Alice 2
In 1995, we (Brad and Lawrence) were living in Sunnyvale, about forty-five minutes south of San Francisco, California.
The Alice 2 was the obvious next step after the Alice 1. We decided to work on it together as a fun educational project. The Alice 1 was a single-board computer with everything hardwired into it. The Alice 2 was designed to be extensible, with a bus and plug-in cards. For simplicity we kept the Z-80 as the CPU, but that's about the only resemblance to the Alice 1.
We used a rack-mounted modem frame that we found at Weird Stuff. It came with its own power supply, but it didn't provide the voltages we needed, so Lawrence made his own, which didn't work. We ended up using a PC power supply, which gave us the right voltages and enough power. The nice thing about ours, though, was the amp-meter, which told us when we had a short circuit in our system.
We used a simple system bus that basically exported all the important pins of the Z-80. Two 25-wire ribbon cables were used for the bus, each with DB-25s attached at regular intervals to plug the boards into. Everything was attached to the back of the frame so that our cards could slide in and connect to the plugs.
We then set ourselves on making the plug-in boards. We decided to first prototype the boards on breadboards, using long 25-pin cables to attach to the bus. We would then make the printed-circuit (PC) boards. We later found that we spent more time manufacturing the PC boards than designing or prototyping them, and in retrospect we should have found another method, such as wire-wrapping.
The first board was the clock board. It was responsible for toggling the clock pin of the bus. There were several switches that allowed the user to set the clock board in one of three states: single-step, variable, and full-speed. The "variable" mode used a knob to go from stopped to 1 KHz, and the full-speed mode went at 1 MHz.
The best advice to anyone designing either hardware or software is to spend time on debugging tools—they are always worth their development effort. So we next built the debug board, which displayed the state of the bus. Six 7-segment displays showed the address and data bytes, and ten LEDs showed various pins of interest. We could not have built the next two boards without this one.
We were torn between building a memory board next or a CPU board. It's a chicken-and-egg problem since they both need each other in order to be tested. We finally built both on breadboards, then built the memory board, followed by the CPU board.
The memory board has 16K of EPROM and 48K of RAM (since the Z-80 can only access 64K). There's a switch on the board to turn 16K of the RAM into memory-mapped I/O. We used this later for the bit-mapped display.
The CPU board has a Z-80 CPU, an 8259 interrupt controller, and some glue chips. We made a big mistake when making this board. We got overconfident with the success of our previous boards, and didn't bother to prototype the 8259 part of the board. We had never used this chip, and it took us over a year to figure out why it wasn't working and to patch the board. We even ended up making the board twice because our first version was so badly botched.
We wanted some input and output, so next came the I/O board. The input of choice was a standard PC keyboard, and the output, for simplicity, was a 2-line LCD display. We wanted serial I/O as well to download code from our PC (burning EPROMs was becoming tedious). We used a PIC chip to interface to all three. Using the PIC was a good idea in the end, but added a level of complication that we hadn't expected. Our computer was essentially a multi-CPU machine! Ironically, the I/O co-processor ran at 20 times the speed of the main CPU.
Our final board was the video board—we wanted to display images on a monitor. It took us a long time to decide on the design. At one point it seemed easier to use yet another PIC chip that would take drawing requests from the CPU. Finally we settled on a simpler memory-mapped I/O scheme, with the CPU writing to memory and a counter reading from it. We used an EPROM to map out the horizontal and vertical sync signals. Although the board worked very well on the breadboard, it stopped working altogether when made into a PC board. After many months of frustrations with analog electronics, we finally added enough capacitors to make it work.
Not a lot of software was written. We wrote the basic ROM functions needed to write to the LCD display and read from the keyboard; the PIC code to interface with the peripherals and the CPU; and a simple C compiler. We even started on a multi-tasking operating system (the PIC board interrupts the CPU 10 times a second). In the end we didn't write any more software because every time we turned the machine on we had to spend two hours finding out why it had stopped working (usually a bad solder connection).
The whole Alice 2 project took about 4 years, with large breaks throughout. We're not sure how much we spent on it, but it's probably on the order of a thousand dollars. We did half of it in Brad's apartment in Sunnyvale and half in his garage in Mountain View on Wednesday nights. We learned a lot and had a great time, but are glad to go back to software, where weenies like us belong.
You may find other interesting Z80 projects at www.z80.info.