Delta DPS-200PP 89G

This power supply was found inside my HP Vectra computer. It was made in 1998 and has apparently died of bad caps. I'm publishing here the full capacitor list, along with some notes.

Capacitor list

Cap Value Volt. Model Diam. Height ESR
C3 470µF 200V Rubycon 200MXP470MA40 22mm 40mm ?
C4 470µF 200V Rubycon 200MXP470MA40 22mm 40mm ?
C17 0.22µF 50V Nippon Chemi-Con EKMA500ELLR22MD07D 4mm 7mm ?
C18 0.47µF 50V Nippon Chemi-Con EKMA500ELLR47MD07D 4mm 7mm ?
C103 2200µF 6.3V Nichicon UPJ0J222MPD 10mm 31.5mm 66mΩ
C152 2200µF 16V Nippon Chemi-Con ELXJ160ELL222MK30S 12.5mm 30mm 39mΩ
C153 2200µF 10V Nippon Chemi-Con ELXJ100ELL222MK25S 12.5mm 25mm 25mΩ
C201 22µF 50V Rubycon 50YXF22MEFC5X11 5mm 11mm ?
C252 470µF 25V Nippon Chemi-Con ELXJ250ELL471MJ20S 10mm 20mm 88mΩ
C253 100µF 25V Taicon TVT1E101MK06110LL3 6.3mm 11mm ?
C302 3300µF 10V Nippon Chemi-Con ELXJ100ELL332MK35S 12.5mm 35mm 33mΩ
C303 3300µF 10V Nippon Chemi-Con ELXJ100ELL332MK35S 12.5mm 35mm 33mΩ
C601 1µF 50V Nippon Chemi-Con EKMA500ELL1R0MD07D 4mm 7mm ?
C603 0.22µF 50V Nippon Chemi-Con EKMA500ELLR22MD07D 4mm 7mm ?
C604 0.22µF 50V Nippon Chemi-Con EKMA500ELLR22MD07D 4mm 7mm ?
C605 0.22µF 50V Nippon Chemi-Con EKMA500ELLR22MD07D 4mm 7mm ?
C904 100µF 25V Nichicon UPJ1E101MED 6.3mm 15mm 390mΩ
C905 10µF 50V Taicon TVT1H100MK05110LL1 5mm 11mm ?
C906 1000µF 10V Nippon Chemi-Con ELXJ100ELL102MJ20S 10mm 20mm 88mΩ
C909 100µF 25V Taicon TVT1E101MK06110LL3 6.3mm 11mm ?

Nominal ESR measured at 20ºC, 100kHz.

C18 is hidden between the heatsink closest to the input stage and a varistor.

C17, C18, C601, C603, C604 and C605 are timing capacitors used in low voltage parts of the circuit for non-critical purposes (power on, power off delay, overvoltage and overcurrent protection). Given their DC bias is low enough, they can be replaced with MLCC capacitors which are easier to obtain nowadays.

C152, C153, C302 and C303 are all part of the output smoothing section, and thus their values are not critical, so feel free to replace them with higher value caps. I've replaced all of them with 3300µF 16V capacitors, because it was cheaper to order five of these than a singular unit of each specific value.

Repair log

After replacing all the capacitors on my board, the damn thing isn't still working, and I'm afraid I've thrown 30€ in high quality Rubycon caps down the sink. My board was indeed running the 5VSB rail just fine, but the rest of the lines are entirely dead.

First, the initial analysis I did was with a thermal imaging camera, which displayed a 600V N-Channel P2NA60 MOSFET getting slightly warm (about 55ºC) with the power supply connected to the grid but with the computer turned off.

MOSFET seen through the thermal imaging camera

Removing this MOSFET from the board and testing it on a simple test circuit proves it to be in good health, switching on and off just fine, so it being bad is out of the question. Furhter looking at the traces on the PCB, I reach the conclusion this is driving the primary of the 5VSB transformer, and thus getting hot even with the computer off from powering the other ICs on the cold side of the board.

After this, I do some other basic tests with my DMU, such as checking for the main MOSFET (a 1000V N-Channel Toshiba 2SK2611) to be shorted, or the transformer to be open, all in vain.

I then decide to ramp the efforts: I dig out my isolation transformer, the dead man's pedal I use when working with deadly voltages, and my old and trusty analog oscilloscope, so I can poke safely the live circuit.

Armed with this arsenal, I power on the power supply again, short the PW_ON signal to ground, and randomly probe around the board.

I quickly find out that, as predicted, the auxiliary MOSFET for the 5VSB rail is ticking perfectly fine, and that the main MOSFET isn't being driven at all. I also discover something I didn't expect: 14 out of 16 pins of the DNA1001D IC (a proprietary, undocumented IC made by Hitachi specifically for Delta) on the hot side are suspiciously dead. Given this is a switching-mode power supply and hence there's ought to be a timing signal always ticking for driving the MOSFET, this is very bad.

After a while on Google, which as usual yielded only useless garbage, I choose to go on the Chinese Internet (aka Baidu), and I find a repair log for the same board by a chap called 四哥, which was experiencing a couple years ago the very same issue I am, and whose report turns out to be extremely helpful for fixing this power supply.

Thanks to the magic of machine translation, I vaguely understand that this IC takes an unregulated voltage of circa 12V on pin 4, and outputs a stable 5V reference on pin 5. This signal is exclusively for the hot side of the circuit to use, and is completely independent to the 5VSB rail of the cold side.

On his PSU, this 5V reference was missing, and he was reading zero volts on that particular pin. He was unable to find a suitable replacement for the proprietary ASIC, and ended up fixing this by bypassing entirely the built-in regulator by hooking up an external LM7805.

In my mind, this sounds really odd: if the IC was going bad, my guess was that it should've gone thermonuclear and ruined the rest of the IC, specially considering we're talking about a linear regulator which are very prone to failing catastrophically. Sure enough, this must be something else going bad on the board. However, it's something quick to test and thus worth a try.

I then try injecting 5V using an external, isolated bench supply and, to my surprised, it powers on.

Just to be extra, super sure there's nothing else bad and that this missing 5V is a symptom rather than a cause, I decide to look further in Baidu, and out of pure luck I find a leaked description of the pinout of the DNA1001D, which contains a basic description of the security features of the IC. Sure enough, none of them are getting triggered, and thus the internal regulator is highly likely the only culprit.

I then attempt to look for NOS replacements on the internet, and just like him, fail to track down one. I then decide to go the same route the fellow Chinese folk, but going in a slightly less overkill way, by using a smaller Holtek HT7150 LDO, given my bench supply said it was using a little over 10mA.

Patched IC

After running for a couple hours, I can guarantee this solution seems to be pretty good, as the computer is back running again and the linear regulator is barely getting warm at 46.8ºC under load.

Regulator seen through the thermal imaging camera