Miniaturization

Digital: FPGAs + software.

Analogue: there have been many commercial attempts at that over the decades, all have failed. I presume, without evidence, the causes are limited configurability and therefore applications, poor/limited parametrics, single sourced.

Mask costs, Large minimum batch size (about 3 wafers if you want to be sure of getting anything out at the end),

Long lead time (3 months),

Horrifyingly expensive CAD tools,

Easy availability of good-enough alternative at much lower cost (quick-turn PCBs and distributor parts),

And so on.

Not to mention that the number of hobbyists equipped to design analogue or mixed-signal chips of any size can probably be counted on the fingers of no hands. ;)

a-Si transistors are super slow because the disordered lattice scatters high-momentum (i.e. short wavelength) electrons very strongly, so the effective mobility is the pits.

Cheers

Phil Hobbs

Tooling costs dominate.

Naaaah! It's easy to get an account there.

MOSIS has so-called "shuttle runs" where multiple designs are put on a single wafer... reducing costs to a few grand.

Most commercial foundries also offer same.

Bull hockey. The only NRE at the foundry is the mask set. Most of the NRE is in designing the chip and the layout (from whence the mask set is created).

Big bucks usually equates with big name simulators and their associated layout tools... and being gullible enough to assume that big name means clever at circuit design... a fool (management :-) and his money are easily parted >:-}

I know that there is academic research into "printed" chips... but the resolution isn't yet fine enough to be cost-competitive.

I think a more important "printing" would be true "printed" circuit boards... run thru a "printer" and a board ready in minutes. ...Jim Thompson

I saw a machine that did that from the 1940s. It did almost every step of m anufacture of radios etc. PCBs were made by metal deposition from an arc wi th a mask. The only thing humans needed to do was add mains lead, put valve s in and screw the panel into the case. Capacitors were made by milling the plastic PCB thin before deposition. Valve holders weren't separate, they w ere pins the machine inserted into PCB holes. The video shows the stages of manufacture for a fairly small domestic reaction set. The whole thing was controlled with electromechanics, and various limit switches detected probl ems & stopped the production line.

I presume no-one ever bought the system, I've never seen 1940s radios made that way. I must have saved a link but I'm damned if I can find it.

I wonder why no-one makes plotters with arc deposition for development PCBs .

NT

Let's get rid of the masks!

Who needs coherent light? Send a DLP projector into a lens. Microns were good enough for the 70s!

Get rid of batch sizes, and open source the CAD tools (or better yet, use existing ones?).

(The DLP thing has been done, for 3D printing with photo-cure resin. Resolution is excellent, though I forget if it's finer-than-PCB-pitch good.)

Phil, how many projects have you done, where the PCB has been nothing but a limitation?

I'm going to guess most of them. :^)

Well, perhaps. What do you mean by "equipped to design"? Anyone who's got LTSpice should be "equipped to design" this!

Do you mean in the elitist sense of "they don't know how to make *good* things"? Who cares, it's not like they're making a million of them, let them waste their money -- that's what makes it a hobby. Or, would make it, if they could.

Yeah, but how slow? I want numbers! Vacuum tube slow?

Transistors (of the single-crystal type) are on the order of 10-1000 times better than tubes, depending on which parameters you count (V/I range, transconductance, capacitance). How much worse are a-Si ones?

And even if they're worse than tubes, the impedance is still lower, and the scale smaller, which means you can use way more of them. You could make shitty computers, neural networks, radios, audio amplifiers...

Tim

A small minority is any good with opamps on PC boards!

There have been analog arrays, a heap of parts on a chip that only needed a top metalization layer to finish off. I don't know if that still happens.

The programmable analog arrays, the analog FPGAs, don't seem to work very well.

Those old tube radios were something else... particularly those phenolic PCB's... just a glance and there'd be an open :-(

A little old lady came into my dad's radio & TV repair shop (when I was 16* ) with an intermittent radio.

Twisting the PCB made it come and go.

I examined to PCB for breaks... none visible... so I bathed it in solder... still no joy.

Playing around I found two IF cans touching... if forced apart the radio worked... Ohm-ing, both cans were grounded.

So I stuffed a popsicle stick between the two.

When the little old lady came for her radio... no charge ;-)

(*) I got all the "dogs" ;-) ...Jim Thompson

Many processes are already E-beam... more costly mostly because it's slower.

I'm sure Cadence will be happy to comply >:-}

Virtually _no_one_ hanging out here is capable of designing at the device-level.

There _are_ companies providing "paste-together" OpAmps, etc, from their IP "catalog"... big bucks, poor chip area allocation.

What don't you start up a company doing that ?? ...Jim Thompson

And use what, crayon?

A DLP doesn't have the resolution to make a good use of chip area. They're only 1-2 megapixels. Plus you have to worry about the geometric distortion if you want to use multiple fields.

Such as?

I don't know. I don't usually complain about circuit boards, except that they're so much slower to build than dead bug protos.

There's no point in a hobbyist designing his own crappy op amp, so I'd be imagining things like ham SDRs, inexpensive sampling scopes, FFT analyzers, that sort of thing.

I've wanted to do a laser noise canceller chip for awhile now.

But LTspice doesn't do chip layout--no floor plans, no routing, no DRC, no nothing. You might be able to trick gEDA or something into doing they physical design. I don't know what the really low budget MOSIS projects typically use.

Elitist is in the eye of the beholder. (Or as we say in NY, "ya mudda." I started out as a hobbyist myself, and I have a lot of time for folks with a fire in their bellies. Give me a motivated and talented amateur over a time-card-punching "pro" any day. (I've never taken a circuits class, other than the old hackneyed "RLC circuits for sophomore physicists", which I could just about have taught at the time.)

However, learning how to do all the steps from concept to working silicon isn't like learning to build radios--it's much more like building your own car from bulk metal. It's not much of a hobby if nothing you build ever works.

(I've often said that it's a weaknesses of mine that my ideal project is building a computer from sand. I understand the ambition, believe me.)

Relay slow. ;) As in, only just fast enough to run a display. Back in the day there was doubt about whether they were fast enough even for that.

Really horrible. About good enough to run an LCD.

But you can make good digital things with an FPGA demo board and free tools. And I doubt you could make a useful radio with TFTs.

Cheers

Phil Hobbs

CMOS didn't even exist when I graduated from MIT, so I'm self-taught on CMOS design.

But there's been a number of projects I've lost because the potential client insisted on "testing" me with school-boy questions with an OpAmp structure no one in their right mind would implement.

... yet I have ~45 successful CMOS and BiCMOS chip designs in volume production.

...Jim Thompson

Difficult to see why you'd want to design a chip for those when you could use a Zynq (dual ARM A9 @1GHz + FPGA). You're obviously thinking of the RF front-end, but I doubt that hobbyists would tackle that stuff on-chip anyhow - too hard to iterate.

Clifford Heath.

I haven't read it but a colleague enjoyed a book about a guy building a toaster from scratch.

George H.

Den mandag den 11. april 2016 kl. 16.52.20 UTC+2 skrev Jim Thompson:

people keep saying that, but I don't see much point.

almost everything needs more than one layer so you need holes and vias doing a schematic and layout takes much more time anyways

and if you can live with simple boards, toner transfer / photo mask and etch doesn't take much time

-Lasse

Well, if you could print insulator too, you could do the Multiwire thing. ;)

Cheers

Phil Hobbs

Actually, a homebrew Multiwire machine would be really cool, and a whole lot more useful than your average home PCB setup. Something like a Vector 8007 board (0.1 inch pitch, pad per hole, ground plane) would be a good place to start.

Cheers

Phil Hobbs

Tom Gardner's answer pretty much says it all.

But, there are two big problems with chips. One is the cost of the masks. PCB master artwork can be pumped out by a laser photoplotter in 10 minutes on $10 worth of film and chemicals (per layer). Phototools for chips are WAY more expensive, a typical mask set now costs $100K to $250K.

Second is the number of steps. PCBs are just copper on a substrate, and them maybe build up for more than 2 layers. Chips are made with ion implants, diffusions, polysilicon, oxide, finally metal/oxide/metal for the interconnect. LOTS of layers, many of them different types, and all VERRRY small features, requiring fancy alignment systems and insanely expensive optics on the step and repeat cameras.

There probably are other systems like this, but MOSIS offers multi-project wafers, where they combine dozens of different designs onto one reticle and make all of them on one wafer. This spreads the cost of the masks and wafer handling over many users, making it more affordable. BUT, this is SURELY not hobby territory, unless the hobbyist has Bill Gates' deep pockets.

YES, this is true. MOSIS offers the old AMI (now ON Semi.) C5 process at

Yes, they have educational runs, which are reserved for students of IC design to get their own desing fabricated and then they get to debug it. For "non educational" projects, like our research systems, where the chips will actually be used in physics experiments, they have a different pricing tier. It still is one of the cheapest ways to go. Have you heard of cochlear implants? The first signal processor chips for those were done through MOSIS. Remember ATM (networking technology?) That was prototyped through MOSIS.

No, not really. They know there are niche projects, like our physics chips, that will NEVER be competition. And, they can make a little money on the side running wafers for us.

And, other projects, like experimental CPU architectures, cochlear implants and other bio-projects, and other research stuff may well turn into production-level silicon at some time.

Jon

I work with a group that develops mixed signal chips. The guy who heads it worked on the original cochlear implant signal processor, and a bunch of other stuff before he did some designs fo us.

The digital stuff is pretty easy. You can pick standard circuits out of a library and Cadence has synthesizers that will generate it from logic equations. The real strength of the tools is to simulate it out the wazoo at all the process corners as well as voltage, thermal and timing corners to make sure it provides the right functions.

The analog stuff is maddening. If you use the stated noise for the transistors from the AMI models, for instance, you will be VERY disappointed that the actual noise is about 10 X worse! So, our guy had to go searching, and there was a guy in Korea who characterized the transistors and published real world models for the process. We are still using those numbers, and now we get realistic simulations - MOST of the time. When you get into the analog realm, there are SO MANY places for Murphy to bite you, it is just a quagmire.

But, then, once you have realistic models, you have to ask the sim for the proper conditions to see what works. Don't just give the ideal voltages and timings, you also have to throw in some edge cases to trip up marginal circuits on your design. We didn't always do a good job in this area, and ended up with chips that had subtle quirks that took a long time to understand. Once we understood how it was behaving in the real world, we could go back to the sim, set up the same conditions, and it was a case of "Oh, WHY didn't we THINK of simulating that???"

Jon

LPKF makes machines that mill away lines to isolate traces on PCB material. They or somebody similar had a machine that took a copper-clad plastic flexible amterial, wrapped it around a spinning drum and then blasted bits of copper away to do the same thing. Not sure that machine is sold anymore. Either of these processes can make a small circuit pretty quickly.

Jon

• We are talking about "shitty transistors" and thus one does not give a damn about resolution. I know almost nothing about 3D printers, but if what i have seen on Youtube is not faked, resolution is at least 100DPI and should be good enough. Cost competitive? If you are doing specials or in-housework, That becomes meaning less.