Login
Back to forumSee the full topicGo to last reply

Posted By

Luca
on 2019-02-12
04:36:28
 Re: Heatsinking or not heatsinking?

Reporting the other side of the discussion, from Ian Gledhill and Bil Herd, which another precious tile of the mosaic.

Ian Gledhill
"I should be happy with this. After all, I run a repair service and have a stock of 8501s. happy
However, there is a minor problem in that this long article misses the point completely (and actually mentions the Spectrum ULA where he also misses the point completely!).
Measuring the temperature of the casing of the 8501 is pointless. Completely pointless. Why? Because the 8501 doesn't overheat in that way. If you measure the plastic part of the case at the end with a heatsink or not, you'll find it doesn't run that hot. So that means you don't need a heatsink, right?
Wrong.
Many chips run hot and are reliable. Why? Because they're designed well to cope with it. The 7501 and 8501 CPUs are the opposite. They have an actual flaw on the die. This is ironically suggested by Bil Herd as well where he mentioned (I screenshotted it in another thread) that the CBM engineers measured a point with just 200R across a voltage rail (which is a really bad idea on a die!).
When you put a heatsink on the 8501, you're not lowering the temperature over the whole chip, you're trying to get the heat away from the one point on the chip which produces too much heat for its size. The rest of the chip is fine; it won't get too hot - but this particular part of the die needs heat conducted away from it as fast as possible. *THAT* is why we apply heatsinks. It's nothing to do with the general temperature of the chip.

As I mentioned above as well, the Spectrum ULA has the same problem on a few revisions. Ferranti made the ULA but made them in different factories. There are the first ones, which have a torch logo. These are made in the UK. These run really hot. They're also very reliable, and it's unusual to see a dead one. Then in 1983 onwards, Ferranti started producing later versions in the Philippines. These ones run cooler. They are also MUCH less reliable. And yet if you put a heatsink on them, guess what? They run reliably. But why? They don't run as hot, right? That's correct. But it's not JUST about heat. It's about how the heat is handled on the die, where it is concentrated and how the die copes with heat cycling. The UK ones are made much better than the Philippines ones.

What we're doing here is running our machines WAY past their sell-by date. And, incidentally, in the documented it mentioned that they would fail within a year. Well, THEY DID. The C16 and plus/4 had an atrocious death rate even then. I've seen many an owner from the time say he got through several of the things.
We cannot in 2019 go around treating our machines like we did in 1984 when spare parts were plentiful and cheap. We have to do everything we can to keep them running and if that means spending a whole $5 or so to safeguard your C16, seriously, who wouldn't do it?

I can only speak from my experience. I have seen MANY C16 and plus/4 machines which have had perfectly good power supplies, perfectly good TED chips, and a stone dead CPU. I would say over half of C16 and plus/4s that are sold as untested have a dead CPU. The C16 even used a standard 9V PSU through a regulator, so although the brick PSU for the plus/4 is not reliable, the C16 one won't cause that trouble because the power goes through a regulator anyway.

My own evidence is very clear:
1) 8501s die. A LOT. They are responsible for nearly all C16/+4 failures.
2) A good heatsink makes them operate much more reliably. I know this because of the amount of people who have sent me C16/+4's for repair, who are still using them much later. No matter how often they're turned on/off (which I agree is bad for them, but not a killer).
3) Heatsinks are cheap
4) Seriously, why WOULDN'T you heatsink it? Even if you're not sure. Even if you think I'm making this up. Why take the chance for a couple of quid/dollars."

Bil Herd
Caps unfortunately, have a limited lifetime due to drying out. They were literally designed with vents so they didn't explode. I tended to pick caps that had 5 year lifetimes. Batteries are the other limited lifetime component that you just lived with what was available.
I wouldn't remove the factory heatsink which uses conduction typically, and replace it with a freestanding one which uses convection, especially without measuring the temperature difference. What if you are making the junction warmer by removing the factory heatsink, would you still do it? Did you measure the difference? The TED shield is an pretty big heat sink and also uses conduction rather than air flow for cooling.

Geometry changes would have been one of the kind of things that caused revs, others for logic issues. The process engineers would lay an LCD temperature sensor (think mood ring) on the chip and identify hot spots and we would correct.

Some quick tidbits:
- If the chips suffered from over heating they would have failed the first year.
- I am not surprised chips failed 20-35 years later, especially for a process that was only a year or two old. (HMOS I HMOS II) We would not have spent a single dollar to make them last more than 3-5 years. They also have smaller geometries to reduce current consumption, but smaller can be less robust.
- To make any analysis of failure rates you would need the statistics back to when they were made and failure rates compared to total made. We would have lived with 5% failure rates as AQL right off the bat.
- We dealt in millions, you would need thousands of samples to be statistically significant.
- Apparently the chips run at 50 degrees C, they are designed for more than twice that with a junction temp of 125 degrees.
100% of all chips will fail at some point.
- The TED series was a consumer product. Think toys. The question I have is why do _any_ of them work 35 years later?
- Yes we made mistakes, we were making these things up as we went. I know most of the people here that have created a new NMOS process and then ran it and sampled it 30 years later would agree.
- Its probably not the heat breaking them, its probably the thermal shock when you turn them on. Think when a light bulb breaks. What if preheating the chip reduced this shock and consequent failures, would you preheat your chip.



Back to top


Copyright © Plus/4 World Team, 2001-2024