AC Analysis of clamping diode effect in 5V to 3.3V mismatch

A lot of people were wondering exactly what testing I did to come to the conclusions from the other post.  That’s a great question!  Here’s what I did that proves out the math from the other article.  Feel free to reference it when you’re double-checking the findings.

My first article on this subject was intentionally written for a non-technical audience. We purposely simplified some of the models, opting for an ideal representation instead of a practical one, in order to produce a message that was accessible to all. Seems like some part of the internet was not happy with that. Therefore, in this video I describe exactly what the clamping diode effect is, measure it on the oscilloscope, show when it happens, and why it’s not such a great idea to design a circuit that does this.

The video starts off with a theoretical explanation, followed by a practical example of the effect at 6:05. The practical example shows the clamping diode effect being measured on D12 of the Sega Genesis while using a Mega Everdrive.

EDIT: The normal logic high voltage of the console, when a real cartrdige was used, is 5V dead on. I’ve seen a few forum posts today that “theorized” the actual console’s output was always closer to 4.6V regardless of what device is connected in the cartridge port.

 

René

Electronics engineer and retrogaming fanatic!

23 thoughts on “AC Analysis of clamping diode effect in 5V to 3.3V mismatch

  • July 13, 2017 at 13:15
    Permalink

    All kinds of testing = didn’t even look at the back of the everdrive to see the ground leads, LOL. Sure sure frenchy.

    Reply
    • July 13, 2017 at 14:50
      Permalink

      I’ll just leave your comment here so everyone can see how rude you are. Also, you are confusing ground leads and ground plane.

      Reply
    • July 13, 2017 at 15:15
      Permalink

      What a brave soul to post anonymously everywhere you go!

      Reply
  • July 13, 2017 at 13:19
    Permalink

    Any chance to add the timing info of the event?
    The hump is it 100nsec? It seemsa round 50% of the full access duration but I can’t make it out.

    Also during /OE high wouldn’t the databus pin of the flash be in HiZ, as such is the ESD clamping diodes all that rest connected?

    Reply
      • July 13, 2017 at 13:47
        Permalink

        I could try to measure that – I need a new and better scope 🙁

        Reply
        • July 14, 2017 at 02:51
          Permalink

          It seems that from the time div (at 250ns) the peak itself lasts close to 200ns … that is a little long .

          Reply
    • July 15, 2017 at 02:08
      Permalink

      The ESD diodes in virtually all chips these days (and on retro machines) are permantly tied to the IO pins.

      They are the main line-of-defense against ESD or voltage transients.

      But, as René said, the diodes are only really intended for very short transients that are above VCC, or below Ground (ie. negative voltage swings).

      The diodes are usually part of the silicon die itself, so made with actual “transistors” of sorts, but often a bit chunkier.

      Yes, the data bus on the cart will (/ should) go High-Z as soon as the /OE or /CS signals on the cart bus as de-asserted, but the protection diodes will still be connected to the console / cart bus at all times.

      Reply
  • July 13, 2017 at 13:58
    Permalink

    The good-ish news is that the current being sourced from the CPU in an earlier Genesis comes a little easier as it’s an NMOS device. For this later 68HC000 CPU, which is CMOS, the risk of damaging the CPU’s outputs are a little higher.

    Reply
  • July 13, 2017 at 15:58
    Permalink

    Appreciate you raising this topic, René. I still see this being contested by the community and since I don’t understand the technical aspect of it I can’t say I know for sure whether to stop using my mega everdrive (but luckily I have a backup machine). I’m glad going forward people will be at least aware of the whole 3.3v/5v mismatch and design with it in mind. Also props for staying classy in the comments section.

    Reply
    • July 13, 2017 at 16:30
      Permalink

      Thanks! I think staying classy is very important. None of this is about hurting people, it’s about engineering.

      Reply
  • July 13, 2017 at 17:59
    Permalink

    I’m an industrial control and automation technician and I can fully appreciate the requirement of “not fudging” the rated specifications of electrical/electronic equipment. Of course, I deal with controls and loads that operate at up to 480 volts, but despite the engineering differences, the discipline is largely the same. In my case, polarization index (winding health) tests are your friends. 🙂 Sure, you can run a compressor motor above full-load amp ratings and it will work “fine”, for a while. But, when it does, don’t crap your pants and then fling it at the people who warned you against exceeding the design guidelines.

    If for no other reason, folks need to be aware of things like this and to have these discussions, even at the risk of out of control trolling.

    Reply
    • July 13, 2017 at 21:28
      Permalink

      This is the best comment I’ve ever read I think. Thanks!

      Reply
  • July 13, 2017 at 22:47
    Permalink

    No, thank you for speaking up! I cracked open my NES Powerpak and Super Everdrive after catching the last Retro Roundtable. Looked as though the Powerpak may be all 5V components, but I could be wrong. Have you looked at it very closely? I picked up a Rigol scope earlier this year, to make carrier frequency plots (15 kHz max) on lower voltage variable frequency drives and speed reference I/O, shaft encoders, etc., but I’m a little out of my element, on what to look for on these rigs.

    Reply
  • July 14, 2017 at 07:03
    Permalink

    Thanks for the video! In general, I was totally unaware of this phenomenon along with the various factors coming into play, and so I am truly grateful to understand more about it.

    While it is interesting to see evidence that the diode-clamp-effect is indeed happening, my understanding from your article is that the real potential for damage to the console comes from over-sized amounts of current flowing where it shouldn’t be. That’s still true, right?

    I posted videos from Krikzz in the comments of the main article showing the difference in total-system power-consumption between a real cart and an Everdrive, with the result being around 30mA as an average in the case of three designs. I understand that there are potentially higher peaks of consumption when zoomed in close enough, but over time, can we say that these are accurate averages? Or is there some other factor at play that we can’t see?

    Krikzz has it that the vast majority of the differences we see when measuring this way are simply due to the normal consumption of the components on the Everdrive via the console’s power rail, where there is plenty of tolerance for that kind of extra draw. This seems to be the next thing to verify.

    For example, it could be that the components on an Everdrive actually consume less than a real cart, and that more than the difference he has measured is being pulled through the bus over time. However, if Krikzz is right, it could be that the amount getting pulled through the bus is a whole order of magnitude less. That’s quite a difference for what seems to be the most important single factor in all of this, as far as the many preservationist retro-gamers with older Everdrives are concerned.

    Of course, how much is really too much is a whole other can of worms.

    Anyway, if all you wanted to do is articulate the potential for danger, then I suppose you’ve done that. If you want to actually prove an existing threat, though, why not just measure the actual power consumption of your Everdrive itself from your Genesis’s power rail? It seems to me that the the total consumption of a console with an Everdrive, minus the consumption of a console with no cart and the consumption of the Everdrive alone, should give us a fairly close guess at how much is actually being pulled through the bus during normal operation.

    Unless, of course, I’m completely missing something.

    Reply
    • July 14, 2017 at 07:39
      Permalink

      The issue has nothing to do with the general power consumption of an Everdrive – that is never what I discussed. Taking an average DC measurement of the console’s, or Everdrive’s, current consumption does nothing to prove or disprove any sort of unnecessary transient power consumption through the bus (i.e. clamping diodes).

      I took the time to explain it in detail here because in my last article, to keep things simple while targeting a non-technical audience, I gave an idealized value for the clamping diode current. This current occurs in shorts spikes as I’ve demonstrated here. Yet some people thought I was predicting that each data and address line would add average DC current that can be measured with a multimeter – this is most definitely not the case. The current through the clamping diodes, as I’ve shown here, can only be measured with an oscilloscope.

      Point being, even if each address and data line on the console side were driving a logic high during many bus cycles you would never be able to measure the sum of their currents with an averaging DC multimeter. The instantaneous current will be very high, but no multimeter can detect that. Furthermore, to get very technical, on bus cycles where there are a very large number of address and data lines being driven to one – the voltage of the 3.3V rail will actually be raised by the additional current being dumped into it, temporarily exceeding the recommended maximum operating voltage of 3.6V for most 3.3V devices.

      Reply
      • July 14, 2017 at 20:37
        Permalink

        http://gendev.spritesmind.net/forum/viewtopic.php?f=13&t=2170#p27335

        In this post, you suppose 7.25mA drawn per pin, times 36 address and data bus pins, divided by a 25% duty cycle for a total of 65.25mA of excess current caused by this diode-clamping effect. Isn’t this an average? If I measured the total current consumption of a Genesis running this Everdrive using a digital multimeter (much like Krikzz did just recently), wouldn’t that 65.25mA be included in the reading? Would it be divided by some other factor that I’m not seeing? Or would it sail past the DMM’s detection completely because of how short the current spikes on logic highs are, or some other reason?

        Maybe this is all beyond my present ability to understand. I don’t know.

        Like I said, if all you want to do is point out bad design, you’ve done it, and I can’t ask anything more of you. If you want to answer the burning question that this has raised, though…namely, how much of a threat old Everdrives really are to irreplaceable consoles…it seems to me that we need to know actual sum averages. Right? After all, isn’t the heat the extra current generates the biggest danger to the console’s parts? Or do spikes of high current cause other problems in the console itself no matter how short and infrequent they are?

        Thanks!

        Reply
  • July 14, 2017 at 12:54
    Permalink

    Still no explanation why even the oldest and having a bad (comparing to the new ones) design devices haven’t killed even single console after 7 years? What’s wrong with those EverDrives? What’s wrong with those consoles? How much time should pass to satisfy the consequences of your advice? Didn’t you think you made any exaggerations? Wouldn’t you like to make a real stress test to prove your horrifications and prove that EverDrive damage anything? So far you don’t know anything about any real damage but you advise people to be afraid of EverDrives. If you see that the current transition scheme is not perfect it’s not the reason to frighten people. Don’t you think so? Why do you spread lies about EverDrives? What’s your motivation?

    Reply
    • July 14, 2017 at 13:03
      Permalink

      Lies? Right, cause I clearly faked those oscilloscope plots 😉

      Reply
  • July 15, 2017 at 22:27
    Permalink

    Hi Rene.
    I would like to clear some things.
    You show current leak and io voltage drops on your oscilloscope, diagram, where extra voltage dissipated at clamp diode, and this is really facts, i completely agree, but right after that you gives your own conclusion about this facts, and your conclusion is unclear a bit. You just saying that we out of specs and we have a problem, what is kinda unclear. Please correct me if i miss something.

    So, according to your own measurements we have around 7mA pulse load on IO, right? But seems like it fits CMOS devices specification, isn’t?
    Also we have 3.9v on IO, what is actually fits in specification for 3.3v devices (VCC + 0.6v).

    First of all i want to clear things about console damage, which outs of your conclusion.
    My questions is:
    1. Why console should be damaged? Transisots will evaporate under 7mA current load, or what will happen? Current in acceptable range, voltage also, then where the problem?
    2. Where exactly we out of chips specs on console hardware side?

    Reply
    • July 17, 2017 at 08:06
      Permalink

      1. There is very little info for I/O sourcing or sinking above 5.3mA in the Motorala 68K datasheet. Typical CMOS ratings are ± 5.2mA.

      2. Vcc + 0.6V is the “Absolute Maximum Rating” at which normal device operation is not guaranteed, which the Flash input is exceeding. The specification to follow is the DC characteristic Vih = Vcc + 0.3V in order to guarantee normal operation in accordance with the datasheet.

      Pulse current load can be very high when more than a few of the data and address lines are logic high.

      Reply
  • July 17, 2017 at 09:06
    Permalink

    1. You saying about typical CMOS ratings which is ± 5.2mA, but where from you get this numbers? According to JEDEC specification CMOS IO ratings should be 20mA at least, datasheets for old 5V memory usually gives same numbers.

    2.No no, at the moment i do not asking about cartridge components, we may discuss about this also, but first we need to clarify situation around console hardware risks. Your article named “The Dangers of 3.3V Flash in Retro Consoles”, and there already lot of scared people which thinks that their consoles in danger. So, i asking again: Where we out of specs for console components? Voltage and IO current notable less than maximum, according to JEDEC interface specification.

    Reply
  • July 21, 2017 at 21:42
    Permalink

    Hi Rene,

    I understand that some of the newer consoles, especially the HDMI-upgraded NES clones, may be using 3.3v NOACs and GOACs. Are there similar dangers to the ROMs of real cartridges with using real cartridges with a 3.3v clone? Should people be using their precious Little Sampsons and Panic Restaurants in their NES clones?

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *