Apple 1 shift registers

I think a replacment using shift registers like the AM1403(2x512) would be more practical (you can combine them for 1024 bits, and because it's a DIP-8 you could hide wiring under the pcb or in the socket)

In post #3, 'Maxplayz' wrote:

" Yeah that could work but the Am1403A is still kinda expensive, about £20 for one of them which is kinda ridiculous. I may have to try and find another way to do this ... "

Uncle Bernie comments:

These 1k x 1 dynamic PMOS shift registers were technically obsolete even back in Y1976 when Woz designed the Apple-1. This is why they were so cheap (back in the day, relative to other semiconductor memory).

I agree that £20 is a ridicolous price, but if you factor in inflation you will see that it's about the same (in purchasing power) as back in the day. IC brokers know that. Imagine, these ICs were sitting in a warehouse for almost half a century, and this space did cost money, and on top of that, the inflation must be factored in.

1 US$ of Y1976, according to the official numbers, which are bogus, is worth $5.58 today - with honest inflation numbers it would be more, maybe twice that, but we have to work with whatever official numbers are available. The Apple-1 was sold for $666.66, which is equivalent to $3720 of today's 'green toilet paper'. Compare which kind of computing power you can buy today for $3720 and stand in awe of the progress in semiconductor technology during these 50 years (but they did run out of steam, and there won't be this kind of progress anymore, unless we re-invent the Laws of Physics).

Just sayin' to get the price perception right.

As for the commercial aspect, the recent 5+ years of Apple-1 building around the world has pretty much cleaned out all the sources for pin compatible 1k x 1 shift registers worldwide. The irony is that the Am2804 is used in some U.S. weapons systems (I don't know which) so a quantity of them sits somewhere in the - you guessed it - "National Stockpile" of the DoD. From time to time the ICs in that "stockpile" are auctioned off due to age, and this is where they become available again in larger quantities (typcally, thousands). The "stockpile" is then replenished, or not. (There are specialized companies here in the USA which are able to run wafer lots using long obsolete process technologies, but I'd bet they use more modern tools, and the prices are ludicrous = military grade, I know that because I did ask them). Here is one link to prove what I say:

https://www.lintechcomponents.com/product/010305280/AM2804HMC/34335

My own journey to source these shift registers for my famous kits (now all sold out) was very time consuming and bizarre, and the AM2804 I had were from this "National Stockpile", at least this is what the IC broker who found them for me claimed to justify his price. Which was below $5 a piece, however. (I don't think the hermetic versions from the link above would be as cheap, so let go any hope).

One thing I remember dimly is that years ago, one IC broker (IIRC, here in the US) had thousands of  type 1404 shift register is stock, alas, in the TO can, which causes awkward wiring to fit into the DIL-8 footprint. This is why I did not buy them. I do not even remember the part number. TO package, being hermetic, these would be much more likely to work after half a century in storage, though.

One thing you should be aware of is that regardless of what kind of 1k shift register you buy, be careful, as they may be wonky and may not work anymore (I had some lots of mid 1970s National Semiconductor MM1404 of which 50% were bad - but they did only cost about $1.30 a piece, so even with 50% loss it was a great deal). The reason for this is that back in the late 1960s and early 1970s the semiconductor industry did not yet understand the problem of Sodium contamination in wafer fabs. Sodium is in human sweat. And a few Sodium ions in a process water bath can cause later failures of MOS ICs due to a phenomenon called "mobile ions". These contaminated ICs may work until EOL test, and then die after weeks, months, years. Intel (yes, that Intel !) was a pioneering semiconductor company who researched the issue down to the root cause - human sweat and skin particles carrying sodium. The problem was considered to be fully understood and avoidable by proper procedures and fab operator training from the early 1980s onwards, everywhere in the semiconductor industry.  Every MOS IC made much earlier must be suspect.

Oh, and better avoid Chinese sellers of such ICs (or any other rare IC): the nice "Signetics 1404" in the vintage looking grey plastic package they sold a few years ago turned out to be re-stamped Signetics 555 timers - which only get hot and melt the IC socket when put into an Apple-1.

I think that the only viable substitution for these 1k shift registers using more readily available ICs is use of SRAMs and a counter which cycles through 1024 addresses. You'd need a little bit of logic around it to separate the input and output bit streams but this is trivial to design. There is a certain irony that modern FPGAs are not very well suited to implement such a primitive function. Sure, it's doable, but with a lot of complications revolving around reduced power supply voltages, level translation, etc.

Guard your Apple-1 well - it's getting more and more difficult to find these rare ICs which were designed in the late 1960s (yes , nineteensixties - these simple and dumb shift registers were the first PMOS ICs produced and sold in larger quantities).

- Uncle Bernie

In post #5, "Maxplayz" wrote:

" if there is a replica for the Signetics 2519 then i dont see why there shouldnt be one for the 2504, AM1403, AM2804PC etc "

Uncle Bernie comments:

The 2519 only has a  shift register length of 40, so we are lucky that CD7XXX family ICs exist which happen to feature a programmable shift register length, programmable down from 64 to 40.

But the 1404/2504 has a shift register length of 1024, and you don't want to cascade these CD7XXX shift register IC to reach 1024 bits - you would need lots of them.

Hence, the only viable way towards a substitute (I've pointed out to you in my post #4) is to use SRAM and a 10 bit counter. It does not even matter how it counts.

You could emulate these shift registers by software, if you use a fast microcontroller, such as a ST32, but once you have it why not emulate the whole Apple-1 on it ?

You see there are many ways to skin that cat. But you - or somebody else - need to do the work.

- Uncle Bernie

I've been buying AM1404APC shift registers on Aliexpress for quite a while and in that time the price has increased 20 times!

In 2019 I paid then about 1$ per piece, and last year I saw an offer already at 19.95$. The latter generosity of Chinese sellers I wisely refused. If I had invested in them, then I think I could sell them for 25$ now. Missed opportunities, as we say, if I had seeing cards upside down I would have lived in Sochi...

Indirectly, this proves that Woz did borrow Don Lancaster's TV Typewriter circuitry for the Apple-1. Or at least "took inspiration", a very fashionable phrase I often hear here and there. Otherwise it is very difficult to explain how a brilliant young engineer like Woz used technically obsolete components in a circuit fundamentally very similar to the one Don Lancaster published in Radio-Electronics magazine in September 1973.

Maxplayz wrote:

Yeah same with me. I paid like £20.48 for a single piece of the MM1404AN. quite ridiculous. But i am planning my own external circuit which uses the LM7660 for -5V on the VDD pins of the shift registers and 7x SRAM ICs for the 7 shift registers that the apple 1 uses.

I can say that the emergence of a 2504 replacement on modern components will definitely bring the hoarders down to earth with their appetites. This was already the case when not expensive replacements for 2513 and 2519 appeared in open source projects.

In post #8, "macintosh_nik" wrote:

" Indirectly, this proves that Woz did borrow Don Lancaster's TV Typewriter circuitry for the Apple-1."

Uncle Bernie comments:

Nope, this does not prove anything. I've studied both the TV Typewriter circuits of 1973 and the Apple-1 circuits of 1976 very thoroughly and they have nothing in common except the overall video data path architecture published by Signetics for video terminals based on their shift register ICs. Here is a snip from their 1972 MOS Databook:

Video_Gen.JPG

You can see that the Signetics 2503, 2504, 2505, 2512, 2513, 2518, 2519 were a family of ICs tailored to be building blocks for video terminals. And this is all the true story to it. Neither Don Lancaster nor Woz did invent that architecture, it was Signetics, and as far as I was told the story from older IC designers who had worked there in the early 1970s,  their MOS venture was largely driven by the bipolar IC wars that begun in the mid 1960s and eroded Signetics market share.

Here is the cover sheet of their 1972 MOS IC Data Book:

Signetics_1972.JPG

in which you can find datasheets for all the Signetics MOS ICs used in the Apple-1, and most of them even come with transistor level schematics, so if you can interpret PMOS and NMOS dynamic logic schematics, you can understand all the inner workings of these ICs, and this certainly could help to make substitutes. The 2504 is a bit tricky how the "doubled pumped" method works, it actually has 2 x 512 bit dynamic shift registers inside and  these are identical except that the two clock phases are reversed between the two blocks, and the resulting two data streams are then combined into one with a final gate, where the clock phases select from which shift register the data bit is taken and given to the output pin driver.

If you understand how this works then it would be easy to design a substitute based on a RAM, a counter, a hex register, and a hex data bus driver. You need to make the single clock for that from both the PHI3 and PHI4 phases, but again, this is trivial to do.

- Uncle Bernie

From time to time you find chips like this:https://www.mouser.com/catalog/specsheets/intersil_hsp9501.pdf

Got 2 pieces below 10€ from china.

I did not test it but one HSP9501 should be able to replace all 2504.

And a second HSP9501 should be able to replace the 2519. 

Unfortunately the PLCC Formfactor is not very helpful.

When I have some free time I will try to build a proof of concept.

Be warned this chip sis as well hard to get.

If you plan to create a solder below the board adapter C4 74157 would the best option it connects to 4 of the 2504 and to the C3 2519.

For the other 3 you would need to create connections to the 2 sockets D14A,B and C11B.

In theory C11A the DS0025 would not be needed as well.

In post #14, 'natas666' wrote:

" From time to time you find chips like this:https://www.mouser.com/catalog/specsheets/intersil_hsp9501.pdf

Got 2 pieces below 10€ from china. "

Uncle Bernie comments:

Right there you admit two problems: First, the part is obsolete, and second, you bought IC from China, which is very very risky because of the counterfeiters.

Alas, I can't help with the obsolescence problem . . . when I looked into 2504 substitute circuits (NOT: drop in substitutes, but some PCB with some ICs), the things called "FIFO"s were my favorite, because on many FIFOs, you can connect the input clock and the output clock together and then the FIFO turns into a shift register of the given FIFO length. No extra logic needed.

But here is the problem: small FIFOs of the 1024 x N size (N >= 7) which can run in 5V TTL systems are obsolete, too, and can be hard to find. Distributors like DIGIKEY and MOUSER don't have them anymore, and if you search for FIFOs of 1024 x N you end up with "marketplace" products where you might need to buy large lots.

Here is the best alternative I found (AM7202 1kx9 FIFO):

https://www.silicon-ark.co.uk/am7202-50rc-fifo-ram-amd

... which appears to be more hobbyist friendly than the DIGIKEY marketplace where minimum order size for them is 36.

But I never experimented with that, and I still think that a multi chip design based on small SRAMs (like the 6116) and a counter and some latches and gates would be more sustainable on the long run. And designing such a SRAM based substitute is certainly more educational than just using a ready made FIFO in a mode it gets turned into a shift register.

- Uncle Bernie

Nice find to substitute the 2504.

There are more chips like the one I found that have been made for digital video applications.

What I like with the Harris/inersil/Renesas HSP9501 you can fully program the lenth and use it as 40 bit or 1024 bit multichannel shiftregister that support recirculation.

You have a very valid point so far I did not test the chips maybe they are fake (they don't look) or they might not work.

There was also a newer version that was also availible with twice the size can't remeber the type code.

It was LOGIC Devices LF9501 and LF9502.

The AM7202A is made as a generic buffer I found some newer compareable chips but most of them had the problem that writing new data into the buffer while is filled is inhabited, in most cases it should be possible to work around with the pins intended for cascading.

https://www.applefritter.com/files/2025/05/13/HSP9501_0.PDF

https://www.applefritter.com/files/2025/05/13/lf9502_324441.pdf

As the newest datasheet is from 2004 there should be somewhere a big a mount of this chips.

As it was manufactured under 4 company names and two product names and more speed grades and two differente capacity versions.

In post #16, 'natas666' wrote:

" You have a very valid point so far I did not test the chips maybe they are fake (they don't look) or they might not work. "

Uncle Bernie has a hint for you:

Use a Q-Tip, drench it with Acetone (aka nail polish remover) and rub on the surface of the suspect IC. If black paint comes off, it's most likely a fake.

If so, you can repeat the process until all the black paint is gone. Then you inspect the surface of the package under magification (5x or 10 x is enough) and if you see a scratch pattern then it's certainly a fake.

This test works because of the way most Chinese chip counterfeiters work: they first grind off the surface with the original type number, then coat it with black paint, and then laser engrave it with the new type number. So any IC with the same type of package can be turned into any other IC type with the same package.

This is the same test IC brokers use. It is not 100% failsafe but it is quick and cheap to do, and it is efficient.

- Uncle Bernie

P.S.: Oh, and plenty of AM7202 or their equivalents on Ebay right now (I just looked this morning). There also are IDT7202 and MK45H12N25 from ST.  I did not check whether these types can do the "common clock" trick I remember from several FIFOs I used decades ago. It is usually advertised in the datasheet. Otherwise you must design some more elaborate control logic for the FIFO.