Hi,
first at all sorry for writing in English. I am the curator of the 8bit-museum.de, a website that exists for 25 years now, and the developer of the memory tester you are talking about. In case you have any questions please feel free to ask me about it, but let me give you some information about it.
I started developing a tester for SRAMs for the Arduino about two years ago to test some memory chips from an EACA Video Genie and the Commodore PET, because the popular TL866 does not support these types. Shortly afterwards I developed a circuit board for 4116 DRAMs, also for the Arduino. Since several supply voltages are required, I added a power supply shortly afterwards.
However, it quickly became apparent that pretty much every computer manufacturer used its own RAM types, so that more RAM types were added, e.g. for the Sinclair Spectrum the half-defective 3732. The Arduino tester quickly became too impractical (too little space), so that I have now reached Rev.7 and Rev.8 over several versions (these two versions are almost identical, but the 8 series allows a ZIF32 socket (I also plan to develop these as 7 series = ZIF28 and 8 series = ZIF32).
I saw that the discussion arose about two things:
1. The accuracy of a memory tester
2. The cost of the boards
Pabbe has already posted a message from me about this. Accurate tests simply cannot be done with an 8 EUR chip. To test RAS/CAS timings exactly, 1-5ns triggering would be necessary, but the ATmega already needs almost 60ns for a command. With the above we would be in the GHz range, i.e. a two-layer board would no longer be sufficient and much more expensive components must be used. Without counting the development costs, we would reach several hundred EUR retail price, with the small quantities and development costs probably even well over 1000 EUR.
What would you have won then? You can then test 0.5%-1% more closely, i.e. in the event of an error, one more chip would be correcty recognized or a false positive would mean that one out of a hundred chips would be rejected. This is unnecessary for a hobbyist and makes no sense.
I write on my homepage that I "deliver the boards at cost price". Strictly speaking, this is not correct. If I actually include all(!) costs (including the failed attempts with faulty designed boards, burned components or unsold boards), the boards would be much more expensive. So the price covers the current costs, i.e. I order a set of boards (I have to buy a minimum amount) and keep one for myself and sell the rest. The demand is not so high that every board always goes away. So I have to add a little buffer to avoid slipping too much, because then I could no longer finance the development. The time invested for the hardware (and we haven't talked about the software yet) is not included in the calculation (small fact: I currently have over 12 different versions of my memory tester which I strictly speaking do not need
)
In total, over 1000 hours have gone into the development of the tester and I am constantly improving it.
In the meantime, he can not only test SRAMs and DRAMs, but also identify EPROMs (very useful if the label is no longer readable, unfortunately only possible from 2732), he can identify (EP)ROM content (e.g. if you find an EPROM and do not know which software is on it), he can test several TTL/CMOS devices.
During the development, I tried to choose a middle way between cheap but still useful and accurate.
With all components it should be less than 50 EUR when you build it. With the documentation and the simplicity, you also have the option of programming something yourself based on it. I think that I have managed to find a good compromise.
