Antec Aria AR300 PSU Repair

Have you ever bought a computer case with a proprietary PSU form factor? What do you do when it eventually wears out? One option is to repair it.

The Antec Aria

When I decided to build my Northwood-based PC back in 2004, I went for style and relative portability instead of ease of building, maintenance or cooling. While I was happy with the visual results, small footprint and portability, the same could not be said about how the processor heat sink's proximity to the side wall and the power supply's enclosure hanging right over it were causing the CPU fan to screech along at 3800 RPM after just a few minutes. This effectively forced me to permanently remove the side panel in order to make the fan noise bearable. The AR300 could have really used a lot more intake mesh area to help move air through the case, and especially out of the CPU area. Airflow inadequacy aside, everything else was fine until about a year ago.

What happened? I wanted to scan some documents. However, my scanner wasn't supported under Windows 8.1 running on a Core i5. So, I decided to boot the old Pentium 4-based box still running XP and get the job done that way. I pushed the power button and waited a few seconds. No noise, no lights, no fans. I checked the connections inside and out, and verified that the outlet I plugged into had power. Then, I unplugged the ATX connector and tried the “paperclip test”. Still nothing. Finally, I connected my digital multimeter to the 5VSB line and read 5.5V, confirming that the power supply was definitely receiving power and that something suspicious must be going on with the 5VSB output.

At this point, you might be thinking the situation sounds similar to the SL300 from a few months ago, and you would be correct. My first definitive proof that something had gone wrong with the SL300 was also increased output voltage on the 5VSB rail. Chronologically though, the AR300 failed a few months earlier. I chose to try repairing the SL300 first mainly because the SL300 could still be coerced into fully powering up, while the AR300's main outputs were completely unresponsive.

What do you do when your proprietary form factor power supply blows up? Either buy a refurbished or used replacement (if you can find a reasonably priced one from a trustworthy source), have it repaired, repair it yourself, rig it to a regular ATX PSU or ditch the proprietary case it came with. I did not feel like shoehorning my old P4 into a new case just to scan some paper, and still lacked the equipment to properly investigate at the time. I ended up rigging it by scavenging an old TigerPlus supply from a friend's PC that had been collecting dust for over a year. I put it on top of the Aria with its top and side panels off, plugged it in and called it done until further notice. The result wasn't pretty, safe or particularly reliable, but it got my scans done.

Since my first power supply repair story was such a success and some of you were curious about what went wrong with the AR300, I decided to investigate and hopefully restore my Aria to its former glory by fixing its proprietary form factor ATX power supply.

Before we get started, let's get the boring stuff out of the way.

A Foreword

I do not normally bother repairing power supplies beyond re-capping the outputs. This is only my second repair job that's more in-depth. You may want to read about my Antec SL300 PSU repair if you have not done so already, since there will be many similarities and references to it.

Disclaimer

As usual when fiddling inside line-powered equipment (and especially on the primary side of it), don't try this at home unless you are a trained professional or have equivalent experience. You assume all risks for whatever it is you decide to do.

Preliminary Observations

While a good multimeter's direct current voltage measurement tells us the average value of a signal over some period, it does not tell us what the signal looks like if it has some other stuff on top of it. You can use the alternating current measurement to get a general idea of how much non-DC voltage might be present, but the RMS, crest approximation or other numeric representation still does not really tell you what is really happening, assuming the meter even has the necessary 10-200kHz AC measurement bandwidth necessary to keep up with the frequencies that might be present on switching power supply outputs to represent a reasonable chunk of total noise and transients. The only way to know for certain what is actually happening is to use an oscilloscope.

What does that 5.8V “DC” look like when observed with microsecond-scale horizontal resolution? Surprise! I thought the SL300 was bad with power-up transients up to 16V, but the AR300 beats it hands-down with steady state transients up to 27V. I am surprised my P4's motherboard survived, and I can already predict with fairly high confidence that there will be no surviving capacitors on any of the auxiliary and 5VSB transformer outputs. The massive spikes are a dead giveaway for equally massive equivalent series resistance on flyback output filter capacitors.

After connecting a good capacitor (1200µF, 16V Panasonic FM) across the 5VSB and GND pins on the ATX connector, the switching transients almost completely vanished. However, the DC voltage rose to an uncomfortably high 10V, which was considerably worse than the SL's 6V or so. How can a motherboard survive having 10V on the 5VSB input? Simple: most circuits connected to the 5VSB supply are powered through point-of-load linear or switching regulators, and those regulators are typically built using parts rated for at least 10V on the input side. Far out of spec, but not necessarily catastrophic.

From the looks of it, the 5VSB rise with the external capacitor attached could either be an indication that the AR300's auxiliary supply capacitor(s) might be deader than the SL300's, or that it has other issues as well.

All of this may sound like déjà-vu if you read my first repair, and perhaps you're questioning the usefulness of this story. Well, let me assure you: there would not be much of a story to tell if this repair turned out exactly like the first one. Expect a twist or two.

It's time for a visual inspection to see how many issues, if any, are apparent.

Top Side

Since the top usually hosts a majority of the components we might be interested in, let's start there. Given how the two heat sinks cover nearly everything of interest, I decided to go with an angled shot. Most of the 5VSB and auxiliary components, the main points of interest, are along the board's front edge anyway.

Hidden under the primary side heat sink on the right, there is a substantial pileup of EMI filtering components, one input filter inductor, a chunky toroidal inductor for active power factor correction, the APFC and main converter MOSFETs in TO-247 packages, the APFC diode in TO-220 packaging and the 400V 220µF Fuhjyyu main input cap poking out in the top-right corner.

On the primary side of the 5VSB/auxiliary transformer, we find a 50V 100µF capacitor that appears to be the main auxiliary output filter cap, a 50V 22µF cap next to it that's presumably part of the primary side's feedback loop based on its proximity to the photocoupler, along with a handful of diodes and resistors.

The secondary side offers little worth noting. Aside from the usual crop of output filter inductors and rectifiers, we do have the following list of visibly strained capacitors:

  • a weepy 10V 4700µF Fuhjyyu TMR on the 3.3V rail – the tall blue one on the left
  • a pair of slightly bulged 10V 2200µF Teapo SEK for the main 5V output hidden under the heat sink, one of them partially visible just above the TMR capacitor with a blob of brown-orange adhesive on top of it
  • a pair of 10V 1000µF Teapo SEK for the 5VSB output near the bottom edge

Does anything else stand out on the top side? Yes, there is one fairly clear sign of something gone wrong if you look closely enough.


While looking around for capacitors that show an obvious need for replacement or other obvious issues, I noticed a discolored area around the diode next to the 5VSB transformer labeled ZD4. The device in question is a 1N4746, a 18V 1W zener diode. Does it get that hot by design? Did it get that hot due to a failure elsewhere? Did it fail? I powered up the unit for a few minutes, turned it off and felt around but could not find anything that appeared to warm up significantly apart from the 5VSB dummy-load resistor, which was getting scalding hot from having four times the power it is supposed to draw going through it due to the 5VSB rail rising to twice what it should be. This is one of those situations where a thermal imaging camera would come in really handy: watching the whole board warm up in real time makes finding unexpected hot spots orders of magnitude faster, safer and easier.

Considering how hot a printed circuit board must get before it starts darkening like this, everything in the vicinity becomes a likely suspect. Even the printed circuit board itself might be in question if the charring got bad enough to start conducting enough to interfere with normal circuit operation.

Bottom Side

Time to look at the other side and see if there is anything new and horrible waiting under there.

Aside from the charred area of interest about half way along the left edge, there does not appear to be anything else obviously wrong with it. I did a quick check on all the diodes around the auxiliary transformer just in case and did not find a shorted or open diode. Even the zener responsible for scorching the PCB still checked out fine. I went through the extra trouble of pulling it out and hooking it up to 24V through a 100Ω resistor and got 18.4V, which means something else must have failed first.

The AR300's main outputs are driven by an UC3845 controller in SOIC-8 packaging, while the APFC function is managed by a PCS01. To cram all the electronics and a 120mm fan in about two thirds the volume of a conventional power supply, the majority of low-power resistors that would have been through-hole back then have been surface-mounted on the bottom. Ceramic capacitors and some small diodes have also gone SMD. While this is often taken for granted in modern supplies and more complex circuit boards out of necessity, it was not as common a decade ago when pressure to build more compact units was not so high.

Taking a closer look at that blackened area under the zener diode, it got so hot that the glue between its copper pads and the circuit board burned off, leaving the copper pads and part of the traces leading to them loose. For this to happen, the temperature had to significantly exceed 130 ºC for some time. I could reattach the pads to the circuit board using high temperature epoxy, but before doing that, I would need to lift the pads, scrape off the charred board area and the bottom of the pads to get reasonably clean surfaces for the epoxy to bond to, then press the pads to the board until the epoxy cures. I doubt the abused traces would survive the bending they might get subjected to during the cleaning process, and a $10 tube of epoxy is rather expensive for a likely one-time use before it expires. So, I will settle for re-soldering the diode with the diode and pads pressed snug against the board to remove as much slack as possible. Securing and sealing everything in place with superglue might not be a bad idea either, assuming the zener does not normally get anywhere near as hot as it had to be to cause this damage.

The large pad on the left connects to the short-leaded end of ZD4 and got so hot that the solder mask on part of its trace sputtered off, leaving the reddish-brown copper underneath exposed. The long lead of ZD4 connects to the other large pad and only shows moderate discoloration. How hot does a pad need to get to make the solder mask not merely darken but actually smolder clean off? I could not find numbers, but I suspect we are talking somewhere north of 200 ºC, possibly hot enough to melt lead-free solder. Now that I think about it, I remember smelling overheated electronics for a few days early last year and not finding what it may have been. I did not think of trying to turn on my P4 as a probable source of the smell since I had no need for it. This must have been it.

Round One

Just like I did with the SL300, I'll replace the 5VSB output caps and the 100µF auxiliary capacitor to see how much that helps. I am not going to bother with the main output capacitors at this stage, since they are not relevant to the immediate issue of the 5VSB output being wildly out of bounds and the unit not showing any signs of powering up whatsoever when PWR_ON# is shorted to ground. Also, most of the main output caps are inconveniently tucked away under the secondary-side heat sink, so I would rather not bother with the hassle of replacing them until the 5VSB and turn-on issues have been addressed.

The new caps are in place (my usual 1200µF FMs for the 5VSB outputs and a 100µF FC for the auxiliary output). As before, they come nowhere close to fitting on the board. But that's alright since all I need them for is getting the power supply to either work or blow up. If it works, then I can add parts that fit.

Another advantage of capacitors that do not fit, or intentionally leaving long leads on temporary components, is that they make convenient test points to attach oscilloscope or multimeter probes to on the top side, rather than soldering wires on the bottom to expose the points of interest.


Don't pay attention to the Anode, Cathode and VAK traces for now; we will get back to them later. They're visible only because I had no screen captures from the first round with only the 5VSB trace on.

The result? Not much change apart from the “10VSB” output (yellow trace) being cleaner. In the SL300, the dead auxiliary capacitor was causing the flyback supply to fail to read the photocoupler's output, but it doesn't help at all in the AR300, which means there is something else to look at.

What could be the issue? A problem with the 5VSB feedback circuit? A busted feedback photocoupler? Dead feedback filter capacitor on the primary side? Busted auxiliary transformer winding? Remember “point four” from my SL300 investigation plan, which I secretly hoped I would not need to get to? Well, I am not getting away from it this time around; I will need to investigate the whole 5VSB feedback loop just to convince myself that it survived the grossly abnormal conditions it got exposed to.

The Plan

Before I start, let me tell you that I am well aware that conventional wisdom says to start troubleshooting at the source, since there is no point in troubleshooting a circuit that receives dirty, incorrect or no power. For this story's purposes, though, I thought it might be more interesting to investigate what feedback signals look like during abnormal circuit operation before fixing it, so I decided to start from the output instead.

What are my main points of interest going forward?

Looking at the signal path backwards from the 5VSB output to the primary side, we find:

  1. the TL431 shunt reference partly hidden by the capacitor, acting as a comparator and also as a driver for the photocoupler
  2. an 817 photocoupler
  3. the 22µF capacitor I did not change, which may have been roasted by the zener
  4. some other issue with the auxiliary output; taking advantage of the new capacitor's exposed leads, I hooked up my multimeter to measure its voltage and read 0V there, which seems obviously wrong

Time to start poking around and see where things break down.

The TL431 And You

What is a TL431? It is one of many widespread three-terminal programmable references, often referred to as programmable zener diodes due to how they mimic the zener diode's shunt mode voltage regulation.

What is a shunt regulator? It is a regulator that works by sinking however much current as it has to or can across its main terminals to prevent the voltage across them from exceeding a set value, similarly to how surge protectors short out surge energy.  In the programmable reference's case, the current it shunts from its cathode to its anode is a function of the voltage difference between the voltage applied to its reference pin relative to its anode and the device's own internal reference, 2.5V in this case. Using a simple voltage divider between the anode and cathode to drive the reference pin makes it possible to program any shunt voltage from the internal reference value (VRef tied to the cathode) up to its maximum voltage rating, provided that total power dissipation also remains within limits. If you read my SL300 repair, you saw my little shunt regulator schematic, which I used to drag the surging 5VSB output down to about 5.2V using a PNP transistor to boost the shunt current.

The TL431 and its countless variants from multiple power management and linear integrated circuit manufacturers are the duct tape of instrumentation electronics. They get used everywhere some sort of input needs to be weighed against a known quantity, and there are many creative ways to use them. In the case of power supplies, they function as comparators to drive the feedback photocouplers due to their flexibility and simplicity: two resistors to set the reference pin ratio, which sets the comparator's threshold, one resistor to limit current through the photocoupler, the *431 and the photocoupler, five parts in total, seven if you include an RC filter to improve transient response. Of course, it has other common uses as well, such as fan speed control and combining (ORing) feedback output from the 3.3V remote sense wire, as was the case in both the SL300 and AR300.

This is the AR300's actual 5VSB feedback circuit. If the 5VSB voltage is below roughly 5.1V, the resistor divider brings the reference pin below 2.5V, the reference stops drawing current and its cathode's voltage rises to rail voltage minus the photocoupler's diode voltage drop. If the output is above 5.1V, the resistor network pulls the reference pin above 2.5V, the cathode current increases, turns on the photocoupler, signaling the primary side to reduce power.

Looking at voltages around the AR300's 431, we see the reference pin at 2.74V and anode at 0V, which means it should be fully turned on. This is further confirmed by the 2.05V cathode-anode (VKA) voltage. The path from cathode to anode within the 431 and direct equivalents has three base-emitter junction voltage drops, and at about 0.65V each, 2V is just about the lowest possible shunt voltage. There is little doubt that the photocoupler and its current-limiting resistor are seeing the balance of the 12V currently present on the 5VSB output.

From the looks of it, the 431 appears to be working perfectly fine. Its state is consistent with a shunt reference trying to pull its cathode down as hard as it can. Time to move on to the next step: the photocoupler's LED.

What does the signal across the photocouper's LED look like in action? I'm glad you asked since I have a screen capture of that--it is that 10VSB picture I said I would get back to later. Now is later.

Here is what happens to the 5VSB feedback photocoupler LEDs as the 5VSB output starts ramping up after the power supply gets plugged in. Until the 5VSB output reaches 5V, voltage across the LED settles at about 1V due to the TL431's minimum operating current of about 100µA. Shortly after the 5VSB supply passes the 5V mark at the vertical cursor, though, voltage across the IR LED increases to 1.5V, indicating a significant current increase through it. You can also see the LED's anode and cathode voltages dropping by roughly 2V when the TL431 turns on. With 5VSB at 12V, the LED's anode at 3.6V and the 75Ω in-between, Ohm's law dictates about 112mA must be flowing through them, which is just over double what the 817 photocoupler's LED is rated for and slightly more than the 100mA the TL431 is specified at.

Despite the high current flowing through the LED, it appears to still be behaving like a typical infrared LED should, so I am inclined to believe it is still perfectly fine after possibly weeks of operating at more than twice its rated power. I did not take measurements while the power supply was connected to the motherboard so I have no idea exactly what the motherboard and 5VSB output were actually exposed to back then, and I had no intention of plugging it back in just to find out. I wonder how much longer this could have continued going on if I had not happened to need to use my Pentium 4 for a while longer. If the photocoupler's LED had failed, there would have been no feedback whatsoever left and things could have become much worse.

The fact that the 5VSB output was indeed 5V (if you overlook switching transients before I added external capacitors or replaced the internal ones) indicates that the primary side of the feedback loop was still at least sort of working, much like the SL300's. After investigation, it looks like it is none the worse for wear.

There's nothing more to see in the relative safety of the secondary side. It's time to start poking around the primary side.

The Great And Powerful Primary

As usual when messing around the primary side, be aware that most circuits there are referenced to the negative end of the input filter capacitor(s), which means up to about -170V might be present on 120V input. An additional detail to keep in mind is that this unit has active power factor correction and when the main outputs are active, the full APFC boost voltage of about 350V will be present. Because my probes and oscilloscope inputs are only rated for 300V to ground, I won't be able to probe the primary side while the main outputs and APFC are active.

Moving Along

Since the last thing we looked at on the secondary side was the photocoupler, let's see if anything interesting is happening across its output terminals. In a photocoupler, the current going through the output phototransistor is roughly proportional to how much current is used to light up its LED. On the secondary side, we saw that the diode was being driven continuously at more than 100mA. So, on the primary side, we can expect the output transistor to be fully turned on with its two terminals nearly shorted out.

As expected from a photocoupler in the fully 'on' state, its collector to emitter voltage is under 1V. The amount of noise here appears to be too much for my AD629 to cope with, and it ends up outputting a mostly useless waveform. From the way the photo-transistor's voltages pulse together, it looks like there is no source filter capacitor on the primary side of the feedback circuit to tide it over between transformer pulses. A quick look at the circuit traces reveals that the photocoupler does connect directly to the positive terminal of that 22µF capacitor labeled EC3 on the board, so I would expect the collector voltage to hold some sort of value above the 60Hz wave it is riding on, not sharp pulses coincident with the auxiliary supply switching cycle. This looks very similar to the auxiliary supply output waveforms on the SL300 before I changed the auxiliary capacitor. While the 22µF capacitor should have seen negligible electrical load, it could still have gone bust from its close proximity with that scorching-hot zener diode.

Another oddity with the photocoupler's collector-emitter voltage is that once it turns on, the voltage across it on the primary side becomes mostly negative, which is really odd. Time to replace that 22µF capacitor and see what happens.

Progress

I replaced the 22µF with a 33µF FC and, guess what?

The negative peaking across the photocoupler's output (VCE) is mostly gone and the 5VSB output came down to a noisy but sensible value. I doubt the feedback signal is supposed to be this noisy and the main outputs are still dead: no response of any sort to tying PS_ON# to ground. Related or coincidental issues?

Where the heck do those negative spikes across the photocoupler come from and why are they still there? Looks like my initial guess about EC3 simply being a feedback filter capacitor may have been wrong. Time for some more reverse-engineering to see exactly how it is wired up.

Interesting. Instead of powering the primary-side feedback circuitry from the same auxiliary output as the main controller chip, the photocoupler has its own dedicated auxiliary output, and EC3 turned out to be the filter capacitor of that dedicated supply. In place of the usual rectifier diode, the circuit leverages the zener's forward conduction to clamp reverse polarity and avoid the extra part, albeit at the expense of a few extra resistors. I could not find a silkscreen designation for the 2.7kΩ resistor, but if you want to find it on the board pictures, it is hidden inside the piece of shrink tubing between a transformer pin and one of R71's pads on the bottom side.

Why does EC3/AUX2 have such an odd circuit configuration? My guess is that the separate high-impedance circuit for the feedback supply capacitor was intended to protect the capacitor from high ripple current so it can prevent potentially catastrophic 5VSB output voltage excursions long after the other outputs' capacitors went bad, but got ruined by ZD4 roasting EC3. Had EC3 been located away from potentially extremely hot components like ZD4, it should have succeeded at preventing the 5VSB output from rising to abnormally high voltages in the presence of good capacitors on the 5VSB output. The engineers at ChannelWell had the right idea but failed on the actual implementation by overlooking a simple board layout detail.

This explains why the photocoupler's output did not seem to make any sense with the busted capacitor in place. Without a decent amount of charge to maintain positive bias between -170V and AUX2, voltage on AUX2 simply ends up following whatever the transformer puts out, and this includes reverse polarity during the forward half of the switching cycle when the primary winding is being driven.

I still do not like the noise across VCE. Based on voltage across the photocoupler's diode, I would expect current through the phototransistor and its VCE to be relatively steady. Also, the 5VSB output got noisier with the replacement capacitor.

Interlude

What do the removed capacitors measure up to? I used my multimeter for capacitance and a simple debounced switch combined with my oscilloscope to estimate equivalent series resistance based on voltage at the 500 nanosecond mark from the step input. The result I get this way is a blended figure of ESR from DC to about 10MHz instead of ESR at a specific frequency.

If you were wondering what sort of rig I am using to measure ESR, the circuit I used in my SL300 repair was only an SPDT microswitch paired with a simple Set-Reset latch (CD4013) for debouncing running off 12V. After putting some more thought into it and not being satisfied with how lazy the CD4013's rising edge was, nor the way it drooped for several microseconds afterwards from driving about 6mA into the capacitor under test and resistors, I decided to improve it by adding a 74HC08 as a buffer and paralleling the outputs for lowered output impedance. Why use a 74HC08? Mainly because I happened to have some on-hand, but also because HC-series CMOS gates are able to drive their outputs within millivolts from either rail under light load, even at 5V.

How much improvement do I get out of buffering the 4013's output through the gates? With a 5V supply voltage, 1kΩ resistor and a 1200µF FM capacitor providing output loading, the CD4013 takes 240ns to rise to only 3V (60%) and becomes current-limited beyond that point (the chip has weak output drivers at 5V). With the AND gates acting as buffers, the test signal rises from 10% to 90% in 19ns and 0% to 100% in less than 40ns. Not bad for a $0.40 part and five-minute tweak.

In principle, voltage across an ideal 10µF capacitor with a 1kΩ charging resistor using a 5V source should be 250µV after 500ns, which can be considered negligible for my ESR evaluation purposes, since it is close to the noise floor of my test setup. With 5V across the 1kΩ resistor, the test current can be considered constant at 5mA for any remotely reasonable amount of ESR. Therefore, the ESR can be estimated as the voltage difference between initial turn-on and 500ns later divided by 5mA, which translates into 0.2Ω of ESR per millivolt. Using this approximation, we are talking about an approximation error in the order of 1% on 10Ω ESR, less for lower, more reasonable values of ESR. Since I am working with a noise floor of about 1mV, my ESR measurement resolution is limited to 0.2Ω while using this method. It's useless for measuring good caps but good enough to spot low-quality or significantly degraded ones. I could achieve better resolution by using a lower-value resistor, but then I would also need much stronger drivers than a 74HC08.

I am not going to post a full set of oscilloscope screen grabs this time around. However, here is a comparison of a fresh Panasonic 16V 1200µF FM as the good cap reference and the first 1000µF Teapo capacitor on the 5VSB output as the bad one. The Panasonic has a specified ESR of 0.018Ω and, as expected, my measurement attempt shows effectively no change in the noise level before and after the step since the ESR is an order of magnitude smaller than my measurement capability. Noise at turn-on is just parasitic inductance causing momentary oscillation due to extremely low ESR, while the 0.5mV bump afterward is caused by wiring resistance in my test rig and is the same with my test leads shorted out. The Teapo, on the other hand, lost nearly 99% of its nominal capacitance and instead of its already passable 0.29Ω specified ESR for “general” applications, the 440mV rise implies an ESR in the neighborhood of 88Ω and the charge curve implies capacitance much worse than the 11µF reported by my meter. In any case, anything beyond 10Ω worth of ESR is already so far gone that worrying about accuracy is like beating a dead and buried horse.

So, how did they fare?

CapacitorMeasured ValueEstimated ESR
Teapo SEK 50V 22µF
(photocoupler output filter)
333nF1128Ω
Teapo SEK 50V 100µF
(auxiliary supply filter capacitor)
43µF35Ω
Teapo SEK 10V 1000µF
(first 5VSB output filter cap)
11µF88Ω
Teapo SEK 10V 1000µF
(second 5VSB output filter cap)
79µF7.2Ω
Fuhjyyu TNR 10V 2200µF
(first 5V output filter cap)
31µF2.4Ω
Fuhjyyu TNR 10V 2200µF
(second 5V output filter cap)
59µF1.8Ω
Teapo SEK 16V 1000µF
(12V output filter cap)
990µFLess than 0.2Ω
Teapo SC 10V 3300µF
(first 3.3V output filter cap)
3250µFLess than 0.2Ω
Fuhjyyu TMR 10V 4700µF
(second 3.3V output filter cap)
3040µF1.4Ω
Panasonic FM 16V 1200µF
(just because I have a bag of them)
1199µFLess than 0.1Ω

As predicted, all capacitors connected to the standby/auxiliary transformer are well and thoroughly thrashed, especially the 22µF cap that used to be next to ZD4. I am surprised to see there are still two seemingly serviceable capacitors on the main output rails, and even more surprised that one of them happens to be the general-purpose Teapo SEK on the 12V rail.

Why does the FM get a result of “Less than 0.1Ω?”  Because I fiddled with my test circuit some more after I finished the repair to find out how much I could lower the test resistor before seeing significant droop on the HC's output. By halving my test resistor, I double the test current and double the ESR sensitivity to a much more useful 0.1Ω/mV.

Where To Go From Now

Now that the out-of-bounds 5VSB output is reined in, we still have the power-up failure issue to fix. Time to revisit that EC4 capacitor.

I hooked up my multimeter to measure voltage across EC4's leads again, plugged the power supply in, turned it on and still got 0V. Something must have blown somewhere between the auxiliary transformer winding, the capacitor, its rectifier diode and whatever else is left. Time for some more reverse-engineering.

Remember when I said the AUX2 output seemed oddly convoluted? Now this is a classic flyback output: an output winding with one of its ends connected directly to the output side's reference point, the other end connected to a rectifier diode, an output filter capacitor and a zener diode across the capacitor to limit the maximum voltage across it. ZD4 is that scorching-hot 18V zener mentioned earlier and R29 must be a current-limiting resistor for ZD4 and, by extension, EC4.

Why is R29's value unknown?

Because its markings are partially burnt off and what remains of them does not seem to make sense based on the previous schematic. Read one way, it it seems to be 5?6, and read in the opposite direction, it would be 9?5. On my multimeter, the resistor reads open, and a value over 100MΩ would not make any sense. A blown resistor in series with the transformer winding would definitely explain why the main outputs still refuse to turn on after changing most of the caps: no current through that resistor means no charge going into EC4 and no power for the APFC and PWM chips.

What do you think the burnt character is supposed to be? The value needs to be sufficiently low to let enough current through to power the main PWM circuit but not so low as to cause the zener diode to become searing hot. My best guess is that the missing middle character must be an R, since that would match what is left of it and 5.6Ω (5R6) makes reasonable sense for this location in the circuit. The only SMD resistors with plausibly adequate value I have in stock are 10Ω in 0805 package. Let's hope the 1208 package on the PCB was selected by ChannelWell more for its qualities as a potential jumper or excess inventory than for power dissipation. Since the 1208 resistor is glued on the PCB, I will have to solder my replacement 10Ω on top of it.

Soldering a 0805 on top of a 1208 is fairly easy: create a small solder puddle on the pads at both ends of the 1208, position your 0805 part on top of it, hold it there with either a fine-tipped tool or glue, dip your soldering iron tip in the puddle, drag it over the 1208 and 0805 terminals to create a solder bridge, repeat on the other terminal, done. If the bridge keeps breaking up, add some more solder.

Do We Have A Fix?

Time to roll the drums, take a deep breath, plug it in, start the camera, duck (just in case) and throw the switch...


It's ALIVE!

Now that the unit is back from the dead, it is time to assess its remaining issues. As mentioned in the interlude, three of the five output filter caps I could easily pull out for measurement were in poor shape and definitively required replacement, two of them from the 5V rail. Predictably enough, with the 5V rail having the two worst capacitors of the main output lot, the effect on output quality is obvious here with periodic pulsing to 6V.

I do not have 10V 2200µF or better fresh capacitors that fit under the heat sink due to circuit board components, including other capacitors, being packed too tightly around the capacitor footprints to accommodate replacements any wider than the original 10mm diameter. However, I do have a HiPro power supply from an IBM ThinkCenter PC that had a catastrophic primary-side failure and was packed with 10mm 10V 2200µF Teapo SC capacitors for the 5V, 5VSB and 3.3V rails which seemed like perfect replacement candidates – definitively nice upgrades for the original Fuhjyyu TNR and Teapo SEK assuming the Teapo SC are still good. Since the donor power supply presumably had a quick demise rather than a slow capacitor decay one, I am fairly confident its secondary caps will be in reasonably good shape. And sure enough, they all tested well within tolerances for capacitance and were indistinguishable from the Panasonic FM on my ESR check. Reasonably convinced that my salvaged caps would perform substantially better than the dead or dying caps they are replacing, I decided to put the SCs in.

I could not find a direct replacement for the 10mm 4700µF Fuhjyyu TMR, so I just slapped one more 2200µF SC in its place. At a glance, it seems 2200µF is the largest capacitance most manufacturers offer in 10mm diameter low ESR format with 10 to 16 volts rating.

My donor power supply did not have 8mm caps for the 5VSB output either, so I had to leave my 10mm 1200µF FMs flapping in the breeze until I order 8mm parts for that. Conveniently enough, the Rubycon 1000µF 10V ZLH I planned to order for the SL300 will fit nicely here too.


After letting my repaired unit simmer overnight to see if the 10Ω resistor would survive with the main outputs enabled and powering a 20W 12V halogen desk lamp, I went back to have a look at power-up transients. Everything looked fine. The 5VSB output overshot by 200mV while the main outputs only overshot by about 100mV, well within ATX's 10% power-up transient tolerances. The main outputs rise from 0% to 100% in 15ms, which also meets ATX's 20ms maximum requirement.

Once these measurements were done, I disconnected the power supply, opened it again and felt around for anything unexpectedly warm. The first thing I checked was ZD4 and components immediately around it. I also took a look at my new R29. Its markings were still white, a sign that if it did heat up, it did not get hot enough to discolor the ink. It seems ZD4 is clearly not meant to dissipate any significant amount of power when good capacitors are present and provide the bulk of flyback clamping for the standby/auxiliary transformer's outputs.

What are those zeners for? They are primarily intended for clamping leakage inductance. While the bulk of flyback energy in the 5VSB/auxiliary transformer ends up on the 5VSB output, a small amount of it will still get dumped on other windings due to leakage inductance. Since the auxiliary windings may have negligible load attached to them, the stray inductance energy can cause auxiliary capacitors to slowly creep up to arbitrarily high voltages. The zeners simply put a ceiling on how high this is allowed to go.

Revisiting The Photocoupler

Remember those noisy-looking feedback photocoupler output waveforms I did not like earlier? Now that I have something that seems like a definitive fix, I decided to go back and have another look:

And sure enough, these look a lot more like what I was originally expecting. Here, VF is the voltage across the photocoupler's LED on the secondary side, while VCE is the voltage across the photocoupler's collector-emitter terminals on the primary side. As with previous photocoupler waveforms, nothing exciting happens before the 5VSB voltage actually reaches 5V. Until then, VF simply follows the TL431's minimum operating current. The VCE ramp merely follows the 5VSB and AUX2 voltage ramps since it is open-circuit. What changed is that we have no more negative spiking and no more 10-15V peak-to-peak noise on VCE, only a relatively smooth control signal with the error output from the 5VSB's TL431 coupled through the 817 on top. Putting some load on the 5VSB rail causes the TL431 to draw less current through the photodiode to maintain its reference voltage, which in turn reduces the phototransistor's current and allows VCE to increase, reducing the 5VSB transformer's triggering rate and power output.

On the unloaded waveform, you may have noticed how the pulse rate increases as the output voltage ramps up. The reason is that it takes longer for inductors to dump energy into low-impedance loads like discharged capacitors. You've probably heard the classic xenon flash capacitor bank charging noise: a low coil whine that ramps up to ultrasonic pitch over a few seconds. It is caused by the exact same principle.

More Complementary Curiosity

I remembered to take data dumps from the scope before and after the fix so I could at least compare 5VSB standby power. The 5VSB rail has a 51Ω dummy load built-in, which translates into roughly 0.7W with the original voltage waveform and 0.5W at the nominal 5V output voltage. What did the primary waveforms look like before and after the 5VSB fixes?

The two waveforms look similar except for the magnitude of the current peak. By replacing the dead capacitors and fixing the busted resistor, standby power draw dropped from 7.9W to 1.6W, a nice 80% reduction. I wonder how much worse it may have been before R29 blew up. In any case, it is good to know that my P4 should draw less than 2W while hibernating. I know I could simply disconnect it, but I prefer wasting $0.10/year on standby power than $1/year on dollar-store CR2032 batteries.

While I was at it, I decided to have a quick look at 5VSB efficiency with some load attached.

Were it not for the different title, this last set of "after" waveforms with a 12V 20W halogen lamp connected to the 5VSB output could easily be mistaken for the "before" ones. The bulb was drawing 1.03A at 4.97V on the 5VSB rail for 5.12W output power while I measured 8.11W integral power on the AC side, an efficiency of 63%.

How does that compare to modern low-power AC adapters? Under the old class-III efficiency requirements, a 10W supply similar to the 2A 5VSB rail must have an idle power draw under 0.5W and efficiency under load of 49%. The AR300's two-transistor design falls tragically short of meeting the idle power draw requirement, while efficiency under load easily beats it. Not bad for a 12-year-old design. However, this is nowhere near the newer class-V requirements countries started adopting a few years ago, which drops idle power draw to 0.3W and bumps loaded efficiency to 76% for a 10W supply.

Trial By Fire

I decided that three loosely-fitted capacitors were not worth leaving the power supply on the bench, so I decided to proceed with the somewhat tedious task of putting the Aria's original power supply back in. I can always defer installing properly-fitting capacitors until I have some other reason to take the Aria apart again, although I am seriously considering just leaving it that way to spare myself the hassle.

Just about everything seems to get into everything else's way while re-assembling the case. The drive cage catches the FireWire cable and connector on my Audigy 2. The HDD's fragile SATA connector brushes against the power supply housing. I actually broke the PSU's original SATA power plug and had to splice new ones in years ago because of this. The first AMP (“Molex”) connector on each of the three cables is too close to the power supply to do any good. And cables get pinched between the HDD or drive tray and the DIMMs, CPU fan, cards, or between the drive tray and chassis. As if having one side of the heat sink blocked off by the side panel was not bad enough, the power supply cables exit down the trench between the GPU and CPU, blocking off most of the stock Socket 478 heat sink's inward-facing side and obstructing airflow across the northbridge. There is also an annoying bracket over the power supply and across the top of the case that makes installing and removing the power supply far more inconvenient than it needs to be.

The Aria may have been cute, and it still is. But I believe it may also have been one of the worst cases of its day to work with. If it had been even only one centimeter taller to increase clearance between the PSU, drive tray and motherboard components, one centimeter wider to put some distance between the CPU heat sink and the side panel, and one centimeter deeper to increase clearance behind the drive tray, it would have been much nicer to work with.

At any rate, all of the outputs look right on the mark except for 3.3V, which is slightly on the high side at 3.4V, and there is no obvious sign of noise being an issue. If I get my hands on a couple of programmable loads or get around to building my own, I may revisit my inventory of old power supplies and give them a little workout.

With the computer booted, its power supply draws 97.5W from the wall with a nearly ideal power factor of 0.98. Not bad considering that APFC was a new requirement in most countries back then, and most countries' PFC requirement is still only 0.9 today.

Another One Doesn't Bite The Dust

This repair required a little more time and effort, but it turned out successful. It would have been a few pages shorter if I had stuck to chasing the lack of voltage across the EC4 auxiliary output capacitor before investigating the secondary-side circuit on a remote chance the issue may have been there. Still, I would likely have ended up checking everything out anyway.

What have we learned?

Common troubleshooting wisdom exists for a reason. Start troubleshooting from the source to avoid wasting time chasing wild geese when you don't have to. I only ignored it here because going backwards was going to be more interesting. Knowing which components to pay attention to when repairing devices is nice; knowing why these components need to be paid attention to is nicer.

As a forum member pointed out in comments on my SL300 repair, those two-transistor 5VSB designs are potentially dangerous. When coupled with low-endurance capacitors, they are downright evil as shown in both my SL300 and AR300 repairs. With integrated standby PWM controllers like the TOP252GN costing less than $0.70 in 1000-unit reels, while requiring fewer support components and less board space, the cost savings from using a discrete implementation with a TO-220 main switch is questionable. Increasing pressure from governments and various agencies around the world for reduced standby power usage and higher standby supply efficiency should also help guide two-transistor standby designs on their way to extinction.

Just like the SL300, the AR300 demonstrated how efficient flyback supplies are at killing their output capacitors evenly, albeit with some extreme heat assist in the AR300's case, and this is not limited to two-transistor-type flyback contraptions.

The AR300 also reminds us of why printed circuit boards, electronic component packages and device enclosures are either made of non-flammable materials or contain flame retardants: they may get extremely hot, but do not catch fire easily and should not be able to sustain a flame. So, most failures result in little more than an unpleasant smell announcing that some of your equipment went up in smoke.

In retrospect, I am a little surprised this power supply managed to give me at least four solid years while powering my 3GHz P4C and Radeon X700 in such an airflow-challenged enclosure through thousands of hours of moderate to heavy use (for a 300W PSU) and nearly 24/7 power-on time. With that said, I am not pleased with how close this unit may have come to ruining one of my PCs through what I like to call engineered failure--using some of the cheapest parts possible in near-critical locations to barely exceed warranty requirements, effectively guaranteeing high failure rates shortly thereafter.

In any case, the original mostly just-good-enough caps lasted nearly 10 years, and with what little use my P4 has been seeing over the past few years, its reworked PSU should easily outlast the PC's utility.

MORE: How We Test Power Supplies
MORE: Who's Who In Power Supplies, 2014: Brands Vs. Manufacturers

MORE: All Power Supply Articles

MORE: Power Supplies in the Forums

Daniel Sauvageau is a Contributing Writer for Tom's Hardware. Follow him on Twitter and Google+.

Follow Tom's Hardware on Twitter, Facebook and Google+.

This thread is closed for comments
15 comments
    Your comment
  • iam2thecrowe
    That was a good read. I have a lot of custom power supplies (not PC atx) that i would like to fix at work instead of replace, would be a lot cheaper (there's some high wattage power supplies that cost $5000 to replace). Need to convince the boss to buy a good oscilloscope.
  • Rookie_MIB
    I have several of those cases - they make nice little machines for office use but the PSUs tend to flake. I've had to replace several...
  • Daniel Sauvageau
    All the standby power supplies I remember looking at are of the good old flyback design and one particularity of the flyback topology is that it has the worst ripple current on the output capacitors of all converter topologies or in other words: it is the harshest on output capacitors. Until I saw the waveforms while poking around the SL300, one little yet important detail skipped my mind: the harsh current waveform behavior applies to ALL outputs.

    Any flyback supply with sub-par output capacitors on any of its outputs will inevitably fail in relatively short order.

    After that 'oops' moment, I decided to pop the lid on my LG 204WT which I thought had dying CCFL tubes, popped three of its auxiliary supply capacitors off the board, measured them at 5-20 ohms of ESR, replaced them, no more flicker or random turn-on issues.
  • gnyff
    I love it when stuff is repaired! It's so sad that almost every consumer device today is made to fail and be replaced. Even in the cases where the designer did not specifically design for failure most commonly it's definitely not cost-effective for the user to have it repaired. :-(

    Anyways, I've also replaced one of those lousy non-standard PSU. Hard to find and depressingly expensive. Thus, I am avoiding non-standard parts as much as possible now! ;-)
    By the way, in one of my systems the standby power was around 12W (!) even after shutting off 5VSB functions (wake on LAN, keyboard etc) in the BIOS. I did on bother to investigate further (thinking "lousy supply"!) and simply added a proper "full off" power switch. Probably other owners should think of doing the same ;-)
  • Rookie_MIB
    335509 said:
    Anyways, I've also replaced one of those lousy non-standard PSU. Hard to find and depressingly expensive. Thus, I am avoiding non-standard parts as much as possible now! ;-) By the way, in one of my systems the standby power was around 12W (!) even after shutting off 5VSB functions (wake on LAN, keyboard etc) in the BIOS. I did on bother to investigate further (thinking "lousy supply"!) and simply added a proper "full off" power switch. Probably other owners should think of doing the same ;-)


    I wonder if a SFX PSU could be mounted in those cases. Might be a viable solution...
  • nukemaster
    Another great one!

    I actually was interested in that case back when it was released, but just kept rocking the Sonata and Sonata II. While my power supplies did not fail. caps leaked after a while(power supplies still worked)and I took the power supplies out of service forever.

    Rookie_MIB,
    SFX would be an interesting idea to see for sure. My SFX is still going after 5 years of 24/7 operation(not bad for a case included 300 watt unit).
  • Daniel Sauvageau
    1373686 said:
    I wonder if a SFX PSU could be mounted in those cases. Might be a viable solution...

    The dimensions might fit but mounting holes would be on the wrong side (for mounting in a way that might fit internally) and the IEC socket would face either the motherboard or the top cover. Nothing some drill-and-tapping with some wire splicing cannot fix for people who would favor this route instead of a repair of the existing unit where that is possible.

    BTW, the total repair cost for the 5VSB side is about $1.50 for the caps and resistor. The rest is basically free since I scavenged the SCs from a dead (catastrophic primary-side failure) PSU and would have cost around $2.50 otherwise. If Asus had specified higher quality parts from Channel Well though, the PSU may have cost less than $0.50 extra to get right on the first try. Evem doing as little as specifying Teapo SC as the minimum requirement would have easily doubled the trouble-free service life over using SEK..

    It always bugs me when only a few pennies worth in parts and materials can spell the difference between a product that might only last 2-3 years and a product that should easily last 10+ years.
  • nukemaster
    That small cost per unit adds up.

    I can not count the number of times I say, gee a xx cent part would have solved this damn issue and yet no one does it.

    Drilling and making it fit is no issue. :)

    I still get my small system fix with an SG05 case.

    Where do you get your parts from? Digikey is not too expensive, but many times the shipping cost more than the parts(especially with resistors that are cheap to begin with.).
  • Daniel Sauvageau
    35532 said:
    That small cost per unit adds up. Where do you get your parts from? Digikey is not too expensive, but many times the shipping cost more than the parts(especially with resistors that are cheap to begin with.).

    The "small cost" may add up but if you charge $1 to put the $0.50 worth of better quality parts in, you still make a 100% profit on that extra cost and you score points in the quality/reliability column. For the most part, those better quality parts are the same physical size, so no fitting or PCB layout concerns there.

    As for where I get my parts, they are mostly Digikey. I either lump my order with parts I need to repair something for someone else (and pass the shipping cost to them) or build my order for multiple non-urgent projects until it qualifies for free shipping.

    Edit: and for things like 0805 resistors and capacitors, I usually order in cut-tapes of 100 since we're still only talking $1-2 there. If I had more storage space, I would order whole reels for $10-15 (2000-3000 units) and likely never have to worry about ordering more of that resistor or capacitor value (for personal use) for the rest of my life.
  • pjmelect
    I often repair computer power supplies and I enjoyed reading this article. One comment, that may help others and may explain some things in the article such as the power reduction of the fixed power supply.
    Capacitors with high ESR always have a very high leakage current near their rated voltage. The leakage current at low multi-meter voltage levels is not normally measurable, but increasing the voltage across the capacitor dramatically increases the leakage. Circuits with capacitors with high ESR in them normally fail not due to the ESR or the reduction of capacitance, but due to the high leakage current. If you do not have a ESR meter you can simply put a suspect capacitor across a power supply at its rated voltage and check the leakage current. A high leakage current is a good indication of a high ESR value.
    I use a in-circuit ESR meter to test all of the electrolytic capacitors (even the small ones) before I even start trying to fault find a power supply.
  • nukemaster
    1736083 said:
    35532 said:
    That small cost per unit adds up. Where do you get your parts from? Digikey is not too expensive, but many times the shipping cost more than the parts(especially with resistors that are cheap to begin with.).
    The "small cost" may add up but if you charge $1 to put the $0.50 worth of better quality parts in, you still make a 100% profit on that extra cost and you score points in the quality/reliability column. For the most part, those better quality parts are the same physical size, so no fitting or PCB layout concerns there. As for where I get my parts, they are mostly Digikey. I either lump my order with parts I need to repair something for someone else (and pass the shipping cost to them) or build my order for multiple non-urgent projects until it qualifies for free shipping.

    Ohh I agree. I would pay more for better parts, but many will not.
  • Daniel Sauvageau
    71303 said:
    Capacitors with high ESR always have a very high leakage current near their rated voltage.

    If there was truly massive leakage, the leakage would effectively clamp the voltage well below 27V and the voltage would decay to 0V instead of 5V. What you are seeing in my first scope picture is the i(t) * ESR of the flyback transformer's output current added on top of the capacitor's ~5V.

    I did measure the leakage of a handful of those capacitors by applying 12V to them and none of them was above 1mA for the 10V ones despite the 2V over-volt. For the 50V ones, leakage was only in the microamps. While this is two or three orders of magnitude worse than spec in most cases, this is nowhere near as much as the 100mA draw from the 51 ohm dummy load resistor on the 5VSB output and would have negligible effect on the 5VSB circuit's operation and is effectively negligible.

    Leakage was definitely not a significant factor in either the SL300 or AR300. Loss of nominal capacitance and high ESR were - not many circuits will work properly on a 5V supply output with 10-22V of ripple+noise from the flyback transformer's output current and capacitor ESR.

    BTW, in my SL300 repair, a "massive current leak" is effectively what I created by using a TL431 + TIP42 to clamp the transients closer to 5.2V, drawing 3-4A peaks in the process to stabilize the 5VSB output and make the PSU turn on..
  • f-14
    great article
    for the missing capacitor you need try the ATI 4850 512MB video cards:
    I could not find a direct replacement for the 10mm 4700µF Fuhjyyu TMR, so I just slapped one more 2200µF SC in its place. At a glance, it seems 2200µF is the largest capacitance most manufacturers offer in 10mm diameter low ESR format with 10 to 16 volts rating.

    as for that cpu cooler problem in the aria would this cpu cooler fit? Thermaltake / Ruby Orb / Socket 775/754/939/940/AM2 / Aluminum ...
    http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=2844209
    Heatsink Dimensions 140 x 140 x 73.3 mm
    RPM 1700 RPM
    Air Flow 77.85 CFM
    Noise Level 17 dBA
  • Daniel Sauvageau
    361277 said:
    as for that cpu cooler problem in the aria would this cpu cooler fit? Thermaltake / Ruby Orb / Socket 775/754/939/940/AM2 / Aluminum ...

    Northwood = S478 with the cam-lock HSF retention mechanism, none of that push-pin rubbish.
  • Rookie_MIB
    1736083 said:
    The dimensions might fit but mounting holes would be on the wrong side (for mounting in a way that might fit internally) and the IEC socket would face either the motherboard or the top cover. Nothing some drill-and-tapping with some wire splicing cannot fix for people who would favor this route instead of a repair of the existing unit where that is possible.


    Well, I have to build out a computer soon and I might utilize that spare case. I have a SFX PSU at home, maybe I'll take a look and see what kind of work it would take to adapt it. I already know that the mounting is going to have to get a bit creative because of the average length of the SFX main ATX connector. I'm just a little miffed that they went and used a non-standard PSU instead of fitting it for a SFX in the first place. Lets be honest, most people wouldn't use these for anything more than a simple, basic web surfing/office machine because of the mATX limitation on the board and the very shallow depth for any video cards.