Making 128mb SIMMs from 64s

Here we are, back with another project that would have been much more useful if I had undertaken it 15 years earlier.

If it's not already abundantly clear, I spend quite a lot of my time working with old computers. I do it for The Hackery, I do it for other collectors, and I do it for fun. But given that most of the stuff I'm working with is a minimum of 10 years obsolete, it can be quite the challenge to source the needed parts. Sometimes that means I have to make my own, as is the case today.


One of the joys of working at the Hackery is getting access to weird and wonderful things which people donate for recycling. Today's project, like many of mine, was inspired by one such donation. Amongst a load of gear from a point of sales company was a large antistatic bag filled with 64 megabyte 72 pin RAM SIMMs. For the uninitiated, this was the standard RAM used by PCs, Macs, Amigas, etc, through most of the '90s. And though sticks as large as 128mb exist, anything beyond 32 is vastly harder to find. So I was quite happy to get enough 64s to keep me stocked into the foreseeable future.


But then I noticed something interesting about the sticks. Though there were only memory chips on one side of the stick, both sides had solder pads. This suggested to me that the stick was wired to support up to 128mb. Given how rare 128mb sticks are, this was an exciting thought. And with so many to play with, it would be no great loss if I broke a few testing this theory. So that's what I did.

To start with I sorted them all out. There were two types, ones with Micron chips and ones with Spec Tek chips. I started with a pair of Micron-based sticks. From there I looked up the datasheets for the RAM chips and the little voltage regulator in the center. The datasheets told me that these were EDO chips, 4K refresh. It also said they ran at 3.3v, which explained the onboard regulator. Most computers which used 72 pin simms ran at 5v. Now, running 3.3v chips in a 5v system isn't ideal, even with a voltage regulator, because the IO voltages will still be a little bit off-spec. This is probably why the Micron-based sticks are actually wired to get ~4.1v from the regulator instead of 3.3v. Again, a bit hokey, but they worked in all the machines I tested them in, so I went ahead.

I didn't take any pictures of the first attempt. I hope you'll forgive me. But as you'll see in a moment, you didn't miss much.

With my hot air pencil, I de-soldered all the memory chips, the regulator, the capacitors and resistor from one Micron stick. Then I soldered them onto the backside of another Micron stick. After washing the flux off and inspecting all the solder joints beneath my microscope, I threw it into a Mac Quadra 605. This was a machine I knew to support 128mb simms. Unfortunately, while it had worked fine with the 64s, it gave the chimes of death when I turned it on with my home made 128mb stick installed. Clearly something was wrong.

First thing I did was pull the stick out and go over it extra carefully beneath the microscope. I could see nothing wrong. No unconnected pins, no shorts, no nothing. So I re-cleaned the contacts and put it back into the machine. Again, the chimes of death. But this time, when I went to pull it out, I discovered that the voltage regulators were very hot. I knew from the earlier tests that the 64s didn't draw enough to heat up the regulator so quickly, and though I'd doubled the number of chips, I'd also doubled the number of regulators. It shouldn't have been a problem. It seemed there was a short somewhere.

So for the next half an hour or so I studied the now blank PCB I took the chips off of, to see if perhaps there was a manufacturing defect that caused the regulator on the backside to be wired incorrectly. It was during this examination that I discovered the odd wiring of the regulator. Rather than having its adjust pin tied to ground to give 3.3v, it was instead tied to its output pin, so that it produced something more like 4.1v. This seemed strange to me, until I considered the misaligned IO voltages that would result from putting 3.3v chips in a 5v system. And indeed, the absolute maximum rating for VCC in the Micron datasheet was +4.6v. So I figured that wasn't the cause of my problem.

Instead I assumed that one of the smaller components, a filter cap or similar, had shorted. It seemed plausible given the amount of heat that the parts were exposed to during desoldering and resoldering. And being the lazy bastard that I am, I preferred to not desolder everything and check them one at a time. So I used my bench power supply to apply 3.3v power after the regulator, then let it run for a moment to allow whatever was shorting to heat up. After that it would be a simple matter of feeling around to find the hot spot.

To my surprise, there was no single spot which got hot. Instead, all 8 chips which I had soldered on were warming up, while the ones which came from the factory were not. Had I really killed all 8?

Well, yes, as it turned out, but not by overheating them during soldering. Instead it turned out there was a simpler explanation. Unlike most RAM sticks I can think of, these 64s were wired up so that the chips on the backside needed to be mounted upside down compared to the chips on the front. And having not considered that possibility, I went ahead and soldered them on right side up. As a result, the power polarity was reversed, causing the chips to bake themselves. Oops.


Here I've removed one of the chips I had soldered on to test it out of circuit. There was no doubt about it, I had installed it upside down and killed it. So I decided to call it a day and went to sleep.

The next day...

The following morning, having learned from my mistake, I decided to have another go at building a 128mb stick. And this time I decided to take more pictures during the process as well.


I started with Spec Tek sticks this time. Again, the first step was to strip all the components off with the hot air pencil.


Then I soldered the smaller components to the backside of another stick using a traditional soldering iron.

I had tried a few different methods of attaching the RAM chips during my previous attempt. It would have been very easy to put them on if I had access to soldering paste, stencils, and a reflow oven. But sadly this was not the case. And though there was just enough space between the chips for me to get the tip of my soldering iron, I couldn't quite touch the PCB pads with it, only the pins. That sounded like a recipe for bad joints to me.


So what I ended up doing was a sort of half-assed version of the solder paste method. Instead of using a paste, I used traditional solder to heavily tin each pad on the PCB. Then I brushed on a thick layer of rosin flux and set the chips on top. Careful nudges were needed to position the chips on top of the bulging solder blobs, instead of allowing them to slip off and sit between them.


But despite the jury rigging, a short pass with the hot air caused the chips to settle neatly onto their pads and adhere. A quick check beneath the microscope revealed joints which were perhaps a little crooked, but electrically sound.

So with the method proven, I proceeded to tin, flux, populate and solder the rest of the stick. With the exception of one chip that went sideways half way through, it was all smooth sailing. Then I used a can of flux remover to blast most of the sticky rosin off, revealing the nice clean stick beneath.


And here it is, on the left. I think I voided the warranty... Can you tell which chip shifted sideways?


Despite the crooked chip, all the joints themselves looked good. And that meant it was time for the test. I grabbed the Quadra again and socketed the chip in. I flipped the switch and watched carefully for signs of smoke. The mac bonged, and then was silent. This was good, no chimes of death. But nothing else happened either. The monitor remained black, there was no hard drive activity. I started to feel around the system to see if anything was getting hot. This time the regulators and RAM chips felt fine. I was just about to assume something had gone wrong when suddenly the monitor lit up. A moment later, the happy mac icon appeared. Yes! It was booting!

Turns out that the RAM test which occurred on every cold boot was simply taking far longer than usual due to having so much RAM to check. And being a mac instead of a PC, there was no way to tell how much RAM was detected until the system had reached the OS. I had to wait impatiently all the way through the test and boot process, wondering if the RAM was actually being addressed correctly. I half expected the system to crash at any moment.


Instead, it worked! All 128mb, plus the 4mb on the motherboard were detected. Success!

So there you have it. If you need some big 72 pin simms for an old computer, all you need to do is find a hot air pencil, a high end soldering station, a Canadian POS company that's retiring old gear, and several free hours over 2 days to solder and troubleshoot with. Then you too can make something that's often available on ebay for $30 in your own home.

Page created October 20th 2013

Memories, memories, sweet memories