@all_repair
Thank you for your generous offer. Unfortunately we don't have a turn-key solution at the moment (Kjelt, MarkL, et al, please double-check me on this), as I believe the others are still searching for a commercial programmer that is qualified to read the Mostek 36000. There are two considerations that must be kept in mind, firstly, that the Mostek device has only 24 pins, whereas most other 8KB devices are in 28-pin packages. Making an adapter is simple, and has been done by Kjelt. The second consideration is that the Mostek device latches the address inputs on the falling edge of /CE, which is also used to enable its output buffers. This means that /CE must be cycled back 'high' and then 'low' with new address values in order to read successive locations. It is possible, perhaps even likely, that programmers would cycle the chip-select, output-enable, etc when reading a device, but it probably isn't something that is specified. It may, or may not, be relevant to note that Kjelt was unable to get anything other than 0xff from his ROM when using the adapter. It may be simply because his ROM is completely dead, or it may be related to the access protocol of the programmer he used. I have successfully read my ROM in-situ by replacing the 8039 micro with an Arduino hookup, described elsewhere. Of course, it would be simple enough to make an adapter board that accepted the de-soldered ROM and which plugged directly into an Arduino.
So, reading the ROM isn't a two-minute job, at least not at the moment.
Now, about the replacement device for the ROM. We have blithely assumed that we would use an MCM68764 or MCM68766 EPROM, because it is directly pin-compatible with the MK36000, and there are undoubtedly a number of programmers around that will handle those devices e.g. an old Sunshine EXPRO-60 of mine has them listed.
However,
Kjelt, MarkL, and anyone else reading this / contemplating replacing their ROM, please check my reasoning and understanding of the 8039 in the following paragraph.
I had unconsciously considered the 8039/ROM implementation in the 3468 as (to me, anyway) a standard external memory system using a multiplexed data/address bus, where the lower 8 bits of address need to be latched, and the higher address bits are provided statically from other pins on the microprocessor. The 3468 provides the necessary latch (U504) and address bits <12:8> come from P14 and P2<3:0>, so I didn't see any reason why the non-latching MCM68764/6 shouldn't drop straight in. It was only this evening that I realized there are no timing specs associated with P2<3:0> when it is used to provide the higher-order address bits for external memory access. For those unfamiliar with the 8039 series, P2<3:0> can be used as normal I/O bits, but during external program memory access they will be replaced by address<11:8> for the access, then restored automatically. The exact words from the MCS-48 manual are:
'... the I/O information previously latched will be automatically removed temporarily while address is present then restored when the fetch is complete'
I would like to hope that this means that those address bits will be present right through to the end of the /PSEN signal, which would then mean that there is no problem with us using the MCM68764/6 ROMS which don't latch any of the address lines. However, the fact remains that the MK36000 latches *all* address lines at the leading edge of /PSEN, and so it doesn't care what happens to them after that.
Adding to my concern is that all the example / system block diagrams for 8039 expansions show the use of the Intel 8355/8755 Peripheral_IO/ROMs which latch all address bits, or show generic EPROMs with an address bus latch (width unspecified).
It should be noted that in the 3468 the outputs of P2<3:0> are latched (into two separate 4-bit latches) for use driving the keypad and the display, although I am not inclined to read too much into that.
Tomorrow I will get the scope hooked up and see if I can resolve it. But until I do, it may be prudent to wait until there is a definitive answer.
As a further thought, it might be worth foregoing the convenience of the MCM pin-compatability, and use a more modern/common EPROM which will also run cooler (the MCMs draw 85mA, vs ~20mA for a 27C64A). The necessary adapter could also carry a high-order address latch, if it turns out to be necessary.