Close

Virtual CPU

A project log for Novasaur CP/M TTL Retrocomputer

Retrocomputer built from TTL logic running CP/M with no CPU or ALU

alastair-hewittAlastair Hewitt 05/23/2019 at 02:370 Comments

The project is 3 months old. The CPU is running and boots up correctly. All the timing has been optimized and operates at up to 32 MHz (requiring a 64 MHz dot clock). The project even made it to the front page.

So far about 40% of the circuit has been built and tested on breadboard. The design is now finalized enough to commit to a proper CAD drawing of the schematic. This took several days to complete, which explains why the breadboard has stalled at 40%. It will take at least another month to finish the build and get the GPU generating a video signal. The initial PCB routing will start in parallel, so that may be completed around the same time.

The software design is also starting to pick up pace. This machine follows a similar design to the Gigatron in using the Harvard Architecture. There are two separate address spaces for both the ROM and the RAM. The ROM owns the Program Space identified in the schematic using the PA[0..15] and PD[0..7] bus labels. The RAM owns the Data Space identified in the schematic using the DA[0..15] and DD[0..7] bus labels. Only code in ROM can be executed and only data in the RAM can be accessed.

The machine code is designed to be more like a microcode rather than something you would use to write general purpose programs with. These low-level instructions can be combined to build up a more expressive instruction set in the context of a virtual machine. This machine would have the more familiar Von Neumann Architecture where both the code and data exist in the same Data Space of the underlying Harvard Architecture.

The plan is to take an existing CPU and build a virtual machine to be binary compatible with it, rather than design a new instruction set from scratch. The main advantage with this approach is to leverage existing tool chains and software for that processor. In theory, an existing C compiler for that processor can be used to build executable code, which would greatly reduce the software development overhead.

My inclination is towards something in the 68XX family. These have a simple elegance that should translate well to the VM. The current target CPU is the 6809 and might include some or all of the 6811. There is also a lot of retro love for the 6502 and it may be possible to have a reduced 6502 mode. All the CPUs in the 8-bit Motorola lineage share a lot in common, so it's just a variation on a theme.

Discussions