First post, by Tevian
- Rank
- Member
12/8/24
V0.5 Prototype testing.
For whoever is interested, I need two or three volunteers to test this V0.5 backplane. If you want to DM me here on Vogons, first send me a message of your interest in participating and a photo of one or more compatible PISA/PCISA, Allen Bradley, or ISA "half-sized" SBC along with the model number so I can verify what you'll be testing with. Within the next week, I'll have a few of the V0.5 boards ready to test with. I will provide these prototypes for free, but I'd ask that the volunteers provide the shipping costs. I can't promise this will move quickly as I do this as a hobby in my free time, but I will continue to make forward progress.
The goal is to have honest feedback about functionality and possible changes for a final version and what should be included in the documentation for the end user. Thx ahead of time for anyone that's interested in helping out.
Project Backwards
I've started this thread as a record of progress and soliciting advice or suggestions on a small universal backplane specifically designed for half-sized single-board computers. The primary edge connector for this involves a now-obsolete 188-pin EISA standard slot and this design is based on the PISA specification here -> https://www.kontron.com/download/download?fil … f&product=87221. While the edge connector is the same on the Allen Bradley SBCs, the physical wiring is incompatible with current PISA backplanes. Making them work requires careful design adaptations in order to accommodate the need for a 3.3V CPU supply in place of 8 total pins with alternate usage in the PISA spec. As a bonus, the EISA slot in the PISA configuration also allows for the use of ISA half-sized SBCs.
EBay is littered with these SBCs in many different forms and prices. Since these were primarily designed for industrial use, the prices can be a bit steep. However, of the examples I've purchased, most were under the $60 range so deals can be found! Since the slot this was designed for has been EOL for a while, the inventory of these cards will have less and less value as the systems they are supposed to go into slowly evaporate. The remaining value of these old x86 SBCs will be found in ancient industrial equipment and/or retro enthusiasts who want to tinker with these cute little boards. 😜
For example, these are a few of the part numbers I'm designing around. These are all half-sized SBCs for reference measuring around 190mm in length.
(ICP PCISA-158HV) PISA Socket 7 Pentium
(IEI PCISA-C400R-RS-R20) PISA Mobile Celeron
(Advantech PCA-6145R) ISA Socket 3 486/586
(Allen Bradley 6189-1CPU233) Proprietary Socket 7 Pentium
Currently, the status of this project is in the design stage and almost ready for the first prototype boards. My affinity for small compact computer systems powers my motivation for doing this. I grew up with computers since I was a kid and always admired smaller systems that were as powerful as the larger bulky stuff. I never had access to industrial single-board computers back in the day. Now, getting into vintage computing, I'm now able to afford to play around with these old systems and it's been super fun. Check out the World's Smallest Voodoo 2 SLI system using these boards... 😜 Worlds Smallest Voodoo 2 SLI Setup!! (Need help PISA SBC)-FIXED-
The focus of this project will remain on the half-sized SBCs. I really like this small size and the design goals are focused on the backplane that's sized to fit these cards. And, while the PICMG is a newer spec that is just as small, it involves PCIe which is a bit too modern as I'd like to stick with early Pentiums and 486 stuff for retro software and games.
Of course, there are many SBCs that are monsters! Most of those use a variety of slot types that get rather lengthy. I'm not trying to reinvent the wheel here and I'm focusing on a short 4-slot backplane that might be useful for those looking at these small SBCs. The overall footprint will be similar to the 4 slot backplanes such as the (PISA) IEI IP-4S2-RS and the proprietary Rockwell Automation A140422-020 Backplane. The latter is rather expensive and in short supply. I don't have much knowledge about the Rockwell backplane other than what I've been able to reverse-engineer from my only example. You can see some of my journey here NEED BIOS HELP Rockwell Automation 140420-030 Allen Bradley 6189-1CPU233 CPU Card SBC.
From what I've gathered, there are a few major differences between the two incompatible cards. The first is the usage of some of the slot pins to provide the 3.3V CPU supply for the card. This is very different from the PISA SBCs in that they all have the 3.3V supply onboard. The next is the relocation of the BIOS battery to the backplane instead of onboard. The BIOS battery feeds through a double diode with the 3.3V supply through what is a ground pin on the PISA spec to feed the super cap on the SBC which runs at 5V when the board is live. Both the supply and BIOS battery use a variety of pins with other important uses on the PISA spec backplanes. Since all these SBCs can be directly powered and function outside of their respective backplanes, I'm confident that careful routing of the ISA and PCI traces will allow for a compatible backplane that omits and/or allows for some of these pins to simply float.
Another reason for only a 4-slot design is because of the Allen Bradley's usage of the CPU supply in place of a few of the 3rd and 4th PCI clock and control pins. While I could simply have an alternate routing for these pins when using a true PISA card, for my purposes, two PCI slots are enough and I wish to keep the overall footprint smallish. This also makes reconfiguring the backplane relatively simple and straightforward. This also means designing some safeguards so as to not ground out the 3.3V and VBAT when slotting a true PISA vs AB card into the SBC slots.
One important design aspect is the allow for the overhang on these cards. In the half-sized category, in order to be efficient, the PCBs are designed to overhang the slot and require about 25mm of clear space behind and 10-15mm of space to the immediate right. This space must be clear of connectors or components that are more than 5mm in height. All the examples I have incorporated this space for jumpers, RAM, headers, etc... In fact, the IEI IP-4S2-RS backplane I have fails because of the SBC fan header colliding with the screw terminals in two examples. The front of the cards also overhang and must be designed around as well.
The current slot configuration is as follows:
Slot 1 (PISA, AB, ISA) SBC
Slot 2 (PISA, AB, ISA) SBC
Slot 3 Dual purpose ISA/PCI
Slot 4 PCI
It's important to note that most of the SBCs take a large amount of space and in most cases take nearly two slots internally. This is for two main reasons, the CPU heatsink and the RAM. Depending on the CPU, a low-profile heatsink and fan can be used. With adequate airflow, you might be able to get away with omitting the CPU fan. The mobile Celeron I have for instance has no CPU fan but does require airflow to cool the CPU. The next height issue is finding the lowest profile RAM possible. For the socket 7 systems, I was able to find some PC-100 RAM that was 24mm in total height and that's about as low-profile as I've seen.
Some possible configurations:
SBC, ISA card, PCI card, PCI card
ISA card, SBC, PCI card, PCI card
ISA card, ISA card, ISA SBC
These will all depend on the total width of the cards. The ISA, ISA, ISA SBC is a less desirable configuration since an ISA SBC will have no use for the PCI slots. This would simply be a viable configuration since the EISA slot edge connector in ISA is compatible with the upper pin layer in this implementation.
Most of the SBCs have a dedicated power port similar to a standard 4-pin Molex or other. The Allen Bradley for example has a power port that's in the style of the old AT power connecter but is a proprietary keying and pinout. I've explained this in the Allen Bradley SBC post linked above. Powering the SBC directly can be important since the EISA slot has a dual-level pin configuration with rather anemic gauge contacts. This can make for weak connections as opposed to the ISA slot which has larger contacts. I've added a 6-pin header that can be used with a custom-made cable to provide power for either SBC. I've also included a -5V regulator for the vintage ISA cards that may require it.
Deciding on the mounting hole dimensions was interesting. While simply adding holes at the extremes, I decided to also add a second set of holes to the right and bottom that allows for this board to be mounted in the lower slot portion of a micro-ATX case. While I imagine I will make a custom case that encloses the system, for practical reasons, mounting this in a micro-ATX case might be required or beneficial.
While the backplane function can be purely passive, I wanted to add something useful. Based on the work from Sergey Kiselev's DIY backplane https://www.retrobrewcomputers.org/doku.php?i … backplane:start, I've added his implementation of a discrete logic POST code decoder. The overall function of the backplane doesn't require populating these parts, but I've used my POST code reader a handful of times to diagnose some of these SBCs. I'm hoping this addition will be worth the time it took to place and route the ICs. 😁 This is untested and I won't be able to declare its functionality until I'm able to test the first prototype.
The PCI MAPPER!
One interesting part of this journey has been understanding how the PCI slots are addressed in the different backplanes. This is clearly laid out in the PISA spec which was very useful! Specifically, this involves the routing of the INT and IDSEL pins to the respective PCI slots. For instance, on the PISA spec, the INT(A-D) pins are wired round-robin. The AB backplane doesn't follow this and has specifically different INT wiring. The IDSEL pin needs to be connected to a specific PCI address pin and is different for each additional PCI slot. While the PISA spec uses AD19-22, the AB backplane I have uses AD27 and 28 for the two PCI slots. I've come across information on another card (can't remember where now), that uses AD30 and 31. So, instead of two configurations, I've decided to add a PCI MAPPER CARD. This card can be configured with any combination of the 4 INT pins as well as the IDSEL from PCI slots 0 and 1 to any address line from AD19-31. The PISA spec clearly shows the passive backplane using AD19-22. I'm taking a guess that none of the SBCs use higher than AD19 and included all the way to 31 for flexibility with other cards I haven't come in contact with.
It is my belief unless someone corrects me, that these hardware wirings are mapped somehow in the BIOS and are configured specifically to the backplane they expect to use. The PISA spec is clear and I assume the SBCs are configured to address PCI cards in line with the spec. The AB card is different, but I've been able to reverse-engineer this and it's clear that a reconfiguration of these pins is required when using the different SBCs. The question of, "Is PCI addressing agnostic at the hardware level?", seems to be a tough question with an elusive answer at least for me. I'm moving forward with the assumption that this wiring configuration is board-specific and held in the BIOS as an address map.
The decision to make this a card and not simply a set of solder jumpers was to bring focus to the issue. The card represents something important to the functionality of the backplane and a physical card brings attention to the user. This is also untested and I'm hoping it works out the way I think it will.
PISA backplanes are still plentiful on eBay and range in price. Since this is just a passive backplane, the "cheapest" 4-slot IEI-manufactured board will provide the same functionality for the PISA cards. The Allen Bradley cards on the other hand require the proprietary board and are much rarer while still being close to their MSRP even 20 years later. Making a universal board seems to be a worthwhile effort as the AB cards are more of a mystery in the absence of documentation. Aside from the 6189-1CPU233 socket 7 cards I have, there is also the socket 370 flavor Rockwell Automation 140573-010. Being Rockwell/Allen Bradley, I'm under the assumption this board also takes the proprietary backplane. However, unlike its socket 7 counterpart, this has a fully implemented CPU supply onboard and it's unclear if it needs the 3.3V dedicated pins. This is an assumption from looking at the photos of this card. It's unclear if this card does in fact use the same backplane and I won't know for sure unless I actually encounter one for myself. Because of the unknowns, I will most likely need to make a list of known working boards for reference.
And finally, I'm very curious about feedback from others that have played with or want to play with these half-sized SBCs. Keeping things in a relatively small footprint, what specific features might you want in a board like this?
Thx ahead for any advice or critiques. Once I nail down the schematic a bit more, I'll gladly post it here. 😄👋