I think I have the MOSFET bug now.

[ilimzn]

Thanks ilimzn for the tutorial!

I only do an average of one post a day, might as well make it a good one

 

[Yamaha B-2]

Let's open a 'sunset' fab and made VFETs!! I can come up with all the process tools and cleanroom ancillary equipment if you come up with the fab building and employees (should be cheap in Croatia). And, we can dream about getting a mask-set. My guess is that the mask-set is even more scarce than is a set of brand-new Yamaha VFETs for a B-1 or B-2. Or, well. Was a good thought while it lasted.

 

[ilimzn]

Let's open a 'sunset' fab and made VFETs!! I can come up with all the process tools and cleanroom ancillary equipment if you come up with the fab building and employees (should be cheap in Croatia). And, we can dream about getting a mask-set. My guess is that the mask-set is even more scarce than is a set of brand-new Yamaha VFETs for a B-1 or B-2. Or, well. Was a good thought while it lasted.

AFAIK the masks are the least of the problem. VFETs use an etching process that is no longer used. Be that as it may, it may well be possible to use different geometry for the same purpose. When VFETs were still being made, the best feature size available on chips was on the order of several micrometers. At the moment, ten times smaller is being phased out. As for etching, HEXFETs and TMOS use a similar process. If one actually wanted to build a VFET today, it would probably be easyer and also a better product.

As far as the Yamaha VFETs, I don't think Yamaha had a semiconductor fab of their own when these were made (but Sony and NEC did). It is quite possible that Yamaha parts were made by someone else speciffically for Yamaha, NEC would have been the most logical candidate, having made that type of semiconductor but having no product to use it in.

 

[Yamaha B-2]

If made today, VFETs would be much better product. Silicon is much better today. Process tools are much better. Control systems are much better. The highly anisotropic 'selective' etch required to make the features of a VFET has become pretty well developed due to all the 'trench' technology developed by IBM and others. Dry etch would be slow, but wet-etch has become very well developed. Really no big deal today. But I am surprised to hear that a mask-set would be no problem. If so, the Japanese definitely must archive their old technology much better than I am used to in the companies I've worked for. Late '70's was still a time for large reduction cameras, red vinyl and eXacto knives for lots of products. Especially those with as large a design-rule as a VFET. This was about the time Perkin-Elmer came up with the first large steppers of any ability. One stepper, two techs. Early technology was so quaint. And unreliable. Although something as non-critical as a VFET probably used contact printing. And 4" silicon.

Think you are correct about Yamaha not having a fab of their own. At least I've never heard of one. Sony had several, including two in the U.S. (think they have gone the way of the dinosaur, though - anyone from San Antonio know if Sony's fabs are still running? They were old in early '90's when I worked in Austin. Never could figure out why they bought them from AMD.), and of course, NEC has several, including a large one in the U.S. (Roseville, CA).

 

[ilimzn]

Dry etch would be slow, but wet-etch has become very well developed.

That's the one I've been thinking of...

But I am surprised to hear that a mask-set would be no problem.

No, I was thinking of something else.
In the late 70s someone in the ex-USSR alegedly reverse engineered a couple of the then popular CPUs (Z80, 6502, 6800) by using very precise grinding (if it could be called that) and micro-photography, then reconstructing the masks from that. Given a functional VFET, something along those lines could be done, considering that the original VFETs had large geometry. Of course, designing something like a VFET from scratch would be far easyer today but still very expensive - but it would be a better product. Which brings to mind another idea - there are manufacturers who specialise in production of obsolete parts - most notably for military and aerospace purposes (which unfortunately means they can get REALLY good prices for said parts, no doubt a large part of making such an undertaking viable). Perhaps such a feat would not be outside of the possible for them?

Think you are correct about Yamaha not having a fab of their own. At least I've never heard of one.

I'm pretty sure they didn't have one then. Still, yamaha made (and still makes) various specialist chips, mostly for their musical instruments division. Of course, it is quite possible that they never owned a fab of their own, and certainly today, when things have gone well and truly digital, there are a fair number of fabs that operate independantly, that Yamaha or anyone for that matter, could use to fabricate their designs.

BTW what little there was of semiconductor manufacturing in Croatia has died sometime in the late 70s and was replaced by just packing die products. Then, during the late 80s that died too. Finally, during the late 90s, the parent companies died as well. So, there is no equipment left and neither is there any expertise available outside of the Uni here in Zagreb, and most of these people were trained abroad, and end up working abroad. Also, cost of labour, though way lower than in the US, is not nearly on the level of Chinese sweat-shops so the financial feasibility is questionable. OTOH, we are talking about a premium product. Still, IMHO the 'street price' of a VFET should be no more than $20, and that would be for the replacement parts. The same die could be packaged in a modern plastic high power package and perhaps be sold to boutique amp manufacturers. But this means we are talking an out-the-door cost of at most $10 per unit and possibly far lower than that.

 

[Yamaha B-2]

The term 'sunset' fab is exactly what you mention. Specialize in making the older chips with low volumes that still demand high prices. Some of them actually run on 3" wafers as the original milspec is for 3" and it is not worth the cost to update the spec's to something larger. Run lots are too small. So buy larger wafers and laser cut them down to the correct size. All power device manufacturers are 'contract' manufacturers to one degree or another (as were we in the silicon end of the business). That is, although they have their own line of products they are constantly looking for devices to make to someones specification to keep their fabs running around the clock. So most anyone of the power device fabs, there are four or five of note still in the U.S. and they do have a group in a couple of them that specialize in making products for audio. The rest are overseas and starting to concentrate, as you have mentioned, in China (take your pick).

Now that I think about the mask set, I would guess that with the use of computers and mask-generation tools today, the development and production of the mask set would be relatively easy. I'll put out some feelers.

p.s. Don't hold your breathe.

 

[rca2000]

It's so nice to have all of the Knowledgable people we do here!

Thank you, Ilimzn, you shared a LOT of interssting information,and answered several questions I had. Most notably:: Mosfet amps seem to idle quite warm(not hot), and, when under power, (say, 40-50%) this heat level does NOT increase by much.I assume this is due to the idle current level, that they tend to idle warmer than Bjt amps.

SO, it is the THIRD order of distortion, that is the most objectionable, and MOSFETS seem to reduce that, am I reading that correctly? (I thought ALL distortion was bad).And, while they are priobably inneficent at idle or no-signal, due to the high bias, MOSFETS probably are MUCH more efficent through the power spectrum, than Bjt amps. Is this correct?

Thank you again for the wealth of information!

 

[ilimzn]

Thank you, Ilimzn, you shared a LOT of interssting information,and answered several questions I had. Most notably:: Mosfet amps seem to idle quite warm(not hot), and, when under power, (say, 40-50%) this heat level does NOT increase by much.I assume this is due to the idle current level, that they tend to idle warmer than Bjt amps.

Yes - though you have to remember that most listening is actually done at an average 1W of power - a typical MOSFET amp will make 10W or more into heat just idling. 1W on top of that will not make much of a difference.

And, while they are priobably inneficent at idle or no-signal, due to the high bias, MOSFETS probably are MUCH more efficent through the power spectrum, than Bjt amps. Is this correct?

If done correctly, yes, but only at the highest power level, and for a typical amp topology the difference is not very large. Below that, but signifficantly above idling, efficiency has very little to do with the specific semiconductor device type, and it's all just Joule's law. When you get close to clipping, that's where the differences can be felt. Here is an example: to get 100W out of an amp on a sine wave, you need a peak voltage of 40V. If your amp can give you just one more V, i.e. 41V peak, you get 5W more power.
A power anp with a single pair of MOSFETs can be made with no series resistors in it's output (these would be needed for BJT linearisation, even though they could be very low). At saturation, a MOSFET used for such an amp would only have some 0.2V across it. In other words, this design can practically use all the power the power supply is theoretically capable off, and the important part here is that this will give you the highest possible headroom. It is far more difficult for BJTs to come this close. Also, since BJTs are current drive, the more output current is required, the more input current is required. For MOSFETs, negligible amounts of current are required in the first place, and this is practically independant of the required output current. And, as we know, it is current passing through anything that creates a voltage drop, that creates heat, and that in turn is lost power and efficiency.
In a typical class AB amp efficiency at full power is theoretically about 70%, but on audio signals it will only be perhaps 5% on average, so the difference a MOSFET makes in this case is not overly important unless we are talking driving an amp as close to clipping as possible. However, for a class D amp, where efficiency is practically the very reason of it's existance, clipping is the way it works, and MOSFETs make a crucial difference. Especially since class D amps are used for the highest power amps (on the order of kW).

SO, it is the THIRD order of distortion, that is the most objectionable, and MOSFETS seem to reduce that, am I reading that correctly? (I thought ALL distortion was bad).

All distortion is bad but depending on the order it's more or less objectionable. In other words, it is percieved differently. Keep in mind that even though high even order harmonic content sounds 'good' it still doesn't sound 'right' - very easy to check using headphones and a low impedance source. Adjusting for equal volume, an A/B test via headphones will invariably show that even an amp with only second order distorsion sounds inaccurate, regardless if it still sounds 'good'. This in a nutshell is the reason I am not one of the subjectivist camp that congregate on AA - to me they use equipment to modify the music the way they like it to sound, even when it clearly was not recorded that way. That may make THEM happy but to me it's not HiFi. The fact that this lot considers tone controls the ultimate insult is an irony that just ticks me off no end

It's not third order distortion per se that is a problem, but odd order (3rd, 5th, 7th, 9th etc. harmonics) that is much more objectionable than even (2nd, 4th, 6th, 8th etc harmonic). 1% even order distortion is quite tolerable while 1% odd order is very objectionable and harsh. It still does not make a 1% even order distortion amp very good!
Because amplifying elements have continious and smooth slope transfer characteristics, the distortion they generate has diminishing ascendent order harmonics. Past some 2kHz, the human ear has diminishing sensitivirty to higher order harmonics as well.
A single MOSFET will have more even order distortion than odd compared to a BJT. However, symetric circuits tend to supress even order distortion, so for BJTs it's not uncommon for the second order harmonic to get nearly completely supressed, leaving the third most dominant - so the most dominant harmonic and the one that catches attention is the more objectionable one. Higher order harmonics tend not to be so supressed, but still the odd order ones are higher in amplitude compared to even, so they dominate. With MOSFETs this is different, you get more even order harmonics and the higher order harmonics (all of them) tend to fall in amplitude smoothly, with no great discrepancy between even and odd. Because of this, our perception of this kind of distortion is that it is far less objectionable, but an A/B test as outlined above will clearly show it is there.

 

[axel]

Thanks ilimzn for the master-class! :yes:

 

[Texas42]

B&K used MOSFETs in their amps for a long time (maybe still do?). I have one of their early ST-140s which has always been my favorite sounding solid state amp (for the price). Looks great, sounds great, built well and can be found on Audiogon and ebay for as little at $200. Highly recommended....