Collection of notes, software and upgrades for Ben Eater's 6502 Computer
I've been having a ton of fun with this project. Probably to the point my coworkers are sick of hearing about it. I had a lot of the parts list for the computer laying around, so I just ordered Ben's Clock kit. While I was waiting for that I used my arduino mega to generate the clock as well as look at the signals. Which was fun because I could clock up to 8Mhz. This was only up through one of the very early videos so does not necessarily run with RAM and the additional decode logic at that speed.
| File | Description |
|---|---|
| hand_asm/ | C code to build bin from hand assembled code and early video examples |
| bounce/ | One of my early bouncing lights examples, modified to use assembler |
| video4/ | Assembler code from video 4 |
| crt/ | A sample CC65 environment that has display and PS/2 keyboard I/O |
| echo_console/ | wokring space for developing console with well behaved input echo |
I'm using the W65C816 instead of the W65C02, which is an almost identical CPU, but can be put into a 16-bit mode if I want to try to extend this project later. It required a slight adjustment to how the control pins are tied, but as of the 6th video it is still working well. If any one is interested:
| Pin | Name | Connection |
|---|---|---|
| 1 | VPB | FLOAT |
| 2 | RDY | HIGH |
| 3 | ABORT | HIGH |
| 4 | IRQB | HIGH |
| 5 | MLB | FLOAT |
| 6 | NMIB | HIGH |
| 7 | VPA | FLOAT |
| 35 | E | FLOAT |
| 36 | BE | HIGH |
| 38 | MX | FLOAT |
| 39 | VDA | FLOAT |
This is incredibly straight forward, but has been a very nice addition. I had a ZIF socket that would fit the ROM chip so I put the ZIF socket on some standoffs to give it a little more room on the board. This makes swapping the program a little easier.
I didn't have a 2 row LCD display, but I did have a 4 row version. The 4 row is actually wired up in the exact same way as the 2 row version. The strange thing about the 4 row versions is that if you just overrun the end of the line it overflows in a not immediately intuitive order row 1, 3, 2, then 4. This should not provide a software compatibility problem as the display will just have blank lines if software only uses 2 rows.
My eventual goal is to get the C runtime for this to properly format newlines by advancing to the next line from the user point of view and having the text scroll off the top as one would expect when calling printf on a short terminal. I will try to writ the software to work with both 2 and 4 row models, but it may require a compile time setting be set. The committed version will probably default to 2 ROw to be compatible with the be6502's as originally spec'd.
The only oscillator on hand that was close to 1MHz was a 3.7152 Mhz oscillator. It wired up exactly the same as the 1Mhz clock and ran without issue.
Ben said that the address layout was some what wasteful of address space. The engineering trade-off goal was to reduce the complexity of the decoder. In fact, all his decoding logic only uses 3 nand gates, a pretty impressive feat.
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x3FFF | Main RAM |
| 0x4000 - 0x5FFF | Reserved |
| 0x6000 - 0x6003 | VIA Interface |
| 0x6004 - 0x7FFF | Reserved |
| 0x8000 - 0xFFF8 | Main ROM |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
The goal of this memory layout is to slightly more efficiently allocate the addresses that are used reserved for the IO area. Our trade space will allow for an increase in the amount of logic, but try to not go too far maybe at most 2 chips. Still keep is fairly simple i.e. a single contiguous RAM area. The final requirement is backwards compatibility so anything designed to run on the original be6502 will still run on this. That means we cant move anything that is used we can only shift around the reserved space.
To that end the first step is to allocate all addresses below 0x6000 to actual RAM going from 16k to 24k RAM. The next priority is splitting out enable signals that each correspond to 1k of I/O address space. Then maybe each of those could be used for a different one of the chips in the future expansion section below. Once a serial loader is available the ROM address space could probably be greatly reduced, but if we wanted to keep the Main RAM section contiguous then we would have to move the VIA and that would make it no longer compatible.
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x5FFF | Main RAM |
| 0x6000 - 0x6003 | VIA Interface |
| 0x6004 - 0x63FF | Reserved |
| 0x6400 - 0x67FF | Reserved for 65C51 |
| 0x6800 - 0x6BFF | Reserved for Device 4 |
| 0x6C00 - 0x6FFF | Reserved for Device 5 |
| 0x7000 - 0x73FF | Reserved for Device 6 |
| 0x7400 - 0x77FF | Reserved for Device 7 |
| 0x7800 - 0x7BFF | Reserved for Device 8 |
| 0x7C00 - 0x7FFF | Reserved for Device 9 |
| 0x8000 - 0xFFF8 | Main ROM |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
The circuit as originally designed only uses 3 of the 4 NAND gates on the 7400. This brings to mind the question can the extra NAND gate be used to get us closer to the target memory map above?
The answer is yes, with the caveat that we have to sacrifice a little robustness in the design. in particular we have to assume no one is going to try to write to the ROM. if they did it could actually mirror thru to the lower addresses of RAM. For 50% more RAM this seems like a fair trade off.
Fundamentally we want to change when the RAM is selected from when A14 is not high to when A14 and A13 are both not high. We can do this by changing the equations for /CS and /OE for the RAM from:
/CS = NOT( NOT(A15) & CLK )
/OE = A14
to:
/CS = NOT( NOT(A14 & A13) & CLK )
/OE = A15
You get this by first changing the RAM /OE(pin22) from A14(RAM pin1) to A15(CPU pin25). The extra NAND gate is input A14(already connected to another NAND gate) and A13(CPU pin23). Finally, where the output from the not of A15 is fed into the adjacent NAND gate to combine with the clock instead feed in the output of the new NAND gate.
Weirdly, having NOT(A14&A13) can free up one of the NAND gates used by the VIA. Changing its CS inputs from:
/CS = NOT( NOT(A15) & A14 )
CS = A13
to
/CS = NOT(A14 & A13)
CS = NOT(A15)
So now the question is can we use this reclaimed NAND gate to either get us back protection on writes to ROM changing RAM? Maybe we get A15 back into the calculation for the RAM chip select.
/CS = NOT( NOT(A14 & A13) & CLK )
/OE = A15
to:
/CS = NOT( NOT(A15) & CLK )
/OE = NOT( NOT(A14 & A13) )
TL;DR So in the end you can use all 4 NAND gates to increase the RAM by 50% without any loss of functionality by changing the bus selection logic from:
ROM:
/CS = NOT(A15)
RAM:
/CS = NOT( NOT(A15) & CLK )
/OE = A14
VIA:
/CS = NOT( NOT(A15) & A14 )
CS = A13
to:
ROM:
/CS = NOT(A15)
RAM:
/CS = NOT( NOT(A15) & CLK )
/OE = NOT( NOT(A14 & A13) )
VIA:
/CS = NOT(A14 & A13)
CS = NOT(A15)
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x5FFF | Main RAM |
| 0x6000 - 0x6003 | VIA Interface |
| 0x6004 - 0x7FFF | Reserved |
| 0x8000 - 0xFFF8 | Main ROM |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
I don't know if this is as useful as it seemed at first. The decoder can take decoder 3 address lines and decode them to 8 output lines. allowing an address space to be easily divided between 8 devices. the 74138 is designed to control active low chip select devices. It has it chip selects to make it easier to turn off a whole set of devices. It has 1 active high chip selects and 2 active low chip selects.
| CS | /CS | A | B | C | Y0 | Y1 | Y2 | Y3 | Y4 | Y5 | Y6 | Y7 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| L | X | X | X | X | H | H | H | H | H | H | H | H |
| X | H | X | X | X | H | H | H | H | H | H | H | H |
| H | L | L | L | L | L | H | H | H | H | H | H | H |
| H | L | L | L | H | H | L | H | H | H | H | H | H |
| H | L | L | H | L | H | H | L | H | H | H | H | H |
| H | L | L | H | H | H | H | H | L | H | H | H | H |
| H | L | H | L | L | H | H | H | H | L | H | H | H |
| H | L | H | L | H | H | H | H | H | H | L | H | H |
| H | L | H | H | L | H | H | H | H | H | H | L | H |
| H | L | H | H | H | H | H | H | H | H | H | H | L |
Additional logic may still be needed to use the CLK line to drive the RAM and the limited number of inputs really causes the chip to drive the layout making it very difficult to build a layout compatible with the original design. But let's see what we can do.
We could use the A15 line on the decoders positive chip select to cycle thru 9 different active low regions, though we still won't have the clock in the correct phase to drive the RAM CS pin.
Decoder:
A = A14
B = A13
C = A12
CS = A15
RAM:
/CS = A15 [Would need to be (& CLK) to work ]
ROM:
/CS = Y7 [Would be nice to OR together some line to get more valid ROM space]
VIA:
/CS = Y0
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x7FFF | Main RAM |
| 0x8000 - 0x8FFF | Device Y0 (VIA Chip) |
| 0x9000 - 0x9FFF | Device Y1 |
| 0xA000 - 0xAFFF | Device Y2 |
| 0xB000 - 0xBFFF | Device Y3 |
| 0xC000 - 0xCFFF | Device Y4 |
| 0xD000 - 0xDFFF | Device Y5 |
| 0xE000 - 0xEFFF | Device Y6 |
| 0xF000 - 0xFFF8 | Main ROM (Y7) |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
If we are keeping the NAND gate in order to clock the RAM CS correctly, adding the decoder does let us split the "wasted" address space allocated to the VIA to other peripheral expansion. Then we can chain the VIA off the decoder in order to get more expansion for more VIA or other peripherals.
ROM:
/CS = NOT(A15)
RAM:
/CS = NOT( NOT(A15) & CLK )
/OE = NOT( NOT(A14 & A13) )
Decoder:
A = A12
B = A11
C = A10
/CS = NOT(A14 & A13)
CS = NOT(A15)
VIA:
/CS = Y0
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x5FFF | Main RAM |
| 0x6000 - 0x63FF | VIA Interface (Y0) |
| 0x6400 - 0x67FF | Y1 |
| 0x6800 - 0x6BFF | Y2 |
| 0x6C00 - 0x6FFF | Y3 |
| 0x7000 - 0x73FF | Y4 |
| 0x7400 - 0x67FF | Y5 |
| 0x7800 - 0x6BFF | Y6 |
| 0x7C00 - 0x6FFF | Y7 |
| 0x8000 - 0xFFF8 | Main ROM |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
This gets the target interface, but requires us to add an extra chip.
There are several clones of this part. I'm using the ATF16V8 that is still in production. This part is nice selection for this project because it can be programmed with the same minipro programmer used for the EEPROM and has a similar footprint to the Quad Nand gate so it will fit on the bread board. I've also independently played around with these. https://github.com/tedkotz/PLD_16V8
The GAL16V8 has 8 I/O pins and 9 input pins. So it can only drive 8 devices without additional logic. Though that can be based on fairly complex internal logic relying on the CLK and the top 8 address lines. This could allow the peripherals to have much tighter allocations, give RAM some Non-contiguous allocations, and change the memory map by just reprogramming the GAL. We could take advantage of this and make separate images for the having each device reserve blocked used or map to RAM.
| Address Range | Description |
|---|---|
| 0x0000 - 0x00FF | Zero Page (6502 Fast access RAM) |
| 0x0100 - 0x01FF | Call Stack (6502 Stack) |
| 0x0200 - 0x5FFF | Main RAM |
| 0x6000 - 0x6003 | VIA Interface |
| 0x6004 - 0x63FF | Reserved |
| 0x6400 - 0x67FF | Reserved for 65C51 |
| 0x6800 - 0x6BFF | Reserved for Device 4 |
| 0x6C00 - 0x6FFF | Reserved for Device 5 |
| 0x7000 - 0x73FF | Reserved for Device 6 |
| 0x7400 - 0x77FF | Reserved for Device 7 |
| 0x7800 - 0x7FFF | Reserved or more RAM |
| 0x8000 - 0xFFF8 | Main ROM |
| 0xFFFA - 0xFFFF | Vector ROM (6502 Vector Table) |
With the bank of LEDs on the outputs of the VIA, This little snippet of code uses the rotate and conditional instructions to create a bouncing light effect
0xa9, 0xff, // lda 0xff
0x8d, 0x02, 0x60, // sta 0x6002
0x18, // clc
0xa9, 0x01, // lda 0x01
0x8d, 0x00, 0x60, // sta 0x6000
0x2a, // rol
0x90, -6, // bcc -6
0x18, // clc
0xa9, 0x80, // lda 0x80
0x8d, 0x00, 0x60, // sta 0x6000
0x6a, // ror
0x90, -6, // bcc -6
0x4c, 0x05, 0x80, // jmp 0x8005
I threw together a simple C program to facilitate building the binary files with specified machine code. I used this instead of the script Ben used, but it is functionally identical.
You can see the code here.
I used the assembler and linker that comes with the cc65 c compiler for the video 4 code. You can see that here.
This started with the code from http://sbc.rictor.org/io/pckb6522.html.
I had to massage it to get it to work with CA65, the pins I was using PA0-1 and interoperate with the C runtime environment I'm using.
It could be used with a USB keyboard that supports ps/2 signalling (most do with this adpater).
https://www.instructables.com/USB-to-PS2-convertor/
- Add USB wiring diagram
- Make it easier to change pin configuration