Hi Fans,
here is a report about the use of early-mid 1970s 4k x 1 DRAMs in the Apple-1.
Most of us are using later 4k x 1 DRAMs produce towards the end of the 1970s or later, such as the ones I furnished with my famous kits (now sold out). Typical examples are:
Motorola MCM4027AC3 Date codes 7927, ...
Mostek MK4027N-4 Date codes 7935, ...
Intersil MK4027J-3 Date codes 7948,8017 ...
It was found that the former needed all the "reliability mods" with the extra bypass capacitors and the six damping resistors, Mostek often was happy with only the damping resistors, and on some PCBs / builds, the Intersil ones needed no reliability mods at all.
A while ago I had the opportunity to acquire a S-100 bus board with 16k Bytes of DRAM, built by Processor Technology, with a (c) 1977 on the PCB, and using white ceramic Intel C2104-6 DRAMs with date codes like 7545 (45 week of 1975 !) or 7601 (1st week of 1976 !). Alas, the date codes are in the bottom, while the markings on the top don't give a clue:
I found out when I desoldered the DRAMs. One out of 32 was dead, probably the reason why this memory card had been put away and then was forgotten. Otherwise it may have ended up in a landfill together with the rest of the S-100 system.
This was the opportunity for me to get my hands on earlier, slower 4k x 1 DRAMs which were of the same vintage as the Mostek MK4096N-11 with date codes 7601,7620,... seen in the Apple-1 originals. These MK4096, like the Intel C2104 were 1st generation 4kbit DRAMs, while all the other ones listed above are 2nd generation 4kbit DRAMs, with revised circuitry and the next process node, and hence, much faster then the 1st generation. After the 2nd generation 4k DRAMs came out, which were faster, had a smaller die size, higher yields, and lower price, the 1st generation DRAMs quickly went the way of the dinosaur, and this is why they are so hard to find nowadays. The 2nd generation 4kbit DRAMs soldiered on and were still in production in the early 1980s, despite 16 kbit and 64 kbit DRAMs soon replaced them in new computer designs.
For me, the curious point was whether the Apple-1 would work with these 2104 DRAMs, so I rigged one of my fourteen Apple-1 as a test bench:
The timing resistor on the 74123 was replaced. If you ever try this, use a a series connected 5kOhm fixed resistor and a 20kOhm trim pot:
Finally, the Wozmon PROMs were replaced with my diagnostic PROMs which contain a powerful DRAM test, and the 6502 was replaced with an early MOS Technology one which is the slowest 6502 I have. All others are "A" or "C" types rated for more than 2 MHz operation, and these have a more relaxed timing when running in the Apple-1. Note that in the Apple-1, the address bus setup time does not matter for this experiment, but the data bus setup time does. If a faster 6502 is used, then the read data may come a bit later, allowing for a slower DRAM.
I then let the machine run and by turning the trim pot, I tested the operating margins for the 74123 oneshot timing (logic high time Th on 74123 pin #12 measured by oscilloscope):
Th < 420ns: DRAM NO GO
420ns <= Th <= 530ns: DRAM GO
Th > 530ns: DRAM NO GO
The "GO/NO GO" decision was based on occurence of the first, rare DRAM errors. So 420ns and 530ns are the points where the DRAM operation gets wonky.
Now, calculate the "center" of the DRAM GO range, which is the mean: (420ns + 530ns) / 2 = 475ns
In the Apple-1 "Processor Section" schematic of the Apple-1 manual, Woz specifies 480ns as the target. Compare to the 475ns result above.
This experiment proves several things:
1. The DRAM timing in the Apple-1 has been chosen carefully and is optimum for these slow 1st generation DRAMs.
2. Any trouble with DRAM errors in the Apple-1 does no come from circuit design mistakes.
3. Woz is a competent digital designer (the "genius" label never applies for mundane timing calculations).
Years ago, when trying to find out why my first Apple-1 clone I just had built did not work, I did some hand calculations on the whole DRAM timing and using datasheet values for these old parts and an 1 MHz 6502, I arrived at an optimum oneshot setting that was very close to the 480ns specified by Woz.
So both the experiment on the actual hardware and the theoretical calculations prove that the 480ns is a good choice to run mid 1970s DRAMs on an Apple-1.
Of course, this quick and crude experiment was only done on one specimen of Apple-1, and the exact speed of my "slow" 6502 is unknown, all I know it's an honest 1 MHz type which does not run at 2 MHz, tried in an OSI Superboard II which had the 2 MHz mod. If I wanted to spend more time on this, I could repeat this experiment with my other thirteen Apple-1 clones, but as far as I'm concerned, all I wanted to find out if these 1975/76 Intel 2104-6 (which are one of the slower speed grades of this type) can run reliably on an Apple-1 without changing the oneshot timing.
The bottom line is, they should just drop into the sockets and all is fine.
Comments invited ! I could write more anecdotes about early 4k DRAMs, but not in this post.
- Uncle Bernie
Interesting article Bernie, thanks for putting it together!
Your DRAMs in the white ceramic case look awesome, I'm even a little jealous of you....
I guessed a couple years ago when your reliability mod came out that it was just an attempt to integrate some not-so-original components into the Apple-1. I even contacted the owner of Schlumberger NTI to check it out. Deep down, I've always been convinced that Woz is a great engineer.
I too have done timing measurements on pin 13 74123. Plus or minus 50n/s from 480 is absolutely not critical. Then I gave up doing nonsense, just started to add a 100K resistor parallel to the 27K resistor at the bottom of the board and always hit the right window.
You are very lucky that out of 32 of your beautiful DRAM chips only one of them turned out to be inoperable. When I bought memory chips at unicorn, I had it much worse. I had a half a cup of 4027s in various packages that didn't work. Probably this Ron or Rob, i want his pockets empty, realized that the buyer from Russia is unlikely to bother with the return and regularly threw me 1-2 non-working on a dozen. Cunning bug! I'm even a little glad that the political situation forced me to look for other sources. Soon I met a guy, if you order 16 pcs from him, all 16 will be working, you don't even have to check. When some doors close, others open.
In post #3, Macintosh_nik wrote:
"I guessed a couple years ago when your reliability mod came out that it was just an attempt to integrate some not-so-original components into the Apple-1. I even contacted the owner of Schlumberger NTI to check it out. Deep down, I've always been convinced that Woz is a great engineer."
Uncle Bernie comments:
It's not as simple as you imply. The Apple-1 in its original form is a marginal design in some respects, expecially the PCB layout has issues, and not so much the digital logic design as such, which is very efficient and mostly clean, with the exception of some minor quirks, such as "gates" made from diodes and resistors, but in my experience, these rarely cause trouble, especially when using Schottky diodes. I can say that with authority as I have analyzed the schematic down to the last gate, and spent many hours over the Apple-1 with an oscilloscope, checking all the signals.
There was a good reason for the "buyback" decision at Apple, they were not happy with the customer experience, and got just too many phone calls distracting Woz from his work on the Apple II. They would not have done that if the Apple-1 was no troublemaker.
What most people writing about how good or bad the Apple-1 might be don't seem to understand is that the robustness (or lack thereof) of Apple-1, originals or clones, comes from many factors and component choices. Even the "recipe" for the PCB plays a role. And you are right, Macintosh_nik, that use of substitute ICs can have a large impact on the functionality of the Apple-1, even when these substitutes were advertised to be "drop-in" replacements back in the day. This boils down to the fact that faster ICs react more allergic to the "hostile" environment on the Apple-1 motherboard, as they can "see" ringing signals which the slower ICs of an earlier generation did not "see", so they would not react to them.
According to my research, the biggest issue with the original Apple-1 was program crashes caused by occasional and random DRAM bit errors. The botched ACI card was another offender, but this is a topic for another thread, and it has been fixed with my Gen2 improved ACI.
Not all the Apple-1 originals had these occasional DRAM bit errors. Some worked fine (the one in the weather station). Others didn't (Lisa Loop's).
WHAT WENT WRONG WITH THE APPLE-1 ?
In short form (I have longer and more detailed writeups on that):
1. Lack of adequate power supply bypass capacitors for the three DRAM power supply voltages (+12V, +5V, -5V).
Not exactly Woz' fault because in their early datasheets, DRAM manufacturers lied by omission: they did not mention the fierce current spikes produced by the DRAMs. Only after too many design-ins failed and DRAMs got a bad reputation, the DRAM manufacturers started to show oscillograms with the spikes and gave examples for proper power supply bypassing and even recommended PCB layouts. This solved the problem and the computer industry started to accept the DRAMs more and more. Soon they did replace the SRAMs (which always had been more robust, easier to design in, but far more expensive per bit).
2. Supply voltage and ground return traces on the PCB being marginal. Inductance too high, exacerbating ringing.
3. On the Apple-1 motherboard, the traces for the multiplexed address bus are too long. They exceed the 11" length limit for Schottky TTL, after which you have to work with transmission line techniques, or you get reflections and ringing. Older, slower 1st generation 4k DRAMs may have been too slow to "see" the ringing as logic transitions. But some of the 2nd generation 4k DRAMs are allergic to them.
4. The "27k" carbon composition resistor for the 74123 timing oneshot is a poor choice. For typical 74123 (note: 74LS123 is different) the 27k Ohms always exceed the 480ns. In most cases, I ended up with values between 21 kOhm and 22 kOhm for 47 pF capacitors. The Mica capacitors were a good choice, as they are very stable and can be bought with small tolerance bands (typical is 2% --- 1% also exist but are much more expensive).
All these mistakes conspire to make the Apple-1 a marginal design in terms of power supply bypassing, power supply rail and logic signal ringing, and DRAM timing. It is highly likely that the slow 1st generation Mostek MK4096-11 were more robust against this hostile environment.
As we now know, since the prototype "Apple Computer A" was found, this prototype did not use the brown ceramic disc capacitors as power supply bypass capacitors, but smaller, "boxy" looking ones. Since I can't desolder one of these capacitors from the original prototype to put it on a network analyzer, I can only speculate that these may have had a significantly better performance in the role of bypass capacitors, compared to the brown disc ones used in the "BYTE SHOP" (aka non-NTI) production run. If it is so, the prototype may have worked robustly and fine, with no crashes, while the production versions were less robust and more crash prone, on an individual basis (example: Lisa Loop's story).
We also know that the later "NTI" versions which they sold directly via their magazine ads had boxy looking bypass capacitors again. Maybe the same type as in the prototype ? To undo the big mistake to change component specifications from the prototype to the 1st production run ?
Such last minute substitutes are super risky. The industry standard practice absolutly prohibits such changes without prior assessment in yet another prototype. I follow the same rules. This is how I ended up with the fourteen Apple-1 clones I built so far !
ABOUT THE "RELIABILITY MODS"
When you build an Apple-1 using 2nd generation DRAMs, it depends on the type of DRAM and the performance of your bypass capacitors whether your build will work without "reliability mods" or not. In my first build (which did not work before I developed the "reliability mods") I had used Motorola MCM4027 and vintage, dark brown ceramic disc capacitors which probably were older than the Apple-1 originals !
I also had used a DS0026 clock generator because I could not find any DS0025. All this conspired against making the build work. The DS0026 just has faster signal edges than the DS0025, which couple into the -5V power supply rail via the 2504 shift registers. And the DRAMs don't like that.
You can use a DS0026 if you add some bypass capacitors at the DRAM -5V pins, or if you put 39 Ohm series resistors in the DS0026 outputs.
Be aware that most "DS0025" sold from mainland China are counterfeit and some are re-labeled DS0026, or, worse, pin compatible MOSFET drivers with the same pinout but built in CMOS, which would crash any Apple-1 (and these types also lack the pulldown resistors on the inputs).
CONCLUSION
There are remedies for all the mistakes (above 1. - 4.) made with the original Apple-1, and we now know how to build Apple-1 which work robustly and without crashes. I've also fixed the botched ACI so now, 47 years too late, we can have Apple-1 systems which really do work.
But back in the day, they didn't. Some were probably OK, others, not so, and drove their owners nuts, generating phonecalls at Apple, which kept Woz from working on the Apple II design and development. You see, there really was a good reason for the "buyback" decision.
- Uncle Bernie
Thanks for the detailed reply, as always it was very interesting to read it all.
Yes, Apple-1 does contain errors.
The schematic from the manual for the Apple-1 has errors.
The silkscreen on the Apple II Rev. 0 has errors as well.
The Apple II schematic from the red book also has errors....
We shouldn't be too hard on the two Steves. They were barely in their 20s at the time, very young guys. In the end, the startup Apple Computer became the most technologically advanced and expensive company in the world. There's no judgment on the winners.