Here is the continuation of the panel discussion among Barrie Gilbert, Ted Hoff, and Bob Pease. The first part of this conversation appears in the Engineering Hall Of Fame section of our September 13, 2004 YOUR Issue, ED Online 8684. Electronic Design brought together these individuals-three of this industry's foremost designers and most respected thinkers-to participate in an informal discussion about the future of electronics engineering in America. We're sure you'll enjoy the rest of their informative, enlightening, entertaining, and sometimes profound conversation. Hoff had to leave early, but Gilbert and Pease talked on. See what they had to say.
BARRIE GILBERT: Not to dismiss these other questions, but let's get back to the technology. There’s another point of departure that I think is also important. Up until now, it has been assumed that the monolithic implementation of everything is the only sound way to go. If you submit a paper to the ISSCC bearing the title “A Monolithic Implementation Of A Very Hard Thing To Do,” it’ll get accepted. But if it says, “A Hybrid Implementation Of A Very Hard Thing To Do,” it won’t. And that’s dumb!
Hybrid implementation is being used again because we’re at a point where automated assembly techniques are now affordable and the technology is advancing rapidly. At Analog Devices, we make parts that contain a large number of individual elements on a tiny little header that looks like a regular IC and sells at IC prices. But inside the package, we can combine a bit of gallium arsenide with a bit of 0.13-micron CMOS along with a bit of complementary-bipolar signal-processing, and other components that are hard to integrate on a single chip. This is no longer an expensive technology. The days of military utilization have given hybrid assembly a bad name, but that’s going to change. We’re going to see an end to the long war between whether “CMOS is better than bipolar”, or “digital is better than analog”. They’re all part of the same world of technical challenge.
Clearly, if something can be done just as well or more robustly in digital, it darn well ought to be. The turf wars and the old partisan views of technology are at last being replaced by a broader and wiser point of view, in which we use the technologies where each is optimally suited. Then we put them all together at very low cost under robotic assembly, just as we would today build a monolithic circuit. Sometimes that doesn’t work well, for obvious reasons. If you want to access cache memory at high speed, you’d better not put it on a separate chip. But that’s a technical issue. It doesn’t mean everything has to be done that way. On the other hand, it doesn’t mean that everything has to be done monolithically either.
TED HOFF: You just touched on one of the main advantages of monolithic technology-minimizing the capacitance associated with interconnections from one chip to another. The issue we kept running into before was how many connections we were comfortable putting on a chip. Generally, each component that we're going to use has to have some number of connections. If we can come up with packaging techniques or interconnection techniques that allow us to do it without trying to print everything on a single piece of silicon or some other substrate, why then maybe we can do some more interesting things there.
Reprints
