Who can clone/manufacture an ASIC chip?

[Rat_Patrol]

I have some custom processors I am interested in feeling out for bringing back into production (for my use).

They are ASIC, obviously, so an example IC needs to be deconstructed and copied. These are old chips from the late 80's.

Any reputable companies that offer this service?

[Benta]

It's a question of your wallet.
A simple gate array NRE is in the range of $1M upwards. Perhaps you should look at FPGAs instead? An 80s chip can't be that complicated.

[KaneTW]

Unless you have significant volumes, don't bother with custom silicon. The NRE costs are huge, even with prototyping services.

[Jeroen3]

Did you ask the company that made them in the first place?
They are probably more willing to make a few thousand if you do not need any warranties or special testing.

[Rat_Patrol]

Did you ask the company that made them in the first place?
They are probably more willing to make a few thousand if you do not need any warranties or special testing.

Originally Motorola :/

[KaneTW]

Huh, that's much cheaper than I thought it would be. Nice to see China reduce costs in that area as well.

[BrianHG]

Did you ask the company that made them in the first place?
They are probably more willing to make a few thousand if you do not need any warranties or special testing.

Originally Motorola :/

Which chip?
We may be blowing things out of proportion by thinking straight to asic.

[evac]

Huh, that's much cheaper than I thought it would be. Nice to see China reduce costs in that area as well.

I'm guessing those are just the manufacturing costs.

I know people doing that business in China. I can hook you up to people knowing the industry better, and he can hook you up to the correct people. I only do forward engineering, but with some help I can find people good at reverse engineering.

Do you know what is the typical flow for IC reverse engineering?

I'm guessing for older chips they can't just decap it, convert the layers to GDSII and re-send for tape-out, since the process may be obsolete and would need porting, right?
How is that done, exactly? I would guess if they can get the layers into GDSII, it's relatively trivial to run layout extraction to get an equivalent schematic, but that would be a flat netlist.
For something as big as a processor, I'm guessing they would need to do some kind of gate recognition to get a gate-level netlist, which they could then use to do PnR in a different process and standard cell library.
Still, without a higher-level understanding of the design, it would be hard to run any functional verification at all, so how can they guarantee that the re-implemented design will work?

This is a very interesting topic to me...

[chris_11]

Regardless, before you have a functioning IC clone in your hands, 7 digit funds are paid and with some luck only 3 years are gone. Copying an IC is nowhere easier than developing a new one, unless you can use the very same technology and copy the mask sets. A pin for pin clone has to fulfil the same timings etc. so expect several small revisions. ASIC IC biz was a hot marketing topic in the late 80s, but is almost gone, because the real costs and project times are prohibitive for most projects.

With 30+ years gone by I am hard pressed to believe, that a new development of the complete unit will not be cheaper than recreating an old clunker.

If it is around an higher end processor of the 80s like 68000 I would ask NXP which bought Freescale which inherited the digital section of Motorola, if they can dust of a mask set or information of that old IC. If obsolescence is imminent, you can make a last time buy. Typical wafers can be stored for a long time under nitrogen, so even if the IC fab is gone you can produce packaged ICs from stored wafers. Next problem will occur if the test system is gone. Then it will most times not be economical to manufacture those parts. Test development alone is a large cost driver.

[Berni]

Instead of asking for a company to reverse engineer a chip, ask the manufacturer instead for documentation on what the chip does.

Something from the 80s is not going to have all that complex of a functionality. So its easier to simply re-implement the chips functionality from documentation. You can then run this implementation on a 10$ FPGA.

And if you really want an ASIC for some reason, then you can fork over some big money to a manufacturer that will take the HDL code you ran on the FPGA and turn that into real silicon. To save money you can go for a semi-costum ASIC design where they use standard mask sets for the lower transistor layers to implement logic gates and then just make you a custom mask set for the upper metal layers that wire up the gates to implement your design. This means you only need to pay for like 1/4 of the total masks. The resulting chips will still run like 100x faster than the original anyway.

[evac]

Because you no longer have access to the exact 80s process, you have to reconstruct the netlist from pattern photos taken from the microscope, then reimplement the design in a modern PDK.
It's strictly speaking a reverse engineering then forward engineering instead of cloning, because it is impossible unless you have unlimited amount of money to revive an 80s process.

That would also be my guess, some sort of image->GDSII->netlist.
But to re-implement things with just a netlist is a nightmare, you would need at least a gate-level netlist.
To some degree they can partition the design based on the images, and use already available software to do gate recognition, but even then, like chris_11 said, to ensure that you are meeting all timing constraints and other specs on the re-implemented design...not trivial.

For something like a processor that is almost all digital IP, I would agree with him that it might be less effort to re-design from scratch, or like Berni mentioned, do the design yourself in HDL, run it on an FPGA, and if for some reason you really need an ASIC, it's easier to find companies that will do the digital implementation part which you can then send for manufacturing.

[coppice]

If it is around an higher end processor of the 80s like 68000 I would ask NXP which bought Freescale which inherited the digital section of Motorola, if they can dust of a mask set or information of that old IC. If obsolescence is imminent, you can make a last time buy. Typical wafers can be stored for a long time under nitrogen, so even if the IC fab is gone you can produce packaged ICs from stored wafers. Next problem will occur if the test system is gone. Then it will most times not be economical to manufacture those parts. Test development alone is a large cost driver.

I agree. Contacting NXP should be the first step. They may well have made provision with someone like Rochester Electronics to ensure a reasonable supply of properly stored die are available. If you don't know who Rochester Electronics are, Google them. Its an interesting niche business.

[Rat_Patrol]

IF we are looking into 6 figures, I'd rather spend the money on redesigning the entire thing.

There is a laundry lists of changes I've noted that should be done, should a redesign take place.

One example is there are actually 3 processors and a large micro on this board, all of which could be done by a single multi-core modern micro. Other circuitry is  compared to what is used today. Lots of thing where TONS of components are used for a function where we now have ICs to handle it all on-chip.

[BrianHG]

IF we are looking into 6 figures, I'd rather spend the money on redesigning the entire thing.

There is a laundry lists of changes I've noted that should be done, should a redesign take place.

One example is there are actually 3 processors and a large micro on this board, all of which could be done by a single multi-core modern micro. Other circuitry is  compared to what is used today. Lots of thing where TONS of components are used for a function where we now have ICs to handle it all on-chip.

That's it?
You do realize even a 5-10$ FPGA could hold and run all 3 processors from the 80's simultaneously and then some...
You also realize that those processors you want are probably already coded at OpenCores.org with exact functionality.
The ram which those can 80's processors can access would probably fit inside the FPGA's memory as well, or if not, on 1 single external ram chip. (IE, the Motorola's entire 6800 and 68000 series VHDL/Verilog can be found online, some even are designed as perfect clones (meaning, wire the IO to the pins, and it will match the IC it's replacing exactly except for the lower modern 3.3v IO voltages), but still yield higher CLK frequencies.)
You can also add any glue logic you like in code too from com ports to video card emulators & any other peripheral devices.

A modern mid range multicore SOC cpu can do this as well which may also grant you a top-governing OS like Linux, however, you will not get the real-time replication of processing and direct peripheral addressing like the FPGA.

[Berni]

Yep an FPGA can hold a whole card cage worth of 80s logic boards.

There are plenty of products out there that actually do this. For example some of the really old HP test equipment that stuck around long enough had the old boards revised to stuff all of the logic crap into a FPGA to simplify things while staying 100% firmware compatible with the old boards.

Big ancient mainframes that run banks also got modern upgrades until eventually they emulated the whole thing using a modern server in a rack.

However at some point you have to just say enough is enough and redesign the thing. If it is getting so bad that you want to clone a microprocessor in order to keep a product going then you clearly have lots of money to spend on the engineering hours to redesign it in modern electronics anyway. Heck something this old could probably be emulated in a ARM MCU if the original firmware is so hugely valuable.

[Andreas]

Originally Motorola :/

Did you already ask at Rochester?
https://www.rocelec.com/manufacturers

they often re-manufacture old chips with the original mask sets.

Motorola has been bought by NXP
https://www.rocelec.com/manufacturers/nxp

with best regards

Andreas

[chris_11]

IF we are looking into 6 figures, I'd rather spend the money on redesigning the entire thing.

There is a laundry lists of changes I've noted that should be done, should a redesign take place.

One example is there are actually 3 processors and a large micro on this board, all of which could be done by a single multi-core modern micro. Other circuitry is  compared to what is used today. Lots of thing where TONS of components are used for a function where we now have ICs to handle it all on-chip.

If you know the functionality of the board, it is a lot cheaper, faster and much safer, to redesign it complete in actual hardware. You own the design and code and you can make all needed changes anytime.
This is a no brainer. You would invest way more money/time in recreating a single processor IC from the 80s.

br
Christian

[coppice]

Originally Motorola :/

Did you already ask at Rochester?
https://www.rocelec.com/manufacturers

they often re-manufacture old chips with the original mask sets.

Motorola has been bought by NXP
https://www.rocelec.com/manufacturers/nxp

with best regards

Andreas

I don't think Rochester work with masks at all. They usually deal with things where the original process is unlikely to be available at any fab to be able to reuse an old mask set. They mostly deal with stocking die in controlled conditions, and doing packaging and final test on demand. I think even those steps can sometimes bite them, when the exact package or testing facilities turn out to no longer be available. This means their ability to provide large volumes is limited.

[Rat_Patrol]

IF we are looking into 6 figures, I'd rather spend the money on redesigning the entire thing.

There is a laundry lists of changes I've noted that should be done, should a redesign take place.

One example is there are actually 3 processors and a large micro on this board, all of which could be done by a single multi-core modern micro. Other circuitry is  compared to what is used today. Lots of thing where TONS of components are used for a function where we now have ICs to handle it all on-chip.

That's it?
You do realize even a 5-10$ FPGA could hold and run all 3 processors from the 80's simultaneously and then some...
You also realize that those processors you want are probably already coded at OpenCores.org with exact functionality.
The ram which those can 80's processors can access would probably fit inside the FPGA's memory as well, or if not, on 1 single external ram chip. (IE, the Motorola's entire 6800 and 68000 series VHDL/Verilog can be found online, some even are designed as perfect clones (meaning, wire the IO to the pins, and it will match the IC it's replacing exactly except for the lower modern 3.3v IO voltages), but still yield higher CLK frequencies.)
You can also add any glue logic you like in code too from com ports to video card emulators & any other peripheral devices.

A modern mid range multicore SOC cpu can do this as well which may also grant you a top-governing OS like Linux, however, you will not get the real-time replication of processing and direct peripheral addressing like the FPGA.

One of the chips I would REALLY like to replace is a 6800 series chip. FPGA's and emulators are new to me, would you mind pointing me in a good direction for working with this? I have the pcb layout file, so I can even change package (originally used a PLCC53 package).

[BrianHG]

This is a video tutorial for Lattice ICE40 FPGA, the entire process from A-Z, all done by yourself, actual coding, so, it's really long:  (He is using an nMigen compiler, not recommended by me)



These are already programmed in standard VHDL or Verilog.
Or, all the programming work already done for you: (Altera/Intel Quartus project written in VHDL)
https://opencores.org/projects/system6801

https://opencores.org/projects/system68

Updated newer 6803/6801 version to System Verilog:
https://opencores.org/projects/mc6803

In fact, just look here:
https://opencores.org/projects?expanded=Processor

Of, google:
https://www.google.com/search?newwindow=1&client=firefox-b-e&ei=KGVWX5XpMNagytMP2JqpiAk&q=MC6800+VHDL+core&oq=MC6800+VHDL+core&gs_lcp=CgZwc3ktYWIQDFAAWABg1ZgCaABwAHgAgAEAiAEAkgEAmAEAqgEHZ3dzLXdpeg&sclient=psy-ab&ved=0ahUKEwiVncfIwNfrAhVWkHIEHVhNCpEQ4dUDCAw

[Rat_Patrol]

If you would be so kind:

Is there a workflow similar to:

Download program from said 6800 series micro
Program new chip with said program
Adjust PCB layout to accomodate new chip and pinouts



I have PCB build files, but nothing for the micro: I can only grab the program off of it.

[Berni]

You posted the datasheet to the micro before:
https://www.nxp.com/docs/en/data-sheet/MC68HC05B6.pdf

THIS is what describes what the chip does.

So in order to have a chip that works just like it you need to implement this. Feel free to outsource this step to someone but again expect them to want a 5 figure paycheck to do it. That's not because they are greedy, but because it takes a lot of work to implement all this. Hence why cloning a chip in a FPGA like this is rarely done commercially unless they really is no easier way around it.

This is why everyone is telling you to just toss the whole PCB and redesign it to use a modern MCU. Its way WAY less work.

[ali_asadzadeh]

It's nice to have a friend in china, specially someone like  blueskull