New GTIA chips!

[atarian63]

I have an industrial blueprint scanner, I'll have to check the make and model, would that help? Looks like 36" width.

If it does I'll donate.

[frogstar_robot]

We just have make sure Bender doesn't get a hold of them before Curt gets them scanned. He might get a tad excited.

[Godzilla]

that is just super awesome. I can't wait to see (and buy,) what comes out of this

[Curt Vendel]

I'm speaking with 2 fab's right now to see if something can be done, what is good is the design is so "BIG" and "SIMPLE" in todays terms, the cost is not as big a factor because the processes are light years beyond what they were in the 90's.

 

 

 

Curt

 

Congratulations on an awesome work, Curt !

(I remember seeing a post from you on DeepChip several years ago, asking for help on how to read those tapes )

 

But I want to know

You did find a fab which is willing to manufacture custom ICs without charging an arm and a leg ? Do they have a specific process, or a way to create "cheap" masks ?

I've always heard that chip manufacturing had extremely high fixed costs (altough I guess that having a very coarse geometry compared to today's chips may lower the cost).

[+pboland]

I just wanted to say thanks Curt to you and your staff. I love the FB1 & FB2 and I look forward to what ever comes of these chip designs. What you guys have up your sleeve I'm sure is going to be great.

[TROGDOR]

Very impressive Curt, great work!

 

I have a few questions about the designs:

 

- What technology node sizes were the chips designed for? (I'm guessing around 2-micron.)

- How many layers are the designs?

- Have you been able to compute any transistor counts?

- What are the file sizes of the gds?

- Is the logic full custom? I can see what appear to be two memories in the center of the GTIA layout, unless those are gate arrays.

- Who owns the rights to the gds?

 

With the full schematics and layouts, it should be possible to perfectly emulate these chips down to the most minute details.

[dmlloyd]

I have suggested this in the past, but once these images are scanned, it would be great to use something like the Google Maps API to create a "map" of the scans which can be navigated and annotated by the community.

[AEX]

 

Put another way, don't you think a 6 foot by 6 foot print is going to be a bit hard to read at 16"x24"? There *was* a reason they printed it so large in the first place.

 

 

 

Aha! My bad, I assumed (which one should never do!) that it was a 6x6 inch print he was talking about

 

Of course, once it has been scanned and stitched nicely at at least 300dpi, you could make some nice copies of the image, a poster a lot of people would like to hang on their wall for sure.

 

Karl

[PacManPlus]

 

Wow - those bring me back to my Unisys days. We had a Unisys 5000/55 that had an external tape backup that used those types of tapes. I was forever converting data between those and the QIC150 cartridge tapes. We used 'dd' a *lot* to convert block sizes in order to read tapes from our clients that we didn't make... Using 'no rewind', and discarding header data...

God, I miss that. It was all so much simpler then.

 

As a side note, I found that many people seemed to use 'tar' a lot to back up, but I tend to like 'cpio' much better. To each his own!

 

 

Great Work Curt!!!!

 

Bob

[Retro Rogue]

As good as Amy could have been, I don't see a lot of point trying to recreate it.

Supposedly, Atari themselves couldn't get it working properly so a modern-day scenario might work out similarly.

 

From the data sheet, my guess as to why Atari couldn't make it work has got to be either lack of manpower or just plain old ineptitude. As far as I can see, it's a pretty simple device.

 

That was Atari Corp. that couldn't get it to work right in what they wanted, not Atari Inc. Atari Inc. had already demonstrated it publicly, and had it planned for use in several different projects. Don't forget, Alan Kay's team had already moved on, as had much of the other advanced engineering groups. Jack had to go outside to get work done with it.

Edited December 15, 2009 by wgungfu

[Retro Rogue]

 

 

Got any pictures of the box you store them in? I want to see that as well.

[supercat]

Are these chips NMOS? Can modern chip fabs and processes handle that acceptably? Most chips today use static CMOS logic and are designed to minimize resistance and capacitance of internal nodes. The 6502 and TIA both use dynamic NMOS logic, and I would expect the other chips would as well. I would be very happy to see a successful build of chips based upon the old designs, but I would worry about whether they'd actually work the same as the old ones. The TIA has at least two race conditions I know about where the results aren't the same on every chip. There may be some others as well.

[Rybags]

Not trying to piss on any parades here, but what is the logic in getting GTIA duplicated anyway ahead of the other custom chips?

 

The demand for Pokey would be much higher among existing users for Stereo upgrades, not to mention arcade and 7800 owners.

Plus, GTIA has virtually been duplicated anyway with VBXE (at least the video part, which is probably 95%+ of the complexity).

[+Stephen]

Not trying to piss on any parades here, but what is the logic in getting GTIA duplicated anyway ahead of the other custom chips?

 

The demand for Pokey would be much higher among existing users for Stereo upgrades, not to mention arcade and 7800 owners.

Plus, GTIA has virtually been duplicated anyway with VBXE (at least the video part, which is probably 95%+ of the complexity).

Rybags, this isn't just about the GTIA. It's also about ANTIC, MARIA, and several unreleased chips. Hopefully the entire line of chips can at least be preserved. Also, these don't need to be reproduced, but we can have comfort in knowing the original chip fabs AND original schematics exist.

 

Stephen Anderson

[Curt Vendel]

Thats been the biggest problem, which may only allow the later CMOS designs to be done as NMOS/HMOS just aren't widely done anymore, but I'm being referred around to a few people who still conquer in the old ways of magic

 

 

Are these chips NMOS? Can modern chip fabs and processes handle that acceptably? Most chips today use static CMOS logic and are designed to minimize resistance and capacitance of internal nodes. The 6502 and TIA both use dynamic NMOS logic, and I would expect the other chips would as well. I would be very happy to see a successful build of chips based upon the old designs, but I would worry about whether they'd actually work the same as the old ones. The TIA has at least two race conditions I know about where the results aren't the same on every chip. There may be some others as well.

[shirsch]

I have finished recovering the GDS stream and reconstructed the GTIA, MARIA, STELLA and several other proprietary Atari chips, now to see if the GDS can be translated to VHDL or if I can find a fab that still does hmos/nmos the original Atari chips can be reproduced at minimum - in SMD sizes...

 

Curt

 

Curt, Do you have anyone who can run that through an LVS tool to derive the netlist? With a proper netlist, synthesis into an FPGA should be relatively straightforward to do. I doubt any existing fab could build usable masks from that layout - too many technology changes over the past 25-30 years.

Edited December 17, 2009 by shirsch

[carmel_andrews]

Now that Curt has done GTIA...Are the likes of ANTIC and Pokey as well as Freddie and PIA (or is that a standard chip) to follow, curt (or are you still hunting through your archives)

 

Just out of curiosity though, as people here are thinking in terms of reproducing such hardware, would it be cheaper/easier to reproduce the specialist/custom chips as is (i.e as separate chips) or as curt said as a System on a chip scenario (i.e All atari's custom chips on a FPGA type device)

[Curt Vendel]

I already have the tools and I think the netlist is going to be extractable, gotta take a few days off to focus on a few things but I'll be back to this full steam again right after Christmas and New Years.

 

 

 

Curt

 

I have finished recovering the GDS stream and reconstructed the GTIA, MARIA, STELLA and several other proprietary Atari chips, now to see if the GDS can be translated to VHDL or if I can find a fab that still does hmos/nmos the original Atari chips can be reproduced at minimum - in SMD sizes...

 

Curt

 

Curt, Do you have anyone who can run that through an LVS tool to derive the netlist? With a proper netlist, synthesis into an FPGA should be relatively straightforward to do. I doubt any existing fab could build usable masks from that layout - too many technology changes over the past 25-30 years.

[ClausB]

Sure would be a good time to add some minor improvements to the GTIA. For example, make the color-lum registers 8 bits instead of 7, i.e. use all 4 luma bits to get 256 colors. Add 7 more color-lum registers to change the 9-color GTIA mode into a 16-color mode.

 

More ambitious would be to double or triple the ANTIC and GTIA bandwidth like we're doing with the LEM project.

Edited December 17, 2009 by ClausB

[ivop]

I already have the tools and I think the netlist is going to be extractable, gotta take a few days off to focus on a few things but I'll be back to this full steam again right after Christmas and New Years.

 

I was wondering, will you be sharing those GDS files somewhere online?

[sl0re]

Not trying to piss on any parades here, but what is the logic in getting GTIA duplicated anyway ahead of the other custom chips?

 

The demand for Pokey would be much higher among existing users for Stereo upgrades, not to mention arcade and 7800 owners.

Plus, GTIA has virtually been duplicated anyway with VBXE (at least the video part, which is probably 95%+ of the complexity).

 

 

One other thing, there seem to be more Pokeys out there on the market. Including quad pokeys. I think because they were used in more than just the computers.

[Rybags]

Arcades. Not sure on total numbers, but I'd guess probably more than half a million - given that many machines used 2 of them.

 

7800 carts - no idea of numbers there but I'd guess significantly less.

 

Plus, Atari always seemed to keep enormous inventories of certain things. Wouldn't surprise me if they just had a million made at a time.

[jkonstan]

Curt,

 

Very nice work ...

Public Atari2600 FPGA implementations have always lacked a complete TIA in Verilog or VHDL where the partially completed FPGA TIA in Verilog or VHDL had lots of issues. Thus, if you could just translate the TIA back into VHDL this would be of great help to those of us who like to experiment with FPGAs and old video game architectures such as the Atari2600.

 

This is the best news yet!! We can finally have the A8 on a chip. How about the mini-mini-mig

 

Stephen Anderson

 

Minimig is an FPGA clone.

 

Reproducing actual chips - this would essentially allow new chips to be run, I have a foundry that can run some samples of the GTIA chip in any package size, so doing a 44 pin tqfp type smt chip:

 

http://www.pulsarprofx.com/pcbfx/main_site/pages/direct_etch/index_support/hoding_smt_300.jpg

 

To move the gds binary format into vhdl may be far more trouble then its worth and is really a backwards approach - you tend to synthesize your vhdl code so that a fab can produce asics based on it, so here is the gds code which a fab can directly take, so it is better to go this route, Christmas and New Years are fast approaching, so I'll look more deeply into this after New Years once I get things rolling on the 1200XL plastics.

 

Curt

[ClausB]

Sure would be a good time to add some minor improvements to the GTIA. For example, make the color-lum registers 8 bits instead of 7, i.e. use all 4 luma bits to get 256 colors. Add 7 more color-lum registers to change the 9-color GTIA mode into a 16-color mode.

Curt, are such enhancements out of the question? We could call it the VTIA!

 

Or are you looking to faithfully reproduce the original?

[Rybags]

One possible enhancement could be RGB, like that done with triple GTIAs.

 

Have a bitsetting that forces all registers to output luma only, then have extra colour registers at relevant addresses +$20 and +$40 that can hold the G and B values, and have 2 extra pins that output those components.