Semiconductor Evolution - John Dunn, Consultant, Ambertec, P.E., P.C.

This is definitely a tale of yore, but it holds a good lesson anyway.

Our product used a push-pull power amplifier made with power darlingtons, type 2N6578, which were operated in class-C service. It was required of each side of the subject amplifier that its transistors achieve current cut-off before the opposing side began to conduct for the next half cycle of operation. The operating frequency was typically 25 kHz. These things really worked very nicely until around 1984 when we began having trouble.

We found that some 2N6578 devices were not getting to their cut-off state quickly enough. They would still be carrying current when the other side started to carry. This led to unwanted overlaps of current flows of the two sides of the push-pull amplifier and sometimes that led to catastrophic amplifier failures with all kinds of smoke and unhappiness.

Since the problem had come about over the course of time, we looked for anything that might have changed over the course of time and we found something.

As a power darlington, the 2N6578 has two internal resistors, one each across the base-to-emitter junctions of the two transistors. We couldn't measure the resistors individually, but we could measure the net resistance between the base and emitter leads of a device.

What we found is shown here:

 

Gradually over the years, the Rbe value was rising. I called the transistor maker, Motorola as it happened to be, and asked about this.

The answer was yes, the Rbe had been rising over the years as Motorola tweaked their manufacturing process to improve product yield. Unfortunately, a side effect of this was to slow down the 2N6578's speed. There was still plenty of margin to the data sheet's switching speed specifications, but the slowing down was rendering the device unsuitable to our application.

It happened that RCA also offered the 2N6578 so we got some of those. They seemed to be pretty much a match to the operating behavior we were seeing with the Motorola devices, so things weren't looking too good. However, when I then tried some RCA devices of a slightly later date code than the first ones I'd looked at, they were utterly superb. They were very fast and there was no current overlap going on.

Out came the hacksaw and I removed the tops of the TO-3 cans of the slower ones and the faster ones whereupon I saw that the semiconductors inside were of very different geometries. This time I got on the phone with RCA.

Yes, there had been a change. RCA had just gone from a homotaxial design (the slower ones) to an epitaxial planar design (the faster ones) and was in the process of converting many devices' geometries in this way since the epitaxial planar designs had numerous technical advantages and also had higher production yields.

Our problem was over, but it was only over because we were just plain lucky about it.

We could not correlate the switching speed specifications of the 2N6578 to the speed requirements for our class-C amplifier service, but we had simply depended on the fact that they worked, at least at the outset, with Motorola parts. However, as the devices evolved, we got into trouble.

The lesson is that you can't expect semiconductor companies to keep making any given part in the same way over the course of time. Internal structures can be changed and sometimes, some of the device properties can change with heaven only knows what effect on your product's design.  

This was one case of that happening, but I have seen other cases too, tales for another day.

Comments

[Account Deleted]

When I apply an external supply from frequency generator to Push Pull Amplifier, Push Pull Amplifire's input gets loaded even if a small amount of supply is given to it..
If the input is sine wave then output is no doubt amplified but sometimes gives triangular like or distorted output..
To avoid this loading effect I had kept the amplitude to its minimum and the supply voltage is hardly measurable on CRO but it gives the correct amplified sinewaves.

[Pedro Perez de Ayala]

Well, I think that a very important lesson to learn from this experience is that you have to design based on the specs as much as possible. I only have 7 years of experience as an electronic engineer, but I have seen many times that a design stops working because a part has degraded some parameter but still within specs. The datasheet is the closer you will get to having a contract on the part's performance, use it as such.

This leads to another conclusion, which is that testing is highly overrated and can be dangerous if not done properly. Sometimes, people are tempted to validate everything through testing, which is better than nothing, but many times, the yield of the test is not clear or simply not enough. On the other hand, a design supported by good theoretical analisys, and proper part dimensioning has many chances to become a rock-solid machine. Of course testing is also necessary, but it doesn't cover everything. Don't feel confident just because there is a test, feel confident because the design is well thought, meticulously dimensioned, carefully implemented, AND properly tested with a good set of tests that give a good enough yield.

I hope to give some light on the subject,
Best regards,
Pedro

[Jonathan Point]

I have seen this issue with CMOS ICs as well. These were hex schmitt inverters - 40106/74C14/MC14584. The circuit was proofed (by others) using MC14584s. At some stage, they had switched to 40106s, but about 6 months into that, the product started showing flaky behaviour and the inverters started 'missing' pulses. Some 40106s worked, most didn't. Back to the MC14584s and no problems at all. It seemed temperature-related, as this was summer at the time.

More proof that it doesn't matter how many prototypes 'work', you MUST stay within the 'minimum' specs for any part, or there'll be trouble somewhere down the line. In this case, the devices were operating close to their maximum frequency from AC-coupled inputs =]

Regards,
Jon Point

[Mark Peacock]

Did you simulate in SPICE using the extreme (W/C) tolerances?