On Home Projects as a Reentry into the Job Market

Those homde-made ICs are logic gates -- not quite up to the CDP1802 level. :-) Jeri's unique for doing this completely at home, but note that until quite recently many universities still had older generation IC manufacturing abilities: I know a guy who finished college in the mid-'80s, and in one of his lab courses they were also replicating standard logic chips, all the way from "bare silicon wafer, some dopants, and a furnace" to "packaged, finished part." These days universities usually just have the job done by MOSIS or a commercial research partner if said partner has a fab of their own (e.g., Triquint, IBM, etc.).

(This evolution with how universities build ICs has largely followed PCB construction: There are a few die-hard schools that insist on cutting and/or etching boards in-house, still, but the majority of them today use one of the inexpensive commercial fabs given how much better the quality is and how inexpensive the pricing has become.)

I actually designed the original SDRAM controller in her "Commodore One" there -- she had it working with SRAM, but figured that SDRAM would be better (and I concurred) given how much cheaper it was. I had just finished a very fancy SDRAM controller for work, so I was able to crank out the simple-minded one she needed in a few days after work and handed her the VHDL when I was done.

Of course, after that it was undoubtedly modified by many parties. Jens Schoenfeld in Germany

Jeri's done an exceptional job of succeeding despite very humble beginnings. I think much of this can be attributed to her tenacitiy -- once she decides she wants to do something, she'll keep at it day and night until it's done. She spent years hounding various former Commodore engineers for schematics, chip layouts, and other historical information they had that she wanted to learn. She's a very sociable person by nature, always wanting to learn.

---Joel

In '74 students were doing good to make NPN transistors. ;-)

We used a lathe to cut circuit boards. AIUI many now use the "CNC router" style prototyping systems. We have one of them at work, too, but it hasn't been used much. It's a little crude.

Just use the FPGA's macro tool. ;-)

Sounds like a fun person to be around.

Hi Keith,

Not bad, not bad at all. :-)

CNC routers are great if you need something more or less immediately... or if for some reason you're going to need a lot of revisions to the PCB before you really know what you want... but these days, if you can wait a week, not only do the commercial guys give far better results, but in most cases it costs less money overall if you have to be paying someone's salary to do it. (We have a pretty nice LPKF machine, but it really doesn't get much use.)

Do you guys do any pick & place/reflow/assembly there? We have a manually operaeted pick & place machine and a small reflow oven, which is convenient at times.

Ha... this was back around the turn of the century, and while I think Xilinx was just starting to have such macros, they sure weren't considered trustworthy! Same with FIFOs at that time -- we had written our own, which "just worked," whereas I remember in the service pack notes how it seemed they were always fixing various empty/full synchronization problems with the silly things. (That being said, I was always impressed with Peter Alfke's abilities -- he clearly knew the parts inside and out, and knew how to make them work reliably. But not everyone there shared his talents...)

These days I would at least take the macro generator for a spin, I suppose. I don't want to become like one guy we had at work then, who didn't want to trust that the synthesis tools could implement addition efficiency, and spent several weeks writing his own basic arithmetic routines... only to find that they performed more slowly they the synthesizer's built-in functions, since the optimizer wasn't able to do as good of a job without "understanding" the overall function of all those 'flops and LUTs he was instantiated one-by-one...

(On the other hand, around the turn of the century was also about the point where you finally could trust the synthesis tools to know how to *multiply* -- I remember people doing a lot of testing, and *finally* the implementations were starting to be reasonably efficient things like Wallace multipliers...)

Oh yeah, she certainly can be!

---Joel

Ditto. We are rarely pressed for time on prototypes and when we are they want it "shippable". Yeah, right.

Sure. We do all our own assembly. Our reflow oven is kinda crude (only five-stage), but other than that the stuff works very well. They're talking about getting an eleven-stage oven but I think the holdup is trying to figure out how to slip one into production without a huge risk of missing ships.

Hint: They still aren't. ;-)

Xilinx' FIFOs worked very well for me, about then. I didn't use the core-gen, or whatever they call it, but I did use their primitive.

I did pretty well implementing wide logic (things like comparators) using Xilinx' fast-carry chains. I could certainly beat Xilinx' synthesis. I checked again a couple of years ago and they were using the carry chains for wide logic. In '98 or '99 Xilinx' synthesis was so bad I bought Synplify Pro.

Does your current reflow oven have a profile for pizzas? ;-)

I'm currently hiring for my company. A hobby in the field is big plus for me, especially for people with short work history. It tells me that they're in the art for passion, not just because of good job.

Getting from Wii to a working embedded system, however small, may show at least capability of learning. Not as good as a hobby, but a plus.

We're constantly doing projects where we do things we've never done before, so the learning capability is an absolute must in my business.

snip

Boy, it's nice to see someone in a position to hire saying that. I've seen all too many people who are trying to do something unique, yet can't gather the courage to go hire someone unless they've done exactly that never-before-done thing.

We do assembly in-house either when time is tight or when the boards are simpler. We have some military products where, while the designs aren't particularly complex -- just a whole bunch of roughly the same circuit repeated over and over -- some boards end up being something upwards of a thousand discrete parts, and the techs really prefer not to have to sit down and pick-and-place that many parts manually... so those go to a contract manufacturer.

Same here (or maybe 6 stages? -- but certainly not 11).

We're planning on building an entirely new shop with a pretty fancy production line. It's been delayed due to the slump in the economy (customers delaying or cancelling orders), although it's now planned to break ground by the end of the year, which is pretty cool. (It is a bit cramped here right now...) There was talk about offering contract assembly services to help pay for all this if we don't have enough of our own goodies to keep at least one shift going every day year 'round; should be interesting to see how it plays out.

Ah, good to know!

We were using Synplify as well -- Xilinx was packaging FPGA Express (aka, FPGA Distress) at the time, and it was really quite awful -- worse than the free synthesis tools that Cypress gave you for their CPLDs!

---Joel

...

I remember interviewing one guy who had programmed a single chip uC and

Grant.

No, but it would probably work better as a pizza oven. ;-)

The motherboard in that widget you were playing with has something like 1700 components.

Five or six is OK, as long as you don't want to do RoHS. If you have any plans to do RoHS, the more the merrier. Five works, if nothing goes wrong but it stresses parts more than I'd like. With nine or eleven stages you can soak everything longer and then spike it for the reflow.

AIUI, we did contract manufacturing at first, too. It's a bigger PITA than it's worth. Splitting allegiances isn't a good position to be in, either.

We abused the software quite a bit. If you do everything by the book they may work. ;-)

I really liked Synplify (Amplify was a sweet idea too). Too bad Synopsis bought 'em. Actually, I'm finding that the freebie Altera stuff works pretty well. I haven't really stressed it, but it's very easy to use. I like the direction they're going, too.

That's a big red flag! When I was interviewing applicants that was an immediate thumbs-down. We weren't allowed to ask technical questions (i.e. give a test), so we talked about their projects. If they could explain what they did and how it worked, I'd move on to something closer to our problem and ask how they'd go about solving it.

Of course the processor isn't really important. The difference is just ones and zeros. ;-)

...with ten years experience doing the never-been-done-before thing with hardware that's two years old.

Was there any explanation given for the "don't ask technical questions" policy? Seems kinda bizarre...!

"krw" is a graduate of politically correct IBM :-( ...Jim Thompson

Hi Keith,

We have some boxes that are something like 3,000 parts... but they're spread over a large handful of boards inside.

So you win for "most parts on a single board." :-)

I'd still be worried that massive ferrite choke on the internal Ethernet cable might get lose some day and start smashing up all those nice audio transformers and relays you have in there!

Hmm, good points. I'll keep that in mind.

Agreed; Synplify was quite good. I was no longer doing FPGA design by the time Synopsys bought'em, at least.

The Cypress VHDL synthesis tool (Warp) was nice in that (kinda similar to LTSpice) it'd been written largely by a single guy up at Cypress's Beaverton, Oregon office -- made for nice "one stop shopping" for tech support if you ever needed it.

---Joel

Law suits. If it's not a sanctioned exam there will be charges of discrimination, sometime, and you *will* lose. There were standardized exams given to programming candidates but engineers were always considered more "professional".

I've found that very few companies gave any sort of test, other than as I suggested above. If all of the companies I've interviewed, only two have given any sort of an identifiable "quiz"; one 36 years ago (NCR - a ball-busting gang mugging) and the other my CPoE (trivial - what's the gain of this OpAmp circuit, ...really, an instrumentation amp).

If I did it again I think I could cut it by almost half, or get rid of the other board completely. Not going to happen, though.

It's still on the rope. ;-) You should see the monster on one of the flat cables in our football product. It's a real Rube. Compliance afterthoughts are seldom pretty. :-(

It brings home some bacon, I understand. Ask your boss, "if it comes down to running our product or theirs..."

I played around with Actel's a bit too. It seems to work, but I don't like their parts as much. LUT-3s do an amazing job of eating up cells (multiply your estimates by at least two). OTOH, Xilinx's new LUT-6 (Spartan 6) is sweet. I'm 95% sure I'm going to stick with Altera, though. Cost and packaging are really what drives the decision.

"Yes, we absolutely discriminate against people who don't have the right technical chops... and we're proud of it!"

:-)

OK, I understand what you're saying though... that's big business today, I suppose.

---Joel