Well ... no. Silicon can only run so hot before it melts or gets really leaky.
Usually the limit is the package, not the chip. Si melts at 1400C so the chip is long dead before this point! Re-diffusion of dopants might happen from 300C, but usually leakage currents are a problem well before that, as are package / bondwire / substrate thermal issues. This is why commercial semiconductors are usually rated to 125C and are much more reliable at much lower temperatures.
Better heatsinking is one way to increase power density - there are exotic substrates that conduct heat better than copper for instance, but they are not economic currently (diamond is one, boron arsenide is another (its even a good semiconductor, but we can't make it pure enough yet, and has toxicity issues)).
Silicon carbide can handle higher temperatures than Si alone, and its use is spreading, but its not replacing Si anytime soon.
Hi Mark,
The max die temperature is given as 150° in data sheets normally, some I've seen are 175°, but I don't believe that figure. That is substantially lower than your figure.
Getting rid of heat is the one major design failure I see today. Failure rates double for every 10° C rise in operating temperature. That is across the board. The new trend for "chiplets" in processors, etc is a disturbing trend, AMD expects a 5 year life as acceptable. The problems are three-fold. Removing heat, inter-die attachment and connection issues (related). Research is one thing (ideal conditions). The real manufactured product is quite another. Power delivery is another can of worms.
Taking it back to a simple transistor, we have the same thing as always. Purity of diffusion gasses and the silicon, plus process control (doping levels) greatly affects the device. Allowing the device to run hot magnifies any production issues in the part. Do you think they take as much care in a transistor as they may in an IC?
Anyway, we have exotic PCB substrates and phase change heat removal, but here we are talking about a part stuck to a chunk of metal to radiate heat. The replacement part will not be treated to any improved heat removal processes. That's if it is installed with fresh heatsink grease or other thermal pad. You won't believe how many people use the original contaminated grease (or nothing)! Getting the part fastened with the correct mounting torque is something unheard of from what I see as well.
Even if top quality, original parts are used, workmanship is an issue.
The max die temperature is given as 150° in data sheets normally, some I've seen are 175°, but I don't believe that figure. That is substantially lower than your figure.
Getting rid of heat is the one major design failure I see today. Failure rates double for every 10° C rise in operating temperature. That is across the board. The new trend for "chiplets" in processors, etc is a disturbing trend, AMD expects a 5 year life as acceptable. The problems are three-fold. Removing heat, inter-die attachment and connection issues (related). Research is one thing (ideal conditions). The real manufactured product is quite another. Power delivery is another can of worms.
Taking it back to a simple transistor, we have the same thing as always. Purity of diffusion gasses and the silicon, plus process control (doping levels) greatly affects the device. Allowing the device to run hot magnifies any production issues in the part. Do you think they take as much care in a transistor as they may in an IC?
Anyway, we have exotic PCB substrates and phase change heat removal, but here we are talking about a part stuck to a chunk of metal to radiate heat. The replacement part will not be treated to any improved heat removal processes. That's if it is installed with fresh heatsink grease or other thermal pad. You won't believe how many people use the original contaminated grease (or nothing)! Getting the part fastened with the correct mounting torque is something unheard of from what I see as well.
Even if top quality, original parts are used, workmanship is an issue.
The power device section takes up some area. Fake 2050’s are usually 2030’s, which are physically smaller. You get more 2030’s per wafer. The fine geometry portion is the same in principle and requires nothing more sophisticated than 1970’s or 80’s op amps. 60V wasn’t groundbreaking either, and certainly not a problem now. The thermal protection schemes on these things were innovative at the time, but probably old hat by now to anybody making ICs that need a thermal supervisor circuit. Lots of other applications for that now besides audio amps.
Distortion profile would be a matter of 2 things. How well the input stage is matched, and the thermal management across the die (layout). You do need to know something about IC design. Fakers probably just got ahold of the 2030’s artwork, perhaps illegally, and tried to fill in the rest. UTC either has someone on design staff with a head on his/her shoulders, or bought the mask set from ST.
Distortion profile would be a matter of 2 things. How well the input stage is matched, and the thermal management across the die (layout). You do need to know something about IC design. Fakers probably just got ahold of the 2030’s artwork, perhaps illegally, and tried to fill in the rest. UTC either has someone on design staff with a head on his/her shoulders, or bought the mask set from ST.
Hi Chris,The new trend for "chiplets" in processors, etc is a disturbing trend, AMD expects a 5 year life as acceptable.
I do wonder where do you get this information from? I am in contact with a couple of guys I used to work with at HP/McData that currently work at AMD. I could ask them about your assertion of reliability. Harold is a fellow at AMD and is a codec expert.
Rick
Last edited:
That's nice to know that you spoke directly to an AMD rep which is promoting throw it away electronics from the factory.
Lots of AMD chips are fab'd at TSMC .
I'll ask my sources on the inside of AMD to comment further when we meet for the next round of beers and get back to you. 🙂
On topic:
Onsemi annouced the EOL on KSC3503 so stock up folks or else find a new part to use or toss that gear/designs which counts on them to tick.
A1381 was EOL earlier this year with a life time buy in progress. What's next on the chopping block for Onsemi? Rochester has old Sanyo metal tab parts in quantity.
Off topic:
For another thread, so what's AI going to do to help out Keysight in the test arena? Keysight is the old HP T&M group iirc. (HP's origins)
Lots of AMD chips are fab'd at TSMC .
I'll ask my sources on the inside of AMD to comment further when we meet for the next round of beers and get back to you. 🙂
On topic:
Onsemi annouced the EOL on KSC3503 so stock up folks or else find a new part to use or toss that gear/designs which counts on them to tick.
A1381 was EOL earlier this year with a life time buy in progress. What's next on the chopping block for Onsemi? Rochester has old Sanyo metal tab parts in quantity.
Off topic:
For another thread, so what's AI going to do to help out Keysight in the test arena? Keysight is the old HP T&M group iirc. (HP's origins)
You can bet whatever is next on On’s chopping block is either a TO-3 or a TO-126. I did my lifetime buy of 3503/1381 a long time ago now. At the rate they were disappearing at the time, I was sure the ON 126’s would have all been gone by now. I just did my lifetime buy of BD441/2 - it’s not a crazy number, just enough to try out some of @cumbb ‘s “theories” and crazy ideas. See what blows up and what doesn’t, maybe get some idea what he thinks he’s hearing. And still have a handful to service old 70’s stereos and 5W integrated amps with the “right” parts.
I do not use BD441/2
Yeah I think more TO-3's will go EOL before the TO-126 do. We'll see about TO-220's, KSC2073 and KSA940 before MJE15032/33
What gets me is why Onsemi continues to sell both MJ21193/94 and MJ21195/96, whereas there is very slight differences between them.
Good luck with " @cumbb ‘s “theories” and crazy ideas" let us know how they turn out.
Yeah I think more TO-3's will go EOL before the TO-126 do. We'll see about TO-220's, KSC2073 and KSA940 before MJE15032/33
What gets me is why Onsemi continues to sell both MJ21193/94 and MJ21195/96, whereas there is very slight differences between them.
Good luck with " @cumbb ‘s “theories” and crazy ideas" let us know how they turn out.
I’m surprised the 2073 is still around. It’s not a sustained-beta type!
I wonder why the 93/4 and 95/6 were even assigned different type numbers. The differences are hFE rank. Why they didn’t just call them O and R like the Fairchild 5200. Even then the R’s went away as the betas got consistently high enough. I can sort of understand the 15022/3 vs. the 024/5. That’s guaranteed breakdown voltage, and they used to do that in the 2N days - 2N5629,30, 31….. It was TI’s idea to grade TIPs with A B and C. In all those cases, gain tended to vary inverse as Vceo but the different type number was for voltage, not for gain. The minimums were usually the same, but typical varied.
I wonder why the 93/4 and 95/6 were even assigned different type numbers. The differences are hFE rank. Why they didn’t just call them O and R like the Fairchild 5200. Even then the R’s went away as the betas got consistently high enough. I can sort of understand the 15022/3 vs. the 024/5. That’s guaranteed breakdown voltage, and they used to do that in the 2N days - 2N5629,30, 31….. It was TI’s idea to grade TIPs with A B and C. In all those cases, gain tended to vary inverse as Vceo but the different type number was for voltage, not for gain. The minimums were usually the same, but typical varied.
Last edited:
Mj2119x series are a mystery but real workhorses.
I assume onsemi still packages the to-3 in Mexico, must be an old factory they keep up and running.
Ksc2073/ksa940 work fine for me as drivers in a lower cost ops, that and 2sa1943n/2sc5200n Toshiba outputs.
I assume onsemi still packages the to-3 in Mexico, must be an old factory they keep up and running.
Ksc2073/ksa940 work fine for me as drivers in a lower cost ops, that and 2sa1943n/2sc5200n Toshiba outputs.
UTC sells a lot of parts with the original numbers, so they do seem to have access to the IP of the original manufacturers, the TDA prefix was mostly used by Philips and ST.
And they have been selling TDA2050 in the EU, which is strict about IP, and home ground of Philips / NXP and ST, without any problems for quite some time.
So I would assume they have permission from ST to make those chips, and possibly got the masks and test jigs and so on from ST.
The Chinese company CRC-Micro mark their chips as 'D2050', and puts their own logo on the chip, so it can be called an 'equivalent' chip, rather than a fake.
They do not use the TDA prefix, nor do two others I have seen in the market, those also have different logos, not the ST logo.
In any case, the UTC parts seem to be more reliable and handle higher voltages than others, so a 'good' replacement choice.
For those who are curious, please take a look at the CDIL and UTC sites.
And once again, I have no connection with the businesses named above.
And I spoke to a parts trader here, he is a relative, who said the the box pack of BEL is 1000 for 2N3055/3773, and minimum assorted order is about 20k, or they will not ship, then you have to buy from their distributors, they too will say one box minimum.
So, you need to be able to but 1000 units 2N3055 at a time, and be able to sell / use them up over say two years.
Or get different makes like CDIL, which makes regular, not high voltage parts, though I expect the 3773 can be used as a higher voltage part there as a substitute for the 2N3055HV
The local main users were audio amps and regulators, amps are now using 1943/5200 pairs or 80N80 mosfets, and are mostly ready plate amps made in Delhi, parts are needed for repair only here in Baroda, which is a city about 1000 km from Delhi, with a population larger than Chicago.
The other users are not much in touch with me, so I have little idea of their volumes.
And they have been selling TDA2050 in the EU, which is strict about IP, and home ground of Philips / NXP and ST, without any problems for quite some time.
So I would assume they have permission from ST to make those chips, and possibly got the masks and test jigs and so on from ST.
The Chinese company CRC-Micro mark their chips as 'D2050', and puts their own logo on the chip, so it can be called an 'equivalent' chip, rather than a fake.
They do not use the TDA prefix, nor do two others I have seen in the market, those also have different logos, not the ST logo.
In any case, the UTC parts seem to be more reliable and handle higher voltages than others, so a 'good' replacement choice.
For those who are curious, please take a look at the CDIL and UTC sites.
And once again, I have no connection with the businesses named above.
And I spoke to a parts trader here, he is a relative, who said the the box pack of BEL is 1000 for 2N3055/3773, and minimum assorted order is about 20k, or they will not ship, then you have to buy from their distributors, they too will say one box minimum.
So, you need to be able to but 1000 units 2N3055 at a time, and be able to sell / use them up over say two years.
Or get different makes like CDIL, which makes regular, not high voltage parts, though I expect the 3773 can be used as a higher voltage part there as a substitute for the 2N3055HV
The local main users were audio amps and regulators, amps are now using 1943/5200 pairs or 80N80 mosfets, and are mostly ready plate amps made in Delhi, parts are needed for repair only here in Baroda, which is a city about 1000 km from Delhi, with a population larger than Chicago.
The other users are not much in touch with me, so I have little idea of their volumes.
Hi Rick,
Yes, two reps from AMD, one male, one female. I specifically addressed reliability with them.
They were not promoting throwaway. But their life expectancy for processors revolved around the logic the parts would be surpassed by newer chips. I pointed out that equipment using those parts in industry had to last much longer. It's stupid to forklift a million $ plus machine because some silicon fails (I used that example). Cisco switches are allowed to overheat and fail for example (know a guy who works for Cisco). None of this should surprise you as a profit model.
Yes, two reps from AMD, one male, one female. I specifically addressed reliability with them.
They were not promoting throwaway. But their life expectancy for processors revolved around the logic the parts would be surpassed by newer chips. I pointed out that equipment using those parts in industry had to last much longer. It's stupid to forklift a million $ plus machine because some silicon fails (I used that example). Cisco switches are allowed to overheat and fail for example (know a guy who works for Cisco). None of this should surprise you as a profit model.
One doesn’t have to clock a CPU to infinity and beyond either. Running a stripped down version of an embedded Windows kernel, not running all that stupid advertising crap and the required antivirus to keep it from getting corrupted is probably a lot easier on the machine itself. More idle time, less average cpu cycles.. less HEAT.
Hi wg_ski,
Completely agree. Try linux. For industrial equipment, don't run embedded Windows. Either run machine code, or run a stripped linux kernel. Test equipment manufacturers are going away from windows embedded to some degree. I flipping hate to wait for my Keysight instruments to boot, they take so long!
No, we were talking about new chip design using "chiplets" which are building blocks (subassemblies) mounted on base chips. Mostly driven by the high computational requirements of artificial intelligence in the data centres. Heat has always been an issue, and this magnifies the problem. With so much heat generated in a tiny space, plus the connection and power delivery requirements, it doesn't take a genius to realise the silicon will be running at elevated temperatures. Then you have silicon to silicon connection issues. At least the expansion co-efficient is the same!
This is "new" technology, a new way of building things. So we are learning. Do not expect reliability.
Completely agree. Try linux. For industrial equipment, don't run embedded Windows. Either run machine code, or run a stripped linux kernel. Test equipment manufacturers are going away from windows embedded to some degree. I flipping hate to wait for my Keysight instruments to boot, they take so long!
No, we were talking about new chip design using "chiplets" which are building blocks (subassemblies) mounted on base chips. Mostly driven by the high computational requirements of artificial intelligence in the data centres. Heat has always been an issue, and this magnifies the problem. With so much heat generated in a tiny space, plus the connection and power delivery requirements, it doesn't take a genius to realise the silicon will be running at elevated temperatures. Then you have silicon to silicon connection issues. At least the expansion co-efficient is the same!
This is "new" technology, a new way of building things. So we are learning. Do not expect reliability.
When I changed the PLC on my molding machine, the engineer tweaked the program using a ladder logic program on a laptop.
There is a USB socket to install program, and copy production data, it has an ARM 266 MHz multi core processor, and the possibility of connecting a modem through the USB socket for remote monitoring etc.
The original Mitsubishi PLC had a 20 or so Mhz 8086 CPU, license made by AMD, worked well for 25 years....
And I am still looking for a competent PLC installer for a KBA Rapida 5 + 1 printing machine, 5 stands with 32 motors (Color control), the + 1 is for the coater, and the other functions are also on the PLC, like registration and drive speed / inching.
A new machine will cost 170 million rupees, about $2 million US.
PlC replacement was quoted about $20,000, one week, but the installation guy refused to take us to customers where he claimed to have done a successful update.
So we got skeptical, no deal.
The PLC in use is from 1998. part of the machine when new, and has like 200 contactors and lots of tiny PCB relays, if it breaks we have to call engineers from Bombay, 400 km away.
The PLC builder has gone broke, taken over later by new investors, and no longer has the source code, we are managing it all here, cannibalizing parts from a pool maintained by the repair people in Bombay, they send spare cards, which are repaired, and returned / replaced into the pool.
One possible replacement model from Siemens has the habit of resetting at random, so kept on hold...
Like Chris says, too stupid to scrap a machine for faulty silicon, a disposable mentality has crept into society.
There is a USB socket to install program, and copy production data, it has an ARM 266 MHz multi core processor, and the possibility of connecting a modem through the USB socket for remote monitoring etc.
The original Mitsubishi PLC had a 20 or so Mhz 8086 CPU, license made by AMD, worked well for 25 years....
And I am still looking for a competent PLC installer for a KBA Rapida 5 + 1 printing machine, 5 stands with 32 motors (Color control), the + 1 is for the coater, and the other functions are also on the PLC, like registration and drive speed / inching.
A new machine will cost 170 million rupees, about $2 million US.
PlC replacement was quoted about $20,000, one week, but the installation guy refused to take us to customers where he claimed to have done a successful update.
So we got skeptical, no deal.
The PLC in use is from 1998. part of the machine when new, and has like 200 contactors and lots of tiny PCB relays, if it breaks we have to call engineers from Bombay, 400 km away.
The PLC builder has gone broke, taken over later by new investors, and no longer has the source code, we are managing it all here, cannibalizing parts from a pool maintained by the repair people in Bombay, they send spare cards, which are repaired, and returned / replaced into the pool.
One possible replacement model from Siemens has the habit of resetting at random, so kept on hold...
Like Chris says, too stupid to scrap a machine for faulty silicon, a disposable mentality has crept into society.
No ties to any names above, the printing machine belongs to a good friend.
The PLC problem has affected sales at the printing machine builder using them, customers have switched from German to Japanese machines, which are 25% cheaper , slightly poorer quality at extreme resolution, but are utterly reliable, PLC has more functions as well.
German machine, 4 color, no coater, is in the $1.25 million range, Japanese is $1 million, about 20 x 30 inch paper size.
That is $250,000 saved, and the PLC would be worth $25,000 at OEM prices.
But a print shop will go through a lot more paper in dollars bthan that annually, a PLC reset means setting up the machine again, with a loss of time and wastage.
And erratic deliveries to demanding customers, with a loss of business and reputation.
That is of the reason why I have seen Japanese PLCs in German origin machining centers, for a patriotic German to use a foreign PLC is very discomfiting...
The PLC problem has affected sales at the printing machine builder using them, customers have switched from German to Japanese machines, which are 25% cheaper , slightly poorer quality at extreme resolution, but are utterly reliable, PLC has more functions as well.
German machine, 4 color, no coater, is in the $1.25 million range, Japanese is $1 million, about 20 x 30 inch paper size.
That is $250,000 saved, and the PLC would be worth $25,000 at OEM prices.
But a print shop will go through a lot more paper in dollars bthan that annually, a PLC reset means setting up the machine again, with a loss of time and wastage.
And erratic deliveries to demanding customers, with a loss of business and reputation.
That is of the reason why I have seen Japanese PLCs in German origin machining centers, for a patriotic German to use a foreign PLC is very discomfiting...
I have noticed original data sheets are disappearing, now only "copy type transistor" manufacturing sheets are common. A lack of details means they haven't a clue what they are selling. That means the parts are garbage. To me, they are the same as remarked parts or factory rejects.
I don't think much of people willing to use that trash.
I don't think much of people willing to use that trash.
Unfortunately expect this more and more as originals evaporate. One of these days, perhaps in our lifetime, whatever becomes of ON’s core business will probably stop making all linear-audio-capable power semi’s and maybe even small signal parts - at least the pet stuff everybody uses. I am already well prepared for that day. But I may have at most 20 years of audio building left, and that’s under the best possible conditions. Realistically I expect 10-12. When they DO exit the market and sell only cell phone, EV, and energy-conversion targeted products, I sincerely hope they sell everything off to someone who DOES know what the f*** they are doing, rather than let the MJW3281 die. Like Motorola did with ON in the first place. When the only place you can get one is ISC, we are all in deep doggie doo.
- Home
- Design & Build
- Parts
- My Transistors, original or copy?