Musings on the Original Atari 8 Bit PCB Designs

[Overange]

Whilst doing fault finding on an 800XL Main board I noticed some interesting design features that are not apparent on the later XE range of systems.

 

There will be more questions but all in good time 

 

Firstly, The Buffered O2 Signal being Flanked by ground traces.

 

GTIA showing Pin 30 B02, aswell as PAL and Pin 28 OSC Traces being flanked with Ground Traces, but for now the BO2 Traces.

Address bus showing B02

Address Bus Trace show BO2 Flanked with GND Traces, before it splits to the right to Main ICs in the following order

 

POKEY

PIA

ANTIC

GTIA

 

The Split trace travelling to the back of the board goes to near inline with the Basic ROM before it Splits again, left it connects to the Cartridge Connector, and upward to the PBI bus Again Both Still Flanked by a Gnd Traces.

 

 

The XE range appears to of done away with these Flanking GND Traces, as shown on the 65XE remake version.

 

Designs by  @mytek  including the 1088XLD, 1088XEL and 576NUC systems, also appear to do away with the flanking GND Traces.

 

Having worked for a company that designed and built its own Signal Conditioning and Processing Systems, I do not recall ever seeing such a technique used??

Bearing in mind this would of been during the 90s.

 

So did they really serve a purpose, are they needed, or was it to try and mitigate noise in the original design with components available at that time?

And with the modern redesigned variants, with out these GND traces, has it improved the reliability of the BO2 Signals.

 

I hope @Piotr D. Kaczorowskican talk about his findings on the subject of the BO2 Signal, Noise and Parasitic Capacitance on the A8 Boards and maybe explain possibly what Atari could of been trying to achieve when they drew in these GND Traces?  

 

Thanks

 

 

[Piotr D. Kaczorowski]

My recommendation is that you take a look at the 600XL board. Everything is best done there. After that, it only gets worse. If you perform a loop probe test using an oscilloscope and its probe with the connected ground at its tip, you will see almost like a radar what the noise level is on the main board. On the 600XL, it's usually around 15mV on the ground plane. On subsequent computers, this noise is much higher. Also, look at the small holes around the ground plane. These are not decorative holes (which are not on the 130XE boards), but electromagnetic wave interference breakers.

 

As for the BO2 line (Buffered Phase 2) itself, ideally on a new board, you would want to attach its twin copies to subsequent circuits (as is possible with O2/Fixer). This matters if you are using original NMOS technology chips. If you start using replacements: PIA -> WDC PIA NMOS, GTIA -> Sophia 2, Pokey -> PokeyMax, etc... the BO2 line is relieved.

 

By the way.. If you take a book about the 6502, it indicates that a maximum of 1 TTL and a maximum of 130pF capacity should be connected to O2. Therefore, for example, if you connect to the O2 signal (pin #39 on Sally) with an oscilloscope probe with a 1x multiplier (which has 135pF on Keysight), the expansions become unstable. SIDE 3.x stops working altogether.

 

In my opinion, it's worth inviting @candle , @Simius, and maybe even @electron can be tempted to join the discussion. These guys have significantly more practical experience. I am refreshing my knowledge after 20 years since finishing my studies (Faculty of Electronics and Information Technology of the Warsaw University of Technology).

[Piotr D. Kaczorowski]

@Overange,

I hope you know that if you want to create a truly professional board from scratch, you're better off with Altium Designer rather than KiCad...

[+mytek]

54 minutes ago, Overange said:

Designs by  @mytek  including the 1088XLD, 1088XEL and 576NUC systems, also appear to do away with the flanking GND Traces.

I used a 4-layer board with power and GND planes in the middle layers. Basically the very best you can ask for when it comes to minimizing noise and droop in the power bus, as well as providing full isolation between the top and bottom traces

 

Now days the added cost in going 4-layer vs. 2-layer is minimal, and in my opinion well worth it for motherboard designs. It also makes it sooooo much easier to do a PCB layout when you don't have to route the power and GND paths.

[reifsnyderb]

18 minutes ago, mytek said:

Now days the added cost in going 4-layer vs. 2-layer is minimal, and in my opinion well worth it for motherboard designs.

Unless you exceed 100mm on a side, then JLCPCB adds a $35.00 "engineering fee" to your order.   😞

 

 

18 minutes ago, mytek said:

It also makes it sooooo much easier to do a PCB layout when you don't have to route the power and GND paths.

I completely agree!

 

Sometimes the options are limited.  When I did the Happy 7.1 recreation, I had to go to 4 layers to squeeze everything into a tight area.  Some of the signal lines had to be on the 3rd layer.

 

When I do a 4 layer board, I have the ground plane on the second layer and the power on the third layer.  However, I route individual power paths, on the 3rd layer, as there are times when it's just a lot easier to have a 3rd layer for some of the signal lines.

 

 

[reifsnyderb]

59 minutes ago, Piotr D. Kaczorowski said:

@Overange,

I hope you know that if you want to create a truly professional board from scratch, you're better off with Altium Designer rather than KiCad...

What's the advantage of Altium?

[Piotr D. Kaczorowski]

7 minutes ago, reifsnyderb said:

What's the advantage of Altium?

I am not yet an Altium user, unfortunately, but it has been recommended to me because of its significantly more advanced features, such as easier multi-layer board design, HDI, and various simulations. These things are probably achievable in KiCad as well, but they are more complicated or less refined.

[+mytek]

18 minutes ago, reifsnyderb said:

Unless you exceed 100mm on a side, then JLCPCB adds a $35.00 "engineering fee" to your order.   😞

Pricey in the prototype stage, but by the time you get to production you'll be minimally ordering 50 boards at a time which greatly dilutes the Engineering fee per board.

[Overange]

1 hour ago, Piotr D. Kaczorowski said:

My recommendation is that you take a look at the 600XL board. Everything is best done there. After that, it only gets worse. If you perform a loop probe test using an oscilloscope and its probe with the connected ground at its tip, you will see almost like a radar what the noise level is on the main board. On the 600XL, it's usually around 15mV on the ground plane. On subsequent computers, this noise is much higher.

Ahh I see how you achieved this now, back in the day, when we had to ensure our systems were EMC Compliant, We built our own probe to perform EMC Leakage tests, it was basically a Biro Pen tube, one end had a BNC the other had a loop of copper and it was connected into a scope, these tests were done in house before it was taken for final approval in an Anechoic Chamber ( If you ever been in one, its the weirdest sound experience you will ever have )

 

But as you say, on subsequent computers the noise is higher, but the XE does not use the flanking GND traces, however on the XE, the route BO2 takes is a lot shorter compared to the 800XL, as well as not being part of the mass address and control lines travelling to the rear of the Board, so maybe it was deemed not needed on the XE systems?

 

1 hour ago, Piotr D. Kaczorowski said:

Also, look at the small holes around the ground plane. These are not decorative holes (which are not on the 130XE boards), but electromagnetic wave interference breakers.

I remember you mentioning this on FB, I can't fully remember your findings, but did you find by drilling more clearance holes made a difference?

 

1 hour ago, Piotr D. Kaczorowski said:

As for the BO2 line (Buffered Phase 2) itself, ideally on a new board, you would want to attach its twin copies to subsequent circuits (as is possible with O2/Fixer). This matters if you are using original NMOS technology chips. If you start using replacements: PIA -> WDC PIA NMOS, GTIA -> Sophia 2, Pokey -> PokeyMax, etc... the BO2 line is relieved.

 

Is that one of reasons why you choose to use individual discrete logic devices on the O2/Fixer, one for each logic element to not only mimic the original 74LS08 IC but adding additional single logic devices to provide the extra Buffered outputs for modern add ons?

 

I see you point on a new main board design as you could add extra O2 buffers as additional hook-up points.

 

However,  lets say the the O2Fixer did not exist, on an existing 600XL board, if you was to fit the 64K expansion, could you not rewire the now unused gate in the 74LS08 as another BO2 source, therefore creating a bodge work around?

But that's only a 'What If?' TBH the price of the O2Fixer is worth every penny  and no point bodging  .

 

1 hour ago, Piotr D. Kaczorowski said:

 

By the way.. If you take a book about the 6502, it indicates that a maximum of 1 TTL and a maximum of 130pF capacity should be connected to O2. Therefore, for example, if you connect to the O2 signal (pin #39 on Sally) with an oscilloscope probe with a 1x multiplier (which has 135pF on Keysight), the expansions become unstable. SIDE 3.x stops working altogether.

I have experienced similar effects to this before when trying to probe the O2 signals.

1 hour ago, Piotr D. Kaczorowski said:

 

In my opinion, it's worth inviting @candle , @Simius, and maybe even @electron can be tempted to join the discussion. These guys have significantly more practical experience. I am refreshing my knowledge after 20 years since finishing my studies (Faculty of Electronics and Information Technology of the Warsaw University of Technology).

Pretty much same here 20+ years ago Plymouth Uni. but via the apprenticeship route, post graduation I was promoted away from the Electronics, Coding and Development side of the business into Technical Support and Onsite Project management, then later moved into different industry's. But deep down, I enjoyed the dev side more, but back then, being a Geek was not cool, however being young and single, travelling to different places... well you can imagine 😜 

Oh and the RF Probe talk gave me a huge flash back  

 

 

[Overange]

3 hours ago, Piotr D. Kaczorowski said:

@Overange,

I hope you know that if you want to create a truly professional board from scratch, you're better off with Altium Designer rather than KiCad...

I see your point, but for the general hobbyist Kicad does the basics  

[Piotr D. Kaczorowski]

@Overange,

 

Quote

But as you say, on subsequent computers the noise is higher, but the XE does not use the flanking GND traces, however on the XE, the route BO2 takes is a lot shorter compared to the 800XL, as well as not being part of the mass address and control lines travelling to the rear of the Board, so maybe it was deemed not needed on the XE systems?

 

I think it was decided on a similar basis that faulty GTIA and keyboards without springs could be put into the XE. For me, it's an obvious cost-cutting measure.

 

It's also not the case that the electrolytic capacitors in Atari are eternal, while for example in the C64 they need to be replaced. Electrolytic capacitors have a typical stable lifespan of 10-15 years. They also dry out and lose their parameters. I replace the capacitors in my Ataris. I replace 470uF with 1000uF (as in XEGS), choosing a polymer super low ESR.

Here I'll add that on XE boards, I always replace the DRAM memory with SRAM, and when installing VBXE, I remove Freddie so that the 3.5MHz signal goes directly into the main board from VBXE. As I wrote about the noise level on the ground plane, in the case of processors it's usually around 30mV, but Freddie generates 100mV. See what an increase that is compared to the 15mV from the 600XL.

 

Quote

I remember you mentioning this on FB, I can't fully remember your findings, but did you find by drilling more clearance holes made a difference?

 

Unfortunately, adding your own holes every 0.4" doesn't yield the same results as the factory implementation. Despite that, I believe they should be introduced. Notice that this is present in all models except for the 130XE boards. The XEGS boards also have these holes, and the noise levels there are significantly lower.

 

Quote

Is that one of reasons why you choose to use individual discrete logic devices on the O2/Fixer, one for each logic element to not only mimic the original 74LS08 IC but adding additional single logic devices to provide the extra Buffered outputs for modern add ons?

I see you point on a new main board design as you could add extra O2 buffers as additional hook-up points.

However,  lets say the the O2Fixer did not exist, on an existing 600XL board, if you was to fit the 64K expansion, could you not rewire the now unused gate in the 74LS08 as another BO2 source, therefore creating a bodge work around?

But that's only a 'What If?' TBH the price of the O2Fixer is worth every penny and no point bodging .

We need to understand how the "F" series works. The technology and design make it so that these circuits, without introducing too much capacitance on the signal lines where they are connected, achieve speeds up to 125MHz. So, when replacing the 74LS08 with 74F08N, it's not about the speed of the gate, but rather about reducing the parasitic capacitance on the main board. An additional effect is that the gate operates faster. Because the Atari design is based on NMOS and is more analog than digital from certain perspectives, everything still works.

The prototype of the O2/Fixer consists of two 74F08 chips, where the first one functions like the 74LS08, and the second one creates four additional twin BO2 signals that can be directly connected to the Atari circuits on top.
You can also reduce capacitance in a completely different way, but it's significantly more expensive. We can imagine replacing the PIA with W65C21N (WDC PIA NMOS), Pokey with PokeyMax, and GTIA with Sophia 2. On the XE board, I also had Freddie removed, so it wasn't connected to the BO2 line. Such a change was enough to sufficiently lower the capacitance of the BO2 line so that VBXE with Ultimate 1MB, Rapidus, and external SIDE3.1 could operate at full speed even in 20MHz mode without interference.

 

I've proven through a reverse slow degradation system test that we can, by replacing elements step by step, increase the system's stability. So with the full installation of add-ons that I talked about, SIDE3 wouldn't start at all. After replacing one element, it started in 6502 mode. After replacing a second element, it also started in 65C816 mode but couldn't read the SD card. After replacing a third element, everything started working beautifully. I then returned all the old chips to the Atari and built a prototype of the O2/Fixer where all the circuits received signals from the twin BO2 signals. Everything worked beautifully.

Switching to 1-gate and AHCT is a matter of productizing lab work, reducing costs, adapting for pick&place production, and miniaturizing without taking up additional space on the main board. This refinement has already been implemented by Lotharek, who adapted the design to conditions allowing for the production of the O2/Fixer.

 

Theoretically, one could also use low-profile sockets, but unfortunately, such gold-plated low-profile sockets would probably triple the cost of the O2/Fixer. I personally didn't negotiate any additional compensation for this project, so as not to increase the product's value. This particular product should be widely available, as it stabilizes the operation of projects that people have often already purchased. In the end, it's all about people having stable computers so that we can continue to have fun, rather than having computers sitting in cabinets waiting for better times.

 

In my opinion, for owners of the Ultimate 1MB, the O2/Fixer device is rather a must-have. According to the assumptions, the memory should be controlled by BO2, not by O2, for example from SRAM 64K V3. This is a matter of further system stability, as well as functioning with extended memory, not only in RAMBO mode but also in CompyShop mode. Remember that in the second mode, Antic has access to a different bank than Saliy. The BO2 and O2 signals are often confused in Atari schematics and descriptions of Atari ICs.