The first time I noticed this, I was camping up in the Sierras. Some people will even read the fine print on a ketchup bottle if there's nothing better to read. But we didn't have any ketchup bottles. So I was reading the information on an envelope of some bean soup—beans, dried onion, salt, spices, etc.—and silicon dioxide? Hey, I know what that is, and I wonder who gives them permission to put sand in the dehydrated bean soup.
When I got home, I read the labels on several other kinds of bean soup from this company. None of the others had sand. Just now I was reminded of this, because I was reading the labels on some United Airlines pretzel snacks. They too had sand added, apparently for the purpose of " preventing caking," but not more than 2% content.
HAVE YOUR SAND AND EAT IT TOO
For about 50 years, we have been using SiO2 to prevent the "caking" of bad stuff on silicon material in the planar process. We use SiO2 to isolate transistors from their surface.
To isolate collectors from each other, we usually use P-type material. At room temperature, it doesn't leak much current. In fact, Bob Widlar's LM12 still uses P-N isolation when it's running a junction temperature of 260°C. The leakage is significant, but Widlar's design allows for that. More recently, we have been using SiO2 to isolate collector regions from each other. The collector's capacitance is nicely decreased. So is the thermal conductance.
When I began to work with NSC's dielectric-isolated process ("VIP-10"), I was told that the heat flow, because of the use of SiO2 in the transistor's side walls, does not allow one transistor to heat its neighbor. Instead, the heat goes straight down to the die-attach.
I thought about that a little. Then I took some measurements of various devices. Sure, some of the heat from one device tends to flow straight down through the floor of the transistor's tub. But that doesn't mean it doesn't heat its neighbors indirectly. So everything gets warm. If you don't believe me, go measure it.
PUT TO THE TEST
So the analysis of transistors' self-heating isn't trivial, and heating their neighbors requires some thoughtful insights. That's because you can't just ask a computer to analyze all those millions of points in a die for many many nanoseconds.
After all, when an LM12 heats up with 800 W (80 V X 10 A) up to 260°C in 90 µs, just before it turns off safely, you can't exactly analyze that with a computer. Throwing SiO2 into the equation doesn't make things any easier either!
We had some engineers who argued that one particularly high-power dielectrically isolated IC must be running up near 165°C, because it was observed running reliably over a weekend. Their analysis showed that the hot-spot must be only a little higher than the (easily measured) temperature-of the electrostatic-discharge (ESD) diode, over near the edge of the die.
Using a number of analytical techniques, we showed that in this rather extreme case, the part was running reliably as well as accurately at a junction temperature of 260°C—over the weekend. So, the simulation and computerized analysis tools were just plain wrong by a big factor.
Skepticism of computer simulation is still in order, as I will be discussing at a panel session at Design-Con in Santa Clara in January.