Hi SY,
The NPD5564 was a big-die monolithic JFET pair with lots of interdigitaled gates. Looks a lot like the LSK389. If memory serves me, the NPD5564 was equal or better than the LSK389/2sK389. National made really good discrete stuff back then. Beware, I think there are "5564" devices available that are not monolithic pairs; not sure about them.
I used the NPD5564 in my MOSFET power amplifier with error correction in 1983.
Cheers,
Bob
Those are about 40 Ohm rbb' @ 1mA the PNP does a bit better. Very good for that, considering they are mainstream cheap and available. Not high FOM but still the little swing won't be bothered in such a stage by Early non linearities, Ib will have to be blocked in SE or trimmed out in symmetric also with any better hfe BJT anyway. 4 are still needed to be shunted for low MC IMHO. Higher capacitance in such very low swings brings down speed but I don't think strongly impacts THD too?
They were sold as low Ron switches, a search now can turn up others but the low freq noise must be checked. The J305 family can be selected for very low noise and costs pennies.
Let me help you out John, you are more right than you think and don't even realize it. The distortion is a wash only at the second order for higher order the paralleled devices win. To make an extreme example, say you had a perfect input stage that clipped at 20mV. Your 100uV Ortophon would probably never clip it, but add a 30:1 step up transformer now you need 0.6V of input range for the same headroom. That's why valves and transformers were made for each other. If you want to stick with linear open-loop stages your's is the way to go (in this case). Now you can buy me a coffee.
BTW paralleling two resistors to get 1/2 R was known a lot longer. At the first mention of rbb being a limiting noise source I would consider the rest obvious.
Well, you are in my playground now. First, let me apologize in advance if I 'correct' a few comments here a little too sternly, it is my 'habit', not my intent.
I discovered the low noise potential of the 2N4401, 2N4403, and the 2N4405 (even better) when I was at Ampex in 1968, with a Quan-Tech noise analyzer. Except for Ampex, it was a well kept secret until 1971, when I sent an LTE to 'WW'. They thought I was some sort of 'nut' but they passed my letter to H.P. Walker, and he added a footnote in his second paper on noise, pointing out my recommendation of the 2N4403. This was the first public mention of a medium to low rbb' device used for low noise input stages.
In 1973, the Levinson JC-1 was released with 4 x 2N4401 and 4 x 2N4403 in noise parallel, giving about 0.4nV/rt Hz input noise, at the NY AES convention. Nobody thought it possible without a transformer.
Later, in 1979, I found that the Fairchild 2A, output transistor pair, especially made for transistor radios, were made on the same assembly line as other smaller devices, rather than with the larger power devices. They were the equivalent of about 4 X 2N4403 or 4 X 2N4401 for each part, and the SOTA head amp was born with just one PE8550 and one PE8050 per channel, again giving about 0.4nV/rt Hz input noise.
As far as jfets are concerned, the 2N3564, and its cousins, is an old military part, just look at the part number. It originally was a dual matched NC geometry by Siliconix, or Process 51 by National. Later, National made a monolithic dual called Process 96. I have at least 1000 of them, here in my lab.
However, these are compromised devices, noise wise, compared to the Toshiba 2SK170 or the 2SK389 which are somewhat quieter, have more Gm, but at the disadvantage of somewhat higher non-linear input capacitance. I used my first NPD5566, (Process 96) commercially, in the Vendetta Research Electronic Crossover, released about 1983 or so. Before this, I used even quieter parts in the Levinson JC-2 input stage such as the J110, in 1973. Unfortunately a process change destroyed the usefulness of these parts for low noise, by 1975.
Toshiba has led the way, up until now, with the best low noise jfets ever actually available for sale in quantity. Unfortunately, they are pulling out of the business.
National is gone, and so is Siliconix.
Linear Systems is trying, and mostly successful in filling the gap.
Other companies are competing to fill the gap left by Toshiba. Let us hope, soon.
I discovered the low noise potential of the 2N4401, 2N4403, and the 2N4405 (even better) when I was at Ampex in 1968, with a Quan-Tech noise analyzer. Except for Ampex, it was a well kept secret until 1971, when I sent an LTE to 'WW'. They thought I was some sort of 'nut' but they passed my letter to H.P. Walker, and he added a footnote in his second paper on noise, pointing out my recommendation of the 2N4403. This was the first public mention of a medium to low rbb' device used for low noise input stages.
In 1973, the Levinson JC-1 was released with 4 x 2N4401 and 4 x 2N4403 in noise parallel, giving about 0.4nV/rt Hz input noise, at the NY AES convention. Nobody thought it possible without a transformer.
Later, in 1979, I found that the Fairchild 2A, output transistor pair, especially made for transistor radios, were made on the same assembly line as other smaller devices, rather than with the larger power devices. They were the equivalent of about 4 X 2N4403 or 4 X 2N4401 for each part, and the SOTA head amp was born with just one PE8550 and one PE8050 per channel, again giving about 0.4nV/rt Hz input noise.
As far as jfets are concerned, the 2N3564, and its cousins, is an old military part, just look at the part number. It originally was a dual matched NC geometry by Siliconix, or Process 51 by National. Later, National made a monolithic dual called Process 96. I have at least 1000 of them, here in my lab.
However, these are compromised devices, noise wise, compared to the Toshiba 2SK170 or the 2SK389 which are somewhat quieter, have more Gm, but at the disadvantage of somewhat higher non-linear input capacitance. I used my first NPD5566, (Process 96) commercially, in the Vendetta Research Electronic Crossover, released about 1983 or so. Before this, I used even quieter parts in the Levinson JC-2 input stage such as the J110, in 1973. Unfortunately a process change destroyed the usefulness of these parts for low noise, by 1975.
Toshiba has led the way, up until now, with the best low noise jfets ever actually available for sale in quantity. Unfortunately, they are pulling out of the business.
National is gone, and so is Siliconix.
Linear Systems is trying, and mostly successful in filling the gap.
Other companies are competing to fill the gap left by Toshiba. Let us hope, soon.
Last edited:
)
Higher noise than I remembered. 50nV/rt Hz.
I had pretty good luck with some old Siliconix switching FETs with low Rds(on). I think they were 2N5114 or 5116. As a demonstration of their quality control, the last batch I bought all had the wrong pinout. Quiet, though, worked great as an input buffer for a phono stage.
The problem with high gm devices like the Toshibas or LS is the high input capacitance. Some MM cartridges don't care but some do (mine unfortunately being one of the latter).
I had pretty good luck with some old Siliconix switching FETs with low Rds(on). I think they were 2N5114 or 5116. As a demonstration of their quality control, the last batch I bought all had the wrong pinout. Quiet, though, worked great as an input buffer for a phono stage.
The problem with high gm devices like the Toshibas or LS is the high input capacitance. Some MM cartridges don't care but some do (mine unfortunately being one of the latter).
One might surmise that a 'power' mosfet
would be very quiet due to the statistical
noise cancellation of the thousands of
cells paralleled.
IIRC that was the method used to make this device:
LM394 - SuperMatch Pair [Obsolete]
(only hundreds of bipolar 'cells' in parallel)
would be very quiet due to the statistical
noise cancellation of the thousands of
cells paralleled.
IIRC that was the method used to make this device:
LM394 - SuperMatch Pair [Obsolete]
(only hundreds of bipolar 'cells' in parallel)
From what I remember, these devices are just 4 transistors in a two by two layout, with diagonal bjt's paralleled. I'll see if I can find the die photo, I have it somewhere...
Cheers, John
That's Bob Pease folklore. You can work out mathematically the optimum multi-stripe geometry, it is definately not ordinary 1974 style BJT's in parallel. If it ever was it would have changed due to the waste of silicon. BTW Philips published the derivation and prototyped .2nV BJT's for an AES paper. I have some somewhere but never could afford a MC to use them on. You will also find some amazing BJT's in early magneto-resistive hard drive heads (not discrete).
Last edited:
> From what I remember, these devices are just 4 transistors
> in a two by two layout, with diagonal bjt's paralleled.
> I'll see if I can find the die photo, I have it somewhere...
If you mean the LM394 then I have the wrong part number.
I remember the spec sheet drawing and the blurb stating
something to the effect of 'statistically matched pair'
I forget the number of xsistors/device, but it was lots.
> in a two by two layout, with diagonal bjt's paralleled.
> I'll see if I can find the die photo, I have it somewhere...
If you mean the LM394 then I have the wrong part number.
I remember the spec sheet drawing and the blurb stating
something to the effect of 'statistically matched pair'
I forget the number of xsistors/device, but it was lots.
Hitachi 2SC2545,46,47 NPN and 2SA1083,84,85 PNP you mean? Toshiba 2SC3329 & 2SA1316 were maybe a wee bit better even.
Attachments
Actually, no. I'm going by memory...used to play with waffle packs and wafers back in the 80's. So I lived dice..
If anything, it's faulty memory. I found some mat-02 stuff on my drive, I may be confusing the two...
hitsware;2508039If you mean the LM394 then I have the wrong part number. I remember the spec sheet drawing and the blurb stating something to the effect of 'statistically matched pair' I forget the number of xsistors/device said:
- Status
- Not open for further replies.