Is it possible to reverse engineer an analogue IC from a physical example?

Question

Lately I've been investigating custom ICs developed by Roland in the late 70s/Early 80s for their analogue synthesizers.

They stopped manufacturing these components around 1989, there are no datasheets available and they will not release, or do not have information on them.

The IR3109 is a DIP16 filter chip made from four cascaded OTAs and buffers controlled by one exponential converter. A rough internal diagram is provided in some synth service manuals, and people have made clones that sound pretty close.

The one I'm interested in is the IR3R01, a DIP16 'envelope generator' chip. Used to create a DC voltage out in response to keys being pressed on a keyboard and applied to the filter or amplifier.

I was wondering if it is possible to examine these ICs somehow and work out what components and values are inside. Maybe exposing the die and evaluating it under an electron microscope? I'm sure if it is possible it would be very expensive.

Answer 1

Of course it's possible. There are many companies that provide these services. The real question is whether or not you could do this at home.

You might get away without needing a SEM (Scanning Electron Microscope), that design might be done in ~3u geometry which would imagable using visible light.

You'll need a wet bench to etch off layers, like HF for SiO2 but you will also have to remove Si3N4, SiON and Aluminum. It's possible you may need a dry etch (Ar plasma in a vacuum chamber) to remove tungsten plugs in vias.

Your main issues will be measuring the exact values of resistors and capacitors (if there are any). Delineating the boundaries of substrate implants (decoration with more nasty chemicals in a wet bench) and determination of doping profiles. The doping profiles are easily obtained in a SIMS unit (Secondary Ion Mass spectrometer) but some of the structural details of implants in the FEOL (Front End of Line) can be subtle.

There will be subtle layer thicknesses that will need to measured before they are damaged or reduced in thickness by the wet etches.

There will be significant topography of the surface of the die (CMP didn't exist then) so depth of focus might complicate picture taking.

It would be unlikely that you'd be able to get the exact transistor characteristics that the original chip had easily. You'd really need to understand not just processing but transistor physics and the role of different implants.

On the positive side, if you had multiple chips (which you would need) you might be able to liberate access to a transistor and be able to put it on a curve tracer to measure directly. The feature size is large enough and being an analog chip it would probably have some large transistors in it. But there is no certainty in that.

The other good news is that you can buy old SEM's for low cost. Only a few $10K and even though they are grainy this chip has large features. Mind you if you have a SIMS unit that also can image (it is a modified SEM) so you might get away without duplicating eqt.

Answer 2

Your last paragraph is basically correct: you can image the chip and then either copy it directly or reverse engineer it to produce a more modern version. The first step might be doable by a university lab with the correct equipment, but the reproduction is not going to be a cheap process (hundreds of thousands of $).

Getting an "exact" sound reproduction might not even be possible with modern IC processes.

The other alternative is to characterise its analog behaviour as a black box, then emulate it with a DSP. People are unlikely to be happy with this solution.

Answer 3

There's a few interesting things in patent documents (I believe there's some great stuff by Bob Moog in his patents) so you might strike lucky there.

As mentioned above, there's ways to "de-cap" a chip and read its structure, although you may have to do a fair bit of reverse-engineering, guesswork, etc. to replicate the thing exactly when it's an analogue part.

You can take the black-box approach of not caring what is inside the box, but qualifying its behaviour and trying to replicate it. This may or may not be a hiding to nothing, it could work great, it could be a huge pain, or you could invent something even better by accident along the way.

There's also (I'm sure) ways/software to qualify the response and then squirt that model into a DSP, microcontroller, or somesuch. Of course, purists won't like that, and it is kinda cheating.

TBH your description sounds like it's a fairly simple function, so the "black box" approach of ignoring the insides and just developing a circuit that does the same thing may be the easiest option.