Etching was fine. I can't remember if the tiny specs were plated up.
The engineer who did the PCB layout did have tight spacing with the ground-pour ~0.2mm which aggravated the issue. (I can't find the pour-trace clearance spec for JLCPCB).

My conclusion was the PCB facility simply wasn't clean enough or had a dust problem that day.
I was quite pissed off after all the drama, it's always very painful dealing with PCB fab problems. But the board house gave me a credit towards the next order to make up for it.

Whether it's the PCB design or JLC, at least give them a chance instead of outright bashing them.

[Rat_Patrol]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

Check your power delivery.

- Fails on 0.5oz, works on 1.5oz
- Only fails when heavily loaded

Points, at least for me, to insufficiently wide power traces or planes/traces necked down by rows of vias or something. Let me also restate that it points to a marginal design. If tomorrow another supplier has a -10% tolerance on the copper thickness or on the copper width of the power trace or... you may well find yourself in trouble again. You need to understand why it works on a 1.5oz board and not on 0.5oz.

If you don't figure it out, it will remain a project that works by luck rather than by design.

Yup, I'll look it over and run some calculations. See what I can see. Maybe I can find something that works fine at 1oz and up but is sketchy at 0.50z pour.

When doing diagnostics, I had kept a scope on the power rail cap while I was going through this, and the 5v plane never moved when the micro would re-boot. I even kept the scope on the power pins for the micro, "perfect" 5 volts at all times (voltage deviated between 4.98 and 4.95v) while the micro would re-boot.

[langwadt]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

Check your power delivery.

- Fails on 0.5oz, works on 1.5oz
- Only fails when heavily loaded

Points, at least for me, to insufficiently wide power traces or planes/traces necked down by rows of vias or something. Let me also restate that it points to a marginal design. If tomorrow another supplier has a -10% tolerance on the copper thickness or on the copper width of the power trace or... you may well find yourself in trouble again. You need to understand why it works on a 1.5oz board and not on 0.5oz.

If you don't figure it out, it will remain a project that works by luck rather than by design.

Yup, I'll look it over and run some calculations. See what I can see. Maybe I can find something that works fine at 1oz and up but is sketchy at 0.50z pour.

When doing diagnostics, I had kept a scope on the power rail cap while I was going through this, and the 5v plane never moved when the micro would re-boot. I even kept the scope on the power pins for the micro, "perfect" 5 volts at all times (voltage deviated between 4.98 and 4.95v) while the micro would re-boot.

did you measure the ground at the micro?

[Rat_Patrol]

Seems to me you're blaming (and shaming) JLC without any actual evidence.

In addition, you seem to discard the possibility there might be a problem in your gerbers or design based on the assumption JLC is to blame. That's a dangerous approach. I personally would not be shipping products unless I knew *exactly* what the issue was. If JLC didn't flag a problem with your design, perhaps you should run it through some other vendors online verification to see if they catch anything. Or double check your design rules and DRC.

Just moving on is asking for trouble IMHO.

My order from PCBWay with the same gerber files sent are coming out just fine. I just put in a larger order from a vendor in South Korea, they didn't contact me yet about any issues with the gerbers.

Maybe there is something in my specs that doesn't get along with JLC that I missed, but as I said: I'm not interested in perusing this further when other fab houses can seemingly do what I need done without issue. If my gerbers were not within the capabilities of JLC, I would have hoped they would have let me know and not put out sub-standard parts.

For me, I don't have interest in "fixing" the issue since other PCB shops seemingly don't have an issue.

Let me get this straight...  You have found an issue with a recent production run your product where using one particular PCB production run variant seems to show intermittent stability issues in the finished units.

Yet, instead of figuring out where the actual problem lies, which could quite possibly be something very marginal or outright defective in your original design that could well make the end products vulnerable to operational stability issues, you're seemingly content to just blame it on a "bad" PCB instead of realizing that something in your design is likely on the verge of totally failing to run properly all the time?!!   

Gee, I hope I never buy one of your gizmos...

My boards are put through extremely rigorous testing prior to shipment, including extreme temps and vibration, along with a couple hours of running.
Even with the testing, all boards from JLC have been tagged with "replace upon warranty instead of repair" in my notes (all boards are serialized and tracked with build info/notes) in case any ever come back, which very seldom happens.

[Rat_Patrol]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

Check your power delivery.

- Fails on 0.5oz, works on 1.5oz
- Only fails when heavily loaded

Points, at least for me, to insufficiently wide power traces or planes/traces necked down by rows of vias or something. Let me also restate that it points to a marginal design. If tomorrow another supplier has a -10% tolerance on the copper thickness or on the copper width of the power trace or... you may well find yourself in trouble again. You need to understand why it works on a 1.5oz board and not on 0.5oz.

If you don't figure it out, it will remain a project that works by luck rather than by design.

Yup, I'll look it over and run some calculations. See what I can see. Maybe I can find something that works fine at 1oz and up but is sketchy at 0.50z pour.

When doing diagnostics, I had kept a scope on the power rail cap while I was going through this, and the 5v plane never moved when the micro would re-boot. I even kept the scope on the power pins for the micro, "perfect" 5 volts at all times (voltage deviated between 4.98 and 4.95v) while the micro would re-boot.

did you measure the ground at the micro?

Excellent point, no I didn't. I have a ground access pin that I used for the scope probe. That said, there is a LOT of ground access. From calculations I just ran, I have enough ground capacity for the internal ground plane for 2.25 amps. The system can only produce 1 amp of power for the +5v ground plane.

Just ran the calculations for 0.50z internal pour to the +5v plane, and the connections have enough capacity for 1.25 amps with a voltage drop of .019v. Not sure that is the issue there either.

I'm wondering if there was enough copper on some vital via? I am running vias at the minimum for JLC.

[langwadt]

Thinking about some of the replies in this thread, I have a thought: copper thickness.

JLC's inner copper thickness is fixed at 0.5oz, where I always order 1.5oz from other vendors.

I wonder if that isn't the issue?

There are a few signal traces that unfortunately had to run on an inner layer.

Thoughts?

The boards that fail would boot up, but fail after running for about 5 minutes once they started "working". If I let them sit idle (without being hooked up to the test bench, just powered up the unit), they would run over-night without the rebooting issue.

Check your power delivery.

- Fails on 0.5oz, works on 1.5oz
- Only fails when heavily loaded

Points, at least for me, to insufficiently wide power traces or planes/traces necked down by rows of vias or something. Let me also restate that it points to a marginal design. If tomorrow another supplier has a -10% tolerance on the copper thickness or on the copper width of the power trace or... you may well find yourself in trouble again. You need to understand why it works on a 1.5oz board and not on 0.5oz.

If you don't figure it out, it will remain a project that works by luck rather than by design.

Yup, I'll look it over and run some calculations. See what I can see. Maybe I can find something that works fine at 1oz and up but is sketchy at 0.50z pour.

When doing diagnostics, I had kept a scope on the power rail cap while I was going through this, and the 5v plane never moved when the micro would re-boot. I even kept the scope on the power pins for the micro, "perfect" 5 volts at all times (voltage deviated between 4.98 and 4.95v) while the micro would re-boot.

did you measure the ground at the micro?

Excellent point, no I didn't. I have a ground access pin that I used for the scope probe. That said, there is a LOT of ground access. From calculations I just ran, I have enough ground capacity for the internal ground plane for 2.25 amps. The system can only produce 1 amp of power for the +5v ground plane.

Just ran the calculations for 0.50z internal pour to the +5v plane, and the connections have enough capacity for 1.25 amps with a voltage drop of .019v. Not sure that is the issue there either.

I'm wondering if there was enough copper on some vital via? I am running vias at the minimum for JLC.

if you have a failing board should be easy to check, scope on gnd/vcc on decoupling cap on the micro, or beef up the gnd to the micro with a bodge wire

[thm_w]

I can only say I've had very good, consistent results with JLCPCB. I've ordered hundreds of boards from them. Some have been unusually shaped, some four layer, mostly two layer.

Of all the many boards we've had, one batch were manufactured "incorrectly". The problem boards had four plated holes in the corners for mounting screws. I decided on a whim to have these holes plating shaped octagonally, rather than circular. When delivered the octagonal holes had been partially rotated by about 30 degrees. This wouldn't have been a problem but for the fact that the shape for the copper planes on the top and bottom of the PCB had not been rotated to match. Thus the points of the hole plating lined up perfectly with the flat sections on top and bottom copper planes. This allowed the points to connect to both top and bottom copper layers, with the through hole plating connecting top to bottom. Top copper plane? 5V. Bottom copper plane? 0V. Thus my 5V and 0V were shorted together. Oops!

JLCPCB remanufactured the PCBs for free after I raised an issue, not before I had changed the hole shape to circular and they were of course perfect. Never used octagonal holes again. I did notice that all octagonal holes had the copper portion rotated by the same amount. In Eagle (version 7 at least) the pads for resistors and capacitors are octagonal and all these pads had been rotated but there was enough clearance between these and the ground planes for it not to be an issue. Just looked at some of the more recent boards I had delivered and they don't have any of the pads rotated.

The octagonal thing was discussed here as well: https://www.eevblog.com/forum/manufacture/jlc-pcb-reorientation-of-octagonal-pads/

[phil from seattle]

Wow, they still haven't fixed the octagonal pad problem.  I changed to round just to use JLCPCB so it doesn't bother me but they really do need to fix that problem.

The gerbers work fine with another manufacturer point.  Gerbers aren't created equal - there are a number of issues you need to deal with.  When I switched to from eagle to kicad, I discovered that gerbers that JLCPCB accepts choke OSHpark.  And the ones that OSHPark accepts choke JLCPCB.  With Eagle, by the way, both companies accepted and build with the same gerber file just fine.

[jmelson]

As I said above though, you have to pay extra for it (as an option) if you want 100% test. Otherwise they will just do random test of a few panels in the batch.

OOOHHHH!  Never again!  MANY decades ago, I disqualified a bunch of US fabricators when I paid for electrical test, and they spot-checked a few 4-layer boards and then sent the lot to me.  Two-layer is mostly OK, I can fix a short on those.  But, when there's an internal short on a 4-layer board, it is a REAL PAIN to fix it after the parts are stuffed.  So, that was my cardinal sin, paying for test and then not actually checking them.  I would never use that company again.

I can imagine getting away with no test on a simple, 2-layer board, but I'd NEVER, EVER do that on a multilayer.

PCBway is slightly more expensive, but they DO the testing, and I look for the prick marks.

Jon