KIT's Spinoff 'Digital Power Systems' Develops Ultralong-lived Industrial Power Supplies:
Switching power supplies are omnipresent in our daily life, may it be in households, offices, or industry. They convert the alternating current supplied into direct current for smartphones, laptops, charging stations of e-cars, and logistics and computing centers. However, conventional power supplies usually have to be exchanged after nine years of permanent operation. Digital Power Systems (DPS), a spinoff of Karlsruhe Institute of Technology (KIT), has now developed and tested power supplies with a lifetime of 50 years.
Conventional switching power supplies are light and compact, but highly susceptible to failure due to the electrolyte capacitors they contain. Film capacitors are far more long-lived. So far, however, they have needed up to ten times more space. "We have now developed a digital control process, by means of which film capacitors can be used on smaller space," DPS Director Michael Heidinger says. [...]
[...] The novel digital control process allows for the use of film capacitors with a slightly increased space requirement only. For control purposes, a microprocessor is integrated in the power supply. It detects disturbing ambient impacts and balances large voltage fluctuations of the film capacitor. As a result, storage capacitors of smaller capacity are sufficient. Heidinger explains that powerful microprocessors have made this possible.
I've had to replace a number of electrolytic capacitors in several of my home appliances over the years, so I can appreciate this development, but do you think the concepts "lifetime of 50 years" and "integrated microprocessor" generally compatible?
(Score: 5, Interesting) by RS3 on Friday January 20 2023, @05:31AM (17 children)
Ah yes, the "capacitor plague" [badcaps.net]. I've repaired countless power supplies in a very wide range of things, including motherboards.
At least when you get really good caps, and I usually try to replace with higher capacitance (uF), you'll get much longer life. I have some motherboards still going 20+ years later (although I don't run them every day.)
Many things I've repaired had unpopulated (empty) spots for more capacitors in parallel, so I usually fill them up too.
As a very long-time electronics hobbyist, technician, EE, it's sad that you can't rely on simple components. Admittedly they're being "worked" pretty hard, but the whole thing with the very bad faked copies is so irritating, especially when all the problems are due to the ever cheapening of everything.
Um, I'm not sure I understand the question. Are you thinking the uP might fail?
I'll add that even film caps can (and will) degrade if they're handling enough AC current to make them heat at all.
A few years ago there was an article where someone was doing a hybrid of electrolytics and a film, or ceramic, or something that would handle the high ripple and would damp out much of the higher frequency spectral components, and the electrolytics with a little bit of resistor and/or inductor would handle the large lower frequency stuff- passively doing with the microprocessor is doing in this article's systems.
As an EE I like this system and I'll be looking into it some more.
(Score: 4, Insightful) by hubie on Friday January 20 2023, @05:47AM (14 children)
You did interpret my question correctly. Do you think these microprocessors will stand up to 50 years of industrial use? I, perhaps naively, assume processors prone to failure easier over time due to the crazy density of components on them, but my electronics expertise comes from the Heathkit age where components are big, soldering was thru-hole, and our traces were large.
(Score: 5, Interesting) by RS3 on Friday January 20 2023, @07:43AM (6 children)
I really don't know a lot about longevity. Generally silicon manufacturing has been getting better, but some things seem less reliable, like RAM. It'll depend on operating conditions, including voltages, vibration, temperatures, etc.
The density is only a problem where there are flaws in the silicon. Others here are more knowledgeable about silicon "chip" manufacturing, but I speculate that smaller feature sizes will cause flaws and imperfections in the silicon to fail more quickly. In other words, you'll know very quickly that it isn't going to last long. (I can expand on that if it doesn't make sense...)
Most electronic parts that have internal flaws will fail in the first 40 hours or so, often called "infant mortality". Every electronics manufacturing / assembly operation I've been part of uses a 40 - 100 hours "burn-in" period. The product is then re-tested and shipped if all tests pass.
In the 70s and 80s a good friend's dad worked building "hybrid" circuits that went into satellites that are still functioning perfectly, even with the radiation up there. "Hybrid" means: ICs, transistors, resistors, etc., in "raw" chip form bonded to a ceramic substrate, wire bonded together, and sealed with a lid.
Other than RAM, I've almost never seen an IC or transistor just simply fail when it wasn't clobbered by some external problem, like voltage "spike", heat, vibration, short-circuit, etc.
I have some things with ICs that are more than 50 years old. I have to think about it, but I know some of my older Tektronix scopes have ICs from the 60s and early 70s and they work very well.
So I think it's very possible to function for more than 50 years if it's done right.
(Score: 5, Informative) by mhajicek on Friday January 20 2023, @11:58AM
It's not uncommon for industrial CNC machines to last 30 years or more. That's not necessarily on all original electronics though; there's a strong market for refurbishing the boards.
The spacelike surfaces of time foliations can have a cusp at the surface of discontinuity. - P. Hajicek
(Score: 3, Interesting) by hubie on Friday January 20 2023, @01:27PM (1 child)
It's funny you mention Tek scopes because I was thinking about those when I typed that comment. We still have some of the old phosphor screen scopes kicking around and I know they'll fire right up and run, but to be fair, they haven't been running all this time. Actually one did run for a very long time, but that was because it was part of a timing setup where we were actually using the plug-in modules for pulse delay and a few other timing things and we still kept a bunch of the old modules around as 1-for-1 swap-ins if any of those pieces fail. (We tossed the last of our Polaroid camera hoods many years ago; I always loved to show those to the newbies to show we saved off our data "back in the day."). We've had newer scopes fail, but the repair there seems to be just give you a new card since scopes seem to be just Windows machines running a very fancy plug-in cards (or is it the other way, very fancy DSP cards with a plugin single board computer?).
(Score: 2) by RS3 on Saturday January 21 2023, @04:26AM
When I was a kid, maybe 10, my dad would borrow a Tek from time to time for me, so I fell in like with them then. I've tried many brands, but I got very spoiled.
Most of them have analog power supplies, so most of the electrolytic caps are only seeing 120 Hz. Like you mentioned, they get minimal use. But many electrolytics, especially very old ones, can dry out just sitting unused, and then they stop working.
I have one of the ones that use plugin modules. Someone gave it to me. It had a dead shorted tantalum cap that I found pretty quickly. They made an amazingly wide array of modules for those things.
I have one USB "Hantek" that's mostly meant for automotive diagnosis, but can work for whatever. It has 8 BNC inputs. I forget the sample rate, but it's high enough for lots of stuff. I need to actually connect it to some sensors in my car and figure out what the heck is going on...
(Score: 4, Informative) by Anonymous Coward on Friday January 20 2023, @01:30PM (2 children)
> some things seem less reliable, like RAM
Just a short comment on this, a good friend used to work at a chip plant. One of his jobs was to analyze details from the production line to determine the potential life of RAM. For a hypothetical example, if the focus of some ion beam is slightly compromised in certain parts of the wafer, the chips cut from that area may work OK and pass burn-in...but won't last forever. The "worst" of these RAM chips got sold at low prices to video game console makers. Slightly better chips went to commodity PCs and the very best (closest tolerance manufacture) went to expensive, high reliability customers.
(Score: 3, Interesting) by RS3 on Saturday January 21 2023, @04:31AM (1 child)
Interesting they actually know there's an ion beam consistency problem. Hmmm. How to get the "good stuff"?
Years ago I learned that Apple labeled hard drives were the best of the drives, so whenever buying used hard drives I look for the Apple logo on the label.
(Score: 1, Informative) by Anonymous Coward on Saturday January 21 2023, @02:36PM
> they actually know there's an ion beam consistency problem.
My conversation was a long time ago, that was a hypothetical. What I do remember is that there are multiple sensors and data recording in every tool for every process step--temps, pressures, voltages, currents, etc. and also special test sites scattered around on every wafer designed to reveal process inconsistencies. It's hard to think of a process that really "wants" to be consistent, it sounded more like a herding cats problem as things drifted (very slightly, but with consequences). Given 6 weeks to process a wafer, a LOT of process data is collected and analyzed.
> How to get the "good stuff"?
From distant memory, I think the best stuff included expensive ceramic packaging...for mainframes. But in general, you get what you pay for--assuming you can find an honest retailer to deal with. I suspect that the good stuff is sold in large quantities, business-to-business. When a computer manufacturer has multiple product lines, the more expensive "business" or "hardened" versions likely have better quality chips too.
Another friend had a favorite saying, "Software is hard, but hardware is harder."
(Score: 3, Insightful) by mcgrew on Friday January 20 2023, @07:25PM (4 children)
Why shouldn't they, provided they were properly engineered? Steam locomotives manufactured in the 1880s had hundred year warrantees, and lots of moving parts!
Planned obsolescence exists today. It didn't 150 years ago. There's no logical reason why any electronic component shouldn't last forever, or at least until it's run over by an automobile or a stray gamma ray hits a transistor. Back then, things were built to last as long as possible. Today, an item's lifetime is the first thing that's engineered.
Only the most evil among us become billionaires.
mcgrewbooks.com mcgrew.info nooze.org
(Score: 0) by Anonymous Coward on Friday January 20 2023, @10:33PM (1 child)
> There's no logical reason why any electronic component shouldn't last forever,
Seems that you haven't worked in product testing and component qualification? Nothing is ever stationary, small changes in temperature make nearly everything grow or shrink. As soon as something changes size it pushes on whatever it's attached to and eventually things might start to crack. There are whiskers, little crystals that grow and short out circuit cards. Plastics out-gas and become brittle with age. Moisture condenses on cold surfaces and corrosion starts. The list goes on and on.
(Score: 3, Interesting) by RS3 on Saturday January 21 2023, @04:37AM
You're absolutely correct. However, ALL of those drawbacks can be overcome. You may not be familiar with spacecraft (and aircraft) design, component specifying, testing, hardening, shock mounting, shielding, conformal coating, and general electronics protection.
The "whiskers" you refer to are a big problem with lead-free (tin-antimony) solders.
Greed is a socioeconomic cancer.
(Score: 3, Insightful) by Azuma Hazuki on Saturday January 21 2023, @04:16PM (1 child)
Nothing can last forever because entropy...or, as it used to be put, "all which has form is void/emptiness." I imagine that even simple solid-state silicon products will eventually get warped or blown out just due to the sheer amount of electricity passing through them decade after decade. Saying that, though, yes, planned obsolescence is an obscenity.
I am "that girl" your mother warned you about...
(Score: 3, Informative) by ChrisMaple on Sunday January 22 2023, @06:47AM
I have a vague memory from about 40 years ago of some engineer from a semiconductor company complaining about the company's NMOS process. He claimed that one of the dopants (samarium?) would diffuse over time, giving the product a lifespan of only a couple of decades.
(Score: 3, Informative) by Unixnut on Saturday January 21 2023, @11:53AM (1 child)
I don't know if it qualifies as "industrial use", but one of my cars turned 40 years old in 2022. It still has the original ECU (8048 based), original caps, ICs, transistors, etc... and they all work perfectly, and an automotive environment is hard on electronics. Voltage spikes, extreme heat, cold, moisture, vibration, corrosion, etc...
So yes, I see no reason why you can't build electrical systems that last 50 years. I don't think they will be built and designed, because the modern world is one of penny pinching bean counters, disposable consumerism and planned obsolescence.
In such a world, building anything that will last 50+ years is seen as a waste of resources. Cars now are designed to last the length of the lease (about 5 years), then you buy a new one. Same applies to mobile phones, or pretty much anything else, simply because business realised they make more money if they force repeat customers every 2-5 years, rather than having a repeat customer every 20-50 years.
(Score: 2) by RS3 on Saturday January 21 2023, @06:49PM
I'm not surprised at all that your "old" ECU still works. It's really not that harsh an environment. Some ECUs are in the passenger compartment (many I've had over the years). Some are underhood, but they're in sealed boxes with very good sealed connectors. If water gets in, of course, things will corrode.
Many ECUs, esp. older ones, were conformal coated (circuit board coating), so even if water gets in they'll probably keep going forever.
A serious voltage spike, or intermittent wire short can fry a driver transistor. But there really aren't many serious voltage spikes happening in cars- the battery keeps them under control, and most electronics (semiconductors) in cars can withstand 60 or more volts.
In electronics production, you might be surprised at how hot everything gets during soldering (wave, infrared, convection oven, etc.), so the heat isn't an issue (unless the ECU somehow lays on an exhaust system part!)
But yes, I agree, it's all about "cost-cutting" - ever cheapening. I'm not sure that throwing everything away in 5 or less years is helping the economy or society in the long-run. We're scrapping things that could have been repaired, esp. if Victorian-era "build it to last" mentality was the rule when they were made.
Plus everyone is addicted to having "NEW!". I feel that we (the whole world) are caught in an economic running-in-place trap. Rather than cure cancer, we're spending money on buying new cars, new everything, and throwing things that are mostly still good in landfills.
Bottom line- as you allude to, short-term profit greed has slowly but surely taken over and metastasized into every corner of life.
(Score: 3, Interesting) by pTamok on Friday January 20 2023, @03:38PM (1 child)
It is reasonably well known that electrolytic capacitors tend to have limited lifetimes. Another approach would be to give people a means of diagnosing a faulty capacitor, or determining that it is near the end of its life, and make the capacitor an easily replaceable module - no soldering required.
(Score: 5, Informative) by RS3 on Friday January 20 2023, @07:30PM
I had started a long detailed answer that I can't finish. Some bottom lines:
Yes, electrolytics have lifetimes, and many good ones have expected lifetime listed in the specs. There are many factors: temperature for sure (heat is bad), compromising (cheapening) the circuit by using fewer microfarads results in the caps working harder than they should, which generates heat, which shortens their lives.
Cheaping out on total microfarads means the whole system is closer to the failure point, so as the caps weaken, system failure happens sooner.
A huge thorn in my side is the whole bad capacitor debacle: southeast Asian industrial espionage, incorrect electrolyte ingredients, etc., all part of the far too much pressure to cheapen everything.
There exist capacitor ESR testers [wikipedia.org].
Your module idea is good, but the "ripple currents" are pretty high and the connectors would need to be substantial. Also, a connector would add length to the conductor (copper) path, which will introduce resistance and inductance. The inductance will cause resonances and reduce the filtering power of the capacitors (which might help the caps live longer, but more uF will be needed (which is a good thing)).
I haven't yet bought an ESR meter, but I typically use an oscilloscope to look at "ripple" and sometimes it's obvious: caps are bad. Sometimes I temporarily add another cap in parallel, and if the ripple goes down significantly, then I know the in-circuit caps are bad.
Bottom line: it's all about cheapening everything to the point of things only lasting a few years. (Much) longer warranties would allow society to put more $ into solving bigger problems like cancer, green energy, and other more important things.
(Score: 1, Interesting) by Anonymous Coward on Friday January 20 2023, @08:27AM (15 children)
Flash cells discharge. They are used all over the place in modern microprocessors as manufacturing-time options.
They will suffer bit rot over time.
(Score: 2) by RS3 on Friday January 20 2023, @08:45AM (14 children)
Do all microprocessors have FLASH cells?
(Score: 3, Informative) by Rich on Friday January 20 2023, @01:09PM (4 children)
In practice, yes. I don't think there are any more mask programmed solutions. You might get some historical 8051 module that will take a piggyback ROM, but mind that any EPROM you might want to plug in also works by trapping electrons.
For really long-term retention, magnetic memories seem to be quite interesting. Avalanche have a graph around their "10 years at 125°C" claim that would imply MUCH longer endurance at lower temperatures (If I interpret their graph right, which extrapolates from four data points at higher temperatures, ~46000 years at 100°C). I haven't seen a uC with MRAM built in, but there are external serial ROMs, and I think the RP2040 would boot from one of those.
(Score: 2) by RS3 on Friday January 20 2023, @06:45PM
So you're saying we shouldn't have given up on Bubble Memory [wikipedia.org]?
What, no likey? How about the IBM Plug Board [columbia.edu]? Guaranteed cosmic ray proof!
:-}
(Score: 3, Interesting) by RS3 on Friday January 20 2023, @07:07PM (2 children)
Silliness aside, and long story short, many years ago I rescued a brand new large expensive production machine whose EPROM lost a bit.
I was doing some consulting and the company had bought 2 brand new production machines. One worked perfectly, the other quit working within the first week. It was probably under warranty, but having it down at all was costing a lot, and warranty repair might have taken weeks or months, so the owner asked me to take a look at it.
Having learned the physics of semiconductors, specifically things like EPROMs, I've never fully trusted them... That fueled a hunch and I pulled the 2 EPROMs, read them, compared the files, and they were different. I had an eraser and burner so I made sure to fully erase the bad one, reburned it, and the machine worked perfectly. I forget if the error bit was a 1 or 0, but my hunch is that it wasn't fully erased at the factory. I'm sure you know that EPROM erasing is a timed thing, typically 15 or 20 minutes or so? All depends on how strong the UV light is, etc., of course. I like to give them a little bit extra just to be sure.
I haven't done enough microprocessor work- some, but not as much as I always wanted to. I thought some were made with OTP (One Time Program) fuse links?
(Score: 3, Informative) by Rich on Saturday January 21 2023, @05:05PM (1 child)
OTP doesn't specify a particular technology. Old PROMs and PALs had these metal fuses, but even back in the days there were OTP-EPROMs that were cheaper to manfucature with the windowless package. I've seen die shots of metal fuses and they look like they don't scale, mostly because that metal splattered around. These days I think there's either old EPROM technology being used for legacy process controllers, or maybe even proper Flash with the erase logic cut away to save die space (like with these few-cent Puolop and Padauk controllers).
I checked the state of magnetic memory, and there's an upcoming technology "SOT-MRAM", with near-infinite endurance, which would have about ideal properties for all controller applications, and could even be used as single main memory.
(Score: 2) by RS3 on Saturday January 21 2023, @07:04PM
Wow, interesting, thanks. I always wondered about those fuses, splatter, and reliability.
Yes, what I was getting at was some kind of very long-life programmable ROM, rather than EPROM or EEPROM (now called "FLASH") that could suffer from "bit-rot", including issues with physics. Stated another way, we really don't know if all the bit cells are erased in an EPROM. We know the circuitry says it's a 1, but maybe it's 0.5001, right on the edge of dropping to a 0 when some thing, like temperature, changes. Which is why I like to "cook" EPROMs a bit longer in the eraser...
(Score: 3, Informative) by kazzie on Friday January 20 2023, @06:30PM (8 children)
Several STM32 microcontrollers support three different boot options using external pins: from Flash, System Memory, or RAM. An application note [stm32.eu] describes the System Memory as being "programmed by ST during production" , so it sounds like mask ROM (or at least one-write EPROM). Granted, the RAM it supports is the built-in SRAM, but the bootloader supports loading code directly into RAM, so it seems you can get a program in there even if the Flash is dead.
(Score: 2) by RS3 on Saturday January 21 2023, @04:39AM (7 children)
Thanks, I didn't know that. I don't even know- can you still buy OTP PROMs?
(Score: 3, Informative) by kazzie on Saturday January 21 2023, @11:19AM (6 children)
In recent decades, what's sold as a one-time PROM was in fact an ultraviolet EPROM in a cheaper plastic package, i.e. without the quartz window above the die to erase it. I assume that this is still current practice.
Parallel (E/E)PROMs are a rarity these days, most manufacturing capacity is dedicated to serial interface PROMs instead.
(Score: 2) by RS3 on Saturday January 21 2023, @07:11PM (5 children)
Arrgh, I didn't know that, thanks. It makes sense though. So theoretically I could open one, UV erase, and seal it up. What could go wrong...
Yes, I've been seeing (and using) them for 30+ years, and I believe they're typically FLASH, and maybe some battery-backed CMOS (or whatever) RAM?
My thought process is to get away from bit-rot vulnerable forms of "permanent" ROM.
Hoping to avoid reverting to paper tape... :)
(Score: 3, Insightful) by ChrisMaple on Sunday January 22 2023, @06:59AM (2 children)
I suppose an EEPROM could be developed that treated each bit like a DRAM treats each bit: after each read, the bit's charge is reinforced by writing back to it.
(Score: 3, Interesting) by RS3 on Sunday January 22 2023, @09:15AM (1 child)
Very good. I like it. Only worry is bit rot if it's powered down for long times. Maybe a battery-backed refresh, but like once a day or something.
My wish for EPROM is a way to really know if all bits are fully erased. You know, if a cell is at 55%, it'll read as a 1, but you'd like it above 90% or so. Not sure if there would be a way to read the analog charge level in a cell. AFAIK they don't do that now, but it might be an interesting idea if you need very high reliability. Now I'm wondering what they use in spacecraft / satellites.
(Score: 2) by kazzie on Sunday January 22 2023, @06:31PM
If you're willing to sacrifice capacity drastically, you could do this by (e.g.) aggregating a byte of output lines to form one bit, maybe by analogue addition of output voltages or maybe digital majority voting. You could then survive three bitflips without repercussion, while still being able to inspect individual bits to check for rot.
(Score: 2) by kazzie on Sunday January 22 2023, @06:37PM (1 child)
Remember that UV EPROMs are also erasable by subjecting them to extremely high heat: a UV window was just considered to be more user-friendly.
How about taking a page out of the Apollo book, and building your own rope memory [wikipedia.org]?
(Score: 2) by kazzie on Sunday January 22 2023, @06:38PM
More detail: https://hackaday.com/2013/10/09/making-a-core-rope-read-only-memory/ [hackaday.com]