The first cut is the deepest!

A while ago I did some temperature tests with my zz9000 graphics card from MNT GmbH in Germany, which turned out to be surprisingly hot. It started out as a test of the CPU temperature of my 68060 rubbing at 100MHz, when I noticed that the graphics card seemed to run hotter than the CPU.

A

A cheap IR temperature tool is super useful

One day later, after installing Linux/m68k from scratch over night, the Amiga 4000 didn’t boot anymore. Of course I thought that I killed the CPU. 100% overclocking can’t be good, right?

I switched Kickstart chips, as Guillaume suggested, and the machine booted again. Or so I thought. But what really fixed it was taking out the daughter board. That’s odd. Because while the card was hot (particularly some capacitors next to the FPGA), it wasn’t electrical death kind of hot. Or so I thought.

But since Lukas Hartmann of MNT fame had mentioned a few boards in the field having heat issues, he helped me debug my board and it turns out that all power rails on the FPGA board were ruined.

Only one way to get the Amiga back to life: exchange the FPGA board on the zz9000. Mouser sells the modules from MYIR.

zz9000 now with eMMC storage 🤯

Meanwhile i also ordered a heat sink for the FPGA which is still on it’s way. But all these measures are of course not really solving the problem of why the card died in the first place. Here I want to send big kudos to Lukas Hartmann for a timely analysis including a simple fix.

Break out the microscope.

The fix exists in two variants. The original fix requires you to drill a hole into the PCB or the carrier board, right next to one of the level shifters because the trace we want to interrupt is on one of the inner layers of the PCB. Alternatively it is also possible, however, to cut the trace right between the socket and the via that’s taking the trace “underground”.

Good bye, 3.3V on the 1.8V line.

This will sever the 3.3V that were bleeding into the 1.8V on the FPGA board. The multimeter confirms we are no longer connected, et voila! The card is successfully reworked.

The microscope was definitely helpful when cutting the trace, particularly because cutting around on such fine hardware makes me very nervous. I used a cheap scalpel from Amazon that came with 10 blades, so I might never run out.

The FPGA die temperature is staying a lot cooler now.

Even after an hour or two of operation the die temperature is not raising above 60 degrees Celsius anymore.

The die temperature measured by the awesome zztop utility (I was hoping this would come to exist) is a few degrees higher than the surface temperature that I can measure. I guess that is expected, but good to keep in mind.

Oh look what the mailman brought.

Funny enough, the mailman brought the Wakefield LTN20069 heatsink right while I was doing my temperature tests.

The Wakefield heatsink comes with sticky thermal adhesive. Very convenient.

And right away, the heatsink cuts another 5 degrees Celsius (9F) out. Pretty awesome.

Now folks, if you’re in the Bay Area and don’t want to do this rework yourself, you can drop the card off with me in Mountain View and I can apply the fix for you. It only takes a few minutes.

Fake flash!

I’ve heard of the whole Chinese relabeling industry, but in the past I have always been lucky with my eBay purchases. Until this last time.

Harmless looking and the correct part on the picture

I kept ordering a number of times from the same folks on eBay, but it seems that they ran out of genuine AMD 27C400-120s and decided to ship relabeled Toshiba -150 parts instead.

My “Made in Germany” fiber glass pen to the rescue

The large window of the flash parts that were actually delivered was immediately making me suspicious though, and luckily my GERMANY made fiber glass scratching pen found a good use as a detective in this case.

Wait, what? Toshiba?

A few light rubs over the inscription reveals the Toshiba TC574200-150 nature of it all. Reached out to the seller for a refund.

… to be continued …

A tale of two diodes…

What happens to long lived electrical engineers? They diode!

I have had trouble with the Amiga 4000’s serial port since I got the machine. At first I got a few bytes through every now and then, but it was pretty much unusable, even at 9600 bps.

When “learning online” about all the magic of open source in GitHub, I stubbled about Keir Fraser’s Amiga Test Kit, John Hertell’s DiagROM and SukkoPera’s Parallel / Serial port tester for the two.

Of course it’s not worth just ordering a single one of anything in the electronics world, and so I ended up joining the market of dongle builders. If you need one, let me know.

This project’s BOM was not as comfortable as some others, it suggested to adjust the resistor value such that the lights are all about equally bright. Conveniently enough I found after the fact that the schematics listed values that roughly matched with my math.

So this was helpful. No -12V light for the serial port. Clearly something is wrong, and this seems to be the evidence of what we already saw on the software side. But where do we go from here? My first suspicion was that something was wrong with the crappy recap that someone had previously done on the machine. But even after I fixed the recap (more pictures soon), the serial port was not working. Working my way up to become a rework tech, I needed all the help I could get, and Paul Resendez (Acill) and John Hertell were extremely helpful. I can only recommend to join Paul’s Discord server. John suggested that I work through the Amiga PCB Explorer. And gave me a pointer or two.

Follow the white rabbit. Or was it the -12V power rail?

When measuring, I could see 0.5V against ground on the line, rather than the expected -12V. How was this possible? Did I measure incorrectly?

All the power comes from … The power connector

Measuring right at the power connector, everything looked great, but even right after the diode D175 things became fishy. Aha, there is that capacitor again. It must be at fault. Or must it? Or could it be, as John suggested, the diode itself that’s broken?

They look so pretty, how can they be broken?

At those point I had no clue that my multimeter has a diode mode for exactly this purpose. So I decided to just give it a shot and exchange the broken diode. And since symmetry is nice and D175 and D176 do similar jobs, D175 got a new colleague at work as well

And bingo! The serial port works again like a charm. (And other things that require -12V? Are there any?)

DiagROM is super helpful when hunting down serial port issues

Keyboard fun!

My keyboard is missing a key. There’s no escape!

Well actually it wasn’t escape that was broken on my nice click-a-dee-clack Amiga 4000 keyboard, but CAPS LOCK! And bit even the key itself, it was the LED that just wouldn’t show the status. No problem, let’s take it apart

So many screws!

At first I thought the LED might have died after my recent diode intermezzo on the Amiga board. But measuring with my multimeter I see that the LED is just not getting any power. Let’s look at the membrane.

It seems I forgot the before picture

There was a piece of the trace missing. It seems it had corroded away. The previous owner probably spilled Coke on it. My first approach was to scratch the green cover of the existing trace so I can connect to it.

Then I covered it with some of my copper foil. I wasn’t expecting it to work, really, but it was worth a try because I didn’t have a conductive pen, and certainly soldering is not an option, unless you want to buy a new membrane (these are the same as the A500 membranes and are available). But my better half is trying to make me frugal, and we the people are creating too much e-waste anyways. So I ordered a nickel conductive pen on Amazon. It arrived today.

I connected the two distinct pieces of the trace to each other with the pen, and after the ink dried, covered it with a piece of copper foil again, for no other reason than that it is shiny. Aaaaand……:

Off and …
Houston, we have continuity

Putting the while thing together again.

Shiny, and good as new!

While testing, I’m actually not using the keyboard on the Amiga, but in my PC with a little adapter that I built with a cheap Arduino Pro Micro clone and tkoecker’s amigakb.

Awesome keyboards for all computers!

I ordered the boards through PCBWay and they’re super easy to solder. I have 9 more, because I didn’t want to pay more for shipping than the PCBs. So if you want one, drop me a note.

Just a small upgrade

Lots of things happening at work, so it’s good to do a little meditation by soldering in the evening. Guillaume sent me a Greaseweazle kit but the STM32 still looks like the next level of soldering to me. It makes 0805 components look large. I’m missing a 74F257 that was lost in the mail, so the 060 adapter has to wait for a while. But there are other ways to treat an A3640 rev 3.1.

First, a socket for the oscillator will make it easier to test a little overclocking.

The 68040@25MHz worked nicely with a 60MHz oscillator, pushing it up to 30MHz.

Since the A3640 provides the clock for the Amiga mainboard, the chipset and memory will get a tad faster, too. This hack is super easy and will cost you 2 bucks for a socket and an oscillator.

The next and a little more involved one is an upgrade of the v3.1 card to v3.2. Gladly this is a simple GAL update, but unfortunately the GALs are not socketed. This leads to a brief intermezzo with my favorite tool ever, the hot air rework station.

The kapton tape job was a bit sloppy and over eager

I had left the temperature at 245 degrees C from a previous experiment. Not good enough to take the part off. Let’s up it to 275 Celsius again (530 F) and the GAL slides off smoothly.

Be careful, this process does not need any force whatsoever. If the pads are slightly corroded or have been messed up before, they might lift off if you’re not careful. That makes for tedious repairs, don’t do it.

Phew, all went well. Let’s clean up

Cleaning up the pads with solder wick and isopropyl alcohol is essential before working on them again. Small unevenness can rip the pads off when cleaning.

I got a cheap adapter off eBay

The Conitec Galep 5 is a beautiful device to program everything from GALs to EPROMs to microcontrollers. I’m glad I still have mine from the beginning of coreboot times, but a TL866 is a more cost sensitive solution.

The best way to put the GAL back is with hot air

I’m using low melting point solder paste and the hot air rework station to put the GAL back on. I’m not happy with the 28 pin sockets I have, so I don’t bother putting one on.

With a little more time I can also apply the speedgeek Wait State mod, but that will require reprogramming another five unsocketed chips.

For now this is the end result:

And it still works and is stable:

Recap of a revised recap

When I got my A4000 late last year, I could not have been a happier kid. I fleas surprised to get it for a fabulous price and it arrived just in time before the holidays.

The A3640 was a good first victim. Notice the through hole caps that ripped off the pads

That joy lasted until I noticed that there was no mouse, a Super Buster r6 (WTH) and a semiprofessionally done replacement of all the capacitors on the board. Unfortunately through hole caps were used in the recap process.

You will find enough discussion about this on the internet, and people will suggest all sorts of craziness, like twisting off old caps or putting through hole caps on SMD pads. These are all terrible ideas. Don’t do it. If you have an Amiga (or any 30yr old computer), do it right or pay somebody to do it right. Recap services are not expensive and are definitely worth it.

One of the tricky parts is that the silk screen on the A3640 is incorrectly showing the direction of three capacitors. Make sure to consult the schematics.

Ruined! It is ruined! Or is it?

When I reached out to the southern California Amiga repair wizard Acill (Paul Rezendes), he suggested that I check out this video on repairing pads. I had to watch the whole thing twice, because it’s mind blowing. But hey, nothing that can’t be done, so here we go

This one is even worse. The pads are gone and the vias look corroded.
Got some copper foil in the mail, 1.4mil

It took me a while to get these right. I bought a cheap scalpel on Amazon
A cheap microscope is worth gold

To glue the pads on I used 2 component epoxy overcoat and let it dry over night.

It’s really tricky to keep this orderly awhile trying to connect the vias to the pads.

I spent a lot of time with my multimeter and the A3640 schematics, measuring the continuity of the pads

But the end result is pretty solid.
And we are online.

This one is certainly not winning a beauty contest. But for a first attempt in fixing a botched recap, I am proud to say that the card is now better than before I touched it.

Time to build a 68040 to 68060 adapter next!?

I’m back.

It has been 10 years since I have moved to the US and therefore had left all my Amiga gear behind. “I can always run UAE” I thought. But some goodbyes are harder than others, and some are not forever. So here I am, proud owner of an Amiga 4000 again.