I am trying to figure out what type of machines could be used to reconstruct these chips nowadays. It doesn't need to use exactly the same machines used to build them, it's just that the product should be very close.

I cannot find much about the original chips though. The 6502 has plenty of documentation, but I cannot find much about the original manufacturing machines used to manufacture it.

The 68000 is harder find, as it was manufactured by many different companies, so, not even I can find what was the original node. Wikipedia says the original was done in 3.5 um and Hitachi later shrank for a 2.7 um node.

It used HMOS, licensed from Intel: https://en.wikipedia.org/wiki/Depletion-load_NMOS_logic#The_HMOS_processes But, there was several versions of this process, so, I am not sure if there was any improvement to the node. (I would like to also reproduce those used in arcade machines and videogames, which used 7.5MHz, 10MHz an and 12MHz vesions).

For these nodes, it seems the g-line of lamp mercury was used(436nm). But some steppers are compatible with the h-line 404.7nm, which seems to be close to violet laser pointers (405nm), which are diode based.

It seems that these thick nodes could use contact steppers, but there is not much information about them. Perhaps their used ceased due to low yields? (that's all I could find).

  • 5
    And unless you are a millionaire and can afford those type of machines, hobbyists can recreate those chips easily and affordably in FPGAs (and there's plenty of existing projects).
    – dirkt
    Sep 29, 2019 at 4:55
  • To clarify: are you wanting to fabricate new 68K and 6502 chips using the original processes?
    – RonJohn
    Sep 29, 2019 at 14:09
  • The idea is to recreate the original chips, or at least as close as possible. The original process doesn´t need to be the same, only the end result, the microchip. But I think knowing how the original process worked would help. 6502 is not hard in this regard, but the 68000 is... It is not a large scale production, so it can a process that is slow. It seems that en.wikipedia.org/wiki/Photoplotter do have enough resolution, like this one: fortex.co.uk/product/photoplotter-fpm25000 1.5 um x 1um.
    – user36088
    Sep 29, 2019 at 16:51
  • 2
    @dirkt: The 0.1uA static current consumption of the 40+-year-old CDP1802 is very low even by today's standards. A modern device which draws more current may not be suitable for all of the applications in which the 1802 would have found itself.
    – supercat
    Sep 29, 2019 at 19:08
  • 1
    The process used to fabricate these 3.5 to 8 micron NMOS chips is likely too dangerous to replicate. Many of the sites where these chips were made, ether were or are still on the Toxic Superfund clean-up list. It's quite possible that a modern fab would not allow some of those old chemicals used anywhere near their modern clean rooms.
    – hotpaw2
    Oct 2, 2019 at 6:07

3 Answers 3


It should be feasible to manufacture NMOS-process parts using 0.6µm equipment that is still in current use. WDC's W65C02S, as well as their other current products, are made at TSMC on 0.6µm, though they are CMOS chips.

You would need to obtain examples of the original masks, have them converted to a format usable by the newer equipment (probably whole-wafer masks with the "stepping" procedure already baked in), and discuss with silicon-process experts how to map the steps of an NMOS process onto those available in a CMOS plant.

Of course, if what you actually want is a 65C02, then just buy a new one direct from one of WDC's distributors.

  • It's the original thing that I want at 8um. W65C02S has 2.5x the transistor count, so it should be substantially different.
    – user36088
    Sep 29, 2019 at 16:40
  • 1
    The 6502 processes some instructions in ways that are usefully different from the 65C02. Two documented differences are the cycle counts for decimal-mode ADC/SBC (annoying but excusable) and JMP indirect (silly and unnecessary, IMHO). Undocumented instances include opcodes to fetch and ignore a value, perform a single-instruction decrement and compare, fetch a value into A and X simultaneously, or store the bitwise AND of A and X.
    – supercat
    Sep 29, 2019 at 18:55
  • Masks for multi-um CPUs could be easily recreated from the photos of CPU die, as it is done here visual6502.org or here zeptobars.com/en/read/…
    – lvd
    Sep 30, 2019 at 19:08

Similar processes/equipment will likely still exist for SSI and especially analog ICs - you could not shrink either of these indefinitely without messing with their robustness/current handling capacity, and analog parts would suffer terrible noise, precision, and drift issues in addition....

  • I am not sure where to find this type of equipment. PCB plotter boards haver precisions which are better than the processes used for 6502 and 68000, but that's all I could find.
    – user36088
    Sep 29, 2019 at 19:33

Small amounts (around 10 chips) are available at some manufactureres for prices of ~5..10k$ per batch.

Here http://www.nedopc.org/forum/viewtopic.php?f=79&t=218 is the report of an enthusiast who successfully manufactured his own trinary logic chips in such a way on 0.5um CMOS process.

I guess that NMOS-style schemes are still available in any CMOS process, probably with bigger area and power.

  • What are these manufactures that can manufacture at 3.5 and 8 um? My idea is a more of a reconstruction, which is as faithful to the original as possible, than simply manufacturing something similar.
    – user36088
    Oct 1, 2019 at 1:32
  • Where is the difference laying between being 'similar' and being 'faithful to the original'? Using smaller process, you still may draw bigger transistors, does it count as being faithful enough? If you'd find correctly sized process, it still may contain differences from the specific process of 80ies, again will that count as being 'faithful' or not?
    – lvd
    Oct 2, 2019 at 17:49
  • It's the same spirit of a historical restauration. The only way of keeping something conserved forever is reproducing either entirely at once or slowly replacing its parts, until it is fully a new object, except that it is pretty much the same as the old one. You need to keep the process as close as the original one, though, or otherwise inaccuracies will accumulate over time. It's also useful to have the cheapest means to achieve that. I am now wondering if you could use the actuator arm of a hard disc to produce a type of mask.
    – user36088
    Oct 2, 2019 at 23:21
  • I would say that computer system is, first of all, a DIGITAL system that is easier to reproduce that completely analog one. For example, reproducing CPUs like 6502 or i8080 (from my previous links) is as easy as replacing actual CPU with a small pcb board, that contains FPGA, some level converters and basically that's all. Things tend to behave well after such a replacement. On the other hand, mad analog things like SIDs (6581 vs 8580) failed to reproduce themselves changing from 6581 to 8580 -- made on the same factory, with slightly different process!
    – lvd
    Oct 3, 2019 at 17:40
  • In the latter case, I don't see any real way to exactly reproduce SIDs, electric circuits simulation is not feasible, while any other method tends to be more or less 'emulation' (what I mean, using ad-hoc models instead of meticulous behaviour reproduction)
    – lvd
    Oct 3, 2019 at 17:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.