New Apple-1 come alive, but sick!

In post #1, "borgmacgb" wrote:

" This video will explain better than me. Look in the middle of the line, the 3 becoming # when the line moves up. "

" Could it be problem with the 2519 ? "

Uncle Bernie comments:

The 2519 is the line buffer and it must work the same for each line on the screen.

Since the problem you observe does not happen on each and every line, the 2519 most likely isn't the culprit - but occasionally I saw very weird behaviour due to bad 2519 you wouldn't believe and I couldn't explain even if I tried.

Diagnosis

When a '3' turns into '#' then the binary code changes from 110011 to 100011 and this means bit #4 flips from 1 to 0 (LSB = bit #0). And the flip happens at the same character location all the time.

There are many possibile culprits. First try this:

Exchange the 2504 at D14A with the one at D14B. If the position of the bad bit moves, the character transformation changes, and then it's a bad 2504.

If this fails to identify the culprit, it will be more difficult to find what causes this, it could be a wonky / marginal signal anywhere in the recirculating screen buffer and associated logic.

Tell us what you find !

-Uncle Bernie

(redited to fix table, grr)

First Uncle Bernie knows more than I do. So take my advice with a grain of salt.

I had a bad 2519 and it showed weirdness. Some of the bits didn't work. So I built a new 2519 board (worked great). 

I diagnosed it with this table

- Clear & Reset worked fine

- SPC ( 0 8 @ H P X are all displayed. These are all first in the row

A0-A2 - Row selectA3-A9 - Column select

|---+---+---+---+---+---+---+---+-------------------------------------|

| @ | A | B | C | D | E | F | G | A7,A8,A9 Do not change <- *PROBLEM* |

| H | I | J | K | L | M | N | O | A0,A1,A2,A3,A4,A5,A6 change         |

| P | Q | R | S | T | U | V | W |                                     |

| X | Y | Z | [ | \ | ] | ^ | _ |                                     |

|   | ! | " | # | $ | % | & | ' |                                     |

| ( | ) | * | + | , | - | . | / |                                     |

| 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |                                     |

| 8 | 9 | : | ; | < | = | > | ? |                                     |

|---+---+---+---+---+---+---+---+-------------------------------------|

I typed in each character, in the same row and noted which ones were correct. As you can see from my notes the @HPX (0 and 8 characters worked. After that some of the remaining characters came up incorrect. I figured out which bits failed and deduced that the 2519 was bad. I have a scope, a pulser and a logic probe (old Radio Shack, my favorite). I used all these tool and the schematic to figure things out. But I think a keyboard, schematic and the above table may help.

Your bit pattern may differ. Also it is possible that another chip inline of the character/video generator has issues. That was difficult circuit to work with and figure out. 

In post #4, 'Borgmacgb' wrote:

" What else could I check to avoid another failure ? "

Uncle Bernie answers:

Thanks for making me laugh ! The truth is, there is nothing you could do, other than avoiding building an Apple-1 clone !

While there are ways to kill your 2504's by having a counterfeit or bad DS0025 clock driver, or by ESD (these early 1970's ICs have very pathetic ESD protection circuits), there is nothing you can do against the "pattern sensitivity" of these early dynamic shift registers.

This is a known industry insider "secret" I wrote about several times here on Applefritter. In a nutshell, these early dynamic shift registers were the first commercially viable MOS ICs. The industry sucked them up in  huge numbers as they were the cheapest way to implement memory at the time being. Now, with these primitive ICs being in high demand, the semiconductor industry knew they could always sell them regardless of the quantity produced. So every wafer fab, if they had a hiccup / loss of yield, would make wafer lot after wafer lot of these shift registers as "pipe cleaners" (industry jargon) to see if the fab problem was fixed.

The consequence of this is that there were many, many lots of these shift registers out there which were manufactured while the wafer fab was still being out of spec, or contaminated. These are "bad" or "dubious" lots.

And guess which are the 2504 which we can still buy today ? These are leftovers which for a reason ended up at IC brokers because the OEM who needed them rejected them as being wonky.

This does not mean that all of the 2504 in a lot are bad. Most will be OK. It just means that there are more bad ones than in the other, accepted, lots. Which of course have been used up and put in products half a century ago.

The 2519 has the same problem. There is only one source left in the world (Manoshevitz Electronics in Israel) and the 1977 date code lot of 2915N they have (with some exceptions) have a very high rate of bad ones. I also wrote about that here on Applefritter. For my early kits (50 or so) I had these 1977 date code 2519N from Manoshevitz (at a good price and not at the usurious $65 they want now) and there were many bad ones. When Manoshevitz started to get greedy and not give me a good price, my American IC broker found yet another stash of 2519B (1976 date codes !!!!) also in Israel and the only downside was that I had to buy all of them and write a biiiiig check. But the failure rate in these 2519B was 10 x less than in the 2519N from Manoshevitz. I got an awesome deal there.

This anecdotal evidence only underlines that we can't expect that these ancient ICs from the dawn of MOS technology have the same quality level than the ICs made later. One example for later shift registers, the AM2804PC shift registers I had in most of my kits were of excellent quality, as they were made by AMD in the 1980s, where Jerry Sanders had pushed the company towards mil-spec quality levels even for "commercial" grade ICs. This was THE decision which catapulted AMD out of the low quality morass which was so typical for the 1980s semiconductor industry. A counter example: I had some MM1404 from National Semiconductors where half of them were bad right out of the tube. But alas, these are the only ones which work with Apple-1 based on LSTTL.

About the "pattern sensitivity":

these early dynamic shift registers were known to have an unfixable flaw, which is "pattern sensitivity".  This means that despite having no manufacturing defects, certain bit patterns shifted through them could cause bit flips. (Ouch !). This effect depends on voltage, temperature, and clock frequency and can't be tested ... 2^1024 bit patterns tested at the specified clock speed would require a test time that makes the test economically unviable ... IIRC (but my memory on this is weak) I once calculated that testing one of these ICs for all possible bit patterns would take longer than the known age of the Universe.

So what do they do ? --- They continued to manufacture and ship these products regardless of the known problem.

Now you have the answer why these long serial shift registers all but disappeared from the marketplace. Even today, with modern wafer fabs, these would be untestable for pattern sensitivity. Just because you can't test all of the possible 2^N patterns. Of course, you could add a test mode to break the long shift register into smaller pieces and scan test them. But would this fix the pattern sensitivity if the IC later runs outside of that test mode ? Hmmmm ... the solution exists and boils down to a lot of extra measures both in the layout design and the process technology.  These "rules" avoid the problem of pattern sensitivity. But back in the day, they had no clue. They were just happy to be able to make MOS ICs at all. The clock speeds in early PMOS processes were pitiful. They could not make any NMOS transistors yet --- despite Damon Kahng had demonstrated it in the early 1960s as an academic exercise (Mr. Kahng, a Korean, who immigrated to the USA in 1955 developed PMOS and NMOS together with Mohamed M. Atalla, an American-Egyptian engineer, it's worthwhile to look up their wiki pages, the are the giants on which all of our current CMOS technology is based).

So what did they do, back in the late 1960s and early 1970s, to make money from the pathetic PMOS technology they had ? They made PMOS ICs for digital wristwatches and early pocket calculators, and the dynamic shift registers seen in the Apple-1. The early microprocessors (Intel 4004 and National Semiconductors IMP-16) all were PMOS. Slow. But there also is the Air Data Computer of the F-14 "Tomcat" figher jet, whose design started in Y1968 and was completed in Y1970.

When Woz designed the Apple-1, all these early PMOS dynamic shift registers were on the way of the dinosaur, towards technical obsolescence, and this is why they were cheap and abundantly available from electronic junk peddlers catering to hobbyists. The early BYTE magazines are full of ads from which you could buy these ICs for cheap. And Woz, always looking for the cheapest possible implementation of his ideas, put them into the Apple-1.

Long story. But I wrote it up for posterity. All these anecdotes will be lost in time once my generation dies off. So I take every opportunity to write about how the early PMOS/NMOS process technologies came on line and which products they spawned.

This also helps Apple-1 builders to understand what they are doing ... building stuff in the 21st Century using ICs made with long obsolete, early MOS ICs which began to be commercially viable more than half a Century ago. This is a bold endavour and I salute everyone who dares to walk that path !

- Uncle Bernie