After figuring out by myself that the Apple-1 needs additional power supply bypass capacitors to ever work reliably (in the sense of "rock solid"), which did cost me just a few months :-(Of course I had tried to find info on it when my build did not work, I did not find any, until I spotted this post by "wsander" here on Applefritter:
https://www.applefritter.com/content/apple-1-mimeo-filtering
which appears to be a pretty complete solution as far as bypassing is concerned. But I wanted to do more perfect ;-) and made this:
You can see I use 10 different kinds of sockets with hidden components, mostly 100nF capacitors with one exception: the socket for the keyboard on location B4 has a 2K resistor from STROBE (pin 14) to GND (pin 9). The reason for this is that I use an 1990s subnotebook running DOS as keyboard which I don't trust to run unobserved due to age, so I like to unplug the keyboard cable and switch the notebook off while the Apple-1 build runs 24/7 with test programs. Having a PC faking the keyboard is really cool as it can also automatically "type in" said test programs once the DRAM works stable enough so as not to crash WOZMON all too often. Adding bypass capacitors is a necessary, but not sufficient condition to get a rock solid build. But this is another topic.
If there is enough interest from other builders lurking on this forum, I will spend some time to describe these sockets and where they go. I just don't want to waste my time on documenting things I found out nobody cares about, like my DRAM address line hammer program I published on this list.
As a proof, here is a photo of my most recent build based on a Newton NTI PCB which has NONE of the unsual 100nF bypass capacitor locations populated and still runs rock solid:
This means that the hidden capacitors can do a complete job in terms of power supply bypassing. This is cool because it enables us to use whatever value, style and color of bypass capacitors we can find or want to populate the usual bypass capacitor locations. Even 100pF would do as long as they have the desired looks ! This added degree of freedom should allow us to "cheat" a bit with the elusive capacitors used in the originals.
You also might note that this build only has 470uF electrolytic capacitors in lieu of the 2400uF/25V ones used in the originals. I proposed the use of the much safer and more abundant 2000uF/50V which I offer on Ebay but got a lot of sceptic messages saying they won't work. These fears are unfounded.Given enough interest, I can elaborate on that topic, too, and why using 25V types may be unsafe, but again, only if there is enough interest in such a discussion here on this forum.
Comments invited !
Uncle.Bernie
... despite of zero interest from the builders lurking on this list so far.
(Maybe there are no builders, just owners of Originals, here ... )
The lab experiment
I have been doing some timing margin testing on the build seen in the above photo, using a trim pot in lieu of the 27K timing resistor on the B3 oneshot and the results so far are:
* with a 6502B (rated for 3 MHz) the timing margins on the 74123 oneshot on location B3 are quite wide. Systems runs rock solid.
* with a 6502A (rated for 2 MHz) the timing window narrows. A position of the trimmer can be found where system still runs rock solid.
* with a 6502 (rated for 1 MHz) the timing window is super narrow, very hard to hit, even with the trimmer. Occasional memory faults, maybe one per day or so running Mike Willegal's memory test. Lisa Loop syndrome or better ?
Conclusions
Apple-1 DRAM timing is tricky and the faster the 6502 is, the wider is the timing window you need to hit by properly adjusting the 74123 oneshot. (wider = easier to hit)
I think with the 27K resistor as specified in the original schematic it's quite hopeless to ever hit the narrow timing window for robust operation.
Recommendation from Uncle Bernie
I recommend to use a 6502B in your builds and initially have a 50K trimmer potentiometer in lieu if the 27K resistor, then find out the go/nogo margins and adjust into the center between these margins. Measure the resistance (easily done after the 74123 is unplugged, measure between pin 16 and pin 15). Then replace trimmer with resistor of appropriate value. Beware that carbon composite resistors are junk (always have been): they are both unstable and have a terrible tempco. After soldering them in, they won't be the same value anymore as they had before, although they tend to slowly creep back, but only so much. This is probably the one place in your Apple-1 build where you want a metal film resistor. The ancient mica cap is fine, though. Silver mica capacitors are known to be very stable, tight tolerances, low tempco, great long term stability. Proper choice here, Woz (if you ever read this). But modern builders be aware that the silver electrodes of your 50+ years old mica capacitors may be corroded or otherwise compromized. At least measure it before you solder it in. It should be very close to the 47pF, any larger tolerance than +/- 2% already is suspect, as these capacitors always have been of tight tolerances (and were expensive). If one particular, old mica cap is off from its value by much more than that, it probably has internal rot and should not be used as a timing capacitor in your Apple-1 build.
Future prospects
I'm still trying to write a memory test program that can discern between read and write failures but so far no success. While a write failure caused by an too early write (before the 6502 has the data ready) causes a permanently wrong byte written to that address, which is easy to detect, there should be a read pattern to find out if repeated reads yield different data (DRAM /CAS too late and hence the 6502 grabs yet unstable data), but so far this has been elusive, although there is still hope.
The failure mechanism explained
If the oneshot pulse width is set too short, /CAS falling edge comes too early and the 6502 may not have set up the write data properly yet, so writes fail.
If the oneshot pulse width is set too long, /CAS falling edge comes too late and the DRAM access time counting from this edge may not meet the read data setup time requirement of the 6502.
Uncle Bernie's ecommendation for Apple-1 builders
Using a faster 6502 version (6502A or 6502B) widens the timing margin as the faster CPU will present the write data earlier (which allows a sooner /CAS for writes which helps to relax the read timing) and, during reads, the faster CPU needs less read data setup time, relaxing the read timing even further. Hence, using a 6502B (or at least a 6502A) in your build is desirable.
Caveat
Just plugging in a faster version of the DRAM chips may or may not help. It may make things worse. There is severe ringing on the too long and non-terminated multiplexed address lines emanating from the 74S257 drivers. This ringing may have enough amplitude even after the falling edge of /CAS to toggle the internal address lines of the DRAMs, violating the address set up time referenced to /CAS falling edge, which, fortunately, for many first generation 4K x 1 DRAMs is zero or even negative. Alas, the ringing seen lasts longer than this added leeway. Not adding lots of power supply bypass capacitors to the DRAM banks and the 74S257 will exacerbate this problem as the DRAM's input buffers derive the logic thresholds from the voltage between VCC (+5V )and GND. Without adequate bypass capacitors, the VCC and GND will not be "tied together" AC wise and may ring independently from each other, so the logic threshold of the DRAM input buffer will bounce around, in an unpredictable way, as the ringing of VCC/GND and the ringing of the multiplexed address bus lines are NOT correlated. There is NO HOPE to ever find a timing oneshot setting for /CAS which mitigates this ringing problem. Only adding bypass capacitors can help. I see one further solution towards more robustness but this involves cutting traces and I won't do that on my valuable NEWTON PCBs - - - Newton Mike (the guy from Hong Kong) has announced he will stop making them due to the recent appearance of the "cheap" open-source-Gerber-file based ones made by JLCPCB in China. Once Mike runs out of the last NEWTON brand PCBs, there will be no more, gone, forever, and the prices for unmolested Newton PCBs most likely will go through the roof. So get 'em as long as you can !
I am interested. Thank you for sharing, and I'm interested in any other tidbits you mgith have.
You might not get responses right away, but you can be confident that others will read your information, benefit from it, and be glad you shared.
As Shrek's mother always says, "Better out than in."
Dave
I have now spent 8 months investigating the quirks of the Apple-1 and putting some real science and engineering into it, and not just some cargo-cult like magical thinking as seen elsewhere at some places in the web. As an example, there have been numerous attempts by desperate Apple-1 builders to find "a better set of ICs". Or to "tune the /CAS timing oneshot". Trust me, none of these measures make any Apple-1 build running rock-solid. None. The reason is that the Apple-1 suffers from several childhood ailments and its "father" - Woz - never had the time to fix them. He was just too busy designing the Apple-2 and the Disk-2 to have any time left to address the issues in his "firstborn", the Apple-1, which mostly are rooted in the terrible PCB layout, and not at all in any poor circuit design, which as far as I have looked into it, appears to be quite sound, although not perfect, except maybe for the complete lack of adequate power supply bypassing capacitors on the -5V rail. No DRAM will ever work reliably without those. How this oversight came about is yet an open question. When Woz designed the "processor section" with these - at the time being - novel 1K x 1 DRAMs that just had come on the market, there were no established engineering practices to design them into a system. Many companies tried to use them, found out they did not work well enough to be any useful, and went back to static RAMs which worked but were much more expensive per bit. DRAM manufacturers trying to hide the existence of fierce current spikes on all the three power supply rails and on the ground rail of their novel DRAM ICs did not help with making them work either. Early DRAM datasheets did not mention those spikes at all. Later, revised datasheets went so far as to actually show oscillograms of how high and how fierce these current spikes could be per single DRAM chip. These spikes looked very frightening when considering that their magnitude would multiply with the number of DRAMs being active ! And some datasheets even went further and showed recommended PCB layouts with all the bypass capacitors placed very strategically.
I dare a guess that Woz had none of these later datasheets when he designed the Apple-1 processor section. And he was unlucky enough that his hand-wired prototype did work well enough to encourage them to make a PCB and "mass" produce them. After the first 50 or so were sold, all hell must have broken loose. But these are my words, my conjectures, and hence, not from the horse's mouth. But my conjectures are supported by the fact that Woz was the only person at Apple who was able to answer technical questions on the Apple-1 over the phone, and this phone rang often enough off the hook to keep him from working on the Apple-2 / Disk-2, so they decided to "buy the Apple-1 back" by offering a discount on the new Apple-2, if the Apple-1 was turned in, and this is how the "official story" goes.
Trouble was, the typical Apple-1 almost worked, most of the time. It typically was running well enough to load BASIC and run some little BASIC programs. And then it crashed. Load the BASIC interpreter again, which was quicker than with other computers of the time, enter that little BASIC program again, RUN, fun, it works ... oops, another crash.
As the story goes, a lady named Lisa Loop ran a sort of education center trying to teach kids programming in BASIC, and she was not amused that the Apple-1 that had been given to her by Woz did crash too often and was frustrating the kids. So Woz took the Apple-1 back and tried to fix it. As the story goes, it worked somewhat better after the master himself had tuned it up. But Lisa Loop still was not satisfied, as it still did crash occasionally. So Woz gave her one of the first Apple-2 that came off the assembly line, again for free. (He's a nice guy). And this one did not crash. So Lisa Loop went on and taught many kids how to program in BASIC. Happy end !
For any 21st century Apple-1 builder there is no such happy end, however. It is a fact that most Apple-1 builds fail and never work well enough to actually use them. Most end up as wall hangers. I call this the "Lisa Loop experience". As an example for all the frustration, pain and suffering experienced by Apple-1 builders, just look at his thread, which is exemplary:
https://www.applefritter.com/content/mimeo-power
The problem is, of course, that unless the DRAM works at least well enough to use the WOZMON to enter diagnostic programs, like Mike Willegal's memory test, there is no hope, and the builder is stuck. This then more often than not is the point of frustration where magical thinking sets in, out of desperation, and chips get swapped around and new DRAMs get bought. But believe me: It. Won't. Work. Doing such cargo cult magic is wasted time.
It has long been known and documented on this forum and elsewhere that the main culprit for all this trouble is the lack of a sufficient number of properly placed power supply bypass capacitors in the original Apple-1 PCB. This problem is worse if the DS0026 is used in lieu of the DS0025 specified in the original schematics. I could not find any DS0025 so I had pulled a DS0026 out of a IMP-16 and put it into my first Apple-1 build. With the dire consequence it did not work. The "terminal section" worked on first power up, however. So I was more lucky than other builders who have to fight with forgetful PMOS shift register chips or the occasional bad or "limping wounded" TTL IC. But my build's "processor section" only came alive after I had mounted additional bypass capacitors on the backside of the PCB. And with "alive" I mean that I finally could enter and run Mike Willegal's DRAM test program only to see it spit out error messages every so often. But my build was finally running well enough to load BASIC and enter and run a few little BASIC programs ... for a few minutes ... crash. Rise and repeat. The "Lisa Loop experience" strikes again. I think with the DS0025 in lieu of the DS0026 it would have crashed less often, but my experiments with slowing the clock signals to the PMOS shift registers down by means of RC networks were inconclusive. The memory errors already happend too rarely to get any meaningful failure rate statistics without spending too much time. And I had the hunch that there must be some further problem lurking deeply within the guts of the Apple-1. This is because I had added lots and lots of bypass capacitors until I had one on every power / ground rail on every single DRAM. And I had checked the /CAS timing to be correct. And the power supply voltages to be in spec. Still there was the occasional DRAM error, leading to the occasional crash. Maybe once per two days or so.
So I dug deeper and found the reflection / ringing problem on the multiplexed address bus to the DRAMs, which is documented here:
https://www.applefritter.com/content/experimental-dram-challenge-program
After the termination network was put in, and the ringing was mitigated, my Apple-1 build finally did work, perfectly, it ran Mike Willegal's memory test program first for days, then for weeks, then for months, 24/7 without fail. Until I switched off the A/C system in my house to repair the inducer blower ... before it was put together again there came a sunny and hot day and my lab upstairs got too hot. Another nasty DRAM error was reported but the Apple-1 did not crash. The problem was traced to the 74S257 running too hot and not driving good "H" levels anymore. A known issue why many practitioners of the art believe that Schottky TTL cannot use any resistive parallel termination. So I moved the common node of the termination network from GND to +5V and added yet another 100nF bypass capacitor between +5V and GND right there, which, at least approximately, has the same remedial termination effect, but does not burden the weaker high side driver transistor of the 74S257 anymore, and the much stronger low side driver transistor instead.
After that mod, I could torture the poor thing with a heat gun and it still didn't get any DRAM errors. In this configuration, I kept it running for months with not a single DRAM error or crash. I think I found the remedies.
My build now is running rock-solid 24/7 for many months.
The remaining question was, how many added bypass capacitors are needed. Not that an excess of them would make it run less reliably, but they are tedious to put in, especially if they are craftly being hidden in the hollow IC sockets, as suggested by "wsander" in this thread:
https://www.applefritter.com/content/apple-1-mimeo-filtering
I did another build with all the bypass capacitors hidden in the sockets, as evidenced by my first post in this thread, but man, it was tedious and time consuming and my poor old eyes and back hurt like hell. The latter was quickly remedied by some glasses of Famous Grouse, which is a universal medicine for all sorts of ailments. Then I started thinking, at least with the few cylinders in my head that still were firing after hunting down and killing all these grouses, how to avoid such an ordeal once and for all. I hatched a plan to systematically try out what is the minimum configuration for bypass capacitors and termination that still would yield a rock-solid build. Alas, I did not want to do this experiment on a precious Newton PCB, for obvious reasons: on these expensive, high-quality PCBs I only want to do the best build possible with every reliability mod being hidden in the sockets, which rules out any experiments involving removal and adding of components. Execution of this plan had to wait for several months, and then came along these cheap "Apple 1 Replica" PCBs based on the open source Gerbers, as seen in this thread:
https://www.applefritter.com/content/apple-1-replica-gerber-files
and I was offered some at a great price (thanks, "mi2k" !) and last weekend I built an Apple-1 clone on one of these cheaper and disposable PCB, which are great for doing all sorts of experiments without much regrets for the PCB itself. BTW, if you want some of these, look on Ebay:
https://www.ebay.com/itm/Apple-1-replica-board-with-ACI-card-Apple-1/203168087819?hash=item2f4dc2ff0b:g:GH0AAOSwoKBflxx-
or search on Ebay with the terms "Apple-1-replica-board-with-ACI-card-Apple-1" to find it. The Ebay nick of the seller is "pinguso" (it's not me) and the price of $35 plus $10 shipping is awesome, considering the added ACI PCB is included, and no Apple-1 build is complete without an ACI, as you can't do much without it. No nasty cassette player is needed, though, just play MP3 files with Apple-1 software to the ACI. But if you decide to buy one of these PCBs, don't fall for one of these price gougers who sell inferior ones with incorrect silk screen at a gouged price that together with shipping, pricewise come so close to the high-quality and more desirable "Newton" PCBs, and don't even include an ACI. Avoid these rip-off artists and price gougers and get yours from pinguso. For any non-experimental build which invests a lot of time into hiding the added components in the sockets and seeks more authentic looks, I still prefer the excellent, but more expensive "Newton" PCBs, which, among other desirable features, also have an gold-plated edge connector (important for long-term reliability of any attachments). But for experiments which may end up in the trash can, or for those builders on a budget who don't seek authentic looks, these cheap "Replica" PCBs from pinguso are OK !
This is the minimum configuration for reliability mods I found so far:
ReliabMods.jpg
All the little yellow things are 100nF ceramic capacitors, and the cluster of six termination resistors uses 390 Ohm ones. The one resistor at the keyboard socket is 2KOhms, and it's not really a reliability mod, but more a convenience mod, which allows to unplug the keyboard cable while the Apple-1 runs, without spurious phantom key entries being provoked. This is very handy in the debug phase. I run my builds 24/7 unattended after a few days running 12 hours each under constant supervision, and only after all temperatures of all ICs and the big blue electrolytics stay normal (as measured with a contactless IR thermometer) I then relax the safety measures and let it run 24/7, but that's me, and my house has smoke alarms everywhere, so I don't worry too much. It's your call and your decision how you run your Apple-1 build, and if your house burns down, your fault, don't blame me. I would not let anything run unattended if there are pets or kids in the house. But even I'm suspicious: I don't trust the 1990s vintage notebook computer (which drives the keyboard emulator cable) to stay connected to the line voltage unattended. I always unplug it from the mains before I leave the house or go to bed. This is because 30 year old electronic components living on the line voltage can't be trusted. Any power outlet can deliver enough power to set things on fire if such an aged component decides to break down. Same rule applies to the ancient 1960s vintage B&W monitor I use. This one I switch off whenever I'm not sitting right at the machine. And yes, I inspected it to see if its power switch really turns it off at the line side. Never trust an old piece of electronics in standby power. Never. Always keep those on a power strip with a master switch and turn that master switch off when not being there.
I almost forgot to mention another reliability mod not visible in the above photo: the 27K resistor for the /CAS timing between PCB locations B2 and B3 was replaced with a 50 KOhm trimmer. I use a special DRAM margining PROM to determine the min/max go/nogo limits, measure the resistance between pins 16 and 15 of the removed 74123 (after power being turned off of course) at these limits and then replace the trimmer with a resistor of the average value just between these two go/nogo limits. So far I ended up with resistors between 22KOhm and 24KOhm, but never with the 27 KOhms specified in the original schematic, although every build so far did work with the 27 KOhms, too. But keep in mind in this case, with 27K, the /CAS timing is not centered in the middle of the go/nogo window. Which, for the 1 MHz 6502, is only 70 ns wide (with typical DRAM speeds). This is why I recommend Synertek 6502B or Rockwell 6502-34 which are rated for 3 Mhz operation, to widen the go/nogo window. These only cost $2 - $3 more than the cheapest pulled 6502 of arbitrary speed grade I could find at the usual surplus places. It's worth to invest these few bucks more to have a higher speed grade CPU. It's not worth, however, to invest in faster DRAMs. These not only are expensive, but also may bring in more vulnerabilities to the remaining, but thanks to the above mods much mitigated, ringing. In some cases, slower is better because slower ICs may not "see" glitches that may throw faster ICs off track. Understand that there will never be a "perfect" Apple-1 with crystal clear signals within. This is all but impossible without a new PCB layout using multilayer. But me (and others) have shown now that with a few cheap mods, we can achieve a robust, rock-solid Apple-1 build, with a MTBF for DRAM failures / crashes in the order of months, a reliability the originals never had. Maybe this is the reason why these originals fetch $1 Million in auctions. Who knows.
Enough of my tips & tricks for Apple-1 builders for today. This became a much longer post than intended, but I did not want to split all this info up and place it here and there and all over the place. If you follow the guidance given in just this one post, in the end your build will succeed and work rock-solid. After you did sort out your bad solder joints and ICs, of course, but all the inherent quirks of the Apple-1 now have been remedied if the above recipes were followed.
Stay safe, and happy Apple-1 building !
Thanks for sharing. I'll likely be doing this to my Apple 1 once I get some time. So many proects, so little time. Retirement just a short time away... ;)
WOW! Thank you very much for your efforts and commentary! I won't be building or using one, but it helped spawn my IIe of the 80's. Thanks for helping it out!
I forgot to mention in my post #4 above, if you use the mods proposed there, please sign up on Applefritter and post your experiences and stories here in this thread, which I hope all are "success stories". But if something does not work, and you had to add some further mods, feel free to contact me via the message system or post your own mods here.
There are still a lot of less important little mods you could do, i.e. modifying the pulldown resistors to "nudge" weak PMOS ICs into doing their job, etc., etc., but I did not get around yet to describe these, as they typically are not needed for every build, but only on rare case by case basis. It's probably easier to just swap such "limping wounded" ICs. The 1K x 1 shift registers being the most notorious, I have had quite a lot of rejects with them, but I also had a few bad 2519, which hurts more, money wise. So far none of my 2513 were defective, and only the occasional bad DRAM had to be widlarized.
Comments invited !
Uncle Bernie
If you have built an Apple-1 clone, you might have noticed one peculiarity: all the building manuals (such as Mike Willegal's) and all the BOMs specify the use of 100nF bypass capacitors, despite the fact that the PCB's silk screen reads "1.0" at the places for them. Meaning 1.0 uF. Now, all the building manuals essentially say something like: don't worry, the 1.0 silk screen is wrong, put those 100nF capacitors in these places.Alas, the nicely drawn schematics in the original Apple-1 manual have the text "1.0" near all those bypass capacitors, here is an example which will play a major role in this post:
A1_1uF_Schem.JPG
So who is wrong ? How did this obvious contradiction come about ?
As it seems from photos of some of the earlier "non-NTI" Apple-1 originals I found on the web, they appear to use 100nF bypass capacitors only, but I'm not sure, as the lettering on these capacitors as seen in the photos typically is too coarse (too few pixels) to be absolutely certain. I did not find any photos of the later "NTI" versions where the lettering on the bypass capacitors was at least somewhat legible or guessable.
Now, here is the story what happend to me. As you might know, from time to time, I do sell a few 100% tested and burned-in, complete IC sets on Ebay (and via private contacts, too), all with the target objective to prove my reliability mods as described above in this thread do work and yield 100% success rate. I do coach these builders who bought my complete IC sets via email so I get feedback on how it goes. This whole hobby activity of course is steered to avoid any profits, as the extra complications for my tax returns probably would cost me more money for the law firm than I could make by selling a few parts to this small crowd of builders. So I don't make complete kits, like Unicorn Electronics other other commercial vendors do. With my fully tested IC kits, the builder has to provide the non-ICs parts which I don't include, although I have begun to add a few of the rarer passive components and those for the reliability mods, to get the results quicker. Now, just a week ago, I had a case where the Apple-1 build did work upon first power up as with the few others before, but this particular one then, after warmup, proceeded to occasionally throw randomly wrong characters on the screen, very rarely, but still considered a failure. It turned out that the MM1404 shift registers acted up and after some head scratching, oscilloscope measurements and experimenting the remedy found was to swap out the 100nF bypass capacitors adjacent to the DS0026 for 1.0uF ones, as seen in the original schematic and in this photo:
PDRM0580.JPG
The orange things on the "inner locations" nearest to the IC are said 1.0uF ceramic capacitors.
Now this Apple-1 build does work perfectly as it should have at the first try. Note that I provide DS0026 in lieu of the "unobtainium" DS0025, which have faster rise times than the latter, but are easy to procure. A substitution that worked well so far, but may have contributed to the phenomenon observed. Unknown properties of the regular 100 nF bypass capacitors provided by the builder also may have contributed. My own clones in which I do the IC test / burn-in also have only 100nF everywhere, so there is no obvious reason why these MM1404 have decided to misbehave in the new build.
It's a mystery to me.
Now, consider this info bite: it is known, or at least told, that some Apple-1 originals have special "BEL" brand ceramic capacitors at the bypass places. It is said that those typically were used in high end audio equipment, and not for humble power supply bypassing purposes, because they were too expensive for that. At least, this is how the story is told. The audio equipment application needing special capacitors even is credible, because some ceramic capacitors suffer from microphony: they may pick up mechanical vibrations, and turn them in an electrical signal, and in the end may act as an unwanted crystal microphone, and having any such parasitic effect in audio equipment is not desired. But having microphony or not is totally irrelevant for digital IC power supply bypassing. It eludes me completely why the two Steves did put these very expensive, special capacitors in some of their Apple-1. If the story is true at all !
These findings / stories / experiences should give rise to some curious questions:
- Did the two Steves possibly experiment with various types, values and brands of bypass capacitors in an attempt to mitigate the reliability issues of their product in terms of random program crashes (the "Lisa Loop Syndrome") ?
- Did the prototype hand-wired by Woz use 1.0uF bypass capacitors that made it into the nicely inked official schematics in the Apple-1 manual, but apparently not into the final PCB based product ? (I don't want to believe that the "1.0" was just an error by the person who drafted these schematics).
- Are the properties of the bypass capacitors really that critical ? (I do know of some perfectly good industrial products who did stop working reliably after nasty bean counters had substituted the originally specified bypass capacitors for cheaper ones, so I do know there could be such a sensitivity. But does this apply to Apple-1?)
These questions could only be answered by Woz himself, if he still remembers the details how they built them in the garage 44 years ago, which is doubtful. Human brains are not perfect nonvolatile memories. For instance, I can't remember all the details of what I did in 1976. I think that in 1976 I designed and built a 8080 based computer with about the same spartanic features as the KIM-1, and I seem to remember the cassette interface did not work in the first iteration, so I had to add a transistor (IIRC) and re-code the firmware on those nasty optical cards a second time, but probably not the whole stack. But I don't remember any details of what did cause the problem and how it was solved. But I still remember when I sat with that fat, but nice and patient lady at at their Data I/O programming system proof-reading against my listing to see if these nasty optical cards had been read correctly, which they weren't, it took more than one hour, and they did not even charge me anything for programming my sole 2708. I still have the punched paper tape we produced on this Data I/O system, which contains my firmware. But alas, I can't read it anymore, these machines are extinct now ! These were the times ... the dawn of the hobbyist microcomputer ... fond memories ... but back now to the Apple-1.
The takeaway of all the above is, that we may have a larger sensitivity of the Apple-1 to the bypass capacitors as was believed until now. And if this is true, then any builder who just uses some random bypass capacitors as found online or in his lab drawers on on his lab floor is in peril, and anyone who sells kits should provide proven bypass capacitors that do not fail to mitigate the inherent problems caused by the terrible PCB layout, which, as I mentioned before at various places, indeed is the root cause of all the pain and suffering and unsuccessful builds. But certain combinations of parts may work better than others, and this still presents a challenge to the builders.
To investigate further, I would love to try of the originally specified DS0025 to compare it to the DS0026 by real science and engineering, meaning oscilloscope measurements in my lab. If anyone reading this can lend me a few DS0025, please contact me via the message system. I promise I would give them back !
I also have another experiment in preparation which involves a PCB based on the open source Gerbers but made with a different and more expensive process flow. The hope here is, of course, to find a way to reduce the number of required but tedious reliability mods. But so far I don't have these PCBs yet. I will publish the recipe for these PCBs on this thread if this experiment turns out to be successful. But beware of the possibility that the random bypass capacitors you found for your build may not be good enough !
Comments invited ! Maybe some "experts" lurking on this forum can shine a light on which types and values of bypass capacitors were used in which production run of the original Apple-1.
Bernie,
The values for C5 and C6 on both NTI and pre NTI on the original boards are .1
As for the types, they were high quality type bypass capacitors (i.e. BEL, "CircleD"and RMC) that would have been used in expensive audio systems at the time, unlike MITS who used the cheapest radio parts they could find.
Cheers,Corey
Corey986 wrote:
" The values for C5 and C6 on both NTI and pre NTI on the original boards are .1 "
This confirms what I had deciphered / guessed from the photos as written in my post #8 above, but now, with the expert speaking who has seen real Apple-1 with his own eyes, we are certain. I don't understand why in the 21st century people still post low resolution photos on the web. Bandwidth and memory should be large enough to have quality photos, but maybe I just did not find those.
I'd think the best course of action for builders who seek authentic looks is to populate the board with 100nF capacitors in all "1.0" locations, and only use the above mod (changing these two for 1.0uF ones) when there are issues seen with the video as described above. Of course, if you have put lots and lots of additional 100nF bypass capacitors everywhere at the -5V and +12V rails, then you should never need this measure. But the purpose of this game is not to use an excess amount of added capacitors, and only the minimum number added to just get rock-solid operation of the Apple-1.
Still, the concerns I have mentioned in my post #8 above about possibly not-so-good bypass capacitors are justified. The problem is, unless you have a network analyzer you can't tell which are the "good" ones and even then you need a sample set of proven "good" ones to compare. So, if you run out of proven "good" ones then any others you may find may not do the job properly, and then you need to add more extra bypass capacitors to compensate for this deficiency. But keep in mind that more capacitance is not always good when it comes to power rail bypassing, this is a LCR network that has resonance frequencies and if the capacitance is increased, the resonance frequencies come down and then the whole network may ring even worse than before. So seeking a solution with minimum capacitance is probably the best approach. Other than throwing that layout away and doing a good one with real ground planes and wide, low resistance and low inductance power supply traces.
Why the expensive "audio" ceramic capacitors should be better as bypass capacitors for fierce digital current spikes eludes me. I would think a low ESL and a reasonably low, but not too low, ESR would make the best bypass capacitors, and for the typical values found for ESL and ESR of 100nF ceramic capacitors, these should not matter at all at audio frequencies. As far as I'm concerned I still have about 500 brown disc ceramic capacitors of 100nF that work fine in my builds. Got them for a song, by the pound, which is good. So "cheap" is not always bad. I just was lucky with them, could have been otherwise, and they could have sucked.
At one point, a a number of years ago, both Wendall and I experimented with a fast WDC 14MHZ 6502 and it turns out that the rise times were so fast that it caused issues. So going with faster processors will only get you so far. The reason for this experiment was the idea of getting a new batch of WDC 6502's packaged in a white ceramic package similar to the original 6502. We never persued this because of the problems.
If you have been using DS0026 clock driver chips, it's no wonder that you have been finding so many issues. The base design is a lot more stable (though not perfect) with the original DS0025. I recommend adding the following fix to any attempt at making a bullet proof Apple 1, especially when using DS0026 chips.
http://www.willegal.net/blog/?p=3668
I also found that stability with my first builds, where I used modern .1 uF decoupling capacitors was better than when I went to using vintage ceramic caps. I never investigated the difference and don't know if this was because of an improved ESR or what.
regards,
Mike Willegal
Bernie,
The higher quality capacitors are generally closer to the stated value (less % off) and tend to be smoother in operation, you can see this using an oscilloscope. Since the Apple-1 really needs more decoupling caps, using the the better ones helps with relaibility.
FYI: When I use vintage ceramic caps on a Mimeo build, I bake them in a special rig in an air oven for about 30 minutes at 320 degrees F to reset their values.
... but allow me to add from 40 years in the profession that for bypass capacitors, a little more or less capacitance does not matter for their efficiency. More important is ESL and ESR. ESL should be as low as technically possible, leaded parts being worse due to the long leads, so those DIL sockets with built-in leaded capacitors can't achieve best results. ESR should be low enough such that the current spikes can be sourced from the capacitor without too much voltage drop on the power rails it sits on to do this job. ESR too low however also is bad as this reduces the dampening of the LCR network, and having a well dampened network is essential. I did a lot of work on bypass capacitors in digital systems and it got worse over the decades due to the CMOS processes getting faster and faster. Sometimes we had to use special sockets made out of multilayer material with SMD capacitors mounted on them, because the DIL sockets with built-in leaded caps could not do the job. I remember CPLDs from the 1990s which were so fast that the bond wires within already had too much inductance, and if you toggled more than X outputs on one side of the PLCC, some flipflops on that bank would flip erratically. No good for your state machine ! Basically, these CPLD were useless junk and everybody who got burned hated them and avoided them.
Modern IC designs use sophisticated CAD tools to tame their power supply grid, some circuit design tricks taming the rise/fall times, and also the modern packaging (solder bump BGA) helps. But for the Apple-1 we need to work with what we got. I bet that with a better PCB layout, and just a few more bypass caps, there would be no problem with any Apple-1, and they would work just fine.
In post #11, Mike Willegal wrote:
" At one point, a a number of years ago, both Wendall and I experimented with a fast WDC 14MHZ 6502 and it turns out that the rise times were so fast that it caused issues. So going with faster processors will only get you so far. The reason for this experiment was the idea of getting a new batch of WDC 6502's packaged in a white ceramic package similar to the original 6502. We never persued this because of the problems. "
Uncle Bernie's comment to this:
Mike -
I never advised to use any CMOS 6502 in the Apple-1 builds, and I would strongly recommend against trying it anyway, as the power supply grid of the Apple-1 PCB layout simply is unfit to take the beating by the fierce current spikes typical for submicron CMOS ICs. IMHO, without a multilayer PCB it's risky to run a submicron CMOS part like the W65C02S6-TPC14 from a 5V supply, especially with the DIL-40 package. Remember the times when 74ACxxx got so fast the "corner pinning" for the power supply did not work all that well anymore ?
But use of higher speed grade NMOS 6502, as I recommend in the above thread and elsewhere, can be proven to be beneficial for reliability and building success both mathematically and empirically. It simply widens the go/nogo window for the /CAS timing, leading to higher robustness of the DRAM operation, without having any ill effects on the power supply grid, because these old NMOS technologies never had a channel length small enough to produce such fierce current spikes, and two phase dynamic NMOS logic never has any shoot-through currents in its dynamic gates, unlike static logic CMOS designs, which are notorious for that, because both the PMOS and the NMOS section of a CMOS gate may be briefly turned on at the same time, essentially shorting the supply rails with low ohmic MOSFETs. This typically happens during transitions of the gate inputs, unless some other input of that gate is in a state that keeps either the PMOS or NMOS side turned off during that transition. This is why micropower digital CMOS design uses reduced internal supply voltages smaller than the sum of the PMOS and NMOS threshold voltages, which eliminates the shoot-through, for obvious reasons.
In case you did not try yet in your experiments, you could try to run your 65C02 in the Apple-1 from a reduced supply voltage, such as 3.0V (or even 2.5V), which would greatly reduce the current spikes produced by the CMOS based CPU. If this reduction is good enough to make it work, I can't say.
Other than that, your only chance to get 6502 die to mount into these desirable ceramic packages is to remove them from plastic packages. If you live near Silicon Valley you should be able to find one of these specialized companies who provide this service on a routine basis. The bigger semiconductor companies can do that transfer process in-house.There is no 100% success rate due to wire bond issues on "used" pads, but they have some tricks to mitigate that. But it ain't cheap. Possibly cheaper than those greedy price gougers who want $500 upwards for ceramic 6502s, though.
I agree that it appears that the Apple 1 desgin has barely adequate bypassing. Your handiwork with hiding ceramic capacitors in the sockets looks very nice. As you may be aware, machined-pin sockets that had such capacitors factor-installed were actually sold back in the day, and some might even still be available today. Of course, they only had a single capacitor between the diagonal pins for power and ground as is typical for TTL, so it wouldn't help with the more exotic chips.
Digi-Key: Mill-Max sockets with capacitors
I'm listening, Uncle Bernie : ) Great info.
Thank you Bernie for all these precious informations!
In the late 70's, Bel capacitors have been used as bypass capacitors in arcade games and pinball machine too, not only in the Apple 1. Could be a choice of the manufacturer I suppose (stock, availability...).
They maybe tried to improve the reliability by using multilayer ceramic capacitor (the green ones) and some yellow Siemens capacitors in the NTI design.
Aurel
Hi aurel6814,
thanks for the info (your post #17) about the BEL capacitors having been used in arcade video games. So, if an Apple-1 builder wants to get even closer to the originals, just buy an arcade game motherboard off Ebay having them (asking the vendor about the capacitor being BEL or not can help). I know that scene, and those arcade motherboards which don't work anymore are reasonably cheap. Alas, I never did pay attention to the bypass capacitors when I repaired those (with varying degrees of success, those arcade boards can be more tricky to diagnose than any Apple-1 disease).
In general, for the bypass capacitors in the Apple-1, I think we can safely assume now that the "NTI" solution worked somewhat better, because they did never go back to the BEL disc capacitors. Proper power supply bypassing is a very arcane craft and even the most experienced electronics engineers run into problems with it. It's a mixture between technician level lab work, engineering science, and art (as in "artist"). Once a good solution has been found, nobody dares to ever change the PCB layout anymore, or any bypass capacitor, or any other component on the board. The design is "frozen" and ready for production. And then, depending on the numbers built, these nasty bean counters of the purchasing department come along and want to buy cheaper substitutes.
I can tell you one thing: at the mainframe computer manufacturer I did briefly work as an intern, once a part and its manufacturer were "qualified", they never ever changed them. And even humble components like capacitors and inductors (or delay line hybrids based on them) went through a rigorous qualification process for which there existed a team of six people (IIRC the number correctly, has been the 1980s) having their own little lab full of fancy measurement equipment, including Tektronix sampling oscilloscopes with GHz bandwidth and HP Network Analyzers.
And now, as a contrast, imagine the two Steves being in a "candy store" aka surplus place in Silicon Valley picking random capacitors out of fish bowls, in hopes to find the "magical" ones that would save their Apple-1 and their fledgling Apple Computer Company ;-)
(I don't know if it really was like that, but I can imagine, and I know their chances were slim, given that terrible PCB layout. I wonder if the thought ever did cross Woz' mind that they should revise the PCB and add more bypass capacitors besides lowering the inductance of the ground and power rails. But if they had done so, that Apple-1 Rev. B would have worked fine, with no quirks, and we won't have this discussion as it would not be famous and originals would not fetch $1 Million at auctions).
Uncle Bernie
Great work and informations!
Question please. Is that importand for the 24x 100nF ceramic capacitors to be with 10% tolerance? Or with 20% tolernce the results will be the same? Cause i have many axial with 20% tolerance.
Thank you!
In post #19, Zijjel asked:
"Is that important for the 24 x 100nF ceramic capacitors to be with 10% tolerance ?"
Uncle Bernie answers:
No, the tolerance is not important. More important is the RF performance for these capacitors. Try to find at least Y class capacitors. The X class, i.e. X7R ceramic is even better. They should be modern multilayer types. You would need a network analyzer to compare their RF performance, which is hideously expensive (even I had to use the one at the company). So it's a hit-and-miss. Try them out. Your chances are good as in the small radial types they could not use the poorer performing and cheaper ceramics used in the disc capacitors back in the day.
Thank you!
It's been a while since this thread was started by Uncle Bernie, but as I had to make the decision to hide the extra decoupling capacitors in IC sockets, I gave one a try, and now I have done most of them. Here are a few pictures. I also show the underbelly so you can see how I did it.
I have been careful to check every solder joint, as well as the distance between the leads and other pins, so I hope this will turn out fine. Is this worth the extra effort? Will it work? Time will tell!
IMG_8737.jpg
IMG_8720.jpg
IMG_8734.jpg
IMG_8717.jpg
IMG_8715.jpg
... for the multiplexed address bus on the DRAMs.
These also are recommended for a robustly running Apple-1. But some types of DRAMs have higher immunity to ringing address lines than others.
It is not necessary to stuff these six into one IC socket. You can use two sockets with 3 resistors each. 1/8W types, 390 Ohms, will easily fit this way.
- Uncle Bernie
Thanks, alas I thought this wouldn't be possible and I have already soldered the sockets :( I might have to put those resistors under the board. I will probably will put the 3 capacitors for the 2504 under the board as well, as they don't fit directly in sockets (no ground). I also have P-Lab's riser board which can fit the 6 resistors, and might try that first (although this might not be as well tested as putting the resistors directly underneath?).
IMG_1390.jpg
In post #24, "ebruchetz" wrote:
" I also have P-Lab's riser board which can fit the 6 resistors, and might try that first (although this might not be as well tested as putting the resistors directly underneath?) "
Uncle Bernie comments:
I did not do measurements with the P-Lab board yet, to see what they do to the reflections on these PCB traces. Which are ill-defined transmission lines, as there is no ground plane underneath. Instead, the electromagnetic shock wave front generated by the fast Schottky drivers (74S257) interacts with adjacent traces of similarly dubious and poorly defined RF properties. These "unknowns" make it very hard to predict what happens when the resistor location is moved. Measurements with a fast enough oscilloscope are necessary. The whole action with adding damping resistors is more an empirically found measure ... even today it would be hard to simulate what happens there, on the Apple-1 motherboard, in terms of reflections and ringing. There are too many unknowns involved. But these unknowns could be measured, models could be made, and then the behaviour of the Apple-1's multiplexed address bus could be simulated in a CAD tool. Costs to do this are probably in the five figures if you have the lab equipment and the software. Six figures if you need to buy the equipment and the software. So you can see it's not what we, as hobbyists, could do in terms of "exact engineering science".
So all we can do is to experiment.
Tell us how it goes !
Maybe the P-Lab PCB fixes the problem with the ringing ... note that getting the ringing below the logic thresholds is good enough.
Otherwise you still could add the resistors at the backside of the PCB, or, if you have good eyes and non-shaky hands, you could also solder 390 Ohm SMD resistors (0805 or 1210 form factor) to the underside of the DRAM IC for the leftmost IC socket. These should be flat enough to hide them from sight, so nobody would notice.
- Uncle Bernie
…if I believe some previous comments in this forums and tips :)
I pondered what to do:
- solder resistors under the board
- use the P-LAB board
- do nothing
- and yes, desolder the sockets!
I had some experience desoldering IC sockets in a video terminal (absolutely destroying the poor quality sockets in the process) with success. I decided to get a little more practice by desoldering the type of sockets I used in this build:
- first, on a random vintage board I had around
- second, on my spare Apple 1 PCB
I used a method I have seen recently on YouTube:
- cut the IC socket in half in place
- use a short length of thick copper wire you solder to all the pins on one side of the socket, using solder and flux liberally
- heat the solder, and the heat transmits to all pins
- pull the half of the socket out
- repeat for the other side
- use a lot of flux and solder wick to clean up the solder mess
- use the hated hand solder pump to empty the vias if needed (I found it works ok for this purpose)
- clean up the flux with IPA
In any case, it worked pretty well, and I now have two sockets each with 3 resistors. I hope I didn't mess up the connections, but I measured the pins after resoldering the new sockets. I am only half satisfied with my fitting of the resistors in the sockets: it was a trickier than the capacitors.
The main drawback of the operation, besides spending the time, is that some of the pads on the top side, which are golden in my ENIG build, got a little gray due to flowing solder on the top side. But you can ony tell if you know what you are looking for.
In case this helps anyone, here are some pictures:
IMG_8766.jpg
IMG_8769.jpg
IMG_8770.jpg
IMG_8771.jpg
IMG_8773.jpg
IMG_8775.jpg
IMG_8783.jpg
IMG_8785.jpg
In post #26, 'ebruchetz' wrote:
" I pondered what to do:
- solder resistors under the board
- use the P-LAB board
- do nothing
- and yes, desolder the sockets !"
Uncle Bernie comments:
I would have tried it without adding the resistors first, and then I'd have run my 'MadHammer' program which was specifically designed to provoke the DRAM errors caused by the ringing on the multiplexed address lines (as it turned out, some addresses are more sensitive to the ringing than others). If there are no errors seen, after running ''MadHammer' for maybe a day, the Apple-1 should be reasonable robust for typical use (but don't control a life support system with it ;-)
Still, your nice photos will help other builders to get it right. This is what this forum is for - exchange of tips & tricks !
It has been my experience (and confirmed by many builders using my IC kits) that not all DRAMs are created equal, and some brands and types are more sensitive to the ringing than others. The most robust ones were the Intersil MK4027 types (most were N-3 grades). These rarely needed the damping resistors. So builders can experiment with what works for them.
For those who want to know more about the technical background, here it is:
The problem with the ringing of DRAM signals was a huge issue back in the days. Because typical TTL designers had no practical knowledge of transmission line effects. STTL and ECL designers did (you can't build anything in ECL100k without using impedance controlled wiring throughout, just see the wiring of the Cray-1, pictures are online. This is still wire wrap, mind you ! It works because each of the "wires" actually is a twisted pair with controlled impedance.)
Poorly understanding what phenomenon they were dealing with, these mid to late 1970s era TTL designers then added series resistors at the outputs of the /RAS, /CAS and multiplexed address line drivers of the DRAMs. Typically 33 Ohms or such. Not an exact impedance match, but it slowed down the rise and fall times enough to mitigate the reflections and the ringing caused by them. Clever IC manufacturers then added these series resistors in their IC meant to drive DRAMs. And the problem largely went away.
I think that Woz knew what he was doing, because the distance between the 74S257 drivers of the multiplexed address bus to the edge connector is about 11", which is close to the maximum trace length recommended by TI application notes for their 74Sxxx family. Alas, on the Apple-1 layout, the multiplexed address bus traces run from these drivers to the edge connector, and near it fork out into the DRAM rows, just to run back from there, and with no termination resistors at the ends of the traces. This increases total trace length far beyond the limit deemed safe by TI. And it leads to weird ringing patterns as the reflected waves travel back to the fork out points and mingle there. It's quite an intractable mess which would be very time consuming (expensive) to model and simulate.
One implication here is that you can't use the Apple-1 "slot" nor the edge connector to extend the length of the bus by any useful amount, the multiplexed address lines being the worst offender, as they are being driven by STTL. You need to put drivers/buffers in to play it safe. On the other side of these drivers/buffers you can then add a "bus board" with many slots, similar to the arrangement seen on the Apple II.
I think this may have been the reason why the P-Lab riser card has the damping resistors there. It may mitigate the problem for additional plug-ins using the multiplexed address bus. But this is only a conjecture at the moment, as I did not measure this arrangement yet. But as the saying goes, there are many ways to skin a cat. As long as any damping resistors added bring the ringing amplitudes down enough to not trip the input buffers of the DRAMs more than once (the wanted signal) then the DRAM operation will not be affected by the ringing. And as every DRAM manufacturer used their own input circuits, the different tolerance for the ringing exhibited by different brands of DRAMs can be explained. I think that using an input latch for the row and column addresses gated by /RAS or /CAS would greatly mitigate the problem, but I can't say if any DRAM manufacturer used that. I can only speculate here based on the setup and hold time specs from the datasheets. Zero (or close to zero) address hold time referred to /RAS or /CAS active edge would hint at a latch being present. What happens without the latch is that the ringing, if it trips the input buffers more than once after /RAS active edge, would discharge more than one row address line in the memory array, and the amount of discharge depends on the duration and number of these ringing events. If two or more row address lines are discharged to their active level, then the contents of these rows (all the colums of all the activated rows!) gets corrupted unless all the bits in a column of said rows are the same. This is a failure mode that is not easily captured by typical DRAM test programs. They may need excessively long runs to spot this type of DRAM error.
- Uncle Bernie