Douglas Tutty wrote on Monday, January 08, 2007 8:02 AM -0600: > On Sun, Jan 07, 2007 at 11:52:57PM -0600, Seth Goodman wrote: > > Douglas Tutty wrote on Sunday, January 07, 2007 4:20 PM -0600: > > > > > Most electronics are designed for an ambient (to them, not the > > > case) temperature of 25 C max. > > > > It would be nice if we could assume that when designing hardware, > > but it isn't realistic. Even for a benign laboratory environment, > > we normally assume 40 degrees max ambient, which means the air and > > all surfaces surrounding the case. Industrial electronics are often > > designed for 50 degrees ambient with the case interior being 15 > > degrees higher. > > > You mean that if you take a number off a chip and look it up in the > manufacturer's datasheet it will say it will tolerate an ambient temp > of 50C? I'm not necessarily thinking of a major heat source like a > CPU/GPU or a power transistor but a simple TTL or CMOS support chip. > What about the clock occilator? They __used__ to go haywire if you > let them heat up too much. > > I'm not saying that the support chips won't work at higher temps but > it shortens the bathtub curve. Its one of those rules-of-thumb: if you > can't leave your thumb on the chip comfortably, then its too hot. > Mil-spec has a different rule of thumb of course. Then its not your > thumb but a delegated thumb:-) Electronics designed from the outset > for extreme environments will of course work in them. E.g. automobile > stuff, Mars rover, pacemakers (water cooled at 38 C). The OP I think > was talking about a regular consumer-grade computer so I limited my > focus to that.
While your rule of thumb is certainly safe, it's overly conservative. Here's some explanation. Integrated circuits today come in a few general temperature ranges: commercial 0 to +70 degrees industrial -40 to +85 degrees automotive -40 to +125 degrees military -55 to +125 degrees These temperature ranges refer to the ambient temperatures of the chip environment, which means the air and circuit board temperatures. It is not the package (case) temperature that you mentioned. Regardless of the ambient temperature rating, the operating limit is 125 degrees junction (die) temperature for silicon. Below this temperature, the failure rate is temperature dependent and roughly doubles for each ten degree increase in junction temperature. While every application has different reliability and operating life requirements, most applications can tolerate junction temperatures of 110 degrees for smaller packages. Manufacturers normally select IC packages to give a junction to ambient temperature difference of around 20 degrees. This gives a maximum case temperature of around 100 degrees and an ambient temperature of 90 degrees. For commercial grade parts, the limit is 70 degrees ambient for the package, so case temperature might be 80 degrees. You shouldn't operate at the limit, but there's no problem with a 60-70 degree case temperature for many applications. This takes some care, but it is not unusual in scientific and industrial equipment. Large scale integrated circuits that dissipate a lot of power are a different matter. These chips are more failure prone because of several factors. The primary reason is that there are so many I/O connections that the probability of a single failure is much higher at a given junction temperature. You use heat sinks to limit the junction to ambient temperature difference and keep the ambient temperature somewhat lower than for the smaller packages. Still, it is overkill to require that case temperatures never exceed 50 degrees. > How can a laboratory environment of 40 C be called benign? I conk out > at 28 C. So do I, but we're soft :) Equipment is still expected to work in a rack, where temperatures are higher than in the room, or in a closet with less than ideal airflow. Laboratories often don't have high-capacity cooling systems with backups. And as Hugo points out, air conditioning is terribly expensive in some places and may not be reliable even when present. -- Seth Goodman -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
