A Practical Discrete 386

There are some chips that no matter how much the industry moves away from them still remain, exerting a hold decades after the ranges they once sat alongside have left the building. Such a chip is the 386, not the 80386 microprocessor you were expecting but the LM386, a small 8-pin DIP audio amplifier that’s as old as the Ark. the ‘386 can still be found in places where a small loudspeaker needs to be powered from a battery. SolderSmoke listener [Dave] undertook an interesting exercise with the LM386, reproducing it from discrete components. It’s a handy small discrete audio amplifier if you want one, but it’s also an interesting exercise in understanding analogue circuits even if you don’t work with them every day.

A basic circuit can be found in the LM386 data sheet (PDF), but as is always the case with such things it contains some simplifications. The discrete circuit has a few differences in the biasing arrangements particularly when it comes to replacing a pair of diodes with a transistor, and to make up for not being on the same chip it requires that the biasing transistors must be thermally coupled. Circuit configurations such as this one were once commonplace but have been replaced first by linear ICs such as the LM386 and more recently by IC-based switching amplifiers. It’s thus instructive to take a look at it and gain some understanding. If you’d like to know more, it’s a chip we’ve covered in detail.

35 thoughts on “A Practical Discrete 386

    1. “not the 80386 microprocessor you were expecting but the LM386”

      They did it on purpose.
      I think that just underlines lack of respect for the readers!
      Let them burn in clickbait hell!

      1. I wasn’t confused. Though perhaps because Isaw this somewhere else first.

        But like I said, I remember when the LM380 was new, and better than what came before. So I knew about the 386 before any 16 bit processor.

        And it’s silly to make ICs put of transistors.Transistors on ICs are close to free, so evenanaudio amplifier will have more transistors than if you were building from scratch. Allkinds of tricks in ICs that are because it’s an IC

    2. It would be more interesting if the linked article actually built a discrete LM386. Bill said he was going to build a discrete LM386, but just designed a push-pull amp, which he admits is only part of what’s in a 386. Dave designed an amp, and made no claims it was based on the 386.

      The schematic in the Hackaday picture isn’t in the linked article, either.

    3. >Is this a clickbait title?

      Of course it is, although the schematic shown was a dead giveaway about which part it was referring to.
      About the LM386, that is not a very good part wrt noise, distortion and power efficiency, it just gets the job done very cheaply. If you already have some in your drawers it’s fine to use it, but if I had to buy it for some projects I’d rather go to newer much better parts.

  1. building expectations – I, as most of readers, was immediately imagining a dense forest of transistors emulating the inner guts of a 386 processor – to raise interest then deceiving people might be fun for the authors, but it´s a flawed editorial practice.

  2. Ha i smiled with the title, being puzzled that a discrete 80386 would be impossible and smiled some more with the joke. Don’t be so harsh and offended at a little play. Great article and yes you got me!

    1. Indeed, nothing wrong with a bit of fun, and its always nice to see the usually hidden analogue circuits revealed so its quite a nice useful short joke article. However it didn’t get me, but that is just because 386 didn’t immediately mean anything to me either way, it probably should have though – so I’m just going to put it down to lack of sufficient thinking lubricant. Clearly need anther cuppa.

    2. 275000 transistors for the 80386 is however not too unreasonable.

      With sot23-6 packages containing two transistors each, it wouldn’t be too huge.
      At 3×3.2 mm (including its keep out zone) in size then 275k of these would just be 2.64 m^2 of single sided board. We can reserve the backside for the “few” passives that will be needed.

      But it isn’t too excessively huge to be fair.

      If we make it a stack of boards, then it can be much more compact. (and likely faster.)
      If using 1.2mm thick PCBs and 1.1mm tall sot23 packages on either side. (passives are likely thinner) Then it is just 3.4mm per board.

      So if we want it A4 sized, 295 x 210 mm per board. Then it will only be 146.2mm tall. Not all that huge to be fair.

      Except, the boards will need to have space for interconnects linking them together. So might need to add a few more boards to the stack up to space things out.

      In the end.
      It isn’t impossible.
      Just silly.

      1. I think the same. It’s quite some work, but doable.

        Size would be that of a wall closet – if regular 2N2222 or BC548 were used.
        SMD types could reduce size to that of a closet.

        A slower clock rate for such a discreet version could be used, also.
        A few hundred KHz to a few MHz, maybe.

        It could still be quicker than an 8086 at same clock.
        Thus, totally usable to run DOS, PC-MOS/386, Windows 3 or PC GEOS.

        But of course, the title/headline was a trap. My first thought was: “386 processor? No wait, that be 80386. They mean an LM386, surely.. Old people often omit the LM prefix..”

    1. This is a complementary output stage, but back then it was not possible to create NPN and PNP with similar characteristics on one chip, PNPs always had higher internal resistance and could not carry enough current. The NPN-PNP pair (also called Sziklai pair) behaves like a PNP, but a better performing NPN does the heavy lifting. This method was also used in many discrete circuits with beefier NPNs like the 2N3055.

  3. Back when the ‘386 amp was contemporary I was always impressed at the incredible hiss level when used as a headphone amp, whereas “pro gear” could have pretty quiet headphone outputs. I was pretty depressed when I found one “pro mixer” produced hiss-less headphone audio by shoving about 10 watts thru a 30 dB attenuator; that’ll kill 30dB of high frequency hiss, LOL. A bit ecologically wasteful LOL, but it worked at the time.

    I’ve always wondered why the 386 was incredibly hissy when driving a high impedance headphone type load. I mean, I know why amps in general are with that kind of load in general, but the 386 performance was just abysmally bad compared to the competition. Any electronic project of the 80s/90s was instantly identifiable as using a 386 if the headphones hissed continuously like a den of snakes.

  4. If anyone can read schematics, they would know this was not about any CPU. The LM386 has been around for a very long time and has been very useful. The cool part about the article is matching the transistors based on temperature. I think adding them to the same heatsink would do the same.

    1. When analog designers first started trying to get accurate transistor pairs, this is what they did: hand-selected and stuck them to a heatsink. Then for a while they started fabricating pairs of transistors out of one piece of silicon. And you can see where THAT led.

  5. With my experience starting with transistors building the 386s, 555s, 78xxs & 79xxs, and then learning math with the 74xxxs… I learned to group the circuits into macro circuits.

    So, yeah. Even the up to 8088s, x86, and some even more complex chips, I learned how the details of what I could do with these.

    I love when younger generations rediscover the development of the previous.

  6. “gain some understanding” I see what you did there! 😉
    The first time I saw an internal schematic in a datasheet I wondered how hard it would be to build a copy from discrete parts. Looking forward to checking out the article.

Leave a 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.