10/29/2015 at 17:14 •
I'm in the middle of a few free days right now, and I figured it's time to take a look at this project. Things aren't good. Let's go over what's wrong with it:
- It's not done
- It's not even close to being done
- The 'not being done' is mostly a result of using a backplane design, using discrete logic, and using extremely old and expensive parts.
- Yes, the way we do things now is better than the way things were done. Imagine that.
So, how am I going to fix this? Rapid iteration. I'm figuring every month I can dedicate 10-20 hours to this project, and going by how long it takes Seeed studio to ship some boards to me, that's just about right for a single hardware revision. Therefore, this project is getting a complete refactoring. I'm going to design monolithic boards (with *everything* on one board - ram, rom, whatever), and get as much as I can working. Iterate on that until it works, and add more stuff. You never expected me to be done with this, did you?
Another factor that has led to this project being stuck in quicksand is the use of 5V parts and discrete logic. This was reasonable for 1982, but it ain't the 80s anymore. It isn't even the 90s. CPLDs are the way to do this, and memories are larger and cheaper if you use a 3.3V part. What does that mean?
The Motorola MC68SEC000.
The *SEC* part is a very late evolution of the 68k family; it's a static design, so it will work even when the clock doesn't. It works at 3.3V, opening the door to some really great parts and programmable logic. It's not available in a DIP part, though, leading me to this problem:
That's a 64-pin TQFP. I can solder that, but it will most likely mean boards with more than two layers in the future. I can deal with that.
With a new part that I created in Eagle, that means I need to verify that part is correct. This means ordering a board with just this package on it, and I think I can do something cool:
That's a board with the *SEC* package just freerunning, with some LEDs attached to the address lines. It's effectively a binary counter in a very small package. And to indulge the racists on hackaday.io, yes, I can solder CPUs.
With such a small board, I think it's time to go over the minimal configuration of a 68000 from one of the best resources around.
Tie to ground:
Tie to VCC:
The *SEC* part, along with the *EC* parts have two additional pins: /AVEC and MODE. /AVEC is an extension of the IPL... pins, so we can just tie that to VCC. The Mode is a bit different, and something that's incredibly valuable in the new design:
The MODE input selects between the 8-bit and 16-bit operating modes. If this input is grounded at reset, the processor will come out of reset in the 8-bit mode. If this input is tied high or floating at reset, the processor will come out of reset in the 16-bit mode.
With one part I'll be able to test both 8-bit and 16-bit data busses, something I've struggled with in finding a proper development environment for this computer. That's cool, several problems down.
As for now, the next update will be a blinkey thing that is a complete waste of transistors. That's cool, though: it's just to verify the part.
04/25/2015 at 06:07 •
Quick story. I brought this computer to the Vintage Computer Festival 9.1 last year to show Bil, Dave, and all the other cool people at the event. At the time, I was freerunning the processor, watching the blinkenlight count up. Great stuff, and proof that the CPU works.
A few hours into the show, the CPU board stopped working. It just wouldn't free run. No idea why, as I was able to get the board working by wacking it against a table once I got home.
Then the Hackaday Prize happened, and then the summer con season happened. I've been working on this off and on in the meantime, but no *real* progress. I made a better RAM card (more on that in a bit), but it's still a bit away from sending characters out over the serial port.
Since then, I got the wire-wrapped CPU card free running again, and I decided to take it to the Vintage Computer Festival X last weekend. When I pulled it out of the box and turned it on.... nothing. This computer hates VCF for some reason.
Therefore, I have decided to build a proper CPU PCB. It's just buffers, the CPU, a clock, and some blinkenlights:
And I have to route that before I leave my house for a month for con season. Great.
I had a front panel milled out of a big ass piece of aluminum. Here are the videos of the milling:
And a picture of the front panel:
That panel was made by the SeeMeCNC guys when I was up there for the Midwest RepRap Festival. I traded a gigantic (5 foot x 8 foot) hackaday flag - made by me - for the CNC work. All the relevant files are up in the Git.
Front panel not included. I'm moving the reset circuitry to the front panel, btw. Another board to build.
Four megabytes of RAM
I hated the wirewrapping on my RAM card. The initial plan for all of these cards was to prototype them with wire wrap, then build a board. I met myself halfway on this one.
That's the layout for the RAM chips, eight 4-Megabit chips. You'll notice there is no control circuitry on this one; there's a reason for that. I might want to change the control circuitry to a PAL or something down the line. Easiest solution is to make a proper layout for the RAM chips themselves, break out the data, address, and control lines, and leave everything else up to wirewrapping. Easy enough.
Since I'm doing all of this from scratch, It would be nice to have a development tool.
That's the Motorola Educational Computer Board, the official 68k trainer from 1981. It works, I have it plugged into a terminal, and it has a great monitor in ROM. I'll be using this heavily once I get a few characters spitting out of the serial port of my project.
Picked that up on eBay for $40, btw.
Also on the eBay Front...
I have most of DTACK Grounded.
DTACK Grounded. the journal of simple 68000 systems, was a newsletter put out by [Hal Hardenberg] on how to design a simple 68000 system.
Most of DTACK grounded covers the development of a 68000 add-on card for the Apple II and a BASIC interpreter. It's good writing, and the issue ever 68000 homebrew wants to read - number 6 - tells you exactly what you need to leave out, what you need to leave in, and what pins to connect to where.
If you want to read these for yourself, here you go.
I did pick something up at VCF...
That's a 12MHz 68000. Five dollars. It's an actual Motorola part, so this is what I'll be using from now on. I don't know if I'll be running it at 12MHz; I specced all my decoding logic for 8MHz. I'll try it - it's just changing a crystal - but I don't expect this computer to run at 12MHz.
That's it for this update.
I'm leaving for NYC next weekend, and LA a few days after that. I won't be home for a month and for some reason that means no Windows, and no Eagle. Don't read too much into that last sentence.
I need to get this CPU card done and sent off to fab. It's exactly like the earlier wire-wrapped CPU card, only this one hopefully won't fail all the time.
So... yeah... it's been embarrasingly long between this update and the last, but hey, I've been busy, and technically, I have...
Read more »
03/26/2014 at 23:12 •
This is something I've written about in a front page Hackaday post, but I think it's time to go over a little more of the theory of what I'm doing here. First, a video:
This is called freerunning the processor. Basically, it executes one instruction, the program counter is incremented, the address is increased by one, and the CPU just sits there, doing nothing, cycling through its address space. Attach a few LEDs to the address pins, and you have an incredibly complex binary counter, also known as blinkenlights.
That's the simple explanation. It's a fair bit more complex in practice. I need to tie a few pins to +5 volts, and ground DTACK. Oh, what about the instruction to freerun the processor? NOP, right? NOPe.
When the 68k first resets, it reads the program counter vector. The program counter vector must be an even address, and the opcode for NOP is $8E71. See that one at the end? That means NOPping the CPU from boot would create an illegal address exception. Then bad things happen.
So, I need an instruction that does nothing, and is even. Inclusive OR Immediate (ORI) does this. Specifically OR.b #0,d0. Bonus, this instruction in hex is $0000, or all zeros. All I need to do to freerun the processor is ground all the data lines.
My first go at freerunning the CPU only used one LED. This LED was tied to the A20 line through an inverter. I hate to waste the five extra inverters on that chip for a single LED, so I added another three.
Now I have status lights for the top four addresses in the computer. Since I'm putting the ROM at $FF0000, the serial port at $FE0000, the video peripherals at $FD0000, the microcontroller at $FB0000, I have a graphic representation of what the CPU is doing with all its peripherals. That's pretty cool. Useful blinkenlights.