By a coincidence i just went through "The art of electronics" 3rd edition reading on JFETs. The autors do not recommend using JFETs at high source-drain voltage (which your circuit is) because above a few volts gate current raises dramatically, by several orders of magnitude. I'd imagine this may have effect on this circuit out V drift, and make it dependent on the primary regulator output voltage which is the JFET input. But im not good enough with JFETs yet to say for sure. Anyway, something you may need to have a look at and perhaps to test.
By a coincidence, I knew about that nuance of JFETs probably before the first edition of "The Art of Electronics" was published (was it what, 1980? - yes, I am that ancient
). Also (not by a coincidence this time, but as a normal design practice), I tend to look at datasheets of devices I plan to use before actually using them. And in the datasheet for the U440 you may find the following diagram:
And I also have in my library an old catalogue from Siliconix where one can find the same graph for the J112 device (and the gate current for it starts to increase substantially only for Vds over 15V .
Cheers
Alex
P.S. for 1ppm effect you would need the gate current at about 1-3nA ( for 1-3mA drain current), by the way. Look at the graphs. Even at 125C and 10V the gate current does not reach that level.
How well this simple and very affordable JFet-based voltage reference compares to the LM399? Let's assume that one creates a simple "oven" by gluing an appropriate resistor to the J112 as an heater element and places tho whole thing into a styrofoam or similar insulator? Let's assume the constant current through the heater resistor will be good enough to keep the temperature within 1 degree of Celcius. Any educated guesses?
How well this simple and very affordable JFet-based voltage reference compares to the LM399? Let's assume that one creates a simple "oven" by gluing an appropriate resistor to the J112 as an heater element and places tho whole thing into a styrofoam or similar insulator? Let's assume the constant current through the heater resistor will be good enough to keep the temperature within 1 degree of Celcius. Any educated guesses?
1) The main question would be the long time stability. It is an unknown. My guess would be that unless the junction is damaged by a static discharge, the stability of the JFET itself (especially in a metal can) should be very good to excellent. After all we are looking at a simple single structure, not a complex one as in the LM399 or any other IC reference.
2) It is reasonably easy to create a temperature control down to 0.1 degree, especially for a small device like that, using something like LM35 for a temperature sensing.
I am not sure that it is sensible to look at this circuit as a replacement for the LM399 before some long-term tests are done on it. My aim was to show a simple way of making a voltage reference in a rather unorthodox way, and frankly I am surprised at the level of performance actually available from this circuit.
Cheers
Alex
I'm beginning to wonder if there is too much emphasis to oven implementations rather than temperature compensation / characterized predictable temperature coefficient in all of these references.
Surely long term drift in the oven temperature controller is as much of an error contribution as drift in the reference itself - particularly in a reference as simple as this one (as opposed to LM399 syndrome as Alex hinted). It just seems a bit 'brute force'.
Wouldn't putting the whole thing in an isothermal metal box, insulated to limit rate of temperature change and avoid thermal stresses work better in a home lab environment? Putting the effort into characterization and compensation instead.
How well this simple and very affordable JFet-based voltage reference compares to the LM399
Perhaps the dealbreaker is in repeatability. For mass production you throw an integrated IC with tight tolerances and no worries every single unit of the product will be as good as the prototype. Whereas JFETs characteristics differ largely between manufacturers and even within the same batch. The datasheet for J112 says cutoff voltage can be anywhere from 1 to 5v. Having said that this circuit operates at near cut-off voltage on the gate, chances are each unit will need to be individually adjusted for the specified output voltage.
But if you only need one off for personal use then not a problem.
How well this simple and very affordable JFet-based voltage reference compares to the LM399
Perhaps the dealbreaker is in repeatability. For mass production you throw an integrated IC with tight tolerances and no worries every single unit of the product will be as good as the prototype. Whereas JFETs characteristics differ largely between manufacturers and even within the same batch. The datasheet for J112 says cutoff voltage can be anywhere from 1 to 5v. Having said that this circuit operates at near cut-off voltage on the gate, chances are each unit will need to be individually adjusted for the specified output voltage.
But if you only need one off for personal use then not a problem.
Exactly. If, however, the long term stability happens to be in the range of few ppm per year (and I have a sneaky suspicion that it might, given a metal case and a good attention to the construction and the resistors used), than it is a different ball game altogether
. For that one may even select the most suitable FETs out of a large batch (I've just bought 100 of J211 from DigiKey - it cost me less than one LTZ1000 would, even including UPS delivery from the US). But we would know that only in a year or so. Or at least in a 1kh after I'll get my hands on the HP3458A
.
Cheers
Alex
Perhaps the dealbreaker is in repeatability. For mass production you throw an integrated IC with tight tolerances and no worries every single unit of the product will be as good as the prototype. Whereas JFETs characteristics differ largely between manufacturers and even within the same batch. The datasheet for J112 says cutoff voltage can be anywhere from 1 to 5v. Having said that this circuit operates at near cut-off voltage on the gate, chances are each unit will need to be individually adjusted for the specified output voltage.
But if you only need one off for personal use then not a problem.
However neither of the 'superstars' ie. LTZ1000 and LM399 is tight initial tolerance. The have quite a large variation requiring adjustments on an individual basis. Stabilty and long term drift are excellent but initial tolerance is poor.
I might be wrong, but I found J112 to be TO92 and SOT23-3 only, no metal can. So humidity sensitivity can cause a big influence, right? A big advantage of your U440.
I might be wrong, but I found J112 to be TO92 and SOT23-3 only, no metal can. So humidity sensitivity can cause a big influence, right? A big advantage of your U440.
Yes, that is correct. However the humidity sensitivity is usually due to the resistors on the chip, here we don't have any. It is possible that for a JFET even in a plastic case it should not be a problem.
Now I need an environmental chamber as well!
Cheers
Alex
Now I need an environmental chamber as well!
I guess it is still not too late the sell those 100 J211s and forget the whole project, and buy the LTZ1000
Now I need an environmental chamber as well!
I guess it is still not too late the sell those 100 J211s and forget the whole project, and buy the LTZ1000
Nah, that would be too easy. I am looking for fun, not for gain
.
Cheers
Alex
P.S. - on the other hand, the LTZ needs a very good board layout and a lot of expensive supporting electronics. Sort of fun too, but it is already a known road
.
however the humidity sensitivity is usually due to the resistors on the chip, here we don't have any. It is possible that for a JFET even in a plastic case it should not be a problem.
Are you sure about that? What about the Source-Drain-channel, not some kind of a resistor though? Not sensitive to mechanical stress?
however the humidity sensitivity is usually due to the resistors on the chip, here we don't have any. It is possible that for a JFET even in a plastic case it should not be a problem.
Are you sure about that? What about the Source-Drain-channel, not some kind of a resistor though? Not sensitive to mechanical stress?
The only way to be reasonably sure is to test thoroughly. I have no data on a single JFET package sensitivity in that respect. On the other hand, everything is sensitive to everything, the only question is to what degree
.
Cheers
Alex
Are you sure about that? What about the Source-Drain-channel, not some kind of a resistor though? Not sensitive to mechanical stress?
Does a metal can maintain constant air pressure?
Are you sure about that? What about the Source-Drain-channel, not some kind of a resistor though? Not sensitive to mechanical stress?
Does a metal can maintain constant air pressure?
AFAIK, strictly speaking, no, as it is though a metal package is hermetic, but not evacuated, and the internal gas pressure should change with temperature.
Cheers
Alex
Supposedly zero tempco for a JFET is near the pinch off voltage plus 0.63V.
There is a 1973 US patent (long expired now)
3,760,199 that details a manner for determining a compensating voltage for biasing any JFET into it's 0-tempco region. The goal of the design in the patent was to determine Vp, the pinchoff voltage, at any temperature, and a compensating voltage Vc, such that Vgsz, the gate-source voltage needed to bias the JFET at its 0-tempco region was Vgsz = Vp + Vc. Since Vp, the pinchoff voltage is temperature dependent, then he develops Vc = 0.64 + a temperature dependent part opposing the tempco of Vp. With proper placement of some PN junctions and some BJTs and resistors, and a little bit of KVL, the inventor develops a voltage loop involving Vp, Vc and 0.64 and places the second JFET's gate-source across the point where those voltages are summed, yielding Vgs = Vp + Vc + 0.64, without ever knowing what Vp was (so it can vary with the standard process variations). The second JFET will now have this temperature compensated bias voltage applied to it's gate wrt source and develop a nearly constant drain current that is, for the most part, temperature independent. Very Cool!
It's an easy patent to read and a good one too. I think it was intended for use at chip scale, so that all the elements are matched and isothermic. It might not work with discrete parts if someone were to try it out.
I might be wrong, but I found J112 to be TO92 and SOT23-3 only, no metal can. So humidity sensitivity can cause a big influence, right? A big advantage of your U440.
Immerse it in mineral oil in a tictac container. No humidity effect and more temperature stability.
Now that I have my scavenged cryocooler getting to temperatures it can liquify air ( ) and keep high temp superconductors superconducting indefinitely, I thought a neat trick would be to be the ultimate volt-nut and combine it with my rubidium standard to make a Josephson junction voltage reference. First step is to get a SQUID working, it would at least show the effect is reproducible with my setup. I think it is unlikely I will succeed on my allotted budget, but I am sure it will be a fun ride anyway.
Well, my 100 of J211 arrived from DigiKey and I've measured the cut-off voltage on the lot at 5V Vgd. Actually, a very consistent batch for JFETs, it looks like the processing did improve since 1980s. All hundred measured between 2.75V and 3.35V at ~3uA current (10M of Keithley 2015 on 100V range between the source and gate). My "experimental" U440 measures 4.6V in the same setup (and the zero tempco point is at ~3.74V, so the "theory" about 0.63 or 0.64V is not working
). One obvious bad point about plastic casing - there is some photo-sensitivity, so the device should not be exposed to light when used as a reference. Not much of a problem though. I am nearly finalised a complete reference circuit, which would take 12-24V supply and provide 10V output with up to 10mA current available and would contain only two FETs, an opamp and few resistors. Not temperature regulated - that would be the next step.
Cheers
Alex
I am not going to publish a complete circuit any time soon, only measurement results
. I plan to include a temperature sweep from 0 to 40C and a 1000h stability test against LM399 (I don't have LTZ1000
).
Cheers
Alex
OK, let's not get all this popcorn wasted
.
I've decided to try a very simple circuit to get 10V output and to use the J211 transistor as a reference. This is a "budget" solution, as it requires only two good quality resistors (R1 and R2 and I've used 15ppm metal film for that, "padded" to the required value by some standard 100ppm/C resistors, so I should not expect much better performance than this value), cheap J211 JFET and an inexpensive INA133 chip connected as a gain 2 buffer with low tempco (~2ppm/C). Altogether is less than £10 in parts (resistors R3-R6 are internal for the INA133).
Here is the circuit which I've assembled on an old piece of a pcb:
And here is a 30-min run on the bench (vertical scale is 1ppm/div):
My estimate is that the temco around 23C is indeed about 10-15ppm/C. The power supply voltage variations from 12V to 28V create less than 2ppm/V change. The load regulation is better than 0.5ppm/mA (a conservative estimate, as a load current up to 5mA, sink or source, creates less than 1ppm voltage change). The supply current is about 2mA.
First R1 is selected for the lowest tempco and then R2 for 10V output voltage.
Cheers
Alex
I've seen this schematic some days ago in the LF355 datasheet, it made me think about you! Your idea is not new.
Thank you, I always thought that it was used before but couldn't find a confirmation!
Cheers
Alex
P.S. - would be interesting to find if that circuit was used somewhere in practice!
P.P.S. - The 2N4118 has a very low drain current (around 100uA for near-zero tempco) - it should make it good for low power but fairly noisy as a reference. A higher current device would make it a very low noise option - much quieter than ~1ppm p-p LF noise of the LM399.