Is this out of the question? Actually, not at all. It all depends upon how much $ you want to spend. Usually, unless you have the small fortune to drop on the price tag of several iterations of photo lithography masks, engineering testing of your revisions, and a run of silicon (you need a LOT of parts sold or used to re-coop your investments), these expeses bar us common enthusiasts from playing in the silicon arena. The engineering expense of creating new custom chip generally is pushing into the millions.
There are, hiowever, some tricks we could use to get some custom silicon created at much reduced cost. But first, you need to decide, analog, digital, mixed signal, MMIC, what are you making? I think for rookies, unless there is a designer around who knows a ton and can contribute, anyways I think CMOS is the best usderstood, easiest to work with, due to the fact that I think it is by far the most used process like ever. Its great for digital stuff, and can be used to make analog as well. the way to get a few parts run for cheap is to pile on to a test wafer with other requestors from who knows, once you have selected your custom house and process geometry. The cost to get on one of these is one heck of a lot less than to try to own the chip and cover entire mask costs. Last I checked, and its been a while, but Maxim is great for doing custom work. They even have online design tools you can use for fee to layout your analog design. I think they probably even have some experts you can get to review your design, and you will want this, have to ask them about costs. I think costs go way down if they believe that you are seriously going to do a run and might hit some volume sales. They are also quite experienced in determining what your interest is., so I would shoot straight. Analog design is as much art as science, but a great reference to study if you have several months full time to devote would be “Microelectronic Circuits” by Sedra/Smith really any edition. This was the text used in my junior level microelectronics class, and I noticed everyone in the analog world has a copy laying around for reference. Pick appropriate topology based on desired functionality, then tune in simulation using process params to determine correct lengths and widths of PN junctions of trannies to obtain desired analog specs.
Digital chips are much easier. Port the resultant edif netlist from your VHDL compile into a place and route tool seeded with your process parameters for selected process, and presto, let the computer drop the millions of gates.
Mixed signal, I dunno. Gonna be some mix of the two. I’d ask from help from the silicon house. If you speak enough of the lingo and they can see it will be quick, they will give you a few sentences of direction. Otherwise, I suspect they will suggest some consultants, possibly even in house. My guess is these guys are not quite what we may call cheap.
But creating a chip just doesn’t make a whole lot of sense unless you need some functionality that you cannot find from presently available options …OR… you have a situation with a high volume product and you have tallied the costs and the savings of creating an ASIC justify the outpouring of cash. Until you can meet either of these criteria, why not use FPGA, DSP, mixed signal programmable chip, hybrid, microprocessor, reference design platform, aruduino, etc? These are always a bit more per unit, but cheap to build with. Hard to beat.
But now, for the last thought rolling about in my grey, and I’m sure where you might like to go, Mr. A. OK, so tech is moved pretty far along down the line since the first CMOS processes started mass producing VLSI for the masses. And makers have gotten pretty good at making automated systems that can print from files. So can we get good with the P/N junction, and devise a system, maybe even its got a rather large process geometiry, but we cooked up a way to form these junctions and somehow ‘print’ our own designs? Well, Mr. A., this is a very interesting topic indeed. Where there is a will, there is a way. What if we could come up with such a system, and drop the price for larger, but fully customizable, methods to create custom circuitry? I’d imagine there is an ocean of profitability somewhere around there… piques my antennae…