Ken Shirriff Breaks Open The Yamaha DX7

For better or worse, this synthesizer was king in the 1980s music scene. Sure, there had been synthesizers before, but none acheived the sudden popularity of Yamaha’s DX7. “Take on Me?” “Highway to the Dangerzone”?  That harmonica solo in “What’s Love Got to Do With It?”  All DX7. This synth was everywhere in pop music at the time, and now we can all get some insight from taking a look at this de-capped chip from [Ken Shirriff].

To be clear, by “look” that’s exactly what we mean in this case, as [Ken] is reverse-engineering the YM21280 — the waveform generator of the DX7 — from photos. He took around 100 photos of the de-capped chip with a microscope, composited them, and then analyzed them painstakingly. The detail in his report is remarkable as he is able to show individual logic gates thanks to his powerful microscope. From there he can show exactly how the chip works down to each individual adder and array of memory.

[Ken]’s hope is that this work improves the understanding of the Yamaha DX7 chips enough to build more accurate emulators. Yamaha stopped producing the synthesizer in 1989 but its ubiquity makes it a popular, if niche, platform for music even today. Of course you don’t need a synthesizer to make excellent music. The next pop culture trend, grunge, essentially was a rebellion to the 80s explosion of synths and neon colors and we’ve seen some unique ways of exploring this era of music as well.

Thanks to [Folkert] for the tip!

23 thoughts on “Ken Shirriff Breaks Open The Yamaha DX7

    1. I would suppose that depends on if it is 100% digital. If it has any analog, I don’t think there are any FPGAs that can do analog, they are gate arrays after all. Some FPGAs have DACs/ADCs but you still wouldn’t be able to build things like filters, mixers etc. You could perhaps build up the digital in the fpga and build the analog in discreet components outside the FPGA. I believe that is the whole problem with people trying to recreate things like the C64 SID, it has analog functions in the chip, and you can’t make that in an FPGA

      1. Cyress makes a PSOC that has a significant number of analog blocks that can be configured with the rest of the fabric. The demo boards are cheap, small and come with it’s own (detachable) USB programmer. The software is free also.

    2. There’s probably a non-insignificant amount of analog circuitry in there, but it would still be easy to have the digital portion in the FPGA and build just the analog circuitry with discrete components.

      1. It seems to be digital right to the very end.

        Makes sense, since it’s much harder to make an analog circuit that will survive process variations without expensive tuning and analog components cost much more chip area. If you can do it in digital logic, do it in digital logic.

        I’ve seen speech synthesis chips which took a similar approach, generating speech with a 10-stage signal-forming pipeline that was entirely digital. Nothing but digital filters replicating all the stages of the vocal tract.

        1. The DX7 is all digital until the output. The original had 10-bit lookup talbes for the internal math, a funny sampling rate, and maybe only 12 bits? on the output DACs, which make them sound a little bit lo-fi. Which a ton of people love.

          The DX7 II, which followed by a few years, doubled the voices, but also improved the DACs. You can _really_ hear the difference in the highs. It’s a lot cleaner, some complain about “sterile”. While they both sound “digital”, it sounds digitaler. (I’ve had both in my house recently, and got to A/B them with “the same” patches on them.)

          Dexed is great. (https://asb2m10.github.io/dexed/) It has a low-fi mode that mimics everything but the output DACs of the old DX7. Plus, it does double duty as a patch-editor and librarian if you’ve got a real DX sitting around.

          I could go on. The DX7 hit a world of lowpass filtered Moogs with a crazy dose of high frequency FM madness, and most of the period patches are waaaaay too bright and jangly for my taste. But that was novel at the time, and it sold units.

          Check out Aphex Twin’s or Brian Eno’s DX7 patches if you’d like a little subtlety. Boards of Canada leaned heavily on them too. It doesn’t always have to sound like Tina Turner’s harmonica solo.

          1. Doubling the voice with a bit of delay will tone down the highs a bit. The E! Expansion had a doubling with random detune, that managed to warm things up. many of the bands I played with complained about that, but once I got the spx90 , it made the synth sound like analog and then some.

    3. [Ken Shirriff] has done a wonderful job reverse engineering a lot of this chip. There is however at least one other custom chip in the DX7 and [Ken Shirriff] still has more to go reversing this chip.

      So on the horizon, yes perhaps we will be able to put most of the DX7 into a FPGA But without the people like [Ken Shirriff] and the great efforts they put in to preserver engineering history we would probably have no chance.

  1. I own the first generation DX7 and the sound is definitely more analog than digital. I know it is digital by heart but the sound is very earthly and warm. Also the noise adds the charm :)

    Dexed VSTI is pretty good too but it is missing some of the vibes.

  2. Great work by [Ken Shirriff] as always.
    It brings back one of my most memorable days, seeing a DX7 in the window of my local music shop on my way home from work, going in with no cash at all and walking out with the DX7, the paperwork for an HP agreement, and even my bus fare home courtesy of the kind sales assistant who doubted my ability to walk the remaining 2 miles home with it :-)

  3. While it was a major technological step forward, the DX7 was difficult to program. That’s why music producers of the 80′ used only a few presets which resulted in that very poor sound creativity at that time : they all sounded the same !

    1. That was true for the analog synths of the 1980s as well. There is a old leaked report, I think from Sequential Circuits, showing that after a big survey of machines getting repaired at SC’s shops, a vast majority still had all of the original factory presets.

      Synth programming is talked about way more than it is done!

      1. Agree !
        The analog Roland TR/TB series were also responsible for the standardization of the sound of the 80′.
        Synth programming was too much time-consuming for mass-produced cheap “music”…
        This brings me memories about K. Emerson patching his massive Moog modular at the begining of a concert and JM Jarre talking about warming his analog synths with hair-dryers before concert to avoid tuning-shift !

  4. I have one of the original, with the e! Expansion. The doubling and detuning of the voices makes it much more warmer. The 12 bit da/10bit wave table look up as well as 35ks makes for some interesting sounds. I once made a demo to get my gtr player away from becoming a malmstein clone (Paganini riffed maniac) . I ran the dx7 thru a scholz rockman fully compressed will speaker melting overdrive. The aliasing from the 10 bit wave table while pitch bending created fret buzz so realistic that my gtr player wanted to know who the gtr player was. And yes I had a bunch of Paganini riffs ( easy to play on keys but not on gtr) anyway after telling him it was me, he started to change his riffing style. Unfortunately the next generation synths (I have the tx802) have 16/12(?) Won’t alias . I still use the old dx7 as well as a dx9, the later for baselines.
    The tx802 also sounds quite a bit different than any of the virtiual clones (fm7 or fm8) cleaner in someways. Not to sure why, except maybe they haven’t perfected the emulation. Would love to buy either the dx5 or the holy grail of fm, dx1 with poly after touch.

Leave a Reply to CMH62Cancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.