Just an idea. JLCPCB offers populated pcb and tools to supply required information (BOM using JLCPCB available parts etc.).
Maybe a SMD version design with just RAM, ROM and possible U37 as well as connectors in THT (so that existing hardware mods will still fit) would open the possibility to have a Gigatron to people not able/or willing to solder hundreds of solder points.
This is way over my very limitted skills in pcb cad design.
(I am just able to make simple eagle designs with autorouter.)
But maybe someone feels interested to jump into it.
Of course, This is not for people who love to solder and work with THT only
This would be for people who will get a Gigatron much populated and only limited solder work needed to get a working system.
There exists a github project with an early smt/smd clone board draft. It seem good pre-developed from monsonite, but it looks unfinished and untested. Maybe this could be a good base for the further development of a gigatron ttl smd clone.
Thx. Interesting.
But one would still need a licensed eagle version, because of the board size.
I would not change the board dimensions and would keep exact the same position of connectors and Rom/Ram socket.
Just change 74 DIL to SMD parts, as well as all Rs and Cs and diodes.
Only because then one could get a 90% assembled board from the pcb fab,
but all cases and mods would still fit.
But who ever will probably do that task has the power to decide
I have a clone of the Gigatron in a slightly smaller form factor. I use some simple 16V8 GALs to replace the diode ROM and integrate the 128k/SD card expansion on board. I'm putting a kit together but also looking at producing an assembled version. The assembled version uses SMDs for the passive components but keeps the chips as through hole DIPs (soldering is done by machine).
I just shipped the kit version and have a few more available soon. The assembled version will be a few more months (end of the year?). Longer term I'm looking at a Gigatron made from discrete logic gates. This would be assembled out of 6-pin SOT23 SMDs, either one or two gates or one flip flop per chip. That's a much bigger project and probably a year out.
petersieg wrote: ↑11 Jul 2023, 13:24
Some pictures and more information would be nice.
I'll start a new thread and post some pictures when I've wrapped up some things. It's basically the "Novatron", but I changed the name (due to a conflict with another project) and switched to a full-size SD card.
petersieg wrote: ↑04 Jul 2023, 08:21
Thx. Interesting.
But one would still need a licensed eagle version, because of the board size.
I would not change the board dimensions and would keep exactly the same position of connectors and Rom/Ram socket.
Just change 74 DIL to SMD parts, as well as all Rs and Cs and diodes.
Only because then one could get a 90% assembled board from the pcb fab,
but all cases and mods would still fit.
But whoever will probably do that task has the power to decide
best, Peter
Well, it won't be the most simple. I mean, you'd likely have multiple voltage domains. Plus a number of chips don't exist in the smaller sizes. So you'd likely need to either cross voltage domains for those or make your own parts. For instance, with no adders, you'd need a bunch of transparent latches, multiplexers, or XOR/AND gates. And if performance becomes an issue, you'd need redundancy (ie., skip-carry adders). And with counters, you'd likely need to make an incrementer and connect that to a register. Even BEC circuits could be used that way.
As for diodes and resistors, I'd say leave the resistor packs at DIP size to make changing the values possible. But with diodes, I'd say use Toshiba diodes. They are 1 ns less in the worst case; 1.x ns is the best case. So a tad better than the BAT42/43 diodes.
I already attempted ones to redesign SMD version with more "regular" components.
But SMD suggests new resolutions.
Good idea to replace 16 bit ROM with two 8 bit flash chips, or may be even 3 of them with elements of VLIW architecture implemented.
With connector on board to flash them. If it's not concentration of how does it work it good to replace those diodes with chips with push-pull outputs promising higher freaquency.
And may be good to have Y also with counters for full VGA video and even Z.
And may be its good to have few more registers including more sophisticated addressing and simplest memory protections.
Also it is good to have at least one interrupt.
And it would be well to rise clock to much higher figure about 50 MHz.
As it could run early Linux after all of that.
Just changing microcode it could rapidly change what hardware it will emulate including variety of different historical machines.
Not bad idea to make them parallel several boards to be working at ones.
With one as for example busy with video, even may be X11Free, and control other one running software.
If you want to eliminate the diodes, options include high-speed transistors if you can find them, or PAL/GAL chips. But if you want just a little faster, use Toshiba SMD diodes. You can get them individually and in multi-diode packages.
One alternative could be to use an ALU chip. I don't mean the '181 or whatever, but one of the 16-bit ones. If you want to go to 16-bit, that is fine, but you can use it as an 8-bit chip if you want. Just use D8 in place of C16. And you'd need to change the diode matrix. But the problem would be how to work in the functionality it doesn't provide to load, store, and jump (the current ALU is involved in those operations).
And really, if you want to go faster, then move to a 4-stage pipeline and learn how to code the ROM to best use 3 delay slots. Do Fetch, Decode, Access, Execute. I'd do Access before Execute because only reads are modified by the current ops, never writes.
But you'd need faster RAM and ROM to go past 14 MHz no matter what you do. If you do use the current parts with a 4-stage pipeline and a faster ALU, the ROM and RAM would be the biggest bottleneck, and mainly the ROM.
And going faster than 6.25 MHz is a challenge unless you split the video output from the CPU.
As for interrupts and DMA, you can simulate those now. If you include a comprehensive controller circuitry, then you can have the ROM to talk to the controller and use spinlocks to buy time for DMA. As long as you are only doing an unimportant read to detect the RAM being available, the ROM can survive without the SRAM during that time. For IRQs, you could poll the input port. To do these ideas, you'd need to move all I/O out of ROM.
But you reach a point where it isn't a Gigatron anymore.
