Amiga.org
Amiga computer related discussion => Amiga Hardware Issues and discussion => Topic started by: freqmax on December 15, 2011, 03:02:24 AM
-
Anyone tried to scan (http://www.chips.5u.com/idxhck.html) the original Amiga chips in any way to recreate the logic matrix?
NMOS, but what nm process and die size has been used for OCS Agnus, Denise, Paula etc?
(This also means that even broken chips are worth keeping!)
-
Probably more trouble than it's worth.
Jeri Ellsworth has the schematics for the OCS chip set. Would be nice if they could be made available to the general public.
-
You'd probably have a hard time finding someone still manufacturing in this process... The inner workings are pretty well known and have been replicated in UAE and Minimig.
-
These guys have done a few chips already (6502, 6800, 68000, SID, etc...), they have decapped, scanned and then recorded the logic layout... so they would be the first place to call :
http://visual6502.org/
-edit- just noticed that these guys are working out the entire logic for the VCS 2600... So given how that machine is related the the Amiga1000 they might be keen to do the Amiga :)
If on the other hand, you happen to have a ton of money, Chipworks Inc will do a private scan of the chips for you and give you everything you need ;)
-
Looks like the visual6502 guys already have Agnus, Denise, Paula and Gary in their chip collection... I guess we just need to harass them into scanning them :)
-
They are scheduled to be scanned.
I am far more interested in the continuing polygonization (is that a word?) of the 68K - which I am working on.
Getting the exact internal microcode (note, it is split into two tables to compress it) will be very interesting and reduce the size of the softcores considerably.
/MikeJ
-
They are scheduled to be scanned.
I am far more interested in the continuing polygonization (is that a word?) of the 68K - which I am working on.
Getting the exact internal microcode (note, it is split into two tables to compress it) will be very interesting and reduce the size of the softcores considerably.
/MikeJ
No idea when they will do them though, I can't be much help to them so I haven't made contact with them.
I also am keen to see the "polygonisation" of these chips not for recreation purposes (since I think MiniMIG and CloneA are probably a better approach), simply so I can see the choices the original design team made all those years ago :)
On the other hand some FPGAs with the original net lists on them would be super awesome... ;)
-
Getting the exact internal microcode (note, it is split into two tables to compress it) will be very interesting and reduce the size of the softcores considerably.
I thought we had the microcode already, I'll ask around in case it's not public yet.
You can find the microcode in the 68000 patent, but we know it's got some wrong values.
-
Let me know if you have it.
I have the details from the patents, but it's not terribly useful.
With the die scan we can see how the ops are decoded. It's a very neat design, all of the "strangeness" comes out as a result of implementation.
/MikeJ
-
Leading Edge Retro, I like it.
-
It seems like Individual Computers did some detailed chip analysis at one time as well. If I am remembering correctly, various historical Clone-A documentation seemed to elude to this. It would seem logical that the data they collected was used to develop the various add-on cards like Indivision and the like, but do not quote me on any of this as my memory is more than a bit foggy on the subject.
-
What I want to know is how much it would cost to do OCS/ECS/AGA in ASICs using the original layouts?
-
What I want to know is how much it would cost to do OCS/ECS/AGA in ASICs using the original layouts?
Jeri told me that it would be in the neighborhood of 100k if I remember correctly. I had 10k so I asked her if it would be enough. Unfortunately it was like 10 times that amount.
-
This begs the question if you committed to producing at least a million units minimum how much it would cost per unit to produce a working 100% compatible classic Amiga motherboard of A500 and A1200 specs?
(no case/keyboard/psu/floppy etc)
-
You'd probably have a hard time finding someone still manufacturing in this process... The inner workings are pretty well known and have been replicated in UAE and Minimig.
There's absolutely no need for anyone to manufacture in the original process. The process size matters for the scanning procedure.
What is known about the inner workings of these chips is a reverse engineering work process, it's very good. But it's not perfect.
This begs the question if you committed to producing at least a million units minimum how much it would cost per unit to produce a working 100% compatible classic Amiga motherboard of A500 and A1200 specs?
The PCB has already been done. So for the chips you seem to need 100k - 200k USD in addition each chip maybe cost 1 USD per chip?
Btw, have a look att EFF DES cracker (https://en.wikipedia.org/wiki/EFF_DES_cracker).
I am far more interested in the continuing polygonization (is that a word?) of the 68K - which I am working on.
How does the polygonization work, and is it reliable in determining the logic function?
Being able to use nitric acid, photograph (using microscope?), and then use algorithms to transform pictures into HDL-code seems really nice.
Is the 68k compatibility a huge issue?, 68000 + 68020?, maybe more critical than the custom chips?
-
Let me know if you have it.
I have the details from the patents, but it's not terribly useful.
With the die scan we can see how the ops are decoded. It's a very neat design, all of the "strangeness" comes out as a result of implementation.
/MikeJ
All I've dug out of my email archives so far is an email from Nick Tredennick.
> The best description of the internal workings of the MC68000 can be found in Chapter 11 of Francois Anceau's
> textbook The Architecture of Microprocessors. It does, however, contain a few errors.
The guy we use for decapping is MIA (we don't pay him anywhere near as much as he gets paid for real work, so it's pretty understandable). So it might still be sitting in his queue all these years later.
You probably don't want to implement the 68000 microcode though as the ALU was 16 bit, so performance wouldn't be that great. Cycle accuracy of the CPU on the Amiga was never that necessary, however for Atari ST emulation you really need it if you want to get all demos working.
If you just want it for inspiration then the book and patent are more likely to be useful. Otherwise it might be better to get a 68020 decapped.
-
The PCB has already been done. So for the chips you seem to need 100k - 200k USD in addition each chip maybe cost 1 USD per chip?
Btw, have a look att EFF DES cracker (https://en.wikipedia.org/wiki/EFF_DES_cracker).
What I meant was how much a single A600 (or A1200) compatible drop-in compatible motherboard with Agnus/Denise/Paula would each cost to manufacture today from scratch inc tooling up for mass production and all R&D. Assume you commit to producing one million units and exclude CPU costs. Go to China, give them the specs for everything and add R&D cost to invoice they charge for producing a million units. £100? £200? less?
It's far from easy to do, the first Amiga Technologies A1200 motherboard that rolled off the production line run at the factory in France was defective in the mid 90s!
-
What I meant was how much a single A600 (or A1200) compatible drop-in compatible motherboard with Agnus/Denise/Paula would each cost to manufacture today from scratch inc tooling up for mass production and all R&D. Assume you commit to producing one million units and exclude CPU costs. Go to China, give them the specs for everything and add R&D cost to invoice they charge for producing a million units. £100? £200? less?
It's far from easy to do, the first Amiga Technologies A1200 motherboard that rolled off the production line run at the factory in France was defective in the mid 90s!
You're making too many assumptions to come up with a per unit cost of a re-production. Do you really expect someone on this forum to research all the variables and give you a discrete price for a product that has no demand in the real world? Or are you just asking a hypothetical question?
-
This begs the question if you committed to producing at least a million units minimum how much it would cost per unit to produce a working 100% compatible classic Amiga motherboard of A500 and A1200 specs?
It doesn't beg the question.
It raises the question. http://en.wikipedia.org/wiki/Begging_the_question
If you wanted to make it as cheap as possible, you want as few components as possible and a board just big enough for the io connectors.
Something like an fpgaarcade in a different form factor.
To make something that looks like an a500 is going to be pretty expensive because of all the sockets and chips. It was designed at a time when commodore made all of their own stuff, so it fit in with what they could do as cheaply as possible (I doubt for instance that the Amiga design included 2x MOS CIA's before commodore bought Amiga Inc).
-
In the real world, a million units is way larger than the actual market size, and ten thousand is more realistic (maybe still too large). Seeing how producing a relatively large quantity of units is not an option, those machines are going to turn out to be pretty expensive :( Makes you wish you were stinkin' rich :(
You're making too many assumptions to come up with a per unit cost of a re-production. Do you really expect someone on this forum to research all the variables and give you a discrete price for a product that has no demand in the real world? Or are you just asking a hypothetical question?
It should be possible to come up with a reasonable estimate ;)
-
How many people are there willing to pay for working Amiga hardware really?
I know it's hard to find hard numbers. But a educated guess based on statistical data?
-
In the real world, a million units is way larger than the actual market size, and ten thousand is more realistic (maybe still too large). Seeing how producing a relatively large quantity of units is not an option, those machines are going to turn out to be pretty expensive :( Makes you wish you were stinkin' rich :(
It should be possible to come up with a reasonable estimate ;)
Then by all means, be my guest. Start your research and give all of us a reasonable estimate. But I think the term "reasonable" is too subjective and any estimate will be just a ridiculous figure that won't be anywhere close to reality.
Your wealth would be better served by investing in a Replay board or a Minimig......or better yet, if you're so enamored with classic Amigas, just go out and buy some. They're still pretty cheap on eBay and other places.
-
Chips that need remanufacturing on the motherboard (http://www.commodore.ca/history/company/mos/Amiga500_board.jpg) is M68k, ROM, Agnus, Denise, Paula, Gary, CIA. The rest is quite standard chips. So lets assume the circuit board (http://www.illuwatar.se/project_pages/gba1000/gba1000.htm) with components (http://www.amiga.org/forums/archive/index.php/t-36440.html) cost 200 USD. Each mentioned chip will likely require 100k USD each. So the first amiga is likely to cost 700 200 USD. Divided by 10 000 sold units gives a unit price of 70 USD. Double chip cost and 1000 sold units gives a 1400 USD/unit price. etc..
So price estimate:
Unit price USD = (200 + 7*100 000) / (units)
-
So price estimate:
Unit price USD = (200 + 7*100 000) / (units)
And what about tax and wages etc. etc. etc?
There is absolutely no point doing it that way these days.
It'd be far, far cheaper to cost reduce the minimig and do a bigger run of these.
Something more like Jeri's C64 in a joystick is much more likely to work and most importantly - actually sell in good numbers.
-
And what about tax and wages etc. etc. etc?
Where is the tax and wages for Minimig?
So it would be impossible for a more ambitious project to do the same at circuit level?
There is absolutely no point doing it that way these days.
Ask @Digiman for the point of the exercise ;)
It'd be far, far cheaper to cost reduce the minimig and do a bigger run of these.
I'd rather cost reduce FPGA Arcade :P
Something more like Jeri's C64 in a joystick is much more likely to work and most importantly - actually sell in good numbers.
These play joysticks are nice but quite limited.
-
Jeez I just wanted an idea of cost today using every possible advance in cost reducing production/technology, as Sony and Microsoft frequently employ, to make a board with surface mounted AGA/OCS reproduction chipset for absolute minimum price.
I never said it was a viable business, it isn't, and 1 million unit is a realistic quantity in general for making a go of a new mass market piece of electronics. I never said the market potential = 1 million. 10k units doesn't give you true mass market product costs in the world of bespoke electronics.
You can run WinUAE to A4000/030 speeds on a complete AMD 1.6GHZ Mini-ITX motherboard today anyway for £90 retail, that's cheaper than Minimig, and both are simulating an Amiga.
To compare, if you ordered a 1 million minimig Chinese production run tomorrow how much could you get the price down to? Assume Peter Jones from Dragon's Den fronted the required cash advance regardless and forget market size. Remember this is about finding out absolute minimum production costs nothing more.
(This is the first thing any investor needs to know anyway, minimum cost is finite....market size is down to talent of marketing depts)
-
Regarding the real purpose of this thread I found a photograph of the Motorola MC68020RC16 CPU die from a blog (http://diephotos.blogspot.com/):
(http://photos1.blogger.com/x/blogger2/7290/770229950760068/1600/109343/gse_multipart67095.jpg)
And from cpu-world.com (http://www.cpu-world.com/CPUs/68000/index.html) Motorla 68000 die can be found (http://www.cpu-world.com/CPUs/68000/die/L_Motorola-MC68000L8.jpg). Neither of above photos are enough for decoding thoe.
FujiFilm S5100 (http://www.dpreview.com/reviews/fujifilms5100/) (4.0 Mpix) with reversed Canon FD 50mm f1.4 lens on homemade macro (https://en.wikipedia.org/wiki/Macrophotography) stage, auto-focus mode, manual exposure, 2-second shutter delay
FujiFilm S6000fd (http://www.dpreview.com/reviews/fujifilms6000fd/) (6.3 MPix) with macrolens Raynox M-250 (http://www.raynox.co.jp/english/dcr/dcr250/indexdcr250eg.htm), camera settings same as above.
Some hints (http://micro.magnet.fsu.edu/primer/index.html) on photography (http://micro.magnet.fsu.edu/chipshots/index.html) and examples of various really microscopic objects (http://micro.magnet.fsu.edu/micro/gallery.html).
-
@Digiman
I think the point is, if you were to make an "Amiga Classic" compatible machine now, due to the complexity of the Amiga (when compared with machines of a similar age), to fully take advantage of technical advances and keep costs down... You would go the MiniMIG/Replay route rather than try and clone the original machine "chip for chip" (so to speak).
-
Yes, it'd be a complete waste to reproduce everything 1:1 on chip level. Even with relative crude (today's) technology it'd be no problem to integrate everything into a single die - just like what C= started with Hombre. Whether you'd use an FPGA or ASIC approach is a matter of investment and volume.
-
@Digiman
I think the point is, if you were to make an "Amiga Classic" compatible machine now, due to the complexity of the Amiga (when compared with machines of a similar age), to fully take advantage of technical advances and keep costs down... You would go the MiniMIG/Replay route rather than try and clone the original machine "chip for chip" (so to speak).
I agree, and chip reproduction is financially only worth it for spares/repairs industry like the company in USA reproducing DeLoreans. They could build a new one from scratch but financially that's a dead end for same reasons.
I was interested in 2 things for comparison ie min technically possible cost for Minimig vs minimum possible cost for a 100% compatible non FPGA method. I know how much an x86 C-USA style x86 emulation based Amiga 600/1200 would cost.....very cheap.
-
What I meant was how much a single A600 (or A1200) compatible drop-in compatible motherboard with Agnus/Denise/Paula would each cost to manufacture today from scratch inc tooling up for mass production and all R&D. Assume you commit to producing one million units and exclude CPU costs. Go to China, give them the specs for everything and add R&D cost to invoice they charge for producing a million units. £100? £200? less?
It's far from easy to do, the first Amiga Technologies A1200 motherboard that rolled off the production line run at the factory in France was defective in the mid 90s!
I expect an entire A1200 could be implemented in a single ASIC on a cheap, but still reasonably modern process (e.g., 180nm). You could probably add on digital video and audio outputs, SATA, USB, etc as well whilst you are at it (so implement a typical expanded A1200 of 2011).
But with mask costs in the hundreds of thousands, you would truly want to be making tens to hundreds of thousands of the chips to make it worthwhile. But after that you would just need a sneaky "amiga in a joystick" product that accidentally left exposed pads on the PCB for all expansions, and you could get Amigas into many many homes again.
-
This is probably a dumb question- but if one had the capital to reverse-engineer, wouldnt it be much cheaper to legitimately buy the original engineering data for an abandoned and obsolete technology?
I don't follow very much anymore, so forgive me if it seems I should know the answer.
(I mean for the MOS ICs, not the 68k, which is obviously still valuable)
-
This is probably a dumb question- but if one had the capital to reverse-engineer, wouldnt it be much cheaper to legitimately buy the original engineering data for an abandoned and obsolete technology?
I don't follow very much anymore, so forgive me if it seems I should know the answer.
(I mean for the MOS ICs, not the 68k, which is obviously still valuable)
I'm sure I've read several times the original chipset designs have been long-lost. Even Commodore had to reverse engineer them to make ECS/AGA if rumours are to be believed.
-
I expect an entire A1200 could be implemented in a single ASIC on a cheap, but still reasonably modern process (e.g., 180nm). You could probably add on digital video and audio outputs, SATA, USB, etc as well whilst you are at it (so implement a typical expanded A1200 of 2011).
But with mask costs in the hundreds of thousands, you would truly want to be making tens to hundreds of thousands of the chips to make it worthwhile. But after that you would just need a sneaky "amiga in a joystick" product that accidentally left exposed pads on the PCB for all expansions, and you could get Amigas into many many homes again.
And there is the issue of software... The floppy disk isn't a sensible distribution medium any more... And the whole thing just isn't practical :)
-
This is probably a dumb question- but if one had the capital to reverse-engineer, wouldnt it be much cheaper to legitimately buy the original engineering data for an abandoned and obsolete technology?
Chips were designed in a completely different way back then so even if you could find them they'd be completely useless.