Tiido wrote on 2024-04-06, 10:21:

I have been trying to find info about this loss of polarity in tantalum over time and my search yields nothing, I'm most curious about it...
But what I did find is that there are filaments growing inside tantalums which can result in catastrophic failures, although in many cases they break down without damage according to a NASA study about failure modes in some types of tantalum capacitors.

I dont understand it myself its only from what I have read from anecdotes around the net, whether they are true or not I dont know, I guess an electrical engineer would be able to explain it better. Perhaps its less of a loss and more it gets shorted internally perhaps by said filaments that allows current to flow in a reverse direction which causes an explosive failure. We call it forgetting as a way for laymen to understand a complex failure mode, I do know for certain that Tantalums dont like reverse current in the slightest.

Trashbytes wrote on 2024-04-05, 09:21:

zuldan wrote on 2024-04-05, 09:05:

The Crystal chip has been ordered. Should I be replacing all the 10uF 16V Tantalum Capacitors on the board? There are 16 of them. Only 1 has a short.

If they check out ok and didn't pop the first time it powered up I wouldn't change them, but others here may have more experience with tantrum caps.

While I don't think I am a believer in the "recap everything" cult, the failure of Tantalum Capacitors is a general problem, not due to single faulty components. Most of the time, we see exploding caps rated for 16V on the +12V rail. If one tantalum blew on a system on power-up, I would recommend to swap all capacitors connected to the same voltage level, in that case, all tantalums that are on the +12V rail or the -12V rail.

As a personal, probably non-representative anecdote, I powered up a Compaq ProSignia server that was ~20 years in storage, and a tantalum on the mainboard popped. I desoldered that one, and tried it again, just to have the next tantalum pop after some more seconds of operation. After the second blown cap, I continued tests with the system outdoors, connected through an extension cord, so the smell stays outside. All three tantalum capacitors connected to +12V and finally the tantalum capacitor connected to -12V popped, but without any kind of collateral damage. At the moment, I have standard electrolytics soldered in place, but I already got replacement tantalums that will get into that system some day.

mkarcher wrote on 2024-04-06, 11:44:

Trashbytes wrote on 2024-04-05, 09:21:

zuldan wrote on 2024-04-05, 09:05:

The Crystal chip has been ordered. Should I be replacing all the 10uF 16V Tantalum Capacitors on the board? There are 16 of them. Only 1 has a short.

If they check out ok and didn't pop the first time it powered up I wouldn't change them, but others here may have more experience with tantrum caps.

While I don't think I am a believer in the "recap everything" cult, the failure of Tantalum Capacitors is a general problem, not due to single faulty components. Most of the time, we see exploding caps rated for 16V on the +12V rail. If one tantalum blew on a system on power-up, I would recommend to swap all capacitors connected to the same voltage level, in that case, all tantalums that are on the +12V rail or the -12V rail.

As a personal, probably non-representative anecdote, I powered up a Compaq ProSignia server that was ~20 years in storage, and a tantalum on the mainboard popped. I desoldered that one, and tried it again, just to have the next tantalum pop after some more seconds of operation. After the second blown cap, I continued tests with the system outdoors, connected through an extension cord, so the smell stays outside. All three tantalum capacitors connected to +12V and finally the tantalum capacitor connected to -12V popped, but without any kind of collateral damage. At the moment, I have standard electrolytics soldered in place, but I already got replacement tantalums that will get into that system some day.

The first protects the second and so on down the rail, I guess they would act much like a fuse when they fail.

Personally, I try to avoiding using 16v types altogether.
For some reason, that's apparently the type of capacitors that fail the most.
Using say 25v types somehow feels better. I guess it's because 16v rating is a tad bit too close to the 12v, leaving little room for tolerance (thinking of ripple, quick voltage spikes etc).
Something similar can be said about 6v caps on a 5v rail, I suppose. Then there's the ESR thing..

Edit: I vaguely remember something. I once heard that ideally, caps should be rated twice the voltage they're going to be used with for.
If true, a 25v cap for a 12v circuit isn't too awkward.
Or a 10v cap for a 5v circuit, too.

I finally had some time to continue with this.

C96 (ceramic capacitor) and CT13 (tantalum capacitor) appeared to be shorted so I removed and tested both of them. However, both capacitors tested fine.

I then replaced the MVA416 with a Crystal CS4216-KL then reinstalled C96 and CT13. Did some hunting for shorts. Ground and 5v is showing 30 ohms. This isn't technically a short but I just find it a bit odd. I tested ground and 5v with a bunch of other ISA sound cards and none of them are showing this.

Any ideas on how to proceed?

Did some voltage injection on the 5v line. The only thing getting hot is the Crystal CS4216-KL. Reaching about 60c on 5v and 0.97 amps.

Going to have a look at the crystal pins to make sure I haven’t shorted any of them.

Not sure what else to check after that.

zuldan wrote on 2024-05-18, 09:52:

Going to have a look at the crystal pins to make sure I haven’t shorted any of them.

Be aware that the pinout you show is for the TQFP version of that chip, but you installed the PLCC version of that chip on your sound card. Installing the PLCC version was the correct thing to do, because the card layout is made for the PLCC chip. The pinout seems identical, but the pin number assignment is different. This is not an issue if you don't get confused by the numbers printed on the card. To reduce the risk of confusion, check the next page of the data sheet. It shows the PLCC pin numbers instead of the TQFP pin numbers. The supply current is way too high. Even if only 0.7A are entering the chip, this is 7 times the maximum operating current as specified by the data sheet. Some ideas for the excessive current consumption:

• A pin shorted to ground or Vcc. Prime candidates are pins that are supposed to be powered by the CS4216, like SDOUT or the but also the REFBYP/REFBUF pins of the voltage reference.
• A supply pin not connected to Vcc/GND while it should. This can cause the internal biasing of transistors to be wrong. Make sure all of REFGND, AGND and DGND are connected to ground. Make sure VA and VD are connected to Vcc.
• A broken CS4216 chip.

mkarcher wrote on 2024-05-18, 11:28:

Be aware that the pinout you show is for the TQFP version of that chip, but you installed the PLCC version of that chip on your […]
Show full quote

zuldan wrote on 2024-05-18, 09:52:

Going to have a look at the crystal pins to make sure I haven’t shorted any of them.

Be aware that the pinout you show is for the TQFP version of that chip, but you installed the PLCC version of that chip on your sound card. Installing the PLCC version was the correct thing to do, because the card layout is made for the PLCC chip. The pinout seems identical, but the pin number assignment is different. This is not an issue if you don't get confused by the numbers printed on the card. To reduce the risk of confusion, check the next page of the data sheet. It shows the PLCC pin numbers instead of the TQFP pin numbers. The supply current is way too high. Even if only 0.7A are entering the chip, this is 7 times the maximum operating current as specified by the data sheet. Some ideas for the excessive current consumption:

• A pin shorted to ground or Vcc. Prime candidates are pins that are supposed to be powered by the CS4216, like SDOUT or the but also the REFBYP/REFBUF pins of the voltage reference.
• A supply pin not connected to Vcc/GND while it should. This can cause the internal biasing of transistors to be wrong. Make sure all of REFGND, AGND and DGND are connected to ground. Make sure VA and VD are connected to Vcc.
• A broken CS4216 chip.

Thanks for the heads up. I grabbed a screenshot from the wrong page. Updated the image now.

Thank you for all the tips. Will do some more digging tomorrow and report back.

mkarcher wrote on 2024-05-18, 11:28:

[…]
Show full quote

• A pin shorted to ground or Vcc. Prime candidates are pins that are supposed to be powered by the CS4216, like SDOUT or the but also the REFBYP/REFBUF pins of the voltage reference.
• A supply pin not connected to Vcc/GND while it should. This can cause the internal biasing of transistors to be wrong. Make sure all of REFGND, AGND and DGND are connected to ground. Make sure VA and VD are connected to Vcc.
• A broken CS4216 chip.

I confirmed REFGND, AGND and DGND are all connected to ground. VA and VD are connected to VCC.

I checked the IC pins for shorts. The following pins were shorted together 29,30,31,33,34,35,36. I thought maybe my soldering had done that so I removed the IC and tested the traces on the board and to my surprise they are still all shorted together.

zuldan wrote on Yesterday, 00:40:

I checked the IC pins for shorts. The following pins were shorted together 29,30,31,33,34,35,36. I thought maybe my soldering had done that so I removed the IC and tested the traces on the board and to my surprise they are still all shorted together.

Looking at the PCB photo, I see all of these pins connecting somewhere using wide traces. This is a hint that these pins are intended to be connected directly to GND or Vcc. It is common to use wider traces for power rails, and just make every trace wide that is connected to a power rail. Taking a more in-depth look at the image, you see pin 32 (SMODE2) connected to a different power rail than pins 29,30,31 or 33,34,35. Your 30 ohm measurement indicates that Pin 32 is at +Vcc, and you still have ~30 ohms from Vcc to GND when you measured the continuity. So it seems the chip is operated in serial mode 3 (SM3), "slave" submode (MF4=0), 64 bits per frame (MF1=MF2=0) or 256 bits per frame with no dedicated SCLK (MF1=MF2=1), data in subframe 0 (MF7=0, MF8=0). In SM3, the only output pins in that area of the chip are pins 37 (DO1) and 38 (MF5 as DO2), which are not connected to any power rail. These connections to GND and Vcc on the card seem intended and make sense.

It does sound like the chip is operating in the wrong mode, with one or more of the multi-function pins operating as outputs when they're intended to be input only. But that's probably just a symptom of VCC being almost shorted to GND; I'd say there's still more bad caps to be found.

jmarsh wrote on Yesterday, 05:51:

It does sound like the chip is operating in the wrong mode, with one or more of the multi-function pins operating as outputs when they're intended to be input only. But that's probably just a symptom of VCC being almost shorted to GND; I'd say there's still more bad caps to be found.

Another possibility which is hopefully not the case here, but nevertheless shouldn't be disregarded: We just know that the printing on the chip says CS4216, and that it consumes more power than a properly working CS4216 is intended to consume on a properly working sound card. We do not know whether the chip actually is a CS4216, or the chip has a fake marking. Furthermore, even if it is the correct chip, maybe it is damaged.

If I'm unsure whether a chip might be fake, I like to plausibilize the chip by testing the ESD diodes in them: Use a meter in diode test mode, red lead to the appropriate ground connection, black lead to a signal pin. On many pins, you will get a reading betwenn 0.3 and .7 volts. Pins with similar functions should have the same drop-out voltage +/- a couple of millivolts. In case you don't get any reading for pin-to-ground, you can also try black lead to the appropriate Vcc connection and red lead to the pin. The key point of this procedure is to apply a "negative" voltage to the pin to make the ESD diodes conducting that will be non-conducting in normal operation. Different input structures (analog input, control input, data input, buffered output) usually have different characteristics, but like pins have very similar characteristics.

This method can also be applied to test soldering: On a typical ISA card with I/O ports, A4..A9 usually are connected to very similar logic structures, and measuring address-to-rail voltage drop at the ISA connector can show you whether some of the address bits is not making contact due to a bad joint. Similar ideas usually apply to all data pins, to the control pins /IOW,/IOR,/MEMW,/MEMR and so on. Especially in old cards with a lot of custom decoding logic, /IOW and /IOR might have different loads, so this is not a clear method, yet it helps getting started on where a problem might be.

mkarcher wrote on Yesterday, 07:06:

Another possibility which is hopefully not the case here, but nevertheless shouldn't be disregarded: We just know that the print […]
Show full quote

jmarsh wrote on Yesterday, 05:51:

It does sound like the chip is operating in the wrong mode, with one or more of the multi-function pins operating as outputs when they're intended to be input only. But that's probably just a symptom of VCC being almost shorted to GND; I'd say there's still more bad caps to be found.

Another possibility which is hopefully not the case here, but nevertheless shouldn't be disregarded: We just know that the printing on the chip says CS4216, and that it consumes more power than a properly working CS4216 is intended to consume on a properly working sound card. We do not know whether the chip actually is a CS4216, or the chip has a fake marking. Furthermore, even if it is the correct chip, maybe it is damaged.

If I'm unsure whether a chip might be fake, I like to plausibilize the chip by testing the ESD diodes in them: Use a meter in diode test mode, red lead to the appropriate ground connection, black lead to a signal pin. On many pins, you will get a reading betwenn 0.3 and .7 volts. Pins with similar functions should have the same drop-out voltage +/- a couple of millivolts. In case you don't get any reading for pin-to-ground, you can also try black lead to the appropriate Vcc connection and red lead to the pin. The key point of this procedure is to apply a "negative" voltage to the pin to make the ESD diodes conducting that will be non-conducting in normal operation. Different input structures (analog input, control input, data input, buffered output) usually have different characteristics, but like pins have very similar characteristics.

This method can also be applied to test soldering: On a typical ISA card with I/O ports, A4..A9 usually are connected to very similar logic structures, and measuring address-to-rail voltage drop at the ISA connector can show you whether some of the address bits is not making contact due to a bad joint. Similar ideas usually apply to all data pins, to the control pins /IOW,/IOR,/MEMW,/MEMR and so on. Especially in old cards with a lot of custom decoding logic, /IOW and /IOR might have different loads, so this is not a clear method, yet it helps getting started on where a problem might be.

Some acetone should quickly tell us if the chip is fake or not, quick dab of acetone on the IC markings and wipe it off. If the markings stay and dont get removed or faded then its a good chance the IC is real, would still need some testing to verify its working though.

Don't forget though the primary fault was the original chip blowing its top, so the replacement acting the same isn't a strong indicator for it being fake.

jmarsh wrote on Yesterday, 18:43:

Don't forget though the primary fault was the original chip blowing its top, so the replacement acting the same isn't a strong indicator for it being fake.

Good point. Possibly the chip misbehaves (and might self-desctruct) if one of VA (supply voltage for the analog section) or VD (supply voltage for the digital section) is way below 5V while the other rail is present ("latch-up"). The datasheet recommends to use an RC lowpass filter to generate VA rejcting high frequency noise from the digital side. If the C in that filter is shorted, the R might limit the current to a kind-of sane level, but the VA will be near zero.

zuldan wrote on Yesterday, 00:40:

VA and VD are connected to VCC.

Please check whether both VA and VD actually reach 5V with the CS4216 still removed. You can do this by injecting 5V into the ISA 5V contacts as you already did when you found that the CS4216 got too hot, no need to test that in system.

jmarsh wrote on Yesterday, 05:51:

It does sound like the chip is operating in the wrong mode, with one or more of the multi-function pins operating as outputs when they're intended to be input only. But that's probably just a symptom of VCC being almost shorted to GND; I'd say there's still more bad caps to be found.

That is true. Might have a hunt around for others that may be bad.

mkarcher wrote on Yesterday, 05:15:

Looking at the PCB photo, I see all of these pins connecting somewhere using wide traces. This is a hint that these pins are intended to be connected directly to GND or Vcc. It is common to use wider traces for power rails, and just make every trace wide that is connected to a power rail. Taking a more in-depth look at the image, you see pin 32 (SMODE2) connected to a different power rail than pins 29,30,31 or 33,34,35. Your 30 ohm measurement indicates that Pin 32 is at +Vcc, and you still have ~30 ohms from Vcc to GND when you measured the continuity. So it seems the chip is operated in serial mode 3 (SM3), "slave" submode (MF4=0), 64 bits per frame (MF1=MF2=0) or 256 bits per frame with no dedicated SCLK (MF1=MF2=1), data in subframe 0 (MF7=0, MF8=0). In SM3, the only output pins in that area of the chip are pins 37 (DO1) and 38 (MF5 as DO2), which are not connected to any power rail. These connections to GND and Vcc on the card seem intended and make sense.

So just to clarify, 30 ohms between 5v and GND may be normal with this card?

mkarcher wrote on Yesterday, 07:06:

If I'm unsure whether a chip might be fake, I like to plausibilize the chip by testing the ESD diodes in them: Use a meter in diode test mode, red lead to the appropriate ground connection, black lead to a signal pin. On many pins, you will get a reading betwenn 0.3 and .7 volts. Pins with similar functions should have the same drop-out voltage +/- a couple of millivolts. In case you don't get any reading for pin-to-ground, you can also try black lead to the appropriate Vcc connection and red lead to the pin. The key point of this procedure is to apply a "negative" voltage to the pin to make the ESD diodes conducting that will be non-conducting in normal operation. Different input structures (analog input, control input, data input, buffered output) usually have different characteristics, but like pins have very similar characteristics.

What a great little test. The Crystal chip reads 0.675 and the original chip reads 0.676 so the new chip is 100% authentic.

I haven't plugged the card into a computer since replacing the IC. I thought I'd inject some voltage slowly into it before and see if anything heats up, since the crystal chip reached 60c, I thought it may be a bad idea to plug into a computer and have another chip blow.

I bought 2x Crystal chips incase my first attempt failed. I also bought tantalums to do a full replacement on the card, if needed.

Trashbytes wrote on Yesterday, 07:48:

Some acetone should quickly tell us if the chip is fake or not, quick dab of acetone on the IC markings and wipe it off. If the markings stay and dont get removed or faded then its a good chance the IC is real, would still need some testing to verify its working though.

They passed the acetone test. Must be authentic.

mkarcher wrote on Today, 00:06:

Please check whether both VA and VD actually reach 5V with the CS4216 still removed. You can do this by injecting 5V into the ISA 5V contacts as you already did when you found that the CS4216 got too hot, no need to test that in system.

VD is 4.96v and VA is 2.2v

I'd also like to point out that the Yamaha YMF262-M chip reaches about 50c and other IC's range from 35c to 45c.

I'm starting to wonder if I should have plugged the card into the machine and tested? or maybe even plug the card into the machine with the Crystal chip removed and see if I can get some basic function from it?