Generally, I need RAM fast enough to keep up with the logic at maximal clock rates. Secondly, I need it all to reside in ONE chip. That's how I could get by with 4 I/O's. Once I use external RAM, at, say, 8 ns, my RAM has to be four times as wide, in order to transport the same amount of data, I have to add pipelines, and so on.

about those BGA's ... So far, I haven't found even one U.S. board/assembly house that can compete in this arena. They have yields which mean that, under the best of circumstances, and with the best of luck, 10% of the boards that are delivered are electrically sound. Once they're assembled, again, if I'm lucky, 1 board in 10 is functional, and it costs plenty to have 'em x-rayed in order to prove that the problem is with the BGA. The result, since the BGA costs more than the combined rest of the assembled board, in a 100x cost increase.

This result forces the manufacturing to China, where they happily produce tested boards in 1/3 the time, including shipping time, at less than 10% the cost, with 100% electrically tested boards, such that assembled board yields exceed 99%. I don't like doing business in Asia, as the culture is so different from what I've learned. Lying, stealing, etc, are not viewed in the same way as here. I've had Asian business people tell me, when I returned defective products for repair/replacment, that I shouldn't expect them to honor their agreements or their warranties. The result was that, while I'd shipped back the defective hardware, I got nothing back.

While a client might consider volume production, I won't. I seldom need more than a half dozen or maybe ten of a given application. Exceptions might be for a "standard" (at least for here) board that I can apply again and again to provide functions for my proof-of-concept work. Since I can't get good boards at reasonable prices here any longer, I stick with what's known and trusted. Perhaps I'll have something made in China in the future, but probably not here. What I get from U.S. board houses is mostly excuses.

I'm, not likely EVER to have another ASIC made. Last time was too painful. Now we just use the easy migration from programmable to fixed logic that most FPGA houses provide and use as much of their help as we can.

RE