Tech Flashback: The SIMMs

As I managed to scrounge up some time, I decided I would document some of the bits and pieces I’ve managed to collect throughout the years. This post will be dedicated to the SIMMs in my collection – and no, I don’t mean the game!

Single In-Line Memory Modules (or SIMMs for short) were a type of memory used in late 286’s, most 386’s, practically all 486’s and even some Pentium/Pentium II class machines (often as a backwards-compatibility measure). They followed the Single In-Line Pin Package (SIPP) memory, which was very similar with the exception of the use of fragile connection pins rather than pads, which would have been very prone to damage.

The SIMMs were succeeded by Dual In-Line Memory Modules (or DIMMs for short), which persists to this day. Which begs the question – what is the difference?

As it turns out, the SIMMs are named “single” for the reason that their contact pins are in a single row. Despite the contact pads being available on the front and the back, both pads are redundant and interconnected. Thus the two variants of SIMMs (30-pin, and 72-pin) have twice as many pads as pins. SIMMs were mounted in most boards in slots that required inserting the module at an angle, and then raising it up to vertical where it “clips in”. Depending on the processor and chipset, 30-pin SIMMs often had to be installed in pairs or (more safely) in a set of four, whereas 72-pin SIMMs could be installed singly (or later, for Pentium class machines, in pairs). SIMMs themselves were built on two different types of memory, the earlier and slower variant being known as Fast Page Memory (FPM), and the later, faster variant known as Extended Data-Out (EDO). EDO memory offered noticeable performance benefits, mainly by extending the time the data output would be valid for, thus allowing for another memory operation to be queued while the data output of the previous operation was being read out.

DIMMs on the other hand, use the contact pads on each side separately, thus allowing many more edge connections to be made in a given space. Aside from that, they also have a data bus twice as wide as the 72-pin SIMMs, and featured a host of new capabilities including serial presence detect (to identify a module’s manufacturer, capacity and timing). It also featured a larger board, smaller contacts, and the “straight down” mounting system. Most often, these could be installed singly, as they provided a wide-enough bus to satisfy the CPU. The DIMMs started off with the SDRAM series, at 66Mhz, 100Mhz and then 133Mhz and progressed through the DDR, DDR2, and DDR3 era.

Nowadays, SIMMs are a rare sight, and many present-day computer builders may have never seen one. Vintage computer owners and enthusiasts can end up paying a fairly big amount for some large capacity SIMMs to keep their computers and synthesizer cards running. In my youth, I had both salvaged SIMMs from ex-corporate machines, but also been guilty of disposing them as worthless. But lately, I have been on a bit of a “vintage” muse, so I’ve vowed to collect any that happen to “fall” my way.

If there’s anything that can make you appreciate the progress of computing technology, it is to look back. Hopefully you will enjoy my collection of SIMMs – this notably excludes any SIMMs within my still-working vintage workhorses for obvious reasons. If you ever need to work out the capacity of a SIMM, this particular page has a good list of part numbers and capacities – very handy indeed.

30-pin SIMMs

The first computer I used as a youth had 30-pin SIMMs. They were 1Mb chips, made by Micron Technology, but unfortunately I had never saved any of them. Instead, here are the ones that are still left in my collection.

DSC_9020 DSC_9021

This one is a Samsung branded module from 1993, made by Samsung with Samsung Korea chips. The main memory chips are numbered KM44C1000BJ-7 (which are 1Mbit x 4), with the parity chip labelled KM41C1000CJ-7 (which is 1Mbit x 1).

As a result, it is a 1 Megabyte stick of RAM – i.e. 1Mbit x 4 x 2 = 1Mb. It also has an extra bit for parity – i.e. a 1Mbit x 9 configuration. Each of the chips has a ceramic surface mount bypass capacitor mounted close to it – many of these chips easily ate 120mA in operation! Most boards would have had eight 30-pin slots, and these modules would need to be installed in pairs, or more commonly, a set of four.

DSC_8194 DSC_8195

This is a Mitsubishi branded module from 1993 using Mitsubishi chips. It appears the module was made in USA. The chips are marked M5M44400AJ-7 (1Mbit x 4) and M5M41000BJ-7. This is another 1Mb stick of ram.

DSC_8192 DSC_8193

This seems to be a much later (1994) “cheap” SIMM. You can tell by the reduced height of the PCB to the bare minimum height, and no real branding. The memory is made of two Motorola MCM44400GN70 with a Texas Instruments TMS4C102DJ-70, making this another 1Mb stick of RAM.

These sized modules were quite common, and I would come across them a lot. It wasn’t particularly exciting, as on an 8 slot computer, that would mean 8Mb tops. Instead, we would salivate over anything bigger. Apparently they even came in capacities as small as 256kB, so I guess having 1Mb is lucky in comparison.

It’s interesting to note that not all systems required parity – it’s presence generally didn’t hurt anything.  The suffix on the memory -5, -6, -7 or -50, -60, -70 were a way of denoting the speed of the chips in terms of the RAS access time, meaning 50ns, 60ns, and 70ns respectively. This was a convention followed by most memory chips, and faster chips generally posed no issues, however using slower chips than required could cause issues.

DSC_9022 DSC_9023

Speaking of larger capacity – this claims to be a 4Mb module. It seems to be sold or distributed through PCDirect, and has an Integrated Memory Systems label on the front (likely, the vendor). The front of the PCB is coded with GT3009. The rear of the PCB has PTC marked on it, and a whole host of horizontal bus traces – it’s pretty unique.

The module itself is made of 9 Micron Technology (think Crucial) chips, each marked with 4C1004JDJ-6 which are 4Mbit x 1 chips. This makes it a 4Mbit x 9 configuration – i.e. a 4 megabyte module with parity.

Larger 30-pin SIMMs were probably available, but rarely seen or used. The number of address lines in 30-pin SIMMs meant that there is a limit to 16MiB addressable. In the case of a regular PC that requires four sticks, the 4Mb modules would reach that. However, in printers and other devices with fewer slots, larger SIMMs may have been used.

It was really a bit of a compatibility issue, as it highly depended on the host system whether it could address all the available RAM installed, and BIOS upgrades were a chip-swap in that era (being EPROMs).

72-pin SIMMs

Late in the 486 era, and into the initial Pentium era, 72-pin SIMMs were king for simplicity. No longer do you have to take care matching sets of SIMMs as much as before. Unfortunately, most boards only featured four 72-pin slots, but the modules were often larger in capacity (and double-sided too) which made up for it. If memory serves me right, even 16Mb 72-pin SIMMs were easily found, and 32Mb were rarer but still available (I never owned any though).

DSC_9004 DSC_9005

If anything, this one 72-pin SIMM is the most hilarious and interesting example in my collection. I salvaged this one from an ex-corporate system when I was back in Primary school, thinking I had hit the jackpot. A module of this size with this many chips would have been quite expensive, or so I thought. The back was even filled with traces. The PCB silkscreen coating suggests it may date back from 1991, but it’s uncertain.

Instead, it has eight Samsung KM44C256BJ-8s and four Samsung KM41C256J-8’s (all 80ns). The datasheets aren’t available anymore, but some quick decyphering of the code suggests it is eight 256kbit x 4 chips, with four 256kbit x 1 chips. This makes 1Mbit x 9 or in other words a one megabyte with parity. It’s hilarious to think this big thing is the same capacity as the small 30-pin SIMM above, but it may have been one of the ways to build a low-capacity 72-pin SIMM while keeping the memory bus width requirement in check.

It’s like welding together four 256kB 30-pin sticks (three chips on each – two 256kBit x 4’s with a 256kbit x 1 parity). In fact, there were SIMM adapters back then which would allow you to adapt 30-pin SIMMs together to make a 72-pin SIMM but I have never used one myself.

That being said, this was the smallest 72-pin SIMM (memory capacity wise) I had met.

DSC_9006 DSC_9007

72-pin SIMMs also came in some oddball configurations, and were able to go dual-sided as well. In many boards, there was an un-even number of chips on the front and back, which seems to be a case of 8-chips (memory) on the front, with four-chips (parity) on the back. In this case, it seems to be made of eight Vanguard VG264400BJ chips (70ns). These are claimed to be 256kbit x 4 x 4 configuration in FPM, which means 512kBytes per chip * 8 = a 4Mb module it seems with no parity (many systems didn’t need it anyway). I don’t know what that 8Mb scrawling was on the back … It matches up when you see the L431B4MB PCB marking, implying it’s a 4Mb module.

A very common capacity at the time, I know I had a few 4Mb modules in my possession in the past.

DSC_9008 DSC_9009

This is a later module, a Hitachi built (note the PCB logo) module with all Hitachi chips. It’s model number HB56D13 6BV-7A and is marked 3rd week, 1994, sold 19th August 1994.

The front of the module is adorned with eight HM514400AS7 chips (1Mbit x 4), dated week 47 of 1993. The rear is adorned with four HM511000AJP6 (1Mbit x 1) chips, dated Week 44 of 1992. All of them were made in Korea. As a result, the capacity is 4 Megabytes with parity.

This one also seems to have its presence detect jumpers (surface mounted, in blue) fitted for the computer to use when detecting the presence and capacity of a given module (pre-dating the SPD EEPROM feature of DIMMs).

DSC_9010 DSC_9011

As time went on, cheaper modules arrived on the market with very few components – notice how there’s virtually no capacitors on this one, with a minimized PCB footprint. This one was dated 15th May 1998 which is a little late in the SIMM game. This one is a double-sided PCB, with space for eight chips on each side (implying no-parity, another money saving measure).

This particular module also seems to have pads for the presence detect feature, but no jumpers (or 0-ohm resistors) were installed, thus it it is not active. It is made of eight Vanguard VG264400BJ (60ns FPM) chips, which is 256kbit x 4 x 4. This makes it 512kB per chip, for a total of 4 Megabytes per module without parity. Using the same PCB and populating both sides, it seems probable that this could have been 8Mb without parity as well.

DSC_9012 DSC_9013

Different manufacturers were also in the SIMM memory package market – this one is based on Fairchild chips. The front is adorned with eight 814400A-70 (70nS 1Mbit x 4 FPM) chips dated week 35 of 1993, made in Malaysia. The rear is adorned with four 81C1000A-70 (70nS 1Mbit x 1 FPM) dated Week 29 of 1993, made in Singapore. This gives the module a total capacity of 4 Megabytes with parity.

Note that the presence detect jumper solder pads are visible but not populated.

DSC_9014 DSC_9015

Other manufacturers of memory existed too – this one was a Seiwa Japan made module, 23216 Rev A, made with 16 Siemens HYB514400BJ-50 (50nS 1Mbit x 4 FPM), made in Germany. Half are date coded Week 34 of 1996, the other half Week 28 of 1996. This board even has the presence detect jumpers fitted, and bypass capacitors for each chip. This makes the module an 8 Megabyte module without parity.

DSC_9016 DSC_9017

This module is a paretty old module, it’s actually quite rare to find a yellow coloured module (in my memory, at least). This was a Kingston KTM-8000/M70 module marked with KTA-CENT/8 which implies it was a compatible 8Mb memory module for an Apple Centris server. Its PCB is etched with a copyright date of 1991.

It is assembled from 16 Hitachi Japan HM514400AS7 (70nS 1Mbit x 4 FPM) chips dated Week 38 of 1992, all arranged horizontally rather than vertically. There are also unpopulated positions for (likely) parity chips, which take on a form factor I haven’t ever seen!

The presence detect jumpers seem to be fitted, with one bypass cap per two memory chips.

DSC_9018 DSC_9019

This one is a module dated 29th November 1995, sold November 1996, made from 16 Hynix Korea HY514400A J-70S (70nS 1Mbit x 4 FPM) memory chips dated Week 43 and 44 of 1995. It does not have its presence detect jumpers populated, nor does it have bypass capacitors for each chip, as it seems to be a cost saving measure. The module itself is hence a 8 Megabyte module without parity.

Many of the SIMM memory makers have been featured in this collection, but I’m sure I have seen OKI, Goldstar and Mistubishi as well, but don’t have them in my possession anymore. It’s been a long time since I’ve last seen them, but there may still be one or two lingering in a dark corner.

Of note was that SIMMs came in gold fingered and “tin” finger configurations, with advice to avoid mixing gold socket pins with tin fingered modules and vice versa due to oxidation concerns. In my experience, I didn’t experience much issue either way, but I didn’t see the intelligence in advice such as “use an eraser to clean the contacts of the SIMMs” – that might develop enough static charge to kill them!

Conclusion

There’s a tour of the collection of spare SIMMs that I presently have, and it covers a good range of chip manufacturers, but not so much in terms of different SIMM capacities. It’s crazy to think that in the early to mid 1990s, the modules numbered in the single digit number of megabytes. In just twenty years, we have modules numbering in the quadruple digits of megabytes. It definitely shows you the level of progress, as well as how “simple” these memory modules were – presence detect jumpers instead of EEPROMs, contacts with wide spacing, redundant on both sides. Even the IC packages themselves are very thick compared to the wafer thin ones of today (not pictured here, but maybe later, if I ever bother with an “SDRAM DIMM” collection). What a world we live in!

About lui_gough

I'm a bit of a nut for electronics, computing, photography, radio, satellite and other technical hobbies. Click for more about me!
This entry was posted in Computing, Salvage, Tech Flashback and tagged , , , , . Bookmark the permalink.

11 Responses to Tech Flashback: The SIMMs

  1. sparcie says:

    I believe many machines that took SIMMS had BIOS options to help use slower RAM as they could not detect speed like modern RAM does. My old 386sx has this in the form of a wait state and RAS-CAS timing configurations. It had its first 1Mb onboard, and it seems to be faster zero wait state memory. This means the BIOS and DOS works even if the expanded RAM is not configured properly. I’ve seen these settings in some other machines, I’m not sure how common they were.

    I didn’t realise that the 72 pin modules were also SIMMS, although thinking about the socket that does make sense. I also thought that they had to be installed in pairs, at least in later systems like pentium II/AMD k6 systems. I have seen some 72 pin memory as large as 16Mb per module, but I think thats the largest as SDRAM took over rather quickly.

    Cleaning with an eraser is meant to be done with a specific type, soft white eraser. I don’t really know if that can generate static. It would be interesting to do some measurements, as I wonder if it is as bad as we think. I have done it once or twice in cases of bad corrosion, but in those cases my prefered cleaning methods were not enough. It did work quite well, the pins were even shiny like new, but I’d suggest it is a last resort when isopropyl alcohol and contact cleaner don’t work.

    Sparcie

    • lui_gough says:

      I remember seeing 0WS and 1WS options in the BIOS before, although some of that applied to ISA I/O as well it seemed (especially for some Graphic Cards), and impacted performance somewhat strongly.

      Later BIOSes (jumper-free) for my Socket 7 board allowed you to set the RAS timings altogether for the 72-pin SIMMs, but it seems the 70ms modules were the most common from the ones I’ve handled. I have heard of ones as bad as 100ms-120ms RAS timings, not that it really mattered if you shoved those into early 286’s, standalone printers as a job-buffer or into a sound card for sound-font caches.

      They needed to be in pairs for the PII and Socket 7 boards as they were already really into SDRAM territory by then (bus width of SDRAM is twice that of a 72-pin SIMM, which is why you needed pairs). The 486’s happily took them singly, as did most Pentium I era boards.

      – Gough 🙂

  2. Maikel. says:

    Growing up with thesame technology, I remember fondly this era when I found the memory manufactor I was working for at the time still had a spare stack of 30-pin SIMMS for RMA.

    The first machine I bought for myself was a 286AT in the lucky later part of 1990, I was fond enough to have DIP and SIMM memorybanks on my board(foxconn), skipping SIPP altogether(lots of branded 386 machines from early 1990 had SIPP-only). the system came with 2x1MB, I later replaced the mainboard with a board(foxconn) with a 386SX soldered on, and purchased an additional 2x1MB for 200Hfl(about 100 USD). I kept thesame memory banks when upgrading to a 486SX(biostar, soldered on) and to a 486DX(socket3, foxconn), where I could mix the 30-pins sims with a 72-pins fastpage IF I switched the parity off, this was no problem at all! I retired the system and the memory banks to a brother who used it untill 1999, when he moved it to his attic in favor of a newer system that would handle the latest games.

    72-pins had several issues, the tin/copper versions matching/not matching your motherboard, Fastpage and EDO, 32MB EDO I even seen in a 72-pins module, it was something we actually had on stock in Apacer for RMA in 2004-2007, but I had never seen them when I worked in a computer store from 1996 to 2002, since they were so expensive to keep in stock and price on memory was unstable. Most systems from the late 90’s had 16/32 MB memory in them.

    DIMMs were a godsend, because of their larger size and the JEDEC standard.

    Mind you, there were also giant ECC SDRAM-DIMMs for HP and IBM servers that were as high as they were wide, and fully stacked with chips, some even “piggybacked” on top of one another…

  3. Zim says:

    Hi, you made a couple mistakes when reading the brand on two modules.
    The one you said Fairchild (Cursive F) is actually Fujitsu (F between two horizontal bars) and the later Hynix is Hyundai. Hynix did not exist at the time.
    Now I’m in doubt about something. I had the 30-pin simm packed with 9 chips like the one on the picture that you called a 4MiB module with parity, fourth one from the top. Funny thing is that I had 4 of them and no computer ever detected it as a 4MB module. The whole bunch on a 386DX summed just 4MiB instead of 16MB. I did not read the chips as you did and now I’m kind of regretting selling them on a computer lot really cheap. 4MB 30 pin simms are a rarity, I never saw or used one in my life.

    • lui_gough says:

      Thanks for the comments. Funnily enough, I did have Fujitsu as the first entry, but then I changed it to Fairchild because I did notice Fairchild made SIMM packages as well. My mistake!

      I did manage to get the 4x4Mb detected as 16Mb on my 486 board with no trouble, it was a very worthwhile swap for me. If you had really old modules, however, it may have been a 1Mb chip arranged as 1Mbit x 1 with 9 packages for parity, which according to the PC Guide, seems to be a configuration that was in use at the time http://www.pcguide.com/ref/ram/packChip-c.html

      I suppose you can decypher the numbers next time, but I hope that gives you some hope :).

      – Gough

      • yuhong says:

        There were issues using the newer 4Mbit based 1MB SIMMs in some older systems, generally either because they have to be refreshed more often or due to the addition of a “test mode” triggered by a CAS before RAS refresh with WE low that some older systems are not aware of.

    • yuhong says:

      What I think happened was Hyundai and LG eventually merged into Hynix.

  4. yuhong says:

    Yea, the history of parity and ECC. One of the reason ECC replaced parity is to allow normal memory chips to be used instead of the four x1 chips shown in the pictures or “Quad CAS” chips. Parity required access to each individual bit per 8-bit byte and the special “Quad CAS” chips had four CAS pins for that purpose, which is why they are bigger than the normal chips. ECC works by generating a 8-bit ECC code per 64-bit word instead and thus no longer require access to individual bits.

    • yuhong says:

      Thinking about it, this is probably another reason why non-parity RAM became common. If PCs still used parity, you would need to produce one 1Mx1 chip per two 1Mx4 chips, and the demand for DRAM was already pretty high back in the 1993-1995 period.

  5. yuhong says:

    “The number of address lines in 30-pin SIMMs meant that there is a limit to 16MiB addressable.”
    That is not true. The way it works is that the memory controller multiplexes the host address into the address pins on the SIMM. There is a limit of 16MiB per SIMM because 30-pin SIMMs only have 12 address pins.

Error: Comment is Missing!