Only to clarify some concepts:

Among the different proposals I read, I think Peter's proposal is the only one that really stops the MCU between instructions. That means that the whole micro is halted, including the embedded peripherals (for example timers do not run), since the MCU time (clock) is frozen. But if you use a standard MCU, you can not inspect your data in internal RAM, because the MCU is sleeping. If you don't need that kind of debugging, this is a good method.

The SW approach is based on interrupts, so the MCU is still working and is able to communicate with a PC. This way you have debugging capability. The drawback is that peripherals do not stop, so real-time debugging is not posible for many applications.

By the way, Peter's proposal can be implemented using a master MCU to generate the slave MCU clock pulses. This gives you more flexibility, you can run a set of instructions in one step, for example. Or you can run in continuous mode at different speeds.

Regards,

Alfredo.